Research On Short-term Wind Speed Prediction Of Wind Farm Based On Model Optimization

At present,the global environment is gradually warming.One of the reasons for this phenomenon is the increase of carbon dioxide emissions.Economic development potentially affects the environment in some ways,energy consumption is often positively correlated with economic development.Conventional coal power generation is the main factor in the production of carbon dioxide.With the rapid development of science and technology,many new energy power generation industries are in full swing,such as photovoltaic,wind,tidal,hydroelectric and other power generation methods,which have made significant contributions to improving the environment.Wind energy occupies an important position in the new energy pattern,and accurate prediction of wind speed is an important basis for predicting wind energy.With the continuous development of deep learning and artificial intelligence technology,these technologies are gradually applied to wind speed and wind power prediction,and have made great achievements.Therefore,accurate prediction of wind speed is conducive to the operation of wind power Grid-connected dispatching.Based on this,it is particularly important to put forward an effective,reliable and accurate prediction method,which has high economic benefits.In the current research technology,the medium and long-term prediction has achieved good results,but due to the uncertainty and discontinuity of wind power output in the short or ultra-short term,wind speed prediction has become an urgent problem to be solved.Based on this background,this paper proposes several hybrid optimization schemes for wind speed prediction,which generally include the selection of Elman network nodes,the selection of intelligent optimization algorithm,the initialization of optimization algorithm and the effective combination of algorithm and network.Group optimization involves particle swarm optimization,ant colony optimization,genetic algorithm.And the wind speed forecasting analysis of the existing forecasting model schemes is also carried out.And then three hybrid optimization schemes,PSOElman,ACO-Elman and GA-Elman,are proposed.After the number of nodes of the network is determined by experimental simulation,the model is analyzed in many aspects,and the statistical indicators before and after optimization are compared.In this paper,we focus on several key problems of network node selection and model optimization in short-term wind speed prediction,and do the following research:(1)For the selection of the optimal model,the model with better prediction performance is determined by comparing its statistical indicators through simulation experiments,which is the object of optimization.(2)For the selection of input nodes,various statistical indicators of the model under different input nodes are analyzed,and the optimal network is determined when the input node is 5.(3)For multi-step prediction analysis,the single-step,two-step and three-step prediction are emphatically analyzed.In one-step prediction,RMSE optimization is 0.7939,0.6225,0.5611 and 0.5087,respectively.Before and after MAE optimization were 0.6436,0.5175,0.4943 and 0.4236,respectively.Before and after the optimization of MAPE were 0.0553,0.0364,0.0478 and 0.0428,respectively.In general,the optimized model is better than that before optimization,and for the same model,the performance of one-step prediction index is better than that of multi-step prediction.(4)For the analysis of different data sampling periods,the prediction performance of different sampling periods every 15 minutes,30 minutes and 45 minutes is studied.RMSE was 0.9241,0.8970,0.7597 and 0.7682 in the 15-minute sampling interval prediction.MAE were 0.7875,0.7536,0.6786 and 0.7450,respectively.MAPE were 0.0621,0.0588,0.0554 and 0.0543,respectively.For the hybrid optimization algorithm model mentioned in this paper,the performance is better than that of the unoptimized model,and the three optimized algorithms have certain advantages in different statistical indicators.