Application Research Of LSTM Network Optimized By LightGBM And DNA-SFLA Algorithm And CFD In MBR

In the process of MBR sewage treatment,membrane pollution will lead to high energy cost and membrane replacement cost,which is the most important factor affecting the efficiency of MBR system.Sludge production is one of the important parameters for measuring membrane fouling.Intelligent simulation and calculation of MBR sludge yield can not only predict the degree of membrane contamination,but also detect the condition of membrane components and get timely replacement.In this paper,the LSTM-MBR prediction model was established based on the LSTM network to calculate the sludge yield,and the following three improved models were proposed:1.The LightGBM MBR membrane pollution prediction model is proposed.Generally,the range of experimental data collected is wide,and the imbalance of data will lead to the decrease of the accuracy of the model.In this paper,LightGBM algorithm was used to first divide the training data into two categories,and then trained the data set generated by classification to obtain the LSTM prediction model with two different weight parameters.Finally,the test data are calculated by the generated prediction model.The prediction results show that the average relative error of the optimization method is 0.05364 lower than that of the LSTM prediction model,and the calculation time of the model is also significantly reduced.2.The DNA-SFLA MBR membrane pollution prediction model is proposed.The initial weight and threshold of the neural network will affect the training effect of the model,so the DNA-SFLA algorithm is introduced to optimize the LSTM network.The DNA-SFLA algorithm has strong global search ability,which can quickly find the global optimal solution and obtain the optimal initial weight and threshold.The prediction results showed that compared with the LSTM prediction model,the optimization method reduced the average relative error by 0.06063,and the calculation accuracy of the model was greatly improved.3.The LightGBM and DNA-SFLA MBR membrane pollution prediction model is proposed.At the same time,two algorithms are used for optimizing the LSTM-MBR prediction model.On the one hand,LightGBM algorithm is used to classify the data.LSTM network is used to train data sets for generating the corresponding LSTM model,which can improve the training speed of the model.On the other hand,the LSTM network is optimized.The initial weights and thresholds of neural networks were optimized by using DNA genetic leapfrog algorithm.The model prediction results showed that the LSTM network optimized by the two algorithms has the highest accuracy and the fastest convergence speed.In this paper,the internal pressure parallel MBR with hollow fiber as membrane module was established by Fluent Ansys software to realize solid-liquid separation and calculate the water yield of the system.Firstly,the forward processor,solver and back processor in CFD were used to establish the geometric model,and then the model was used to simulate the sewage to achieve solid-liquid separation and obtain the water yield.The data obtained from the internal pressure parallel MBR membrane module model established by CFD is basically consistent with the actual data,indicating that the calculated results of the prediction model are relatively accurate and can be applied to the actual production of sewage treatment.This can save a lot of engineering design and implementation cost of MBR and has certain reference value to the research of MBR field.