(2) 본 論文을 崔正惠의 工學碩士學位論文으로 認准함.

전체 글

(1)工學碩士學位論文. N eural N et w ork s an d M ole c u lar A n aly s e s of Circ u lat in g P ig g ery S lurry T re at m e nt S y s t e m 인공신경회로망과 분자 생물학적 기법을 이용한 순환식 돈분페수 처리 시스템의 분석. 指導敎授. 高星澈. 2000年. 12月. 韓國海洋大學校. 大學院. 土木環境工學科崔. 正. 惠.

(2) 본 論文을 崔正惠의 工學碩士學位論文으로 認准함.. 2000년. 위원장 : 김 인 수. (인). 위. 원 : 송 영 채. (인). 위. 원 : 고 성 철. (인). 12월. 한국해양대학교. 토목환경공학과. 22일. 대학원. 최 정 혜.

(3) Table of Contents TABLE OF CONTENTS.................................................................. ⅰ LIST OF TABLES..................................................... … ................... ⅲ LIST OF FIGURES........................................................................... ⅳ ABSTRACT...................................................................................... ⅵ Ⅰ. INTRODUCTION.......................................................................... 1 Ⅱ. LITERATURE REVIEW............................................................... 4 2.1 Swine Wastewater Pollution and Treatment Technology.......... 4 2.1.1 Pollution Status and Characteristics of Swine Wastewater......................................................................... 4 2.1.2 Treatment Technologies of Swine Wastewater................... 6 2.2 Modeling Using Neural Networks............................................. 6 2.2.1 Characteristics of Neural Networks and Principal Component Analysis................ … ....................... 6 2.3 Mechanisms of Biologycal Nitrogen and Phosphorus Removal … … … … … … … . . … … … ...................... 12 Ⅲ. Materials and Methods................................................................. 15 3.1 System Overview..................................................................... 15 3.2 Isolation, Identification and Quantification of Microorganisms............... … … … … … … … . ....................... 17 3.3 Analytical Methods for Piggery Slurry samples from the Treatment System...................................................... 18 -i-.

(4) 3.4 Modeling of the Treatment System Using Neural Networks............................................................ 18 3.5 Ammonium Uptake and Utilization Test................................. 21 3.6 Extraction of Total DNA.......................................................... 21 3.7 PCR Amplification of Glutamine synthetase Gene and Southern Blot Hybridization......... … ................................. 22 Ⅳ. Results and Discussion................................................................. 24 4.1 Microbial Identification, and Analyses of the Population Dynamics and Piggery Slurry Treatment.......... … ................... 24 4.2 Wastewater Treatment Efficiency of the Treatment System.... … .................................. … … … … … … … … . ............ 31 4.3 Principal Component Analysis of Input Data.......................... 38 4.4 Modeling of Treatment System by Neural Networks.............. 42 4.5 Molecular Analysis of Ammonium Removal.......................... 43 Ⅴ. Conclusion.................................................................................... 46. FUTURE TASKS REFERENCES ACKNOWLEDGEMENT CURRICULUM VITAE. - ii-.

(5) List of Tables Table 2.1 Status of the waste, number of heads and percentage rate of the livestock farms in Korea... … .............................. 5 Table 4.1 Differential characteristics of the gram-negative bacterial species isolated from circulating treatment system … … .... 26 Table 4.2 Differential characteristics of the gram-positive bacterial species isolated from circulating treatment system … . ....... 27 Table 4.3 Wastewater treatment efficiency of the piggery slurry treatment system........ … … … . … … … … . ................ 32. - iii-.

(6) List of Figures Figure 2.1 Typical structure of a muti-layer neural network.......... … . 8 Figure 2.2 Data filtration pattern using PCA (principal component analysis)................. … . .......... 11 Figure 2.3 Ammonia incorporation in bacteria.................................. 14 Figure 3.1 Schematic diagram of the circulating treatment system for piggery slurry................................ … . ............. 16 Figure 3.2 A schematic diagram describing training strategy for the neural networks in this study................. … ........... 20 Figure 4.1 Population dynamics of a heterotrophic bacterium (Brevundimonas diminuta) in the circulating treatment system.......................... … ............... 28 Figure 4.2 Population dynamics of a heterotrophic bacterium (Abiotrophia defectiva) in the circulating treatment system............................................ … . .............. 29 Figure 4.3 Population dynamics of a heterotrophic bacterium (Alcaligenes faecalis) in the circulating treatment system.............................................. … ............. 30 Figure 4.4 Dynamics of ammonium removal in the circulating treatment system............................................. … .............. 33 Figure 4.5 Dynamics of chemical oxygen demand (COD) removal in the circulating treatment system......... … ....... 34 Figure 4.6 Dynamics of ortho-phosphorus removal in the circulating treatment system................................. … ........ 35 Figure 4.7 Dynamics of total phosphorus removal in the - iv-.

(7) circulating treatment system....... … .................. … ............ 36 Figure 4.8 Dynamics of suspended solid removal in the circulating treatment system... … ....................... … ........... 37 Figure 4.9 Principal component analysis of input (A) output (B) of aeration tank data during the 47 days' running period.................................................. … ............ 41 Figure 4.10 Perdiction of various treatment parameters COD (A), NH4+-N (B), ortho-P (C) and SS (D) by the neural network modeling.................... … ............. 42 Figure 4.11 Amplification of glutamine synthetase (GS) gene using GS-L and GS-R primers and Southern hybridization ................................ … .............. 44. - v-.

(8) A B S T RA CT. 유기물질과 질소·인 등의 영양물질이 고농도로 포함된 축산폐수는 수계의 주요 오염 물질 중 하나다. 본 연구에서는 혐기성 및 호기성 조 건에서 연속식으로 돈분폐수를 처리하고 처리된 방류수의 일부를 축사에 재이용 할 수 있는 순환식 돈분폐수 처리 시스템의 처리기작에 관한 연 구를 수행하였다.. 이 시스템은 축사나 처리조에서 발생되는 악취와. BOD를 포함한 기타 오염물질을 효과적으로 제거할 수 있으며, 돼지 1000두 이하의 소규모 농가에 적합한 처리공정 중 하나로 알려져 있다. 시스템내에서 우점종으로 분리된 타가영양박테리아는 호기성 및 혐기성 조건에서 질산화 및 탈질작용에 관여하는 것으로 밝혀진 A lcalig enes faecalis. (T SA - 3)와. B revundim onas. dim inuta. (T SA - 1),. 그리고. A biotrop hia d ef ectiva (T SA - 2)가 분리되었고, 혐기성 조에서는 유산균 박테리아 MRS - 1 (미동정) 외 2종(S trep tococcus sp. : MRS - 3)이 분리 되었다. 그 중 최우점종은 A lcalig enes faecalis (T SA - 3)로 나타났다. 중합효소 연쇄반응(polymer ase chain reaction ; PCR) 등의 분자생물 학적 기법을 이용하여 시스템으로부터 분리된 타가영양박테리아에 의한 암모니아성 질소 제거 기작을 밝혔다. 여기서 암모니아 흡수 및 이용에 관여하는 글루타민 합성효소(glutamine synthetase: GS )유전자의 존재를 확인하였다. 또한 신경회로망을 이용하여 순환식 돈분처리 시스템의 실시간 모니 터링을 궁극적으로 구현할 수 있는 새로운 방법을 제안하였다. 즉 미생 물 군집내의 개체군밀도에 따른 각 처리조(유입수, 발효조, 폭기조, 1차 침전조 및 4차 침전조)에서의 폐수처리 기작에 대한 모델을 시도하였다. 측정 데이터에 대해 우선 주요 요소 해석(principal component analy sis :.

(9) PCA ) 방법을 적용하여 각 처리조에서의 입력(미생물 밀도와 처리요소) 과 출력간의 상관관계를 파악하고, 각각의 처리조마다 독립된 신경회로 망을 적용하여 폐수처리 과정을 모델링하였다.. 신경회로망의 입력으로. 현재 탱크에서의 미생물의 개체군밀도를 직접 이용하는 대신 PCA 분석 결과를 이용함으로써, 비교적 적은 수의 데이터로 효과적인 모니터링 시 스템을 구현할 수 있었다. 즉, 각 처리조별로 학습된 신경회로망들을 연 결하여 분석한 결과 2일 동안의 폐수 처리 변화를 비교적 정확히 예측할 수 있었다..

