Prediction Of Dissolved Oxygen In Water Quality Based On Temporal Decomposition Characteristics

The construction of ecological civilization is an important content of socialism with Chinese characteristics.Among them,river environmental governance is very important to the construction of ecological civilization.Dissolved oxygen is one of the key parameters in river water quality.Its content has a very important impact on the survival of animals and plants in river water quality.Accurately predicting the changes of dissolved oxygen in water and establishing real-time and accurate prediction models are of great significance to river environmental governance.Dissolved oxygen in river water quality has many characteristics such as large time lag,nonlinearity and uncertainty.These characteristics affect the stability of the aquatic system.Therefore,the dissolved oxygen needs to be decomposed to facilitate further observation and analysis of its characteristics.Based on this,this paper proposes a variational modal decomposition(VMD)hybrid prediction model based on the characteristics of dissolved oxygen.First of all,this article needs to preprocess the acquired data to deal with outliers and missing values in the original data.And use the wavelet threshold method to reduce the noise of the data.Next,this article uses VMD to decompose the dissolved oxygen sequence.According to the kurtosis curve,the number of decompositions is determined,and the stationarity of each component is judged.Among them,the non-stationary sequence uses a prediction model that combines principal component analysis and a long short-term memory neural network(LSTM),and uses a cross-validation grid search(Grid Search CV)algorithm to optimize the parameters in the LSTM.Stationary series are predicted using the generalized autoregressive conditional heteroscedasticity model(GARCH).Finally,the prediction results of each component are reconstructed using the cumulative method to obtain the final dissolved oxygen prediction results.The model in this paper is applied to a monitoring point in the Changzhou section of the Beijing-Hangzhou Grand Canal for verification.Experimental results show that Grid Search CV optimizes LSTM parameters better than particle swarm optimization(PSO)optimizes LSTM parameters.And using the manual tuning method to verify that the parameter combination obtained by Grid Search CV optimization LSTM is the optimal parameter combination.The performance of LSTM method for non-stationary sequence prediction is better than many models.GARCH is better than AR and MA models for stationary series prediction.The accuracy of the hybrid forecasting model established in this paper is higher than that of many forecasting models.The root mean square error(RMSE)of the model is 0.2303,the mean absolute error(MAE)is 0.1611,and the mean mean relative error(MAPE)is 4.1713.It shows that the model has high prediction accuracy and good prediction performance.