适应风电功率复杂概念漂移的在线概率预测方法研究

With the increasing wind power penetration, effective online probabilistic wind power forecasting becomes the key to improve capability of safe operation of wind power integrated systems, however, complex concept drifts from diverse sources have brought a serious challenge for the sustained stability of online forecast performance. Given that the existing research cannot effectively solve this problem, theoretically based on regression methodology adapting to concept drifts, this project will carry out systematic research in three stages successively: (1) Establish feature information database for wind power concept drifts, then accordingly construct and test a series of concept drift detection methods that have different natures and meet the requirements of probabilistic forecasting; (2) Taking advantages of existing online probabilistic forecast models, improve and generalize the fast incremental regression/model trees, then establish diversified online probabilistic forecast models where explanatory variables and concept drift detection methods are diverse from each other; (3) On the basis of meta-learning theory, construct online probabilistic forecast model selection system where a manual intervention mechanism is designed to easily confront expectable events, which will completely adapt to the complex concept drifts in wind power. The completion of this project has the important theoretical and practical significance, which can not only enrich the regression methodology adapting to concept drifts, but also provide reliable references on safe operation and optimal scheduling for power grids.

随着风电穿透率的不断增大，有效的风电功率在线概率预测成为提高风电并网电力系统安全运行能力的关键，但来源多样的复杂概念漂移为在线预测性能的持续稳定带来了严峻挑战。鉴于现有研究不能切实解决这一问题，本项目将以概念漂移回归方法为理论基础，分三个阶段依次开展系统的研究：（1）建立风电功率概念漂移特征信息库，据此构建并测试满足概率预测要求、性质各异的系列概念漂移检测方法；（2）充分利用现有在线概率预测模型的优势，完善并推广快速增量回归/模型树，建立解释变量和概念漂移检测方法各异的多样化的在线概率预测模型；（3）基于元学习理论，构建在线概率预测模型的优选系统，并为其设计人工干预机制以便应对可预期事件，最终实现对风电功率复杂概念漂移的完整适应。本项目的完成不仅可以丰富概念漂移回归预测理论，同时为电网安全运行和优化调度提供可靠的参考依据，具有重要的理论和现实意义。

风电的间歇性和波动性给风电并网和消纳带来了严峻挑战，可靠的风电预测是解决问题的有效途径。本项目致力于构建可靠的预测模型、提高电力和风电预测准确性，为提高电气化水平和可再生电力穿透率、减少温室气体排放提供参考依据。项目组从预测指标自适应选择、模型优选、概率预测和动态预测等角度提出了适应多种预测环境的多个预测模型，同时通过效应度量方法预测了电气化长期影响、评估了电力行业能源拥挤现象，提出了降低弃风现象、提高可再生能源占比等建议。本项目是预测理论与方法、能源经济与管理的交叉研究，对丰富预测理论、评估可再生能源发展的影响具有一定的理论意义和应用价值。