(10) Ⅰ. INTRODUCTION. The swine wastes may cause a serious degradation of water quality such as eutrophication and spread of pathogens in water bodies (i.e., lakes, rivers and groundwater as water supply sources) (39). The daily volume of livestock wastewater in Korea reached 197,000m3, and 50% of the volume was generated from dairy farms that were not target for a legal pollution control. The amount of wastewater is relatively small compared with total wastewater including industrial and domestic wastewater (7% of the total), but contributes significantly to the pollution of the receiving waters because of its high organic nutrient concentration (>BOD 20,000 mg/L) (32). According to the environmental protection law, the large size farms (more than 1,000 heads) are subjected to regulations for treatment facilities whereas small or middle size farms (less than 1,000 heads) are exempt from the regulations. While an activated sludge system has been proven to be effective in the treatment of piggery slurry at large scale farms (more than 1,000 heads), the system may not ensure the effect in small or middle scale farms (less than 1,000 heads) in terms of its operation cost. The number of swine heads under the regulatory control, therefore, takes only 31% of the total number of heads (33). Recently a cirulating reactor system operated under sequential oxic and anoxic conditions for the treatment of swine wastewater has been developed, in which piggery slurry is fermentatively and aerobically treated and then its effluent recycled to the pigsty (9). -1-.

(11) This system appears to significantly remove offensive smells (at both pigsty and treatment plant) and BOD, and turns out to be cost effective in the relatively small scale farms. There are several treatment steps in the system.. For its. successful operation, it will necessary to monitor microbial population density and treatment parameters.. Modeling relationships among. these variables will be useful in predicting treatment effects and managing the system. One of the best known models applied for wastewater treatment system so far is the activated sludge model NO. 1(ASM 1) introduced by International Association for Water Quality (IAWQ) in 1987 (15).. Application of the model to the field. treatment system, however, may have some limitations because the model usually requires many operational parameters and has quite variable kinetic characteristics within the treatment system over time (25). On the other hand, neural network models that imitate the functions of our human brain have been successfully used to resolve many engineering problems such as complex pattern classification and control of highly nonlinear dynamic systems (4, 26, 31, 42). Those models have the characteristics of massive parallelism, many degrees of freedom, and adaptive learning. It was recently well known that the multi-layer neural networks can approximate a function in L. p. within an arbitrary accuracy (18), and generalize a new data that are not used in learning process (5). Recently a progress has been made in application of neural networks to controlling the biological and chemical engineering processes. There has been, however, no report -2-.

(12) dealing with a neural network modeling for biological swine wastewater treatment system, to the best of our knowledge. This study was carried out to elucidate mechanism of the circulating piggery slurry treatment system using such as variables population dynamics, activity of heterotrophic bacteria, and treatment effects based upon suspended solids (SS), ammonia nitrogen (NH4+-N), total phosphorus (T-P), ortho-phosphorus (o-P) and chemical oxygen demand (COD) as input or output variables. These variables were used to establish a non-linear model emulator using multi-level neural networks that could eventually allow real time monitoring and prediction of the treatment system.. We also tried to elucidate a. mechanism for ammonium removal using molecular biological techniques.. -3-.

(13) Ⅱ. LITERATURE REVIEW. 2.1 Swine Wastewater Pollution and Treatment Technology. 2.1.1 Pollution Status and Characteristics of Swine Wastewater. The wastes cause a serious degradation of water quality resulted from spread of pathogenes and a eutrophication in the receiving waters such as lakes and rivers as water supply sources. The daily volume of livestock wastewater in Korea reached 197,000m3, while 50% of this volume was generated from dairy farms that are exempt from a legal pollution control (Figure 2.1).. The amount of. wastewater is relatively small compared with total wastewater including industrial and domestic wastewater (7% of the total), but contributes significantly to the pollution of the receiving waters because of its high organic nutrient concentration (32). The degree of pollution caused by swine wastewater may depend upon the region of water bodies and sometimes reaches almost 20% of the total pollution load (33). In other word, domestic farms of Korea know almost small or middle size farms are exempt from the regulation. Raw wastewater contains substances to give malodorant smells such as nitrogenous compounds, sulfates, volatile fatty acids, aldehydes and so forth.. Besides, the wastewater can allow the. growth of health pests so that the living environments would be degraded and use of agricultural water contaminated by swine wastewater could reduce harvest of crops. It has become a target of -4-.

(14) public grievances (12). Physical and chemical characteristics of swine wastewater may be different depending on storage period and treatment brocess, but BOD concentration may be as high as about 20,000～60,000 (mg/L). Nitrogen. concentration. as. TKN. can. reach. the. range. of. 4,000～6,000mg/L (32).. Table 2.1. Status of the waste, number of heads and percentage rate of. the. livestock. farms. in. Korea. (Environmental. Management Research Center, 1998) Criteria Farms Heads Wastewater Volume (m3/d). Permit. Report. No. required 3,743. required 24,118. regulation 539,863. 567,724. (0.7%) 3,424,000. (4.2%) 3,726,000. (95.1%) 3,211,000. (100%) 9,911,000. (34.5%). (33.1%). (32.4%). (100%). 46,700. 51,456. 93,861. 197,017. (23.7%). (26.1%). (50.2%). (100%). -5-. Total.

(15) 2.1.2 Treatment Technologies of Swine Wastewater. Conventional treatment technologies available so far are activate sludge treatment, trickling filter, oxidation lagoon, oxidation pond, percolation, composting, and landfill disposal. New technologies are HAF (Hyundai Anaerobic Filer), BIMA, and Bio-Ceramic. These treatment technologies, however, are relatively difficult to operate and have comparatively high running costs. One of the reasons for this may be the characteristics of the wastewater itself: its high organic and nutrient content as well as unbalanced carbon, nitrogen, and phosphorus ratios (28). Treatment plants for swine wastewater using activated sludge and methane fermentation technologies have been disseminated. These methods are effective for removal of BOD and COD but not for nitrogen and phosphorus (29).. This is because swine wastewater. contains a large amount of nitrogen which corresponding to 20～40% of BOD (28).. 2.2 Modeling Using Neural Networks. 2.2.1 Characteristics of Neural Networks and Principal Component Analysis. In this study, we used a multi-layer neural network with error back propagation algorithm to model the complex relationship in the -6-.

(16) circulating system (10).. Artificial Neural Network is a system. loosely modeled on the human brain. It attempts to simulate within specialized hardware or sophisticated software, the multiple layers of simple processing elements called neurons. Each neuron is linked to certain of its neighbors with varying coefficients of connectivity that represent the strengths of these connections.. Learning is. accomplished by adjusting these strengths to cause the overall network to output appropriate results (14). Designing a neural network consist of: ·Arranging neurons in various layers. ·Deciding the type of connections among neurons for different layers, as well as among the neurons within a layer. ·Deciding the way a neuron receives input and produces output. ·Determining the strength of connection within the network by allowing the network learn the appropriate values of connection weights by using a training data set. The process of designing a neural network is an iterative process; the figure below describes its basic steps. As the figure 2.1 shows, the neurons are grouped into layers. The input layer consists of neurons that receive input form the external environment.. The output layer consists of neurons that. communicate the output of the system to the user or external environment. There are usually a number of hidden layers between these two layers; the figure 2.1 shows a simple structure with only one hidden layer.. -7-.

(17) Wi. Input layer. Wh. hidden layer. Output layer. Figure 2.1. Typical structure of a multi-layer neural network. -8-.

(18) The brain basically learns from experience.. Neural networks are. sometimes called machine learning algorithms, because changing of its connection weights (training) causes the network to learn the solution to a problem.. The strength of connection between the. neurons is stored as a weight-value for the specific connection. The system learns new knowledge by adjusting these connection weights. The learning ability of a neural network is determined by its architecture and by the algorithmic method chosen for training. The training method usually consists of one of three schemes: Unsupervised learning, Reinforcement learning, Back propagation. Back propagation method is proven highly successful in training of multi-layered neural nets (27).. The network is not just given. reinforcement for how it is doing on a task.. Information about errors. is also filtered back through the system and is used to adjust the connections between the layers, thus improving performance.. A. form of supervised learning (3, 13, 14). Principal Component Analysis (PCA) is a technique to find the directions in which a cloud of data points is stretched most. These directions represent most of the information in the data and are thus important to know. Knowing these directions allows us to store the data in a compressed form and later reconstruct the data with a minimal amount of distortion. PCA is used in statistics to extract the main relations in data of high dimensionality. A common way to find the Principal Components of a data set is by calculating the eigenvectors of the data correlation matrix. These vectors give the directions in which the data cloud is stretched most. The projections -9-.

