基于神经网络的活性污泥系统建模及其控制

针对松江污水厂污水处理活性污泥系统，采用神经网络技术进行建模试验研究，在对实际运行数据剔除异常数据后，将样本数据随机分成训练样本、检验样本和测试样本。用试凑法确定合理的神经网络隐层节点数，用检验样本实时监控训练过程从而避免“过训练”现象，用多次改变网络初始连接权值求得全局极小点，从而建立了泛化能力较好的基于神经网络的活性污泥系统数学模型。利用建立的神经网络模型，对活性污泥系统运行情况的仿真与控制进行了分析研究。示例研究表明：神经网络技术能较好地应用于活性污泥系统的建模与控制，有很好的理论与实践意义。     

基于神经网络的活性污泥系统建模及其控制

针对松江污水厂污水处理活性污泥系统,采用神经网络技术进行建模试验研究,在对实际运行数据剔除异常数据后,将样本数据随机分成训练样本、检验样本和测试样本.用试凑法确定合理的神经网络隐层节点数,用检验样本实时监控训练过程从而避免"过训练"现象,用多次改变网络初始连接权值求得全局极小点,从而建立了泛化能力较好的基于神经网络的活性污泥系统数学模型.利用建立的神经网络模型,对活性污泥系统运行情况的仿真与控制进行了分析研究.示例研究表明:神经网络技术能较好地应用于活性污泥系统的建模与控制,有很好的理论与实践意义.     

基于BP神经网络的垃圾焚烧过程故障诊断

应用BP神经网络方法,建立了垃圾焚烧过程故障诊断的模型.该方法采用梯度下降法训练网络权值,利用最优停止法对网络进行了优化,避免了过拟合现象.提高了BP神经网络的训练速度和泛化能力.并以实际焚烧过程工况数据进行性能检验,检验结果表明了该BP网络具有较高的诊断精度.     

Elman神经网络在抽水地面沉降预测中的应用

将Elman神经网络应用于南京奥体中心地铁站抽水试验初期对地面沉降的预测，提出了一种预测模型。该模型采用对已知数据样本进行分段的方法，把分段后的数据作为网络输入和期望输出训练网络，从而达到动态建模的目的。从预测结果来看，该模型误差较小，基本上能够反映2002—07—21当天地面沉降的真实情况，具有一定的可靠性和实用性，因此，该模型可作为后期由于基坑开挖和防渗排水所引起地面沉降的分析工具。     

基于BP网络的水质综合评价模型及其应用

讨论了BP网络模型存在的不足及建模条件，提出了建立合理的BP网络模型的基本原则和步骤。针对水质评价问题，通过在各类水质污染指标浓度区间内生成随机分布样本的方法，组成足够多用于BP网络训练、检验和测试用的样本，建立了辽河水质综合评价的BP网络模型；给出了区分不同类别水质的模型分界值样本和模型输出分界值。     

基于BP神经网络的多壁碳纳米管复合CdS-TiO_2光催化剂优化合成研究

基于CdS-TiO2/多壁碳纳米管(MWCNTs)光催化剂降解甲苯的正交实验数据,采用反向传播(BP)神经网络训练并建立了光催化剂合成条件设计的神经网络模型。以正交实验确定的4个主要影响因素作为输入层参数,以甲苯降解率作为输出层参数,将全部实验数据分为训练样本集和预测样本集。运行网络,系统误差为0.000 724,网络预测值与实验数据值相关系数达到0.989,说明该网络具有较好的训练精度及泛化能力。并利用训练好的神经网络预测得到CdS-TiO2/MWCNTs光催化剂的最佳合成条件:焙烧温度为460℃,MWCNTs复合量为1.5%(质量分数),活性组元摩尔比(TiO2/CdS)=80∶1,水加入量为12%(体积分数)。     

基于BP网络的水质综合评价模型及其应用

讨论了BP网络模型存在的不足及建模条件,提出了建立合理的BP网络模型的基本原则和步骤.针对水质评价问题,通过在各类水质污染指标浓度区间内生成随机分布样本的方法,组成足够多用于BP网络训练、检验和测试用的样本,建立了辽河水质综合评价的BP网络模型;给出了区分不同类别水质的模型分界值样本和模型输出分界值.     

