Wind Power And Ramp Prediction Method Based On Semi-supervised Generative Adversarial Network

In order to actively coping with climate change and strive to achieve the long-term goal of "carbon peak and carbon neutral ity",large-scale wind power integration has gradually become the development trend of the global energy structure.However,the strong randomness and intermittent d efects of wind energy are easy to cause large fluctuations in wind power in a short time.T he resulting wind power ramp events will always affect the balance between the power supply and the load terminal,and brings huge challenge to the safe and reliable operation of the power grid.Therefore,the realization of high accuracy wind power ramp events prediction can minimize the impact of wind power fluctuations on the power grid.It is beneficial to improving the peak regulation and accommodation capability of the grid,and making full use of the advantages of wind energy.The research work of this paper is as follows:A hybrid prediction model based on Semi-supervised learning and Generative adversarial network(GAN)is proposed to solve the short-term prediction problem of wind farm power generation.By variational mode decomposition of initial wind data preprocessing,and then a generative-discriminative adversarial model is designed for the prediction task.The generative model can generate virtual samples to improve the generalization ability of the model.The discriminative model is used to extract nonlinear features in all subsequences.The logistic regression laye r at the end of the discriminative model integrates nonlinear features and gives the prediction value.The simulation and comparative analysis of real wind data verify the effectiveness of the proposed semi-supervised generative adversarial network(SSGAN)prediction method.A prediction model based on multi-step self-tuning rolling window algorithm and multi-label learning is proposed to solve the problem of wind power generation power ramp forecasting.The self-tuning rolling window algorithm is used to optimize the iterative multi-step prediction based on SSGAN,and the iterative cumulative effect of prediction error is eliminated to the maximum extent through multiple“training-testing” processes,so as to improve the accuracy of multi-step prediction of wind power.Then,the ramp classification method based on multi-label learning is established.According to the ramp intensity and ramp direction,wind power ramp events are divided into five categories.Finally,by comparing the ramp prediction results before and after self-tuning,the advancement of the proposed self-tuning ramp prediction method is verified.A prediction method based on SSGAN and Long Short-Term Memory(LSTM)is proposed to solve the problem of wind power ramp prediction considering the spatial-temporal correlation among wind farms.The correlation of time se ries data of multiple wind farms is analyzed.The SSGAN model is utilized to collect spatial features,and the LSTM storage unit is employed to collect temporal features.Then,the spatial-temporal characteristics among multiple wind farms are comprehensively acquired by the SSGAN-LSTM to obtain the implicit mapping relationship and infer the possible ramp events.The effectiveness of the proposed hybrid forecasting method is verified by comparing the results of multi-wind farm prediction and single wind farm prediction.In summary,this paper focuses on prediction of wind power prediction and wind ramp events.The GAN is used as the main body,and combining variational mode decomposition,semi-supervised learning,self-tuning strategy and other algorithms are employed to enhance the generalization ability of GAN so as to improve the accuracy of wind power prediction and ramp prediction.It is beneficial to the reasonable arrangement of wind farm maintenance and shutdown plans,promote the effective utilization of wind power,improve the friendliness between wind power and power grid,and reduce the negative influence of wind power ramp events on power grid reliability,which provides a theoretical basis for dispatching personnel to make corresponding dispatching plans.