(19) of the data on the eigenvectors are the Principal Components. The corresponding eigenvalues give an indication of the amount of information. the. respective. Principal. Components. represent.. Principal Components corresponding to large eigenvalues represent much information in the data set and thus tell us much about the relations between the data points (19).. -10-.

(20) (A). (B). Figure 2.2. Data analysis using principal component analysis (PCA) ((A) measured data and (B) predicted data). -11-.

(21) 2.3 Mechanisms of Biological Nitrogen and Phosphorus Removal. Alcaligenes faecalis, commonly found in soil, water, and wastewater treatment systems, was dominant in piggery slurry treatment system.. Since denitrifying bacteria were facultative. aerobes, a sufficiently high concentration of DO prevents their use of NO3-N as the terminal electron acceptor. In general, Alcaligenes faecalis was far less sensitive than nitrification bacteria Nitrosomonas sp. was, reducing or denitrify using nitrate such as electron accepter in anaerobic conditions. Alcaligenes faecalis can denitrify both under anaerobic conditions and nitrify under aerobic condition (2, 35). The possibility of aerobic denitrification by Alcaligenes faecalis implied that inhibition of denitrification by oxygen did not always occur and hence. that. nitrification. and. denitrification. may. take. place. simultaneously under aerobic condition. This strain has no nitrate reductases but is able to reduce nitrite to dinitrogen (34). Figure 2.3 shows a process of amino acid synthesis and pathway of ammonium utilization by bacteria.. The amino group of amino. acids is often derived from some inorganic nitrogen source in the environment, such as ammonia (NH3).. Ammonia is typically. incorporated during the formation of the amino acids glutamate or glutamine by the enzymes glutamate dehydrogenase and glutamine synthetase, respectively (Figure 2.3 a, b). Here the ammonium can be incorporated into α-ketoglutarate by the transamination reaction of glutamate dehydrogenase, generating glutamate. -12-. Glutamate is.

(22) subsequently transformed to glutamine by incorporationg ammonium via glutamine synthetase (30). In addition, this ammonium removal mechanism could be more effective compared with ammonia oxidation bacteria.. Therefore, this. study focuses on the ammonium uptaking and utilizing organisms in the treatment system and tries to elucidate the ammonium removal mechanisms using molecular biological techniques. Phosphorus removal mechanisms can be affected by oxygen concentration.. It is typical in the biological phosphorus removal. process that the sludge releases P i (with concomitant uptake of wastewater organic carbon) in the anaerobic phase and takes up Pi rapidly in the aerobic stage (16, 22).. The P removed from. wastewater is accumulated as a form of polyphosphate (poly P) in the sludge bacteria.. Removal of a portion of the growing biomass. (waste-activated sludge) results in the net removal of P from the wastewater (7).. -13-.

(23) NADH. (a) α-ketoglutarate - NH3. NH2. NH 2. Glutamate Glutamate dehydrogenase ATP. NH2. (b) Glutamate + NH3. NH2. Glutamine Glutamine synthetase. NH2. NH2. α-ketoglutarate + Aspartate. (c) Glutamate + Oxalacetate Transaminase. NH2. NH2. NADH. (d) Glutamine + α-ketoglutarate. NH2. 2 Glutamate Glutamate synthase. Figure 2.3. Ammonia incorporation in bacteria. Two major pathways for NH3 assimilation in bacteria are those catalyzed by the enzymes (a) glutamate dehydrogenase, and (b) glutamine synthetase. (c) Transaminase reactions simply transfer an amino group from an amino acid to an glutamate synthase, two glutamates are formed from one glutamine and one α-ketoglutarate (30).. -14-.

(24) Ⅲ. MATERIALS and METHODS. 3.1 System Overview. A scheme for the circulating treatment system at a pilot scale is shown in Figure 3.1.. Detail description of the reactor operation has. been shown in the previous report (41). Piggery slurry and treated effluent used as a washing water were collected in tank 1, and this influent then flows into the fermentation tank (tank 2; 15L). There is a channel between tank 2 and an aeration tank (tank 3; 15L) so that the fermented wastewater can be transported into tank 3 where oxidative treatment occurs under aeration condition (7.8v/v/m). The treated water then goes through sedimentation process at tanks 6 and 7, and finally is stored at tank 8. A portion of the effluent was used to wash the pigsty. The wastewater used in this study was collected from a mixing and storage tank at Kimhae Piggery Slurry Treatment Plant (Kimhae, Kyungnam, Korea) and carried COD (ca. 4000 mg/L), BOD (ca. 7000 mg/L), T-N (ca. 2100 mg/L), and T-P (ca. 172 mg/L). The influent consisted of piggery slurry (33% v/v), effluent (57%) and tap water (10%) and was supplied every 4 days. Glucose was added to the formulated influent to make a C/N ratio as 100:15 (28) and a microbial agent was also added up to 1 %(w/v).. The hydraulic. retention time of the system was 4 days and it operated for 47 days.. -15-.

(25) 5. 2. 3. 1. 4. 9 10. 6 7 8. D. C. B. A. Figure 3.1. Schematic diagram of the circulating treatment system for piggery slurry. 1 influent; 2 fermentation tank; 3 aeration tank; 4 blower; 5 antifoaming device; 6 sediment tanks (A, B, C and D); 7 reservoir; 8 storage tank; 9 recycling flow; 10. for fertilizer).. -16-.

(26) 3.2 Isolation, Identification and Quantification of Microorganisms. Heterotrophic bacteria potentially involved in the piggery slurry treatment within the system were isolated using the appropriate media (23). To isolate and grow lactic acid bacteria (LAB), MRS medium were used. LAB were grown at least 2 weeks before identification and counting were performed. The ingredients of the medium were: Bacto proteose peptone NO. 3 5g/L, yeast extract 2.5g, dextrose 10g, Tween. 80g,. (NH4)2HC6H5O7. 1g,. CH3COONa3 ·H 2 O. 4.14g,. MgSO4 ·7H2 O 0.1g, MnSO4 ·5H2O 0.04g, Na2PO4 ·12H2O 2.5g, Beef extract 10g; pH 6.5.. After autoclaving trace amount of. bromophenol blue was added as an indicator. Other heterotrophs were grown on TSA ( Trypticase Soy Agar, Difco) at least 1 week, and then identified and counted. The bacterial communities in the system were quantitatively analyzed based on isolation, identification and measuring their colony forming unit (population density) of dominant populations in each medium.. Identification of the bacteria was performed based on. cultural, physiological and biochemical characteristic described by Smibert et al. (40) and Holt et al. (17). Utilizations of sugar, amino acids and organic acids by Gram negative bacteria were tested using API Kit (bio Merieux sa, France) and its accompanied protocol.. -17-.

(27) 3.3 Analytical Methods for Piggery Slurry samples from the Treatment System Monitoring parameters such as SS, T-N, NH4+-N, T-P, o-P and COD were measured for piggery slurry samples taken daily following Standard Methods for the Examination of Water and Wastewater (1): COD by closed reflux, titrimetric method, T-P and ortho-P by ascorbic acide method, suspended solids by total suspended solids cride method, and NH4+-N by indol phenol method.. 3.4 Modeling of the Treatment System Using Neural Networks. For an optimal treatment of piggery slurry, it is important to understand the physiological characteristics of microorganisms and their relationships, but may be difficult to identify the complicated relationship by a linear analytical method.. The relation between. population densities of microorganisms and the treatment efficiency may have a nonlinear dynamic characteristic. In this study, we used a multi-layer neural network with error back propagation algorithm to model the complex relationship in the circulating system. Since the multi-layer neural network is able to approximate an arbitrary nonlinear function with sufficient input and output data, the modeling of the circulating piggery slurry treatment system can be realized by the neural network in complex dynamic systems. For modeling the circulating system, we considered cause and effect relation in each -18-.

