Wind Farm Power Prediction Algorithm Based On NWP Data Optimization

At present,the traditional energy-based power generation structure,supplemented by new energy sources,is widely used in China to reduce coal consumption and greenhouse gases and air pollutants produced by primary energy.Effective forecasting of wind power has a better referen-ce value for reducing primary energy consumption.Compared with the traditional single algorithm used for wind power forecasting,the combin-ed model algorithm can reflect the higher superiority,so this paper studies the long-term and short-term forecasting combined model of wind power respectively for overall optimization.The findings are as follows:(1)In order to reduce the error of the initial data,a reasonable predic-tion model is established and the traditional one-to-one input-output stru-cture is improved.Based on the initial data,the NWP model is establish-ed to modify the wind speed,and the forecast model is more accurate after checking the NWP model.(2)The optimized NWP model is used for data input and BP neural net-work algorithm is used for prediction,but the calculation speed of BP ne-ural network is slow and the result is unstable,so the memory unit is add-ed into BP neural network algorithm,forming long-short Time Memory Network(LSTM).Compared with BP neural network,the long-short time memory network is more advantageous,but there are still serious local errors,which need to be further improved.(3)The network model and the mathematical model are improved by combining various meteorological factors,and a new combined model is form-ed by adding the Convolutional Neural Network to the LSTM,which optimizes the wind power data in the reverse propagation process,comp-ared with BP,LSTM,CNN-LSTM,the results show that the accura-cy of CNN-LSTM is obviously better than that of the other two single str-ucture algorithms,it has a good power forecasting property in the local short-term wind power forecasting model.(4)In the study of long-term wind power forecasting model,genetic operators are needed to increase the proportion of the better solution by using FUN function.The mean square error and mean absolute error of GA-CNN-LSTM algorithm are obviously smaller than those of CNN-LSTM in the last example analysis,so it has obvious superiority to add GA operator in the long-term prediction.In this paper,wind power prediction models are established to verify the practicality and feasibility of the two models.Namely,the superiority of CNNLSTM in short-term prediction and the superiority of GA-CNN-LSTM in longterm prediction.It provides some ideas for the effective utilization of energy and wind power in the future research work.