Wind Power Climbing Recognition And Short-term Power Forecasting Technology Based On Deep Learning

Wind energy is affected by many factors,with the characteristics of randomness and volatility.The research of wind power prediction technology and the prediction of wind turbine output have a vital impact on the flexible dispatch of power grid and the improvement of wind energy consumption capacity.Therefore,this paper studies the technology of wind power ramp recognition and short-term power prediction based on deep learning and transfer learning is studied in the paper.And abnormal data recognition,wind power ramp recognition based on bat algorithm and improved revolving door algorithm,wind farm similar scene matching and short-term wind power prediction based on similar scene matching are included in the paper.The abnormal data recognition of wind power was proposed.The characteristics of wind power output were analyzed from three aspects of wind speed power,time and space;the abnormal data of wind power were classified,and the identification criteria of different types of abnormal data were given;based on the data of Jilin,Ningxia and Heilongjiang wind farms,the abnormal data were identified and corrected,which provides effective data support for the subsequent part of this paper.A method of wind power climbing recognition based on bat algorithm(BA)and improved revolving door algorithm(SDT)was proposed.In this paper,Bat algorithm was used to optimize the tolerance coefficient ΔE of SDT algorithm,and a wind power climbing recognition model based on ba-sdt was constructed.The optimal tolerance coefficient ΔE was determined by iterative calculation.The effectiveness of ba-sdt model was verified by comparing with standard SDT algorithm through example analysis.The similar scene matching of wind farm was studied.Based on the recognition results of wind power climbing events,the optimal search method and pattern matching criteria for mining the spatio-temporal correlation characteristics of Meteorology / power were studied,and the spatio-temporal high-dimensional large-scale meteorological forecast data and largescale new energy output data were analyzed by association,clustering and regression.The weather / power similar scene matching results were obtained,which is the basis for subsequent migration learning based on similar scenes.A transfer learning short-term wind power prediction model based on similar scene matching was proposed.Firstly,based on the source wind farm data of similar scenes,the multisource data migration method was studied,and the multi-source migration model was established;based on the multi-source migration learning model,the model integration method was studied,and the short-term wind power prediction model of multi-source integrated migration learning based on similar scene matching was established,so as to solve the problem of wind farm power prediction modeling difficulty due to the lack of data Question.