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。     

活性污泥系统神经网络建模与仿真研究

采用神经网络技术对松江污水厂污水处理活性污泥系统进行建模试验研究，在对实际运行数据按机理准则和范围准则剔除异常数据后，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，以避免采用过大或过小的网络结构，在训练过程中用检验样本实时监控从而避免“过训练”现象的影响，较好地解决神经网络方法建模的两大难题，从而建立可靠、有效的活性污泥系统神经网络模型。并应用建立的网络模型对活性污泥系统的运行情况进行了仿真研究。建模研究表明，神经网络技术能较好地应用于活性污泥系统的建模，模型具有较好的泛化能力，有很好的实用价值。     

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.     

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.     

用分子连接性指数预测苯衍生物发光菌EC_(50)

从文献中收集了101种取代苯对发光菌的EC50,筛选出90种取代苯为样本集并计算了其价键分子连接性指数。按照2∶1将样本集随机分为训练集和预测集,用逐步回归建立了分子连接性指数对取代苯类的发光菌毒性EC50的QSAR方程(R2=0.874,S=0.224)。Jackknife检验结果表明方程具有较好的稳健性,预测集检验证明方程具有较好的预测能力。     

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。     

自动控制技术在垃圾渗滤液处理中的应用研究

在国内垃圾渗滤液处理工艺的基础上,对其自动控制系统进行研究.主要介绍了该自动控制技术的设计方法及应用实例,并对自动控制及手动控制运行进行了对比研究.研究结果表明,该自动控制技术能较好地适应水质波动大的垃圾渗滤液,同时较好地解决了设备的防腐蚀问题.     

T-S模糊神经网络在厌氧反应器预测中的应用

3个厌氧反应器运行稳定后,用三氯甲烷和2、4-二硝基酚作为毒物负荷对它们进行了冲击试验.利用负荷冲击试验所得的数据集建立了T-S模糊神经网络,并用其预测了反应器的容积产气率、挥发性脂肪酸和CH4体积含量.研究结果表明,基于某一反应器建立的T-S模糊神经网络可以很好地预测毒物负荷冲击下该反应器的容积产气率、挥发性脂肪酸和CH4变化规律,实测值与预测值的相关系数均＞0.850;但是基于某一反应器建立的模糊神经网络用来预测其他反应器时,其预测能力较差,预测值和实测值的相关系数基本上＜0.500.     

氯化钙控制污泥膨胀

污泥膨胀是活性污泥工艺运行中经常遇到的最棘手的问题之一。本实验以人工合成污水为底物,采用序机式活性污泥法（SBR）;研究投加无机混凝剂氯化钙控制污泥膨胀的情况,同时研究丝状菌和菌胶团的变化。研究发现,投加氯化钙后,丝状菌数量明显减少;形成较多大而密实规则的菌胶团,污泥膨胀得到控制。污泥容积指数（SVI）由最初的309.5 mL/g降到67.1 mL/g,污泥沉降性能改善。停止投加氯化钙后又运行了18周期,活性污泥没有发生再次膨胀。投加氯化钙对COD去除率没有明显影响。研究结果表明,投加氯化钙是一种有效的污泥膨胀应急控制措施。     

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.     

生物脱氮除磷活性污泥系统复合模拟方法

为避免繁琐的参数校核工作,提出了活性污泥2 d号模型(ASM2d)和人工神经网络(ANNs)相结合的复合模拟方法。考察了复合方法在某污水处理厂生物脱氮除磷工艺中的应用情况。研究表明,ANNs能够准确地模拟出水实测值与未经校核的ASM2d机理模型的估计值之间的差值。利用Levenberg-Marquardt算法,对出水氨氮、总氮和总磷分别建立网络结构为5-12-1、5-8-1和5-8-1的ANNs子模型,将这些子模型输出同ASM2d机理模型输出相加便得到复合模型输出。复合模型估计值对前10.4 d(ANNs子模型训练数据时段)出水氨氮、总氮和总磷浓度的拟合平均绝对百分比误差分别为0.267、0.055和0.048;其对后2.6 d(ANNs子模型测试数据时段)出水氨氮、总氮和总磷浓度的预测平均绝对百分比误差分别为0.332、0.083和0.069。均方根误差、平均绝对误差等评价指标也表明复合模型能够给出合理的模拟结果。     

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.     

快速生物活性测定仪的发展

在监测污水处理厂活性污泥系统的状态 ,控制其运行时 ,人们广泛采用了呼吸测量法。根据呼吸测量的原理开发的快速生物活性测定仪 (RBAT)是一种有效的仪器 ,能够可信地评价污泥中微生物的活性 ,它的应用已成为污水处理厂监测、运行的重要发展趋势。     

Mechanism of effective nocardioform foam control measures for non-selector activated sludge systems.

