Research on multi-party joint electric data value evaluation method based on optimized decision-making

With the increasing intelligence and flexibility of the power system, various operators have joined the models of the power market and grid dispatch to predict future electricity demand or power generation, providing data support for safe and stable operation of the power system and load demand prediction. However, due to the complexity and uncertainty of the power system, as well as the heterogeneity and distribution of operators, there are significant errors in load forecasting results. As for to improving the performance of federated learning models, it is an important and difficult problem to evaluate the value of multi-party electricity data. In this paper, we propose a method based on gradient ascent model unlearning to evaluate datasets in probabilistic load forecasting scenarios for various flexible resource operating entities in the power grid. By checking new model performance after removing each client, we evaluate the contribution of each client dataset to the global model. It can effectively measure the impact of each client dataset on the global model while protecting client privacy. We conducted experiments on different datasets to verify the effectiveness and feasibility of our method. Our experimental results reflect the adaptability and sensitivity of our method in four electric datasets.