The dynamic urban river water quality prediction based on hybrid model

The choice of prediction model plays a decisive role in prediction performance, and hybrid models contribute to enhancing both model performance and prediction accuracy. This study aims to improve the predictive performance of models in time series data processing and further explore urban river water quality prediction models. The hybrid ARIMA-LSTM model was proposed the to simulate and predict water quality indicators in urban rivers of Zhongshan City. The hybrid model firstly used the linear prediction capability of the ARIMA model to extract linear components from water quality data, thereby generating residuals as inputs for subsequent steps. Then the nonlinear prediction of the LSTM model used the results from the previous step as the input to obtain nonlinear data. Finally, the prediction results of the ARIMA model and the LSTM network were integrated to obtain fitting evaluation results, thereby improving the accuracy of prediction outcomes. Furthermore, three traditional individual prediction methods (CNN, LSTM, ARIMA) along with hybrid ARIMA-LSTM model were conducted and compared to evaluate the performance of the hybrid model. The results showed that the hybrid prediction model was superior to individual prediction methods in terms of accuracy. By integrating LSTM's ability to capture complex nonlinear temporal features with ARIMA's proficiency in modeling linear trends and seasonal components, the hybrid model exhibited superior comprehensive performance in characterizing dynamic water quality variation patterns. The study proved that using the hybrid model could effectively improve the accuracy of short-term urban water quality prediction, and that the prediction accuracy of different pollutants varied under the same prediction method.