利用复介电常数和BPAT神经网络进行环境监测的可行性研究

土壤中的污染物成分复杂，其含量与复介电常数之间具有很强的非线性关系。以土壤样品复介电常数的实部、虚部分别作为输入，以其含水率、体密度和所含6种已知离子的浓度分别作为输出，建立BP人工神经网络。把吉泰兰地区的土壤样品数据分为训练样本集和检验样本集，网络训练后，其学习效果显示模型的性能很好，检验样本的预测结果也与实测值较好吻合，说明利用复介电常数和BP人工神经网络进行环境监测是一种好的方法。     

RBF与Elman神经网络在人工湿地复合基质去污效果预测中的应用

人工湿地的去污机理复杂、呈高度非线性,故利用神经网络技术构建模型预测其长期运行效果。通过构建人工湿地复合基质模拟槽系统进行为期4个月的实验,监测得到56组COD去除率数据样本,经Matlab小波去噪后分别利用RBF和Elman网络构建动态神经网络模型,预测该系统对生活污水中COD去除效果。结果表明,RBF和Elman神经网络预测值的均方根误差分别为0.0186和0.0163,精度较高,该系统后期的COD去除率在49.4%～59.0%之间。     

遗传神经网络对肼类化合物的定量构效研究

应用遗传算法和反向传播（BP）神经网络相结合的方法，研究了胼类化合物的定量构效（QSAR）关系，构建了遗传神经网络QSAR模型。对30种肼类化合物的6个量子化学参数进行相关性和主成分分析，利用遗传神经网络QSAR模型对肼类化合物的毒性参数进行预测。结果表明，与常规BP神经网络建立的模型相比较，遗传神经网络QSAR模型有效解决了常规BP神经网络模型存在的过训练和过拟合问题，并且具有很好的预测效果。     

Improved computational model (AQUIFAS) for activated sludge, integrated fixed-film activated sludge, and moving-bed biofilm reactor systems, Part I: Semi-empirical model development

Research was undertaken to develop a model for activated sludge, integrated fixed-film activated sludge (IFAS), and moving-bed biofilm reactor (MBBR) systems. The model can operate with up to 12 cells (reactors) in series, with biofilm media incorporated to one or more cells, except the anaerobic cells. The process configuration can be any combination of anaerobic, anoxic, aerobic, post-anoxic with or without supplemental carbon, and reaeration; it can also include any combination of step feed and recycles, including recycles for mixed liquor, return activated sludge, nitrates, and membrane bioreactors. This paper presents the structure of the model. The model embeds a biofilm model into a multicell activated sludge model. The biofilm flux rates for organics, nutrients, and biomass can be computed by two methods--a semi-empirical model of the biofilm that is relatively simpler, or a diffusional model that is computationally intensive. The values of the kinetic parameters for the model were measured using pilot-scale activated sludge, IFAS, and MBBR systems. For the semiempirical version, a series of Monod equations were developed for chemical oxygen demand, ammonium-nitrogen, and oxidized-nitrogen fluxes to the biofilm. Within the equations, a second Monod expression is used to simulate the effect of changes in biofilm thickness and fraction nitrifiers in the biofilm. The biofilm flux model is then linked to the activated sludge model. The diffusional model and the verification of the models are presented in subsequent papers (Sen and Randall, 2008a, 2008b). The model can be used to quantify the amount of media and surface area required to achieve nitrification, identify the best locations for the media, and optimize the dissolved oxygen levels and nitrate recycle rates. Some of the advanced features include the ability to apply different media types and fill fractions in cells; quantify nitrification, denitrification, and biomass production in the biofilm and mixed liquor suspended solids; and perform dynamic simulations.     

ASM No.2d模型模拟SBR工艺同步脱氮除磷效能

SBR工艺由于处理上的高效性和操作上的灵活性在世界范围的污水处理领域得到广泛应用.采用国际水质协会1999年提出的ASM No.2d模型,利用matlab作为程序开发工具编制计算模型,并利用该模型对一实验室规模SBR系统进行模拟.模拟过程中动力学和化学计量参数采用ASM No.2d给出的典型参数值,并结合实际SBR系统进行了修正.结果表明,该模型能够较好模拟SBR工艺同步脱氮除磷效能,说明应用ASM No.2d进行SBR系统的模拟能够对SBR系统的优化和控制起辅助作用.     

