Medium and long-term runoff prediction model based on multi-factor and multi-model integration

Improving medium and long-term runoff prediction accuracy is vital for optimal water resource operation. Based on the 130 climate factors obtained from the National Climate Center of China, the Pearson′s correlation coefficient, maximum information coefficient, and variance increment index are used to screen the main factors for runoff prediction. Then, a multifactor synthesis method based on the Dempster-Shafer (DS) evidence theory is proposed. The random forest, BP neural network, and Bayesian network are used to establish medium and long-term runoff prediction models using the screened hydrometeorological teleconnection factors. Finally, an integration model for the runoff prediction results is proposed based on the DS evidence theory. Considering the Three Gorges Reservoir as the case study, the results show that the use of hydrometeorological teleconnection factors can effectively improve prediction accuracy. Moreover, the multifactor synthesis method based on the DS evidence theory can screen the factors with better synthesis and stability, thereby mitigating the shortcomings of single-screening methods. The multifactor and multimode integration model based on the DS evidence theory has higher runoff prediction accuracy than the single-screening models, with the certainty coefficient increased to 0.823 and the average relative error reduced to 23.2%.