Research On Source-Side And Load-Side Scenario Generation Method Considering Data Privacy-Preserving

The rapid development of the global economy has led to an increasing demand for energy,and renewable energy is characterized by abundant resources,low environmental impact and high economic efficiency.In order to achieve the goal of "carbon peaking by 2030 and carbon neutral by 2060",China needs to control the total amount of fossil energy and promote renewable energy replacement actions.At the same time,the voltage stability of the grid is closely related to load changes and the power load is crucial to the grid operation planning.In order to effectively describe the uncertainty of renewable energy or load in multiple regions,collaborative modeling of multi-regional data is required.In collaborative modeling,there may be various reasons such as different parties with different interests,leading to the problem of willingness to share data and bringing the potential of data leakage during data transmission,which is not conducive to modeling using artificial intelligence methods.Therefore,how to generate scenarios from source side and load side separately while considering data privacy protection is the main problem studied in this paper.In this paper,scenario generation is introduced to model the uncertainty characteristics of renewable energy and residential electricity loads,establish source and load uncertainty probability models respectively,and sample to form scenario sets.Firstly,the scenario analysis method is used to study the uncertainty characteristics of renewable energy and residential electricity load.Through source and load continuous seven-day curves and year-round curves,the wind power output scenarios are analyzed to be intermittent,volatile and random,and the residential electricity load is volatile and random.Then,this paper introduces federated learning into multi-regional renewable energy day-ahead scenario generation and proposes a renewable energy scenario generation method based on federated conditional generation adversarial networks.Taking wind power as an example and based on a large amount of actual operational data,the safety and feasibility of the proposed method are demonstrated.When the renewable energy historical data in the target domain is insufficient,the method can also effectively improve the accuracy of scenario generation in the target domain by using the collaborative modeling of multi-area data and provide a reliable method for the planning of new renewable energy areas.Although federated learning is trained only by exchanging the neural network parameters of each participant without directly transmitting the original data,malicious participants can use the transmitted gradient information to invert the original data.Finally,due to the higher privacy requirements of residential electricity load data,this paper proposes a residential electricity load scenario generation method based on improved swarm learning which combines with federated learning and verifies the effectiveness of the proposed method.Compared with swarm learning,the selection of local gradients for up-linking learning can significantly reduce the training time.Compared with federated learning,the risk of leakage of raw data is further reduced.