Improved computational model (AQUIFAS) for activated sludge, integrated fixed-film activated sludge, and moving-bed biofilm reactor systems, part II: multilayer biofilm diffusional model

Research was undertaken to develop a diffusional model of the biofilm that can be applied in lieu of a semi-empirical model to upgrade an activated sludge system to an integrated fixed-film activated sludge (IFAS) or moving-bed biofilm reactor (MBBR) system. The model has been developed to operate with up to 12 cells (reactors) in series, with biofilm media incorporated to one or more of the zone cells, except the anaerobic zone cells. The values of the kinetic parameters for the model were measured using pilot-scale activated sludge, IFAS, and MBBR systems. The biofilm is divided into 12 layers and has a stagnant liquid layer. Diffusion and substrate utilization are calculated for each layer. The equations are solved simultaneously using a finite difference technique. The biofilm flux model is then linked to the activated sludge model. Advanced features include the ability to compute the biofilm thickness and the effect of biofilm thickness on performance. The biofilm diffusional model is also used to provide information and create a table of biofilm yields at different substrate concentrations that can be used in the semi-empirical model.     

Model development and application for surfactant biodegradation in an acclimatising activated sludge system

A model for the biodegradation of non-ionic surfactants in an activated sludge system during acclimatisation was developed, based on respirometric and titrimetric experimental data. The data were obtained in a sequencing batch reactor (SBR) using a non-ionic surfactant as sole carbon source and sludge previously acclimatised to a different surfactant. The model was successfully applied to successive SBR-cycles of sludge acclimatisation processes subjected to two ethoxylated surfactants. The model was validated using the corresponding total organic carbon data. The evolution of the model parameters along the acclimatisation process was examined. An acclimatisation model was developed using the evolution of two of these parameters (affinity constant and inhibition constant), supported by two independently calculated acclimatisation indicators. This acclimatisation model was then applied to determine an optimal surfactant concentration sequence to feed non-acclimatised sludge during a period of 41 days, in order to induce pre-acclimatisation to this surfactant before it replaces another one in the wastewater. The model was also useful in the economical assessment of this and alternative procedures to cope with frequent changes in activated sludge feed composition.     

Estimation of autotrophic maximum specific growth rate constant--experience from the long-term operation of a laboratory-scale sequencing batch reactor system

The autotrophic maximum specific growth rate constant, muA,max, is the critical parameter for design and performance of nitrifying activated sludge systems. In literature reviews (i.e., Henze et al., 1987; Metcalf and Eddy, 1991), a wide range of muA,max values have been reported (0.25 to 3.0 days(-1)); however, recent data from several wastewater treatment plants across North America revealed that the estimated muA,max values remained in the narrow range 0.85 to 1.05 days(-1). In this study, long-term operation of a laboratory-scale sequencing batch reactor system was investigated for estimating this coefficient according to the low food-to-microorganism ratio bioassay and simulation methods, as recommended in the Water Environment Research Foundation (Alexandria, Virginia) report (Melcer et al., 2003). The estimated muA,max values using steady-state model calculations for four operating periods ranged from 0.83 to 0.99 day(-1). The International Water Association (London, United Kingdom) Activated Sludge Model No. 1 (ASM1) dynamic model simulations revealed that a single value of muA,max (1.2 days(-1)) could be used, despite variations in the measured specific nitrification rates. However, the average muA,max was gradually decreasing during the activated sludge chlorination tests, until it reached the value of 0.48 day(-1) at the dose of 5 mg chlorine/(g mixed liquor suspended solids x d). Significant discrepancies between the predicted XA/YA ratios were observed. In some cases, the ASM1 predictions were approximately two times higher than the steady-state model predictions. This implies that estimating this ratio from a complex activated sludge model and using it in simple steady-state model calculations should be accepted with great caution and requires further investigation.     

Effects of chlorination on the adhesion strength and deflocculation of activated sludge flocs.

A side effect of the application of chlorine for controlling filamentous bulking is deflocculation of floc-forming bacteria, which may cause unacceptable effluent deterioration depending on dosing. It was assumed that chlorine may adversely affect the adhesion ability of floc bacteria, promoting their erosion in shear flow. The effect of chlorination on the strength of activated sludge flocs was investigated. The adhesion-erosion (AE) model developed by Mikkelsen and Keiding was used to interpret results from deflocculation tests with varying shear and solids concentration. The AE model yields the adhesion enthalpy (deltaHG/R) of cells in sludge flocs and parameters from the model were used to quantify the sludge in terms of floc strength. Two activated sludges with different initial characteristics were studied. The resulting model parameters showed that the AE model was suitable for quantifying the bond energy of particles to the activated sludge exposed to chlorine. For one activated sludge, adhesion of cells was largely unaffected by the applied chlorine doses. A second sludge showed reduced adhesion strength with chlorine, leading to increasing deflocculation. The simple batch test and AE model proved valuable for assessing the effect of chlorination on the flocs in activated sludge. By use of these procedures, it is possible to determine acceptable chlorine dosing to avoid excessive deflocculation and effluent deterioration.     

