Simplified Design Of LSTM Neural Network Structure And Its Application In Effluent BOD Prediction

In recent years,the research of long short term memory(LSTM)neural network has developed rapidly and made great progress.LSTM neural network introduces memory cells to track the historical information and three gate architectures to control the information flow.It can capture the short-term / long-term correlation of sequence data and has been widely used in the field of time series prediction.However,due to its complex structure,the LSTM neural network has many problems,such as large amount of calculation and storage,which affects the performance of the model to a certain extent.Therefore,how to design a simplified LSTM neural network structure is an urgent problem to be solved.This thesis focuses on the simplification of LSTM neural network structure.The main research work is as follows:(1)Design of a simplified LSTM neural network.Aiming at the problems of complex internal structure,time-consuming and high complexity of the standard LSTM neural network,a method for the structural design of a simplified long short term memory(SLSTM)is proposed.Firstly,the gate architectures of standard LSTM neural network is simplified by coupling input gate and forget gate,with its structure composed of one memory cell and two gate architectures.Secondly,the input weight matrix and bias matrix are removed from the gate control equations to simplify the parameters.Finally,through the experimental verification of UCI benchmark data sets,the results show that the designed model can shorten the training time and reduce the computational complexity of LSTM neural network without significantly reducing the prediction accuracy.(2)Design of a pruning SLSTM neural network based on PLS pruning algorithm.Aiming at the problem that the structure of SLSTM neural network is redundant and difficult to determine,a pruning SLSTM neural network based on PLS pruning algorithm(PLS-PSLSTM)is proposed.Firstly,the partial least squares(PLS)regression coefficient is used as the metric to evaluate the importance of LSTM blocks.Secondly,the redundant hidden layer of LSTM is pruned by merging the unimportant blocks with their most relevant blocks,and the backpropagation through time(BPTT)algorithm is utilized as the learning algorithm to update the network parameters.Finally,the UCI benchmark data sets are used to evaluate the performance of the proposed PLSPSLSTM.The experimental results show that PLS-PSLSTM neural network can achieve the trade-off between good generalization ability and a compact network structure.With the compact network structure,the calculation efficiency is improved,but the prediction accuracy is not reduced.(3)Design of an effluent BOD prediction model based on PLS-PSLSTM neural network.Aiming at the problem that it is difficult to accurately predict the effluent biochemical oxygen demand(BOD),an effluent BOD prediction model based on PLSPSLSTM neural network is proposed.Firstly,the feature selection method based on mutual information is used to determine the input variables.Then,the effluent BOD prediction model based on PLS-PSLSTM neural network is established.Finally,it is applied to the prediction of effluent BOD in the actual wastewater treatment process.The experimental results show that the effluent BOD prediction model based on PLSPSLSTM neural network can effectively predict the effluent BOD concentration in the future.(4)Development of the effluent BOD prediction software in wastewater treatment process.Aiming at the problems that the traditional effluent BOD detection method is difficult to accurately predict the effluent BOD concentration in wastewater and the results can not be visualized,an effluent BOD concentration prediction software is designed and developed.Based on MATLAB GUI function,the software is of low cost,realizing the rapid and accurate prediction of in effluent BOD concentration and the visualization of the prediction results.This provides a reference for scientific decisionmaking of relevant departments,and has good guiding significance for wastewater treatment process.