Water quality prediction based on IGRA-ISSA-LSTM model

It is essential to make an accurate prediction of the concentration of dissolved oxygen (DO), hydrogen ion concentration (pH), and potassium permanganate (KMnO4) in order to ensure the quality of the drinking water. The lack of monitoring data and the large fluctuation increase the difficulty of predicting DO, pH, and KMnO4 in the tide-sensing estuary. In this research, improved grey association analysis (IGRA) was provided to determine the correlation between DO, pH, KMnO4, and other water quality indicators, thereby resolving the dimension disaster problem of long short-term memory (LSTM). Furthermore, LSTM based on the improved sparrow search algorithm (ISSA) was established, and five LSTM parameters—learning rate, batch size, training times, hidden layer nodes, and fully connected hidden layer nodes—are automatically optimized, which could accurately predict the concentration of DO, pH, and KMnO4. Using the data from the Qiantang River Gate Observation Station from November 8, 2020, to June 27, 2021, 70% of which were training sets and 30% of which were test sets, predicted data for day 4. The results show that the coefficient of determination (R) of the IGRA-ISSA-LSTM model for DO, pH, and KMnO4 were 0.92, 0.93, and 0.726, respectively, which are greater than that of IGRA-BP model, IGRA-LSTM model, and IGRA-SSA-LSTM model. Therefore, this research provides technical support for water quality management in tidal estuaries.