Immobilized cell augmented activated sludge process for enhanced nitrogen removal from wastewater.

The immobilized cell augmented activated sludge (ICAAS) system combines a cell immobilization technique and an offline enricher-reactor for the bioaugmentation of the activated sludge system to improve treatment performances. In this study, enhanced nitrogen removal using ICAAS was investigated. Laboratory-scale, offline, batch enricher-reactors were used to maintain nitrification and denitrification activities of coimmobilized nitrifiers and denitrifiers used to augment a laboratory-scale completely mixed activated sludge system (CMAS) treating synthetic wastewater. Cellulose triacetate was the media used to entrap nitrifiers and denitrifiers at a 2:1 mass ratio. The ICAAS augmented with the coimmobilized cells between 5 and 20% by volume gained 24 +/- 5% higher nitrogen removal than a control CMAS, which provided nitrogen removal of 28 +/- 7%. The ICAAS scheme is a viable alternative for upgrading existing activated sludge systems to gain better nitrogen removal. .     

Simultaneous nitrification-denitrification and clarification in a pseudoliquified activated sludge system.

This paper describes results from a pilot study of a novel wastewater treatment technology, which incorporates nutrient removal and solids separation to a single step. The pseudoliquified activated sludge process pilot system was tested on grit removal effluent at flowrates of 29.4 to 54.7 m3/d, three different solid residence times (SRT) (15, 37, and 57 days), and over a temperature range of 12 to 28 degrees C. Despite wide fluctuations in the influent characteristics, the system performed reliably and consistently with respect to organics and total suspended solids (TSS) removals, achieving biochemical oxygen demand (BOD) and TSS reductions of > 96% and approximately 90%, respectively, with BOD5 and TSS concentrations as low as 3 mg/L. Although the system achieved average effluent ammonia concentrations of 2.7 to 3.2 mg/L, nitrification efficiency appeared to be hampered at low temperatures (< 15 degrees C). The system achieved tertiary effluent quality with denitrification efficiencies of 90 and 91% total nitrogen removal efficiency at a total hydraulic retention time of 4.8 hours and an SRT of 12 to 17 days. With ferric chloride addition, effluent phosphorous concentrations of 0.5 to 0.8 mg/L were achieved. Furthermore, because of operation at high biomass concentrations and relatively long biological SRTs, sludge yields were over 50% below typical values for activated sludge plants. The process was modeled using activated sludge model No. 2, as a two-stage system comprised an aerobic activated sludge system followed by an anoxic system. Model predictions for soluble BOD, ammonia, nitrates, and orthophosphates agreed well with experimental data.     

Maximum nitrogen removal in the step-feed activated sludge process.

This paper presents a mathematical framework that can be used to determine the flow distributions for a step-feed activated sludge process that result in maximum nitrogen removal. The model indicates that nitrogen removal efficiency in a step-feed activated sludge process is highly dependent on the ultimate biochemical oxygen demand (BOD(L))-to-total Kjeldahl nitrogen (TKN) ratio of the wastewater. For typical domestic wastewater, which has a relatively high BOD(L)-to-TKN ratio, the step-feed process will outperform the Modified Ludzack-Ettinger process for nitrogen removal, when the flow to each step is optimally distributed. Using plant-specific water quality data and operating conditions from a 1-year period, nitrogen removal performance for four step-feed activated sludge plants operated by the Sanitation Districts of Los Angeles County (California) was calculated using the developed model. The calculated nitrogen removal efficiencies match well with the actual plant performance data. These results validate the model as a useful tool for predicting nitrogen removal in a step-feed activated sludge process. Other analyses revealed that improvements in nitrogen removal at existing facilities are achievable by adjusting the split of primary effluent flow to each anoxic zone several times during the day. The timing of the adjustments and the optimal flow splits can be determined from data on diurnal fluctuations in BOD(L) and TKN concentrations. An example is provided to illustrate the application of such an operating strategy and the potential enhancement of nitrogen removal.