基于改进粒子群优化算法的污水处理过程模拟及故障诊断新方法研究

With global freshwater resources gradually exhausted and water pollution more serious, wastewater treatment has been paid more attention. Process simulating of wastewater treatment process is significant and valuable for ensuring the normal operation, improving the efficiency, and reducing the energy consumption as it can provide scientific idea and decision for monitoring, controlling and fault diagnosing of wastewater treatment process. Based on previous study, this project will optimize parameters of radial basis function neural network (RBFNN) and support vector machine (SVM) using improved particles swarm optimization (PSO), forming two novel algorithms, PSO-RBFNN and PSO-SVM. The performances of these two algorithms will be compared on simulating wastewater treatment process under steady state or unsteady state, in order to find out a better method for process simulating. Then variable importance and network structure of the built models will be analyzing using the Sensitivity Analysis, the Connection Weights method and the Network Interpretation Diagram method. After the important influent parameters and control parameters influencing the effluent were identified, these could provide scientific basis and technical support for the process control and fault diagnosis of wastewater treatment.

随着全球范围的淡水资源日益枯竭和水体污染问题的日益严峻，污水处理受到当今世界的普遍关注。污水处理过程模拟可为污水处理过程的监控和故障诊断提供科学思维和决策，对于保障污水处理过程的正常运行，提高处理效率，降低能耗具有重要的意义和实际应用价值。本项目在前期研究工作基础上，拟采用改进的粒子群优化算法(PSO)优化径向基函数神经网络(RBFNN)和支持向量机(SVM)的模型参数，并将改进后的PSO-RBFNN和PSO-SVM算法应用于稳态和非稳态污水处理过程模拟，探寻较优的污水处理过程模拟的新方法；应用灵敏度分析、连接权值法和网络图形法可进行变量重要性和网络结构的解释，考察和分析影响出水水质的重要进水参数和过程控制参数，为污水处理的过程控制和故障诊断提供科学依据和技术支持。

污水处理过程模拟可为污水处理过程的监控和故障诊断提供科学思维和决策，对于保障污水处理过程的正常运行，提高处理效率，降低能耗具有重要的意义和实际应用价值。本项目首先研究了典型污水处理工艺对印染废水的处理过程和效果，为过程模拟提供基础数据；其次构建了典型污水处理装置，并对典型污水处理装置内的流场进行了数值模拟，为污水处理过程的模拟和故障诊断提供基础；再次，在MATLAB 7.1平台上，采用非线形惯性权值调整策略、模糊惯性权值策略、随机惯性权值策略等方法对标准的粒子群优化算法(PSO)进行了改进；构建PSO-径向基函数神经网络（RBFNN）算法和PSO-支持向量机（SVM）算法，对典型污水处理工艺和污水处理厂的实际处理过程中的COD、BOD5、氨氮等水质指标进行了模拟，考察了PSO-RBFNN和PSO-SVR模型在适用范围、预测效果、样本要求、结果解释以及模型稳定性等方面的性能，并根据模拟结果对上述算法进行改进和完善，探讨污水处理过程模拟和故障诊断的新方法。