An improved particle swarm optimization and its application in long-term streamflow forecast

An improved PSO based on the Metropolis criterion, called MPSO, is proposed and applied in long-term streamflow forecast. MPSO focuses on the inertia weight of PSO, and leads the inertia weight to adjust in accordance to the direction of global best value based on Metropolis sampler. In the case study, MPSO is employed to estimate the coefficients of multivariable linear regressive model (MLRM). MLRM is widely used in engineering project for its simple expression. It was built here to give annual streamflow of Fengtan reservoir, but the inadequate historical records got bad forecasting results. MPSO is used based on MLRM and the objective function changes to total relative errors, not least square method. The final results show that, MPSO can deal with the problem well, and compared with other methods like PSO and LPSO (inertia weight of PSO is calculated by linear decreasing), it has satisfying results and least relative errors.