Modified Transformer Model-based Short-term Power Forecasting of Wind Generations
In view of the low accuracy of power forecasting of wind generations,this paper establishes a denoising Trans-former model based on multiple features.First a Shapley Additive exPlanations(SHAP)model is proposed,which ex-plains the forecasting results of machine learning models by calculating the contributions of features,so as to describe the contributions of various features to the model output.Second the structure and forecasting principle of Transformer model are analyzed,and modifications are made to address the defect of large amount of noise in the original data by adding a de-noising part to the conventional Transformer model,thereby establishing the modified forecasting model.Using the noise learning process to obtain effective data,the robustness of the forecasting model is improved.Finally the measured data of a wind farm in Jiangsu province are introduced to compare the performances of various models,and the effectiveness of the proposed method is verified.
wind power generationpower forecastingSHAP analysisdenoising Transformer model
NETL
NSTL
万方数据