Solids retention time (SRT), biological scum trapping and recycle, and the dynamic equilibrium between Nocardioform populations in the foam and the mixed liquor are the controlling factors in activated sludge foaming events caused by Nocardioform bacteria. For the operating modes described in this paper, a cured mixed liquor foaming condition (filament counts of approximately 10(5) intersections/g volatile suspended solids) was only achieved when SRT control, selective wasting, and polymer addition were in effect. Solids retention time control, with the SRT remaining below 1.5 days, and selective wasting will cure a severely foaming mixed liquor, but effects will only be observed after 3 or 4 months after implementation. The combined wastage of Nocardioform bacteria from selective wasting and SRT control can ensure long-term foam control to the operation of a pure-oxygen activated sludge system with foam-trapping features. An SRT of 0.3 days will result in the complete washout of Nocardioform bacteria from the activated sludge system, which can then operate at an SRT of 3 days free of Nocardioform. Polymer addition to mixed liquor is only effective for foam control when a large portion of the system biomass exists as a heavy layer of foam above the mixed liquor.     

原位臭氧氧化对活性污泥系统硝化和反硝化作用的影响研究

将厌氧序批式间歇反应器(ASBR)和序批式间歇反应器(SBR)串联组成污泥减量新工艺,着重探讨了对SBR段进行原位臭氧投加时,臭氧氧化作用对系统硝化和反硝化能力的影响,并以不投加作为对照。结果表明,将臭氧原位投加到ASBR—SBR组合工艺的SBR段,臭氧投加量为0.027g(以每克MLSS计),每隔3个周期再次投加、连续运行40d,试验组SBR段臭氧投加当期出水COD去除率为86%,比对照组下降了9百分点,但臭氧氧化细胞内大量有机物进入混合液中,为反硝化作用提供了外加碳源,对污泥反硝化能力的提高起到了一定的促进作用;试验组部分硝化细菌由于臭氧的强氧化作用而失去活性,但是随着剩余污泥量的减少,系统的污泥龄延长,有利于硝化细菌的生长,使得系统的硝化能力基本未受影响;试验组臭氧投加当期SBR段出水NO2--N平均浓度比对照组的高18.9%,但经过3个周期的运行后,其SBR段出水NO2--N平均质量浓度降低至7.57mg/L,基本与对照组持平;试验组臭氧投加当期SBR段出水NO3--N的平均浓度高于对照组,但经过3个周期的运行后,试验组出水NO3--N平均浓度低于对照组;试验组臭氧投加当期SBR段出水TN和对照组的出水TN平均去除率分别为65%和75%,但试验组再经过3个周期的运行后,出水TN平均去除率可以达到72%。可见,原位投加臭氧并未对SBR段的硝化和反硝化能力产生明显的影响。     

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.     

Oxidation-reduction potential changes in aeration tanks and microprofiles of activated sludge floc in medium- and low-strength wastewaters.

Real-time control of aeration tank operation is key to high-efficiency pollutant removal and energy savings. One of the aims of this study was to examine the potential for using redox potential (oxidation-reduction potential [ORP]) to indicate wastewater quality online in aeration tanks treating medium (chemical oxygen demand [COD] of 70 to 150 mg/L) and low (COD of 15 to 30 mg/L) pollutant-concentration wastewaters. The field-scale data provide a good relationship between ORP values and nutrient removal along the length of the aeration tanks. The ORP values increased dramatically as organic matter was removed along the aeration tanks, indicating the improvement of the bulk liquor redox status. Dissolved oxygen higher than 1.0 mg/L was necessary for good biodegradation and improvement of the liquid redox status. Nitrification occurred at higher ORP values (380 to 420 mV) than was the case for organic substrate oxidation (250 to 300 mV). The microprofiles obtained from microelectrode measurements substantiate the heterogeneity of the microbial processes inside activated sludge flocs. Because of microbial oxygen utilization, the aerobic region in the activated sludge floc was limited to the top layer (0.1 to 0.2 mm) of the activated sludge aggregate present in medium-strength wastewater, with an anoxic zone dominating inside the flocs. When dissolved oxygen in the bulk water was higher than 4.0 mg/L, the anoxic zone inside the floc disappeared. At low wastewater pollutant concentrations, the ORP and dissolved oxygen inside the activated sludge aggregates were higher than those from medium-strength wastewater. The prospect of using ORP as an online control approach for aeration tank operation and the potential reasons for activated sludge floc size varying with pollutant strengths are also discussed.