(28) tank that was serially connected. As independent parameters in each tank, population densities of heterotrophic microorganisms MRS-1, TSA-1, TSA-2, and TSA-3 were considered because those could dominantly affect the piggery slurry treatment efficiency.. Also,. COD, total-P, ortho-P, SS and NH4+-N were considered as treatment parameters in each tank.. Thus, we designed multi-layer neural. networks in which the input nodes consisted of 4 independent parameters in the current tank and 5 treatment results in previous tank, and the output nodes generated the 5 treatment results in the current tank. For modeling the circulating system, there were two ways to use the neural network.. One was to use a single neural network for. modeling the characteristic of whole tanks in the circulating system. The other was to use the neural network for modeling the characteristic of each tank, and then serial connection of each neural network that modeled each tank could allow a monitoring of for the circulating system.. In this study, it was difficult to model the overall. characteristic of whole tanks by a neural network because each tank of treatment system has different role and characteristic. Thus, we used each neural network that was able to model the characteristic of each tank, and the overall model of the whole treatment system was obtained the connecting each neural network. Figure 3.2 showed a proposed modeling protocol for the circulating system.. We used. principal component analysis (PCA) as a preprocessor of the neural network.. Input of the neural network was reduced to 3 principal. values from 9 independent variables. -19-. The output values of the neural.

(29) network were COD, total-P (T-P), o-P, SS and NH4+-N in the current tank. To accomplish a successful modeling, the connectivity within neural networks in the current tank were adjusted to best predict the measured values to be obtained at the next treatment step using SS, NH4+-N, T-P, o-P and COD as input variables.. MRS-1 TSA-1 COD. TSA-2 TSA-3 COD T-P. D1. PCA. D2 D3. T-P. Neural Network. o-P SS NH4+-N. o-P SS NH4 +-N. Figure 3.2. A schematic diagram describing training strategy for the neural networks in this study. MRS-1, TSA-1, TSA-2 and TSA-3 denote the population density of the bacterial strains. COD (chemical oxygen demand), T-P (total phosphorus), o-P (ortho-phosphorus), SS (suspended solid) and NH4+-N are parameters for the wastewater treatment. PCA, D1, D2 and D3 denote principal component analysis and dimensions obtained after the analysis, respectively. -20-.

(30) 3.5 Ammonium Uptake and Utilization Test Ability of the isolated heterotrophs to uptake ammonium (NH4+) was measured to understand an ammonium removal mechanism in the treatment system.. The dominant organisms (TSA-1, TSA-3 and. TSA-4) were grown in the mineral salts medium (21) containing glucose (0.4% w/v or 3.2% w/v) as a sole carbon source. Unless the organisms were grown in the medium, they were subjected to growth at citrate mineral salts medium (34). Nitrogen source for these media was (NH4)2SO4. The inoculated media were incubated at 26℃ and under rotary shaking (190rpm), and the growth was measured spectrophotometrically (525nm).. The ammonium concentrations. before inoculation and at stationary phase were measured and the ammonium removal efficiency was calculated.. 3.6. Extraction of Total DNA Cells used for the ammonium removal test were collected by centrifugation and subjected to total DNA extraction that was performed according to Maniatis et al (38). The centrifuged cells were washed once with phosphate buffer and then resuspended in 0.6 ml of lysing buffer (0.15M NaCl, 0.1M EDTA, and 15 mg of lysozyme per ml).. After incubation at 37℃ for 3 hr, 0.06 ml of 10%. sodium dodecyl sulfate was added and the mixture was incubated at 65℃ for 10 min, and then -70℃ for 5 min. The freeze and thawing procedures were repeated twice.. The mixture was then extracted. -21-.

(31) with phenol-chloroform three times and with chloroform once. The alcohol precipitated DNA was resuspended in TE buffer containing RNase A and incubated at 37℃ for 3 hr to remove residual RNA.. 3.7. PCR Amplification of Glutamine synthetase Gene and Southern Blot Hybridization. Degenerate oligonucleotide primers (forward primer GS-L and reverse primer GS-R) targeting glutamine synthetase genes from the isolated organisms were designed from the conserved GS protein sequences of Bacillus sp. including Bacillus subtilis 168 (KCTC 1326; ATCC 33234 Spizizen strain 168). The protein sequence alignment and analysis were accomplished using the sequence databases of Gene Bank and the Blast sequence analysis protocol available at National Center for Biotechnology Information (National Institute of Health). Their sequences were custom-synthesized by GenoTech (Taejon, Korea):. GS-L: 5. - GTG-AAG-TAT-ATC-CGY-CTT-C-3. GS-R: 5. - ATA-YTG-WTC-GCG-YTC-CCA-3. One to 3.3 ng of the extracted total DNA were used as a template. Positive control DNA was from Bacillus subtilis 168.. The PCR. procedures for this study were modified based upon the previous report (36).. Each PCR reaction mixture (20㎕) contained the. following reagents: 10 X Taq buffer, MgCl2 (1.5 mM), dNTPs (250 -22-.

(32) M, each), forward primer GS-L (10 pM), reverse primer GS-R (10 pM), Taq polymerase (1.25 U) (Promega). PCR was performed in a DNA thermocycler (Perkin Elmer model; GeneAmp PCR System 2400).. The PCR conditions were denaturation (94℃, 5min), 30. cycles of the standard PCR (94℃ 1 min; 50℃ 1 min; 72℃ 1 min), and a final chase reaction of (72℃ 5min). Amplification products were eletrophoresed on 1% agarose gel, and stained ethidium bromide. For Southern blots DNA was blotted from the gels to nylon membranes and cross linked to the membranes by dry to 80℃, 1 hour. The expected PCR product (1269bp) from GS gene of Bacillus subtilis 168 was identified and nonradioactively labeled using Nonradioactive Labeling and Hybridization Kit (Boehringer Mannheim, Germany). A GS gene probe was prepared by gel-purified 1296bp GS amplification product from Bacillus subtilis 168 with the Random Primed DNA Labeling Kit (Boehringer Mannheim, Germany). Membrane was hybridized overnight at 65℃ in 10% sodium dodecyl sulfate (SDS); 5× SSC; 0.1% Nlaurylsarcosine; 1% blocking Agent and then washed twice for 15min each time at 65℃ in buffer contained 0.1× SSC; 0.1% SDS.. All. the following Southern hybridization procedures were done according to the previous report (21) except using CSPD (Disodium 3-(4methoxyspiro{1,2-dioxetane-3,2´-(5´chloro)tricyclo[3.3.1.1 3,7 ]decan}-4-yl)phenyl phosphate) (Boehringer Mannheim) as a chemiluminescent substrate for the alkaline phosphatase. -23-.

(33) Ⅳ. RESULTS and DISCUSSION. 4.1. Microbial Identification, and Analyses of the Population Dynamics and Piggery Slurry Treatment The most dominant heterotrophic bacteria in the treatment system were 4 aerobic bacteria and 3 lactic acid bacteria (LAB).. The. identified organisms were TSA-1 (Brevundimonas diminuta), TSA-2 (Abiotrophia defectiva), TSA-3 (Alcaligenes faecalis) and MRS-3 (Streptococcus sp) (Table 4.1, 4.2). Population dynamics of isolated bacteria were shown Figure 4.1, 4.2, 4.3. One of the most dominant aerobes was Alcaligenes faecalis TSA-3. The most dominant species of LAB was strain MRS-1 that was yet to be characterized. Population dynamics of the representative aerobic bacterium Alcaligenes faecalis TSA-3 during the 47-day running period was shown for each tank (Figure 4.3). Interestingly, TSA-3 was a predominant species among aerobes in the aeration tank (107～108 (c.f.u./ml)) but was also observed in the influent and fermentation tanks (Figure 4.3). Thus the strain appeared to survive and grow under low oxygen tension and anoxic condition.. A. reported species of Alcaligenes faecalis could oxidize ammonia under aerobic condition and denitrify nitrate ions via NO and N2O gases under anoxic conditions (2, 35). Alcaligenes faecalis was found to accumulate NO2- during exponential growth (34). Population of the strain MRS-1 was more dominant in the influent and fermentation -24-.

(34) tanks than aeration and sedimentation tanks, indicating its facultative anaerobic characteristics. The overall population density was in the range of 104～107 (c.f.u./ml).. -25-.

(35) Table 4.1. Differential characteristics of the gram-negative bacterial species isolated from circulating treatment system Characteristic Gram staining Cell shape Oxidase Catalase Anaerobic growth * Acid from : D-Glucose Mannitol Inositol Salicin D-Melezitose L-Fucose D-Sorbose L-Arabinose D-Ribose D-Sucrose Rhamnose Maltose * Utilization of : Valerate Citrate 2-Ketogluconate 3-Hydroxybutyrate 4-Hydroxybenzoate Itaconate Suberate Malonate Acetate DL-Lactate 5-Ketogluconate 3-Hydroxybenzoate Glycogen * Decomposition of Histidine L-Proline L-Alanine L-Serine. TSA1 rod + -. Brevundimonas** diminuta + + -. -. -. -. + + + -. -. -. + -. TSA3 rod + + -. Alcaligenes ** faecalis rod,coccal + + -. -. -. + + + + + + -. +. + + + -. Symbols: +, 90% or more positive; -, 0-10% positive *. Tested using API identification program (ID 32 GN: bio Merieux sa, France) -26-. -. +.

