A Water Quality Prediction Model Research Based On WT-CNN-LSTM-HHO

Accurate water quality prediction helps to grasp the law of water quality change,which is of great significance to water resources protection.How to establish a water quality prediction model with higher prediction accuracy and stronger generalization ability is an important topic of current research.River water quality data is time series data that changes nonlinearly with time.Most of the existing studies ignore the complexity of water quality data,only using a single model to predict water quality,and the constructed model prediction accuracy is low.In this paper,a WT-CNN-LSTM combination model was proposed,in which the spatial relationship of water quality monitoring points was considered,and the parameters were optimized by Harris Hawk optimization algorithm.the main contents are as follows:(1)Wavelet transform was used to reduce the noise of the data.Different wavelet basis functions were used to decompose and reconstruct the data respectively,and suitable wavelet basis functions were selected according to the mean square error of the data before and after noise reduction.Then,the optimal wavelet transform was used to reduce the noise in the time series data.(2)CNN-LSTM water quality prediction model was established.CNN-LSTM model,WT-CNN-LSTM prediction model based on single monitoring point water quality data and WT-CNN-LSTM prediction model based on multi-monitoring point water quality data were established by using the pre-processed data.After adjusting the parameters many times,the parameters of the model were determined.The experimental results show that the prediction effect of WT-CNN-LSTM prediction model based on multi-monitoring point water quality data was the best among the three models,with RMSE of 1.4689,MAE of 1.1361,and MAPE of 0.2463.(3)Optimization algorithms were used to adjust WT-CNN-LSTM model parameters.Considering the strong subjectivity of artificial parameter adjustment,it is easy to fall into local optimum.In order to improve the efficiency of neural network parameter adjustment,Hyperband,particle swarm optimization algorithm and Harris Hawk optimization algorithm were used to optimize the historical sequence length,convolution kernel number,LSTM layer neuron number,hidden layer neuron number and learning rate in the WT-CNN-LSTM prediction model based on multi-monitoring point water quality data.Among them,the running time of Hyperband was shorter,but the optimization effect was poorer.Under the same setting of the number of particles and iterations,the Harris Hawk optimization algorithm outperformed the particle swarm optimization algorithm,with RMSE of 1.3807,MAE of 1.0300,and MAPE of 0.2397.Compared with the particle swarm optimization algorithm,the RMSE and MAE were respectively reduced by 3.49%and 6.36%.Compared with the WT-CNN-LSTM prediction model based on multi-monitoring point water quality data,the RMSE of WT-CNN-LSTM-HHO was reduced by 5.48%,the MAE was reduced by 9.65%,and the MAPE was reduced by4.17%.The experimental results showed that the Harris Hawk optimization algorithm could effectively improve the prediction accuracy of the neural network,further illustrating the effectiveness of the WT-CNN-LSTM-HHO model proposed in this paper.