大规模风电并网中的异常值检测与预测研究

Large scale wind power integration not only bring theoretical and technical challenge to the electrical power system, but also make a severe test to the accuracy of the wind power anomaly data detection, wind energy potential assessment for different areas as well as wind power forecasting. This project focuses on studying the above problems by using time series forecasting models, statistical theories and intelligent optimization algorithms. The research contents are as follows: (1) Detect the anomaly data in large scale wind power integration based on cohesive properties and distribution characteristics of the wind power data, and verify the effectiveness of the detection model by examining whether the masking effect can be avoided, thus provide effective early warning with respect to the possible problems in the wind power integration. (2) Make up defects in the existing probability distributions which are applied to wind power distribution fitting forecasting and improve the accuracy of wind energy assessment models based on the uncertainty theory. (3) Correct the longitudinal forecasting results for different decomposition sequences of the wind power data through the transverse time error vector, so as to put forward new approaches for wind power forecasting. The completion of this project has important theoretical significance and application value, which can not only enrich the anomaly detection, wind energy assessment and forecasting theories, but also provide reliable reference for risk management decision making in large scale wind power integration.

大规模风电并网给电力系统带来理论和技术挑战的同时，对风电异常数据检测、不同地区风能潜力评估以及风电预测的精度也提出严峻的考验。本项目将利用时间序列预测模型、统计学理论和智能优化算法，对上述问题开展研究。主要内容包括：(1)充分分析大规模风电数据的凝聚属性和分布特征，实现大规模风电并网中异常数据的检测，并对检测方法的有效性通过是否可以免遭屏蔽效应进行检验，对风电并网中可能存在的问题进行有效预警；(2)弥补现有概率分布进行风电数据概率分布拟合预测的缺陷，完善基于不确定理论的风能评估模型；(3)基于时间序列分解模式和横向时间误差向量，在风电数据不同分解序列纵向预测结果的基础上，对其进行横向时间误差校正，提出适用于风电预测的新方法。本项目的完成不仅可以丰富异常值检测、风能评估和预测理论，同时为大规模风电并网风险管理决策提供可靠的参考依据，具有重要的理论意义和应用价值。

风电缺失和异常数据的处理、风速预测以及风能评估的准确性对大规模风电并网具有重要的意义。本项目致力于利用时间序列预测模型、统计理论和优化算法对风电缺失和异常数据进行处理并提高风速预测及风能评估的精度。项目组首先利用频谱分析和长短期记忆网络对风电缺失数据进行了预测优化填充；其次采用统计学相关理论、神经网络以及混合预测策略对风速中异常数据进行了处理并对风速进行了优化预测；最后利用具有自适应截断点的截尾分布和新构造的损失函数对风能评估中的关键因素—风速的概率分布进行了优化估计。实验模拟结果显示，本项目组提出的优化模型具有较好的鲁棒性且对缺失和异常数据处理、风速预测和风速的概率分布估计具有较高的精度。本项目的完成对丰富风电缺失和异常数据处理、风速预测和风能评估的理论以及辅助大规模风电并网中的风险管理决策具有一定的理论意义和应用价值。