(36) **. Data from Bergey's manual of Determinative Bacteriology (Ninth ed.). Table 4.2. Differential characteristics of the gram-positive bacterial species isolated from circulating treatment system. Characteristic. TSA2. Gram staining + Cell shape coccus Oxidase + Catalase Anaerobic growth + Motile Hemolysis α Growth at + 10℃ 45℃ pH9.6 + 6.5% NaCl Voges-Proskauer Acid from Xylose Lactose Sucrose Rhamnose Melezitose Adonitol Glycerol Raffinose + Sorbitol Salicin + Trehalose + Mannitol -. Abiotrophia **. defectiva + coccus + + -. MRS3. Enterococcus gallinarum** + coccus + + +. α. + coccus + + + -. ND ND ND ND ND. + + + + -. + + + + ND. + + + + -. + + + + + -. +. +. d. d + + -. α,β. Symbols: +, 90% or more of strains are positive; -, 90% or more of strains are negative; α, usually α-hemolytic; α,β,may be α- or β- hemolytic; d, 2179% of strains are positive; ND, not determined. **. Data from Bergey's manual of Determinative Bacteriology (Ninth ed.) -27-.

(37) 8. 6. Log. 10. (c.f.u./ml). 7. 5. 4 0. 10. 20. 30. 40. 50. Days. Figure 4.1. Population dynamics of a heterotrophic bacterium (Brevundimonas diminuta) in the circulating treatment system (•Influent tank; ο- Fermentation tank; ▼- Aeration tank; Sedimentation tank A;. - Sedimentation tank D). -28-. ▽-.

(38) 8. Log 10 (c.f.u./ml). 7. 6. 5. 0. 10. 20. 30. 40. 50. Days. Figure 4.2. Population dynamics of a heterotrophic bacterium (Abiotrophia defectiva) in the circulating treatment system (•- Influent tank; ο- Fermentation tank; ▼- Aeration tank; A;. - Sedimentation tank D). -29-. ▽-. Sedimentation tank.

(39) 9. L o g1 0 (c.f.u./ml). 8. 7. 6. 5 0. 10. 20. 30. 40. 50. Days. Figure 4.3. Population dynamics of a heterotrophic bacterium (Alcaligenes faecalis) in the circulating treatment system (•- Influent tank; ο- Fermentation tank; ▼- Aeration tank; A;. - Sedimentation tank D). -30-. ▽-. Sedimentation tank.

(40) 4.2 Wastewater Treatment Efficiency of the Treatment System. The ammonium removal during the extent of the experiment is shown, for tank, in Figure 4.4. The ammonium removal efficiency reached 41% as a maximum (Table 4.3). The reason for this rather low efficiency was not clear but unbalanced (presumably, lower) C/N ratio would be one of the causes. Here, however, offensive smells appeared to be significantly reduced for the effluent. The COD removal during the extent of the experiment is shown, for tank, in Figure 4.5. The overall COD treatment efficiency was about 54% (Table 4.3).. The COD removal may be mostly. accomplished by biological oxidation or absorption (or uptake) of organic compounds derived from livestock feeds that carried abundant carbonaceous, nitrogenous and phosphorus materials, since livestock wastewater contains generally little recalcitrant compounds. The ortho or total phosphorus removal during the extent of the experiment is shown, for tank, in Figure 4.6, 4.7. The ortho or total phosphorus removal effect was also obvious in the aeration and sedimentation tanks (at least 40%). The possible mechanism for the phosphorus removal would be an uptake of phosphorus by cells under aerobic condition and a subsequent sedimentation of the cells. Surplus phosphorus to be uptaken may be transformed to polyphosphorus as a storage material within the cells (21). A discharge of phosphorus is known to occur under anaerobic conditions (7, 37). -31-.

(41) In this study the best removal effect of suspended solids (SS) (63%) was first observed in the aeration tank. This seemed to be due to a transport hole between fermentation and aeration tanks, which screened out most of the sedimented solids. The suspended solids (SS) removal during the extent of the experiment is shown, for tank, in Figure 4.8.. Table 4.3. Wastewater treatment efficiency of the piggery slurry treatment system Efficiency for 40 days Parameter. COD SS T-P Ortho-P NH4+-N. Efficiency Influent Fermentation tank % (%) (㎖/ℓ) (㎖/ℓ) 18.5 54 3341 (2723.76) 25.9 65.5 0.54 (0.4) 42 33 39. 45.24 43 1431. Aeration. Sedim.. Sedim.. tank %. tank-1 %. tank-4 %. (㎖/ℓ) 45.5. (㎖/ℓ) 51.1. (㎖/ℓ) 56. (1819.5) 63. (1634.2) 63. (1461) 63. (0.2). (0.2). (0.2). 17.5. 45.2. 46.3. 46. (37.3) 18.8. (24.8) 39.5. (24.3) 39.1. (24.5) 38. (34.9) 26.4. (26) 33.5. (26.2) 37.8. (26.8) 41. (0.4). (952). (890). (848). -32-.

(42) 1800. NH4+ -N (mg/L). 1600. 1400. 1200. 1000. 800. 600 0. 10. 20. 30. 40. 50. Days. Figure 4.4. Dynamics of ammonium removal in the circulating treatment system (•- Influent tank; ο- Fermentation tank; ▼- Aeration tank;. ▽-. Sedimentation tank A;. - Sedimentation tank D). -33-.

(43) 4000. 3500. COD (mg/L). 3000. 2500. 2000. 1500. 1000. 500 0. 10. 20. 30. 40. 50. Days. Figure 4.5. Dynamics of chemical oxygen demand (COD) removal in the circulating treatment system (•- Influent tank; ο- Fermentation tank;. ▼-. Aeration. tank;. ▽-. Sedimentation tank D). -34-. Sedimentation. tank. A;. -.

(44) 70. Ortho-P (mg/L). 60. 50. 40. 30. 20. 10 0. 10. 20. 30. 40. 50. Days. Figure 4.6. Dynamics of ortho-phosphorus removal in the circulating treatment system (•- Influent tank; ο- Fermentation tank; ▼- Aeration tank;. ▽-. Sedimentation tank A;. - Sedimentation tank D). -35-.

(45) 70. T-P (mg/L). 60. 50. 40. 30. 20. 0. 10. 20. 30. 40. 50. Days. Figure 4.7. Dynamics of total phosphorus removal in the circulating treatment system (•- Influent tank; ο- Fermentation tank; ▼- Aeration tank;. ▽-. Sedimentation tank A;. - Sedimentation tank D). -36-.

(46) 1.0. SS (mg/L). 0.8. 0.6. 0.4. 0.2. 0.0 0. 10. 20. 30. 40. 50. Days. Figure 4.8 Dynamics of suspended solid removal in the circulating treatment system (•- Influent tank; ο- Fermentation tank; ▼- Aeration tank;. ▽-. Sedimentation tank A;. - Sedimentation tank D). -37-.

(47) 4.3 Principal Component Analysis of Input Data. The neural networks can estimate output about the ignorant input data that were not used to learning neural network. This is called the generalization capability and important characteristic of neural networks. For the effective generalization, the number of training data must be more than 100 times the number of input dimension, at least.. But it was rather difficult to obtain enough training data. because the biological process by microorganisms usually takes long time to reach the stable treatment efficiency. Moreover, the input and output dimensions of the neural networks was 9 and 5, respectively. Training data measured for 47 days, were not enough to figure out the complex correlation between input and output in each tank, and also it was a little hard to expect a generalization. Moreover, there were some noises in data due to a measuring error or unstable bioprocess. In order to reduce the input and output dimensions and also remove the noisy data, we first used the principal component analysis (PCA) method to analyze the training data. PCA projects high dimensional data onto low dimensional coordinates that consist of principal component axes. In other words, PCA finds the first component that can best express variation for high dimensional data and other orthogonal component of first component and represents high dimensional data on the new orthogonal coordinates with low dimensional ones by projection. PCA method is applied to analysis statistical data because the projected data on 2 or 3 dimensional coordinates are visual and can be classified easily. -38-.

