Research On Particle Swarm Optimization Extreme Learning Machine And Its Application In Precipitation Forecast

With the development of human society,the destruction of the environment by humans is becoming more and more serious.Natural disasters occur frequently.If some climate-related disasters can be predicted in advance,however,the casualties and economic losses can be reduced significantly.In particular,precipitation prediction can provide information for the occurrence of flood disasters,reduce unnecessary losses,and at the same time provide guidances for related departments of agriculture.Nevertheless,the precipitation is affected by many factors,such as air pressure,temperature,humidity,etc.,which make it very difficult to predict.Therefore,it is of great significance for precipitation prediction.This topic uses an optimized extreme learning machine(ELM)to predict the precipitation.The main research contents are as follows:Traditional extreme learning machine algorithm has the problem of being sensitive to parameter selection.This thesis uses three optimization algorithms: particle swarm optimization(PSO),genetic algorithm(GA),and simulated annealing(SA)algorithm to extreme learning machine separately for the initial weights and offsets optimization.Overcoming the sensitivity to initial parameter selection problem can speed up the convergence of the extreme learning machine and improves its prediction accuracy.Next,the precipitation data are preprocessed,mainly including regression analysis,correlation analysis.Then,ELM,PSO-ELM,GA-ELM,and SA-ELM are used to constructed precipitation prediction models.Finally,experimental results show that the error of PSO-ELM is the smallest.Therefore,the precipitation forecast proposed in this thesis is designed based on the PSO-ELM model.Compared with the traditional ELM model,the PSO-ELM prediction model proposed in this thesis can reduce the mean square error of 50%.The optimized PSO-ELM model has feasibility in the prediction of maximum daily precipitation.The optimized combined model PSO-ELM proposed in this thesis provides not only an efficient method for the prediction of maximum daily precipitation,but also a new way of thinking for similar research.