基于集成学习的分布式光伏发电功率日前预测

对于光伏发电功率精准的日前预测有助于电网设计未来调度计划，降低新能源发电对电网的冲击，提高消纳率。提出一种Boosting集成学习框架下的光伏发电功率日前预测方法。首先，根据光伏出力主要受天气影响的特点，通过皮尔逊系数获得相关性强的气象因素，利用k-means++对与光伏发电功率相关性极强的总水平辐照度进行聚类以获得相似日数据集；然后，将极限学习机（extreme learning machine，ELM）引入Boosting框架，构建光伏出力日前预测模型（B-ELMs）；最后，利用真实光伏电站运行数据验证模型有效性，该模型在试验过程中展现出良好的适应性，最高决策系数（R²）达0.9819。实验结果表明，由于集成学习框架的存在，B-ELMs能对复杂天气下的规律性弱、波动性强的光伏出力曲线提供较为精确的预测结果；同时，相较于深度学习网络，B-ELMs的收敛速度更快，在维持较快训练速度的同时保障更为精确的预测结果。

Mechanism Design and Trial Operation Analysis of Shandong Power Spot Market

In order to summarize the trial operation experience of the first batch of pilot areas for power spot market construction, and promote the construction of power spot market in China, the market mechanism of Shandong power spot market is introduced, and the existing problems and some feasible solutions are studied based on an analysis of its trial operation situation. The operation features of Shandong power grid, such as the high penetration of clean energy, have dictated that the centralized power spot market mode may be more suitable for Shandong. Shandong power spot market consists of two timing markets, i.e., a day-ahead market and a real-time market, and an intra-day unit commitment, which is used as a cohesive mechanism. During the trial operation period, the overall operation of Shandong power grid is stable, and the power spot market clearing price changes with the relations between power market supply and demand. However, there are still some problems that need to be solved, including the market monitoring ability, the peak shifting compensation mechanism, and the prevention and control measures of users' trading risks. The research results show that the Shandong power spot market has basically met the conditions for continuous operation, but the improvement of market mechanism still needs to be continuously strengthened.