Day-ahead Typical Scenario Generation Method Of Wind And Solar Power Generation And Its Application In Power System Unit Commitment

The grid connection proportion of wind and solar under the carbon neutrality target is rising year by year,but wind and solar has strong volatility and randomness,which brings great challenge to the safe and stable operation of the power system.How to accurately describe its uncertainty is the key problem that needs to be solved in a high percentage of new energy power system.Therefore.a wind and solar power generation day-ahead typical dcenario generation method based on deep learning is proposed in this paper,which can simultaneously take into account the computational efficiency and fitting accuracy,and apply the typical scenarios to the combination of power system unit commitment to verify the effectiveness.The main work carried out and the results obtained are as follows:(1)Aiming at the problem that the existing methods do not take into account the autocorrelation of wind and solar power generation sequences at different time scales,resulting in large differences between the generated data and the actual data,a wind and solar power generation day-ahead scenario generation model based on multitimescale coupling and Generative Adversarial Network is proposed.By building multi-scale convolutional layers in the generators and discriminators of the Generative Adversarial Network,this structure achieves comprehensive extraction of temporal feature information and distribution information of wind and solar power and then generates the initial scenario set accurately and efficiently.The results of cases show that,compared to traditional models,for wind farms,CRPS values are reduced by 0.89%-12.70%and Wasserstein distance values are reduced by 2.92%-9.62%;for solar plants,CRPS values are reduced by 2.68%-9.05%and Wasserstein distance values are reduced by 2.05%-15.16%.(2)Aiming at the problem that the existing methods only rely on the generated data information for scenario reduction and ignore the actual data,this paper proposes a wind and solar power generation day-ahead scenario reduction model considering the actual data information.By adding another loss function on the original AutoEncoder and combining the kmedoids algorithm to build an improved Auto-Encoder for scenario reduction,the proposed model can achieve effective reduction of scenario sets and reduce the error caused by the data generation model.The results of cases show that,compared to traditional models,when the number of typical scenarios is 10,for wind farms,CRPS values are decreased by 2.71%-23.14%and Wasserstein distance values are decreased by 8.52%-42.86%;for solar plants,CRPS values are reduced by 2.53%-68.70%and Wasserstein distance values are reduced by 10.84%-67.42%.(3)With the optimization objective of minimizing the total operating cost of the power system,a day-ahead unit commitment model is constructed based on wind and solar power generation typical scenario set and the effectiveness of the proposed day-ahead typical scenario generation method is further verified.The cases applies the wind and solar actual power,the deterministic power prediction results and the typical scenario set to the power system day-ahead unit commitment model respectively,The results show that:compared to the the total operating costs based on the deterministic prediction results,the total operating costs based on the typical set scenarios is closer to reality;when the number of typical scenarios is 10,the operating cost error can be reduced by 2.25%-2.71%based on actual operating cost.