基于数据学习模型的风电场机组分布式滚动优化调度

The project focuses on the practical application of wind power industry and researches on distributed rolling optimal scheduling of wind farms based on data learning model. Through deep mining the information contained in wind farm history data, considering the wake effect between units to estimate the effective wind speed by the statistical learning method and combining wind field operation data and associated meteorological data, a high precision wind power ultra short term forecasting model is established based on deep learning algorithm. Then for the unit health assessment problem, the project will make an in-depth analysis on the correlation between the state parameters, design a reasonable health evaluation index and construct the health assessment model of key components and the whole unit in wind turbine through statistical learning algorithm. At last, based on unit power forecasting and health learning, an optimization index is developed to reduce the number of start-ups and stops. Also, it minimizes the unit output deviation and optimizes the health of working units. Non-analytic model based distributed rolling optimization method is explored and the performance of algorithm is analyzed to realize the optimal scheduling of wind farm. This project has forefront on theory and provides new methods and ideas in ultra short term wind power forecasting, unit health assessment and distributed rolling optimal scheduling based on data learning model. It also proposes scientific solutions for efficient scheduling of grid connected wind farm.

本课题面向风电产业实际应用，针对基于数据学习模型的风电场分布式滚动优化调度展开研究。通过深入挖掘风电场历史数据蕴含的丰富信息，考虑机组间尾流效应，采用统计学习方法对机组有效风速进行估计，并结合风场运行数据与相关气象数据以深度学习算法为核心建立高精度的超短期风功率预测模型。针对机组健康评估问题，深入分析状态参数间的相关性，设计合理健康评价指标，利用统计学习算法构建风电机组中关键部件及整机的健康度评估模型。基于机组功率预测和健康度学习模型，建立以减少启停次数、最小化机组出力偏差以及出力机组健康度整体最优为目标的优化指标，设计系统结构分解方法，探索在非解析模型形式下的分布式滚动优化求解方法，并就算法性能进行分析，实现风电场的优化调度。本项目在理论上具有前沿性，将在超短期风功率预测、机组健康度评估、基于数据学习模型的分布式滚动优化调度方面提供新的方法和思路，为风电场机组的高效调度提出科学解决方案。

本课题基于数据学习，针对风电场分布式优化调度问题展开研究。首先，结合风场运行数据与相关气象数据分别讨论了风速多步预测、风功率曲线建模和多模型组合风功率预测等问题，在此基础上建立了高精度的风功率超短期预测模型。其次，分析了风电机组健康评估与故障诊断问题，研究了时序数据下的风机子系统健康评估，然后对时序数据和不平衡数据下基于深度学习模型的风机故障诊断展开研究，继而考虑无标记数据集，进一步研究风机主故障诊断问题。再者，基于机组功率预测和健康评估模型，以最小化机组出力偏差以及最优化出力机组健康度为优化指标，分别就成本函数为二次型、凸函数和非凸函数等情形求解风电场分布式滚动优化调度问题。随后，在理论层面上将优化调度问题延伸为分布式泛函最优化问题，结合最优控制，设计了分布式强化学习算法进行求解。最后，将上述研究成果结合实际场景进行应用，进一步验证了所提方法的有效性。本项目为风功率超短期预测、风电机组健康评估、基于数据学习的分布式优化调度以及多智能体泛函最优化等问题的理论与应用研究提供了新的方法和思路，并为风电场的高效调度提出了新的解决方案。