基于集成学习的区间型电力负荷预测技术研究

Interval-valued load series, which includes peak loads and valley loads, has the advantage of reducing the amount of random variation relative to that found in classic point series. Accurate forecasting of future interval-valued load can not only be useful for load flow analysis and transmission network expansion planning, but also provides an important risk management tool in making power system planning and operational decisions in the electricity industry. To address the interval-valued load forecasting problem that is emerging in power system operation and management, Ensemble Learning is proposed to establish a novel prediction modeling framework by taking full advantage of its good generalization ability. Specially, the main work of this proposed project includes: 1) Analytics and preprocessing methods used for interval-valued load series and influencing factors; 2) Systematic research of modeling of interval-valued load forecasting based on ensemble learning by focusing on many key technical issues, i.e., selection and optimization of base model, generation of individual forecasting models, strategies of ensemble, and evaluation of forecasting models; 3) Prototype development of ensemble learning based forecasting system for interval-valued load by integrating the research achievements of the former two points. Generally speaking, this project can make contributions to theoretical developments in field of interval-valued load forecasting and ensemble learning based forecasting, and provides valuable implications for practitioners.

区间型负荷数据，由给定时间窗口内的峰值负荷和谷值负荷构成。准确的区间型负荷预测指出了未来给定时间窗口中负荷波动的极大（小）值，不仅为电力系统错峰支援与峰谷置换、负载潮流分析、输电网扩展规划提供重要信息，还为电力企业制定系统规划和运作决策提供非常重要的风险管理工具。针对区间型负荷预测这一电力系统运营管理中新兴的基础预测问题，为克服现有单一模型及简单混合模型在预测中的稳定性及准确性差的缺点，本项目引入集成学习技术思想，以准确预测区间型负荷为目的，系统地研究基于集成学习的区间型负荷预测理论与方法。具体的，本项目在对区间型负荷波动特征及气象因素影响关系分析与处理的基础上，结合先进的文化基因算法，系统研究基于集成学习的区间型负荷预测建模方法，重点关注元模型选择与优化、个体预测模型的生成策略、集成策略、预测评价等关键技术细节，构建和完善系统的区间型负荷集成预测模型，并完成智能预测系统的原型开发。

准确的区间型负荷预测指出了未来给定时间窗口中负荷波动的极大（小）值，不仅为电力系统错峰支援与峰谷置换、负载潮流分析、输电网扩展规划提供重要信息，还为电力企业制定系统规划和运作决策提供非常重要的风险管理工具。本项目围绕区间型序列预测方法、集成学习技术及其在负荷预测等领域的应用展开研究。通过项目研究，构建了区间型多层感知器和区间型长短记忆网络等多个神经网络模型，以及基于混合抽样、区间聚类的集成预测模型，还构建了基于计算智能的电力负荷预测框架和预测支持系统体系结构。本项目累计出版专著1部，发表论文9篇。相关成果对于丰富区间型序列预测和集成学习理论与方法、提升电力负荷预测性能、保障电力系统的安全高效运营，具有重要的理论意义和应用价值。