Prediction of River Water Quality Based on Neural Network Model

Abstract

River water quality prediction is an important way to protect river water resources, the existing water quality prediction models have low prediction accuracy and are difficult to deal with complex water environment changes. In order to predict the variation trend of river pollutant concentration more accurately, based on CNN model and GRU network model, a CNN-GRU hybrid model is established to predict the concentration of river pollutants. Taking the pollutant concentration, river flow and river flow velocity as the input data of CNN model, the feature vector is extracted by CNN, and the high-dimensional vector of time series is constructed. Then, it is input into GRU for model training, and attention mechanism is used to optimize the model. Finally, pollutant concentration prediction is completed. At the same time, GRU model, BP neural network and ARIMA model are used to train and predict the same training set. The experimental results show that the prediction accuracy of CNN-GRU hybrid model is 3.15%, 4.72% and 10.81% higher than GRU model, BP neural network and ARIMA model, respectively.