(48) To find new orthogonal coordinates for the measured data, we should get eigenvalues from correlation matrix of the measured data, and then selects a few eigenvalues after ordering in magnitude. Finally the new orthogonal coordinates become eigenvectors for the selected eigenvalues.. Appearing the values to be represented by. eigenvectors on planes makes easy to understand the correlation of each high dimensional data. If N dimension of the training data can be expressed by vector X and also C, N×N correlation matrix is given as following.. C =< X T X >. (1) Where X is N dimension input vector and < > means expectation. The eigenvalue and eigenvector of this correlation matrix C can be obtained by following equations. CX = λ X. (2). det(λI − C) = 0. (3) Where λis eigenvalue and X means eigenvector of C corresponding to λ. Principal components are obtained by taking eigenvectors of a few eigenvalues whose magnitudes are more than any other eigenvalues and the projected data onto the low dimension coordinates -39-.

(49) are given by inner production with these eigenvectors. In this study, we used three axes as orthogonal coordinates. These axes were obtained by analysis of the PCA analysis to remove the data with one-to-many mapping that gives different output of the same inputs. Figure 4.9 showed the PCA results for the measured data in aeration tank. The X-axis denoted the mapping result by the first eigenvector, and Y-axis denoted the result by the second eigenvector.. The number of each point in Figure 4.9 (A) indicates. the passage from starting day.. Figure 4.9 (B) showed the PCA. results for target data measured in sedimentation tank 1, and each axis was same with that of Figure 4.9 (A). As shown in Figure 4.9, there were several data with one-to-many mapping property but we removed these values in the training process.. -40-.

(50) (A) 1.5. 9. 2. 41. 47. 38 1. 35 43 4. 26. 0.5. 0. 16 34. 31 10. 44 3 24 2. 20 39. 40. 19 1 32. 17 4 52 5. 37. 46. 22. 33. 27. 171 21. 8 30. 2188. 29. 14. 36 23. 24 -0.5. 3. 15. 5. 6. -1. 1 -1.5 -2. -1.5. -1. -0.5. 0. 0.5. 1. (B) 0.8. 40 43. 18 0.6 16 0.4. 47. 19 15. 22 32 28. 0.2. 37 4. 24 4 63 6. 34. 33. 45 21. 0. 44 3286 42. 13. 30. -0.2. 41. 39. 7. 20. -0.4 9 -0.6. 35 1 01 2 11. 1. 14. 31 27 3. 56. -0.8. 23. 8 -1. 2. 25. 17 29. -1.2 -2. -1.5. -1. -0.5. 0. 0.5. 1. 1.5. Figure 4.9. Principal component analysis of input (A) output (B) of aeration tank data during the 47 days' running period -41-.

(51) 4.4 Modeling of Treatment System by Neural Networks. Among 47 training data, we reversed the 6th, 11th, 16th, 21st, 26th, 31st, 36th, 41st, 46th, 47th data for training phase, which were randomly selected and used as test data to evaluate the generalization performance of the neural network.. The neural network has one. hidden layer with 30 nodes that were determined by an ad hoc method and nonlinear function of each layer except that input layer has a sigmoid.. The weight values were adjusted by error back-propagation. algorithm. Through computational experiment we could assure that the learned neural network successfully imitated each tank of treatment system and approximated the target values of the input pattern well. Figure 4.10 showed the graphic estimations of COD, NH4+-N, o-P and SS values based upon the neural network analysis.. The X-axis. represented the tanks from 46th days influent tank to 47th days sedimentation tank 2. As shown in Figure 4.10, the proposed neural network could successfully monitor the treatment results according to the population densities of microorganisms. A dramatic increase of the measured SS in the influent at 47th day (Figure 4.10 D) was an outlier due to a sampling error.. -42-.

(52) 4500. 1800. A. Measured Estimated. 4000. B 1600. 3000. NH4+-N (mg/L). COD (mg/L). 3500. 2500. 2000. 1400. 1200. 1000 1500. 1000. 800. 500 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 600 1. Tank Number. 2. 3. 4. 5. 6. 7. 8. 9. 10. Tank Number. 60. 1.0. D. 50. 0.8. Suspended solids (SS) (mg/L). ortho - Phosphorus ( o - P) (mg/L). C. 40. 30. 20. 10. 0.6. 0.4. 0.2. 0.0 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. Tank Number. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. Tank Number. Figure 4.10. Perdiction of various treatment parameters COD (A), NH4+-N (B), ortho-P (C) and SS (D) by the neural network modeling. COD and NH4+-N data were dited from reference 25. The serial numbers in X-axis indicate samples taken from tanks of influent, fermentation, aeration, sedimentation-1 and sedimentation-4 at 46 and 47 day's running in order, respectively.. -43-.

(53) 4.5 Molecular Analysis of Ammonium Removal. The isolated heterotrophs TSA-1 (Brevundimonas diminuta), TSA-3 (Alcaligenes faecalis), TSA-4 (not identified) were tested for their ammonium uptake. These three strains could utilize (NH4)2SO4 as a sole nitrogen source for their growth. Ammonium appeared to be almost utilized since little ammonium was detected at the stationary phase.. Phosphorus removal efficiency was observed up to 60%.. NH4+-N and ortho-phosphorus utilization rates appeared to be species or strain specific. This indicates a direct utilization of NH4+ by a heterotroph and hence removal of nitrogen from the system by circumventing nitrification process that is energy and oxygen consuming pathway.. It, therefore, appeared that the ammonium. uptake and utilization could contribute to the nitrogen removal in the treatment system (particularly aeration tank). Amplifications of GS gene with the GS-L and GS-R primers from the isolated heterotrophic bacteria were performed.. The GS gene. product amplified from Bacillus subtilis 168 could hybridize with one of the PCR products from the isolated bacteria (Figure 4.11). Therefore, the presence of ability of ammonium utilization and GS gene in the heterotrophs isolated from the treatment system indicates the possibility that ammonium removal in the system occurs via GS system of these organisms being involved in amino acid synthesis.. -44-.

(54) (A) D 1 2. (B) 3. 4. D 1 2. 3. 4. Figure 4.11. Amplification of glutamine synthetase (GS) gene using GS-L and GS-R primers from the total DNA of heterotrophic bacteria (A) and southern hybridization with the PCR products using the putative GS gene product (1269 bp indicated by arrow heads) of Bacillus subtilis 168 (B) as a DNA probe. Hybridization and washing were done under a stringent condition (65℃). D: DNA 1kb ladder; 1 Bacillus subtilis 168; 2 Alcaligenes faecalis TSA-3; 3 Brevundimonas diminuta TSA-1; 4 TSA-4. -45-.

(55) Ⅴ. Conclusion In this study a novel monitoring system of piggery slurry circulating treatment system has been proposed.. Multi-layer neural. networks combined with PCA successfully modeled the tank characteristics. It was possible to train the neural network with the given data by reducing the input dimension with minimal loss of information and removing the noisy data with one-to-many mapping property.. The proposed model may be useful to develop a reverse. neural network model that could be used to determine optimal microbial densities critical for a desired quality level of the treated wastewater. The following conclusions can be drawn based on the results from this study: 1. Alcaligenes faecalis appeared to survive and grow under low oxygen tension and anoxic condition and may oxidize ammonia under aerobic condition resulting in the ammonium removal.. 2. LAB were dominantly observed in anoxic condition, indicating the BOD removal under anaerobic condition.. 3. Multi-layer neural network combined with PCA successfully modeled the treatment system using relatively small amount of data.. 4. The proposed model may be useful to develop a reverse neural network model that could be used to determine optimal microbial -46-.

(56) densities critical for a desired quality level of the treated wastewater.. 5. A significant removal of ammonium may be attributed to an uptake by glutamine synthetase.. -47-.

(57) Ⅵ. FUTURE RESEARCH. The following needs to be considered for a further research:. 1. The C/N ratio could be adjusted to increase the treatment efficiency.. 2. The application of reverse neural network modeling to the system will allow an elucidation of optimal microbial species and population densities to optimize the treatment system.. 3. Biosensors and chemical sensors will be highly useful for the collection and analysis of data.. 4. Molecular sensors such as gene chips for heterotrophic and autotrophic ammonium oxidizers and other degraders will be quite useful for a rapid and sensitive monitoring of the treatment system.. 5. Transformation of nitrate to ammonium and its removal under anoxic condition.. -48-.

