Application Of Artificial Intelligence Method In Wind Speed Prediction

As a basic element in the atmospheric environment,wind research has important implications for weather,climate science,environmental science,clean energy,and meteorological disasters.However,because of the influence of temperature,air pressure,altitude,terrain,latitude and many other factors,wind is characterized by randomness,intermittency and volatility,making wind one of the most difficult weather forecasting factors.Therefore,the study of wind speed prediction methods plays a key role in improving weather forecasting,research on environmental pollution,development of wind resources,and prevention of wind disasters.Wind speed data has large volatility and randomness.Directly using a single prediction model to make predictions usually results in relatively large errors.At the same time,single model without parameters optimization still has problems in model parameters such as unstable model parameters and low reliability during training.Has become a hot spot in wind speed prediction research.The dissertation firstly proposes a new improved swarm intelligence algorithm(AFSA-ACO)to solve the problem that the convergence speed of fish and ant colony algorithm becomes slow at the later stage of optimization.The optimization performance is verified by various test functions.The improvement of the swarm intelligence algorithm was preliminary discussed.Based on the high noise and nonlinear characteristics of wind speed,based on the methods of EEMD,WNN,and CSO,etc.In accordance with the hybrid idea,a new hybrid prediction model(EEMD-CSO-WNN)is proposed;linear and nonlinear characteristics of wind speed,data preprocessing,multiple linear and nonlinear prediction methods,and improved swarm intelligence algorithm Based on the combination,a new combination forecasting model based on singular value decomposition and improved swarm intelligence method(AFSA-ACO)is proposed.Using the observation data of wind field in Shandong,the prediction and prediction accuracy of the proposed hybrid and combined forecasting model were systematically evaluated and analyzed.The main findings are as follows:(1)The optimization effect of the improved swarm intelligence method(AFSA-ACO)was evaluated.For the Sphere function,the average number of iterations of the optimized AFSA-ACO algorithm and the traditional ACO was 7.4 and 154,respectively,which increased by 95.19%;For the Rastrigin function,the average value of the optimized AFSA-ACO algorithm and the traditional ACO optimization iterations are 148 and 351,respectively,an increase of 57.83%.Therefore,the improved AFSA-ACO algorithm has faster convergence rate and stronger optimization ability than the traditional ACO method.(2)Analyze the prediction effect of empirical model decomposition and wavelet neural network optimization based on cuckoo algorithm optimization.The proposed EEMD-CSO-WNN model is compared with BP,RBF,WNN,CSO-WNN,and EEMD-.For WNN and other forecasting methods,the average percentage error MAPE decreased by 68.89%,69.08%,66.48%,63.32%,and 27.00%,respectively.It shows that the proposed hybrid prediction model has smaller prediction error and higher prediction accuracy than other single prediction methods.(3)Conduct a systematic evaluation of the predictive effects of linear weighted nonlinear methods and improved group weighted intelligence algorithm(AFSA-ACO)-optimized combined weighted forecasting models,and consider that the proposed combined algorithm has a greater improvement in predictive power.Compared with BPNN,ENN,GRNN,WNN,ARIMA and ES prediction methods,the average percentage error MAPE for the proposed combined model increased by 4.83232%,37.4302%,29.3375%,52.5424%,14.6667%,and 56.2927%,respectively.Analysis and research show that the prediction accuracy of the proposed combined forecasting model has been greatly improved,which illustrates the superiority of the combined method.