A seasonal ARIMA model based on the gravitational search algorithm (GSA) for runoff prediction

• Published in: Water science & technology. Water supply Vol. 22; no. 8; pp. 6959 - 6977
• Main Authors: Zhang, Xianqi, Wu, Xilong, Zhu, Guoyu, Lu, Xiaobin, Wang, Kai
• Format: Journal Article
• Language: English
• Published: London IWA Publishing 01.08.2022
• Subjects: Accuracy; Algorithms; arima model; Autoregressive models; Correlation coefficient; Correlation coefficients; Flood forecasting; Flood irrigation; Flood predictions; gravitational search algorithm; Gravity; History; Hydroelectric power; Hydroelectric power generation; Hydrologic data; Hydrology; Hypotheses; Mutation; Optimization; Prediction models; Rain; River discharge; River flow; Rivers; Runoff; runoff prediction; Search algorithms; seasonal differencing; Seasonal variation; Seasonal variations; Time series; Trends; wei river; United States > US; China
• ISSN: 1606-9749; 1607-0798
• DOI: 10.2166/ws.2022.263

Abstract The prediction of river runoff is crucial for flood forecasting, agricultural irrigation and hydroelectric power generation. A coupled runoff prediction model based on the Gravitational Search Algorithm (GSA) and the Seasonal Autoregressive Integrated Moving Average (SARIMA) model is proposed to address the non-linear and seasonal features of runoff data. The GSA has a significant local optimisation capability, while the SARIMA model allows for real-time adjustment of the model using historical data and is suitable for analysing time series with seasonal variations. Consequently, the GSA-SARIMA model was developed and applied to the runoff prediction of the Xianyang section of the Wei River. The results suggest that the GSA-SARIMA model achieves a linear correlation coefficient of 0.9351, a Nash efficiency coefficient of 0.91, a mean relative error of 6.57 and a root mean square error of 0.21. All of the evaluation indicators of this model outperform the other models developed, and its application to actual runoff prediction is feasible, which creates a new path for runoff prediction. HIGHLIGHTS The Mann-Kendall trend test is applied to ascertain the separation point between the training and prediction datasets. It avoids too little data in the test set, while effectively improving the generalisation of the model. The SARIMA model is an improvement on the ARIMA model and allows for convenient real-time adjustment of the model. The GSA algorithm is applicable to parameter search optimization of the model and has great global search capability.