Research On Design Of Self-organizing Modular Neural Network And Its Applications On Wastewater Detection

Human brain is a complex biological tissue which composes of hundreds of different types of cells.It has extraordinary information processing and decision making ability.It can effectively deal with complex information,such as image,sound,semantics and so on.In the process of information processing,each neuron cooperates and coordinates with each other to form a number of neural network modules,but the modules are all independent and collaborate with each other,making the nervous system in different cerebral cortex regions obtain specific functions and performing the independent task in the region.The modular neural network is based on the characteristics of human brain structure and function.From the point of intelligent information processing of artificial neural network,it adopts the idea of "divide and conquer",and decomposes complex practical problems into several independent but interrelated sub problems,and establishes a neural network for each sub problem.The neural network model should be divided into blocks for dealing with different problems.Although the modular neural network can appropriately simulate the human cerebral cortex information processing mode,but there are still many difficulties in the structure design and algorithm learning process,such as the construction of adaptive neural network structure,the design of a stable sub network,and the dynamic integration of sub networks.According to the information processing and transmission mode of human brain,the dissertation focus on the issues of self-organizing modular neural network architecture design,including sub-networks structure design,parameter learning algorithm and dynamic integration of the sub-networks,which based on the achievements of neurophysiology and neuropsychology.This paper analyses the performance of subnet in the modular neural network and discusses the characteristics of dynamic modular neural network.Therefore,the major contributions of this dissertation are specifically stated as follows:1.According to the information transfer pattern of cerebral cortex and the model of biological neuron,this paper proposes a self-organizing recurrent radial basis function(SR-RBF)neural network based on the spiking mechanism.The hidden neuron in the recurrent radial basis function can be added or pruned by computing the spiking strength of the connections between hidden and output neurons of recurrent RBF neural network.Meanwhile,to ensure the accuracy of SR-RBF neural network,the parameters are trained by improved LM algorithm.The SR-RBF neural network is used for approximating the time-series prediction and classical non-linear functions.Finally,comparisons with other methods demonstrate that the SR-RBF neural network is more effective in terms of accuracy,and network structure.2.In order to solve the problem that modular neural network sub-network output can not be integrated optimally,this paper proposes an adaptive modular neural network based on adaptive particle swarm optimization algorithm.Firstly,the sample distribution can be identified and the data center can be updated by computing the data density.Secondly,the corresponding sub-network is activated according to the input data,then the best output weights is computed via APSO algorithm.Finally,this peoposed method integrates the output of modular neural network.Based on the experiments of approximating the time-series prediction system and classical non-linear functions,it is proved that the number of subnetworks can be dynamically adjusted,and the integrated weight of neural network can be optimized by APSO algorithem.Comparisons with other algorithms demonstrate that the APSO-DMNN is more effective in terms of accuracy and adaptive ability.3.Wastewater treatment process(WWTP)is a typical complex dynamic industrial system.This process includes microbial biochemical reaction,the influent flow and influent components,which make the WWTP are strong nonlinear,large time-varying and strong coupling characteristics.Based on the RBF neural network and modural neural network,this paper adapts the method to measure the effluent parameter concentration in a real wastewater treatment process.Simulation results show that this method can effectively realize the online prediction of ammonia nitrogen concentration,improve the prediction accuracy and adaptive ability.4.The characteristic analysis of biochemical reaction process is the basis for studying the process control of wastewater treatment process.The analysis of its mechanism and characteristics can provide a reasonable basis for the application of intelligent optimization algorithm.Based on energy saving and emission reduction,this paper analyzes the wastewater treatment biochemical reaction of primary treatment process and the two layer pool physical settlement model,then establishes a simulation platform of sewage treatment and configuration system based on an intelligent algorithm.Besides,this system eliminates the abnormal data,and transfers the data to the intelligent algorithm platform.Thus,this system can predict the key parameters of WWTP effectively,and prevent the occurrence of pollutants exceeding the standard.