Research On Effluent BOD Prediction Based On Modular Neural Network

Biochemical Oxygen Demand(BOD)is an important parameter to evaluate whether sewage is purified to meet the standard or not.Nevertheless,the conventional detection method cannot meet the engineering needs due to the long reaction time.Then it’s effective to solve the above problem by using a fully interconnected neural network to predict effluent BOD,Whereas the prediction accuracy of effluent BOD by fully interconnected neural network is affected by the large number of parameters and high non-linearity in the process of wastewater treatment.In order to make the result of the prediction of effluent BOD more precise,this paper proposes a modular neural network(MNN)algorithm for effluent BOD prediction with greater accuracy.The main content of the article is divided into the following points:(1)Research of data pretreatment towards wastewater treatment technology.In order to improve quality of modeling data,a systematic three-step preprocessing method is adopted through analyzing the process and mechanism of the sewage treatment system: method to remove abnormal data points;wavelet used to denoise data;data normalization used to eliminate dimensional effects.As the specific data preprocessing system is completed,the basic conditions for the subsequent modeling are created.(2)Research of modular neural network design.Firstly,according to the mutual information and expertise,the concept of "modularization" is introduced,and the numbers of modules are determined by the correlation of the data.Then,the radial basis function(RBF)neural network as well as the long and short memory(LSTM)neural network are taken as subnetworks through the process of analyzing the characteristics of each module.Then,error feedback and sensitivity analysis are used to improve the RBF subnetwork which can adjust the(AD)network structure dynamically.Modal decomposition and verification(CEEMD)method are used to reduce the complexity of data in corresponding modules of LSTM subnetwork so as to improve the accuracy of LSTM subnetwork,and particle swarm optimization(PSO)algorithm is used to optimize the subnetwork.Finally,combining both the time and space complexity of the algorithm,the competitive output evaluation index is designed to calculate the output priority of each sub-network.Therefore,the modular system can figure out the output of the sub-network with the highest priority as the output of the whole system.Simulation results show that the accuracy of the improved subnetwork is promoted by nearly 20%.(3)Research of effluent BOD prediction based on MNN neural network.The prediction model of effluent BOD is established with the actual sewage treatment data,and the sewage data is pretreated by the three-step method,then the modular neural network is used for training and prediction.Experimental results show that the comprehensive prediction performance of the proposed algorithm is significantly superior to other advanced neural network prediction models,and the prediction accuracy is improved as well.(4)Development of effluent BOD cyber-physical prediction systems.According to the actual operation condition of wastewater treatment plant,based on the concept and the framework of the Cyber-Physical Systems,the mobile terminal with the Web application is designed,including the forecast module,outlet monitoring module,inlet monitoring module,operation module,etc.Consequently,it’s the realization of the real-time prediction of effluent BOD and sewage treatment system of digital management.