Local-pattern-aware forecast of regional wind power: Adaptive partition and long-short-term matching

• Published in: Energy conversion and management 2021-03, Vol.231, p.113799, Article 113799
• Main Authors: Liu, Chenyu, Zhang, Xuemin, Mei, Shengwei, Liu, Feng
• Format: Article
• Language: English
• Subjects: Adaptive partition; Algorithms; Cluster analysis; Clustering; Forecasting; Hierarchical clustering; Local-patterns; Long-short-term matching; Matching; Partitions; Renewable energy; Spatial distribution; Weather; Wind farms; Wind power; Wind region
• ISSN: 0196-8904; 1879-2227
• DOI: 10.1016/j.enconman.2020.113799

•An adaptive partition-based regional wind power forecast method is proposed.•Partitions reflect the spatial distribution of local-patterns in the wind region.•Long-short-term matching captures precise adaptive partition for multi-step forecast.•A new clustering method reduces the time cost by removing the invalid partitions.•Superiority and robustness of the proposed method are proved in real-world cases. Importance for the accurate forecast of wind region with multiple wind farms is gradually emerging. As influenced by the geographical features of the wind region, the power output from each wind farm is closely correlated to the local-patterns of its covered weather. However, modeling the highly time-varying nature of the local-patterns’ spatial distribution remains the key challenge to regional wind power forecast. For this purpose, a sub-region is proposed to represent the spatial scale of wind farms covered by the same local-pattern. All wind farms in the wind region are divided into multiple sub-regions. This classification is defined as the partition which represents a typical state of the wind region. To deal with the time-varying nature, partitions are considered on the adaptive process. In this paper, a regional wind power forecasting method based on adaptive partition and long-short-term matching is proposed. First, a refined partition set of wind region is determined by the Regional Hierarchical Clustering algorithm. Second, to identify the current states of the wind region, the partition with minimum forecasting error is chosen as Optimal Partition. Third, the long-short-term matching strategy is proposed to find the adaptive partition among the refined partition set with the indication of recent and historical Optimal Partitions. Eventually, for each time horizon, the forecasted power of each sub-regions in the adaptive partition is aggregated to achieve the final regional wind power forecasting results. The superior performance and robustness of the proposed methods are validated with actual wind generation data from a wind region which contains nine wind farms in China. The ability to capture wind farm local-pattern of the proposed method is also approved.