A Novel Closed-Loop Clustering Algorithm for Hierarchical Load Forecasting

Hierarchical load forecasting (HLF) is an approach to generate forecasts for hierarchical time series. The performance of HLF can be improved by optimizing the model and structure. Previous studies mostly focus on while structure usually formed clustering natural attributes like geography, customer type, or similarities between profiles. A major limitation these structures mismatched objectives forecasting. Clustering aims minimize dissimilarity among customers a group their errors. two independent optimizations could limit overall performance. Hence, this paper attempts integrate novel closed-loop (CLC) algorithm. It links feedback mechanism return goodness-of-fit as criterion clustering. In way, enhanced re-assigning cluster membership parameters models are updated iteratively. method comparatively assessed with existing methods. Using same model, proposed outperforms bottom-up 52.20%, ensemble-based 26.89%, load-profile 19.90%, respectively.