Research On Water Prediction Model Base On Netural Network

Water is one of the essential resources on Earth and has a significant impact on human productive activities.The rapid development of China’s industrial economy has led to frequent water pollution incidents.Water quality prediction,as a fundamental work in protecting water resources,is one of the important methods to solve the current water resources crisis.However,the water quality conditions of rivers are complex and diverse,and water quality prediction involves massive nonlinear water body data processing.Traditional water quality prediction methods cannot capture complex nonlinear relationships when performing nonlinear operations,leading to low prediction accuracy of water quality prediction models.However,the powerful nonlinear modeling ability of neural networks effectively solves this problem,and can improve the accuracy of water quality prediction.Water quality data is a type of data with time series characteristics,and LSTM networks have been widely used in the field of water quality prediction due to their excellent time series modeling ability.Therefore,this paper aims to improve the prediction accuracy and generalization of the LSTM prediction model.Using the historical data of the San Yi Bridge in the Qingbaijiang Basin of Xindu District,Chengdu City,Sichuan Province as the research object,we construct a water quality prediction model based on the LSTM network,with the following specific work:(1)To address the problem of reduced prediction accuracy caused by the increased data load on the LSTM network,a CNN-LSTM hybrid prediction model based on convolutional neural networks is proposed.First,water quality data is input into the convolutional neural network for local feature extraction and data dimension reduction.Then,the water quality data features are input into the LSTM network for learning and training,and the prediction results are output.After experimental verification,the prediction accuracy of CNN-LSTM model is obviously better than that of LSTM model.Taking dissolved oxygen data as an example,the CNN-LSTM model achieves RMSE of 0.02452,MSE of 0.0006,MAPE of 0.00244,and MAE of 0.02152,demonstrating the effectiveness of the model.(2)To address the sensitivity of CNN-LSTM network models to network parameters,a Quantum Particle Swarm Optimization(QPSO)algorithm is proposed to optimize the CNNLSTM water quality prediction model.The QPSO algorithm is used to optimize the convolutional kernels,LSTM neurons,and learning rate in the CNN-LSTM model.By iteratively computing the optimal parameters in the search space,the QPSO-CNN-LSTM water quality prediction model is constructed.Experimental results show that the QPSO-CNN-LSTM model outperforms the PSO-CNN-LSTM model in terms of prediction accuracy.Taking dissolved oxygen data as an example,the model achieves RMSE of 0.01246,MSE of 0.00016,MAPE of 0.00127,and MAE of 0.0111,validating the effectiveness of the model.(3)To address the issue of slow iteration speed and decreased prediction efficiency caused by increased data load in Quantum Particle Swarm Optimization(QPSO),a CNN-LSTM water quality prediction model based on Bayesian Optimization(BO)algorithm is proposed.The BO algorithm is used to train and optimize the convolutional kernels,LSTM neurons,and learning rate in the CNN-LSTM model.The Bayesian algorithm considers historical evaluation information and continuously updates the prior values using Gaussian processes to output the optimal parameters after iterations.Experimental results show that the BO-CNN-LSTM model achieves slightly higher prediction accuracy than the QPSO-CNN-LSTM model.Taking dissolved oxygen data as an example,the model achieves RMSE of 0.00738,MSE of 0.000054,MAPE of 0.0007,and MAE of 0.00616.Moreover,the model demonstrates faster prediction speed,validating its superiority.