Optimizing Partner Selection for Cooperative Solar Generation Forecasting

In our previous work we introduced the ANN-Based Large-Scale Cooperative Solar Generation Forecasting method, where we showed that the forecasting accuracy on large-scale models can be greatly improved with the introduction of cooperation between partners. In this work, we delve into the importance of said cooperation partners and the possible methods of their selection. We tested and compared five different approaches - random selection, temperature based K-means clustering, generation based K-means clustering, temperature based Pearson correlation grouping and generation based Pearson correlation grouping - in two different scenarios (with and without added generation data to weather data), utilizing an Artificial Neural Network for prediction. The optimization of cooperation partners yielded error reductions even as high as 70% for the best-performing solutions, with clear differences outlined between the different approaches. We additionally observed that when utilizing previous generation data, we could not only improve the overall forecasting accuracy, but this also boosted the benefits of the cooperation method as well, as in the case where no generation data was used, the error reduction due to cooperation ranged from 7.92% to 50.74%, whereas in the case of added previous generation data, the error reduction due to cooperation spanned from 22.6% to 69.93%.