Deep Relevance Perception Network Based on Multi-task and Multi-scale for Power Generation Prediction Model of Distribution Network Source

As the global energy and environmental crises become more and more serious, wind power, as a large- scale and easy-to-use clean energy source, plays a more important role in power generation. For wind power, one of the multiple sources for the grid, it is difficult to accurately predict the power generation. To solve the problem, we propose a multi-task and multi-scale framework, DRPN, for wind power generation trend forecasting. It has three modules: a multivariate factor fusion for influencing wind power generation, a multi-scale GRU-based time-series feature extraction for wind power generation, and a multi-task update strategy with local and global forecasts. The proposed framework is evaluated on a public wind power prediction dataset, SDWPF, showing that the accuracy is improved compared with baselines. Also, our framework effectively optimizes the source-network cooperative operation of an active distribution network.