(58) REFERENCE [1] American Public Health Association, American Water Works Association and Water Environment Federation. 1992. Standard Methods for the Examination of Water and Wastewater, 18th ed., Washington, D.C., U.S.A. [2] Anderson, I.C., 1993. A comparison of NO and N2O production by the autotrophic nitrifier Nitrosomonas europaea and the heterotrophic nitrifier Alcaligenes faecalis. Appl. Environ. Microbiol. 59, 3525-3533. [3] Avelino J. Gonzalez & Douglas D. Dankel, "The Engineering of Knowledge-based Systems", 1993 Prentice-Hall Inc. ISBN 0-13334293-X. [4] Barto, A.G., R.S. Sutton, and C.W. Anderson. 1983. Neuronlike adaptive elements that can solve difficult learning control problems, IEEE Trans. Systems, Man and Cybernetics. 13, 834846. [5] Baum, E.B., and D. Hausser. 1989. What size net gives valid generalization? Neural Inform. Process. Systems 1, 81-90. [6] Blouin, M., J.G. Bisaillon, R. eaudet, and M. Ishaque. 1990. Nitrification of swine waste. Can. J. Microbiol. 36(4):273-278. [7] Bond, P.L., R. Erhart, M. Wagner, J. Keller, and L.L. Blackall. 1999. Identification of some of the major groups of bacteria in efficient and. nonefficient biological phosphorus removal. activated sludge systems. Appl. Environ. Microbiol. 65, 40774084. -49-.

(59) [8] Chen, H.J., C.M. Liao, and P.Y. Yang. 1999. Development of a prefabricated treatment plant for diluted pig wastewater. J. Environ. Sci. Health B. 34(6):1009-1021. [9] Choi, J.H, Y.R. Chung, and S.C. Koh. 1999. Microbial treatment of swine wastewater by recycling reactor system with sequential oxic and anoxic conditions. Abstracts of Korea Environ. Eng. Soc., Kwangju, Korea. [10] Choi, J.H., J.I. Shon, H.S. Yang, Y.R. Chung, M.H. Lee, and S.C. Koh. 2000. Modeling of recycling oxic and anxic treatment system for swine wastewater using neural networks. Biotechnol. Bioprocess Eng. Vol. 5 No. 5: 355-361. [11] Chung Y.G. 1990. Proper management plan of swine wastewater for the protection of potable water resources. Korea Advanced Institute of Science and Tech., Korea. [12] Chung, Y.R. et al. 1997. Development of Piggery wastewater treatment. technolnogy using microbial agent. Chinju,Korea.. [13] Fatemeh Zahedi, "Intelligent Systems for Business: Expert Systems with Neural networks, 1993 Wadsworth Inc. ISBN 0534-18888-5. [14] Haykin Simon, "Neural Networks", 1994 Macmillan College Publishing Company Inc. ISBN 0-02-352761-7. [15] Henze, M., C.P.L. Grady, W. Gujer, G.V.R. Marais, and T. Matsuo. 1987. Activated Sludge Model no. 1. Scientific and Technical Reports no. 1. IAWPRC, London. [16] Hiraishi, A., Y. Ueda, and J. Ishihara. 1998. Quinone profiling of bacterial communities in natural and synthetic sewage activated -50-.

(60) sludge for enhanced phosphate removal. Appl. Environ. Microbiol. 64, 992-998. [17] Holt, J.G., N.R. Krieg, P.H.A. Sneath, J.T. Staley, and S.T. Williams. 1994. Bergey's Manual of Determinative Bacteriology. 9th ed. Wilkins Co., Baltimore, MD. [18] Hornik, K. (1991) Approximation capabilities of multi-layer feed forward networks. Neural Networks 4, 251-257. [19] Jolliffe, I.T. Principal Component Analysis Springer Verlag, 1986, ISBN 0-387-96269-7 or ISBN 3-540-96269-7. [20] Kim, M.Y., S.Y. Bang, S.Y. Lee, J.H. Lee, H.T. Joun, H.S. Jung, and H. Jung. 1992. Easy-to-understand Neural Network Computer. 61-63. Electronics Newspaper Publishing co., Korea. [21] Koh, S.C., J.P. Bowman, and G.S. Sayler. 1993. Soluble methane monooxygenase production and trichloroethylene degradation by a type I methanotroph, Methylomonas methanica 68-1. Appl. Environ. Microbiol. 59, 960-967. [22] Kortstee, G.J., et al. 1994. Biology of polyphosphateaccumulating. bacteria. involved. in. enhanced. biological. phosphorus removal. FEMS Microbiol. Rev. 15(2-3):137-153. [23] Krieg, N.R. and P. Gerhardt. 1994. Solid, liquid/solid, and semisolid culture In: Methods for General and Molecular Bacteriology, Am. Soc. Microbiol. pp. 216-223. Washington, D.C. [24] Kuai, L., and W. Verstraete. 1998. Ammonium removal by the oxygen-limited autotrophic nitrification-denitrification system. Appl. Environ. Microbiol. 64(11):4500-4506. -51-.

(61) [25] Lee, D.S. and J.M. Park. 2000. Modeling for industrial wastewater treatment process using hybrid neural networks. Abstracts of the Korea Soc. for Biotechnol. Bioeng., Taejon, Korea. [26] Lee, M.H., S.Y. Lee, C.H. Park 1992. Neural controller of onlinear dynamic systems using higher order neural networks. Electronics Letters 28, 276-277. [27] Lee, S.B. 1999. Fuzzy-Neuro Control System. 141-149. Kyohaksa co., Korea. [28] Liao, C.M., T. Maekawa, H.C. Chiang, C.F. Wu. 1993. Removal of nitrogen and phosphorus from swine wastewater by intermittent aeration processes. J. Environ. Sci. Health. B28, 335374. [29] Liao, C.M., T. Maekawa, and S.D. Feng. 1995. Ntrogen and phosphorus removal for swine wastewater by ammonium crystallization and intermittent aeration process. J. Environ. Sci. Health. B30, 733-758. [30] Madigan, M.T., J.M. Martinko, and J. Parker 2000. Brock Biology of Microorganisms, Ninth eds. 131-132. Prentice-Hall, Inc., New Jersey, USA. [31] Morgan, D., and C. Scofield. 1991. Neural Networks and Speech Recognition, Kluwer Academic Publishers, Boston. [32] National Institute of Environmental Research. 1986. A Study of the Current Status of Livestock Waste Discharge and its Environmental Impact. National Institute of Environmental Research, Seoul, Korea. -52-.

(62) [33] Office of Environment (Korea). 1988. Environmental Protection Act and Wastes Management Act. [34] Otte, S., N.G. Grobben, L.A. Robertson, M.S.M. Jetten, and J.G. Kuenen. 1996. Nitrous oxide production by Alcaligenes faecalis under transient and dynamic aerobic and anaerobic conditions. Appl. Environ. Microbiol. 62, 2421-2426. [35] Papen, H., R. Berg, I. Hinkel, B. Thoene, and H. Rennenberg. 1989. Heterotrophic nitrification by Alcaligenes faecalis: NO2-, NO3-, N2O, and NO production in exponentially growing cultures. Appl. Environ. Microbiol. 55, 2068-2072. [36] Park, Y.I., J.S. So, and S.C. Koh. 1999. Induction by carvone of the polychlorinated biphenyl (PCB)-degradative pathway in Alcaligenes eutrophus H850 Its molecular monitoring. J. Microbiol. Biotechnol. 9, 804-810. [37] Rimmerman, M.W. 1984. Biological phosphorus removal in wastewater treatment. Microbiol. Sci. 1, 149-152. [38] Sambrook, J., E. F. Fritsch, and T. Maniatis 1989. Molecular cloning: a laboratory manual, 2nd ed. Cold Spring Harbor Laboratory, Cold Spring Harbor, N.Y. [39] Shin, H.S., J.K. Koo, J.O. Kim, H.K. Shin, Y.K. Jeong. 1990. Management alternatives of livestock wastes for water resources conservation. J. Korea Solid Waste Eng. Soc. 7, 45-57. [40] Smibert, R.M., and N.R. Krieg. 1981. General characterization, p. 409-433. In P. Gerhardt, R.G.E. Murray, R.N. Costilow, E.W. Nester, W.A.R. Krieg, and G.B. Phillips (ed.), Manual of Methods for General Bacteriology. Am. Soc. Microbiol., -53-.

