基于自组织模块化神经网络的污水处理过程出水参数预测

Prediction of effluent parameters in wastewater treatment process using self-organizing modular neural network

针对城市污水处理过程关键出水水质一些参数难以在线测量的问题,提出了一种基于经验模态分解(EMD)的自组织模块化神经网络(MNN)出水参数软测量模型。首先设计一种基于EMD的任务分解方法,将复杂的时间序列分解为若干子序列,并采用样本熵和欧氏距离分别计算子序列的复杂性及相似性,自适应调整子网络模块。然后针对子网络模块初始结构难以确定的问题提出一种前馈神经网络的结构自组织算法,实现子网络模型根据分配的子任务动态调整自身网络结构,更有效地对各子序列进行预测。最后通过基准时间序列预测和实际污水处理厂中出水水质参数检测实验验证了所提出的模型具有较好的预测精度和自适应性。

It is well known that some key effluent quality parameters are difficult to measure online in the urban sewage treatment.To solve this problem,this paper proposes a new soft-measurement model using empirical mode decomposition and modular neural network(EMD-SMNN)for effluent quality parameters.First,a task decomposition algorithm based on EMD is proposed,which can decompose a complex,multi-frequency time series of effluent quality parameters into several sub-time series,and it can adaptively adjust subnetwork modules according to the complexity and similarity of sub-time series calculating by the sample entropy and Euclidean distance.Then,a novel self-organizing algorithm of FNN is proposed to solve the problem that the initiating structure of subnetwork is difficult to given,which can dynamically adjust the structure of subnetworks and predict subtasks effectively.Finally,through the benchmark time series prediction and the actual effluent water quality parameter detection in the sewage treatment plant,it is verified that the proposed EMD-SMNN has a good prediction accuracy and self-adaptability.