Monthly-scale runoff forecast model based on PSO-SVR

Abstract

The current methods used in the Lubbog reservoir runoff forecast generally have shortcomings such as low forecast accuracy and low stability. Aiming at these problems, this paper constructs a PSO-SVR mid-and-long term forecast model, and it uses the particle swarm optimization algorithm (PSO) to find the penalty coefficient C, the insensitivity coefficient ε and the gamma parameter of the Gaussian radial basis kernel function of the support vector regression machine (SVR). The results demonstrates that the average relative errors of the PSO-SVR forecast model is relatively small, which are all within a reasonable range; the qualification rates for most monthly forecasts are above 80%. Experimental results indicate that compared with multiple regression analysis, the PSO-SVR model has a higher forecast accuracy, a stronger stability, and a higher credibility. It has a certain practical value and provides a reference for related research.