(63) Washington, D.C. [41] Sohn, J.I., M.H. Lee, J.H. Choi, and S.C. Koh. 2000. Monitoring of recycling treatment system for piggery slurry using neural networks. J. Korean Sensors Soc. 9, 127-133. [42] Weigend, A.S., and N.A. Gershenfeld 1994. Time Series Prediction: Forecasting the Future and Understanding the Past, Addison-Wesley Publishing Co., N.Y.. -54-.

(64) A CK N OW LE D GE M E N T. 길게만 느껴졌던 2년이란 시간이 어느새 흘러 그 결실을 거두게 되었 습니다. 지난 시간들을 돌이켜 생각해 보니, 조금 더 열심히 하지 못한 것이 아쉽기만 합니다. 대학원 생활을 하는 동안 많은 실험과 다양한 분 야를 경험해 보았지만 아직도 제 자신은 부족하기만 합니다. 그러나 시 작이 반이라는 말처럼, 지금부터 시작이라는 생각으로 더 넓은 세상에서 제 자신을 발전시켜 나아가겠습니다.. 2년의 대학원 생활동안 도와주신 많은 분들께 감사 드립니다. 석사 과정동안 돌봐주시고 학문적으로 많은 조언과 충고를 아끼지 않으셨던 고성철 지도 교수님께 감사 드립니다. 교수님께서 일깨워 주신 맡은 일 에 대한 책임감과 시간 관리의 소중함은 학교를 떠나더라도 제 자신에게 유용한 재산이 될 것입니다. 그리고 바쁘신 와중에도 관심을 보여 주시 고 학부때부터 석사과정에 이르기까지 자상하게 지도해 주신 박상윤 교 수님, 김인수 교수님, 송영채 교수님께도 감사드립니다. 학교를 마치고 이제 새로운 환경을 접하게 되는 저에게 교수님들의 가르침은 큰 힘이 될 것입니다. 이 논문이 완전한 모양새를 갖출 수 있도록 균주를 동정해주신 경상 대학교 미생물학과의 정영륜 교수님과 양현숙님, 그리고 신경회로망 모 델링을 도와주신 경북대 센서공학과 이민호 교수님, 손준일님께도 깊은 감사의 마음을 전합니다. 아울러, 환경 미생물 실험실 가족들에게도 감사 드립니다. 궂은 일도 마다 않고 함께 실험을 도와 준 후배 향이, 권숙, 상조, 그리고 저에게 많은 도움과 조언을 준, 병혁이를 비롯한 환경공학과 95동기들에게 고마.

(65) 움을 전하며 저마다 앞으로 하고자 하는 일에 많은 발전이 있길 진심으 로 바랍니다. 학부동기이자 대학원 후배인 경자도 열심히 연구하여 좋은 논문으로 잘 마무리하길 바랍니다. 그리고, 환경공학과 석사과정에 함께 입문한 성우오빠, 성진오빠, 진석, 미경, 함께 한 2년동안 정도 많이 쌓이 고 즐거웠던 일들, 힘들었던 일들이 많았지만 자기 일처럼 아껴주고 큰 힘이 되어준 동기분들께 감사 드리며, 학교에 남아있는 동기분들은 열심 히 연구하셔서 원하는 성과를 이루고, 특히 저에게 든든한 힘이 되어준 미경이도 희망하는 분야에서 뜻하는 바를 꼭 이룰 수 있기 바랍니다. 그 리고 제가 힘들고 지칠 때 언제나 아낌없는 사랑과 조언을 해준 헌희에 게 감사와 애정을 드립니다. 앞으로의 대학원 생활을 끝까지 잘 마무리 할 수 있도록 충실히 임하고, 즐거운 마음으로 학문을 탐구하는 아름다 운 모습을 늘 간직해주길 바랍니다.. 무엇보다도 이 결실을 사랑하는 부모님과 동생들에게 바칩니다. 타지 에서 공부하는 하나 뿐인 딸에 대한 그리움으로 가슴 저려 애태우시던 아버님, 어머님께 자랑스러운 딸로 설 수 있도록 어려운 시기에도 사랑 과 정성을 아끼지 않고 저를 돌봐주신 은혜에 감사드립니다. 그리고 이 제는 어른이 되어 오히려 저를 걱정하여 뒷바라지도 마다 않겠다는 동생 정진이와 정환이에게 고마움을 전하며, 항상 건강하기를 바랍니다..

(66) CURRICULUM VITAE Name: Jung- Hye Choi Date of Birth: Oct 3, 1976 e-mail Address : [email protected] Education M.S.. in Division of Civil and Environmental Engineering. Korea Maritime University, Korea March 1999-February 2001 B.S.. in Department of Environmental Engineering. Korea Maritime University, Korea March 1995-February 1999 Publications 1. Thesis: Neural Networks and Molecular Analyses of Circulationg Piggery Slurry Treatment System, Korea Maritime University, 2001. 2. Journal Papers: 1. J. I. Sohn, M. Lee, J. H. Choi and S. C. Koh (2000) Monitoring of recycling treatment system for piggery slurry using neural networks. Journal of the Korean Sensors Society. Vol. 9 No. 2:127-133..

(67) 2. Jung-Hye Choi, Jun-Il Shon, Hyun-Sook Yang, Young-Ryun Chung, Minho Lee, Sung-Cheol Koh. (2000) Modeling of recycling oxic and anxic treatment system for swine wastewater using neural networks. Biotechnol. Bioprocess Eng. Vol. 5 No. 5:355-361. 3. Jung-Hye Choi, Jun-Il Shon, Minho Lee, Sung- Cheol Koh. (2000) Neural networks and molecular analyses of recycling piggery slurry treatment system. J. Biotechnology. (Submitted) 4. Jung-Hye Choi, Jun-Il Shon, Hyun-Sook Yang, Young-Ryun Chung, Minho Lee, Sung-Cheol Koh. (2000) Analyses of recycling oxic and anxic swine wastewater treatment system using neural networks and molecular techniques. J. Industrial Technology Research Institute. (Submitted) 3. Conference papers: 1. J. Choi, Y. Chung, S. Koh (1999) Microbial treatment of swine wastewater by recycling reactor system with sequential oxic and anoxic conditions. Korean Society of Environmental Engineers. May 7-8. Seoul, Korea. 2. Jung-Hye Choi, Jun-Il Shon, Min-ho Lee, Sung-Cheol Koh. (1999) Modeling and Optimization Through Neural Networks of Recycling Piggery Slurry Treatment System. Korean Society of Environmental Engineers. November 5-6. Kwangju, Korea. 3. Jung-Hye Choi, Jun-Il Shon, Hyun-Sook Yang, Young-Ryun Chung, Min-Ho Lee, Sung- Cheol Koh. (2000) Monitoring of.

(68) Recycling Treatment System for Piggery Slurry Using Neural Networks. The Korean Society for Biotechnology and Bioengineering. April 8. Taejon, Korea. 4. Jung-Hye Choi, Jun-Il Shon, Hyun-Sook Yang, Young-Ryun Chung, Min-Ho Lee, Sung- Cheol Koh. (2000) Modeling and Optimization Through Neural Networks of Recycling Piggery Slurry Treatment System. American Society for Microbiology. May 21-25. L.A., U.S.A. 5. Choi JH, Sohn JI, Yang HS, Lee MH, Chung YR, Koh SC. 2000. Neural Networks and Molecular Analyses of Recycling Piggery. Slurry. Treatment. System.. Fifth. International. Symposium on environmental Biotechnology. July 9-13. Kyoto, Japan. 6. J. I. Sohn, M. Lee, J. H. Choi and S. C. Koh (2000) Monitoring of recycling reactor system using effective microorganisms by neural. networks.. Internation. Conference. on. Electrical. Engineering. July. Kitakyushu, Japan. 7. Jung-Hye Choi, Jun-Il Shon, Minho Lee, Sung- Cheol Koh. (2000) Neural networks and molecular analyses of recycling piggery slurry treatment system. The Microbiological Society of Korea. Oct 27-28. Taejon, Korea. 8. Jung-Hye Choi, Jun-Il Shon, Minho Lee, Sung- Cheol Koh. (2000) Neural network modeling and molecular analyses of recycling piggery slurry treatment system. Conference of the Korean Sensors Society. November 17-18. Seoul, Korea..

(69)

(2) 본 論文을 崔正惠의 工學碩士 學位論文으로 認准함.

(2) 본 論文을 崔正惠의 工學碩士學位論文으로 認准함.