A Distributed Approach for Wind Power Probabilistic Forecasting Considering Spatiotemporal Correlation without Direct Access to Off-site Information

Using off-site predictors to capture spatio-temporal correlations among geographically distributed wind farms is seen as one solution to improve the forecast accuracy of wind power generation. However, in practice, wind farm operators are usually unwilling to share their private data with each other because of competitive reasons and security considerations. To address this issue, this paper presents how wind power probabilistic forecasting using off-site information could be achieved in a privacy-preserving and distributed fashion. Wind power probabilistic forecasts are created by means of multiple quantile regression. The original large-scale forecasting problem is first decomposed into a large number of small-scale subproblems. The subproblem can be computed locally on each farm. Then, the closed-form solution to the subproblem is derived exactly for achieving high computational efficiency. The proposed approach offers a flexible framework for using off-site information, but without having to exchange commercially sensitive data among all participants. It relies on the alternating direction method of multipliers algorithm to achieve the cooperation among all participants and finally converges to the optimal solution. Case studies with real-world data validate improvements in the forecast accuracy when considering spatio-temporal correlations. Distributed approaches also show higher computational efficiency than traditional centralized approaches.