耦合KL理论与调度特征的大规模水电站群优化调度降维方法

大规模水电站群优化调度计算效率是水电及电力系统运行面临的最棘手问题之一,是突破超百座水电站高维复杂系统求解的理论和技术障碍。本文提出一种耦合KL理论与调度特征的水电优化调度降维方法,通过对水电站群长系列调度样本进行主成分分析,识别调度过程中的库水位变化特征值与其对应的特征函数,采用KL理论将库水位描述为多个水位变化特征项的组合函数,引入Kullback-Leibler散度以根据问题特点确定调度特征项随机系数的概率分布及初始值;构建了两阶段逐步迭代寻优策略,通过动态搜索水位特征项的随机系数实现大规模水电站群优化调度的高效求解。提出的方法以云南电网超百座水电站群调度问题进行了验证,通过不同算例对方法的有效性、高效性、随机系数概率分布和参数敏感性进行了分析,与经典动态规划及其改进算法相比,KL方法在有效保持结果精度的条件下显著提高了计算效率。

Dimension reduction method for optimal operation of large-scale hydropower plants with Karhunen-Loève theory and dispatching features

The computing efficiency of the optimal operation of large-scale hydropower plants is one of the toughest problems in the operation of hydropower and power systems, which is a theoretical and technical barrier to solve the complex system with more than 100 hydropower plants. This study presents a dimension reduction method for optimal operation of large-scale hydropower plants with Karhunen-Loève (KL) theory and dispatching features. Based on a long series of historical operation records, the principal component analysis is used to identify characteristic value of water level change in the front of the dam and corre-sponding characteristic function. The KL expansion method is then introduced to represent random variables of the water level at any period as a linear function of characteristic terms of reservoir level change. Thus, the operation decision of a specified reservoir inflow process can be determined by a combination of ran-dom coefficients of all characteristic terms. A framework of iterative optimization is constructed,where a so-lution procedure for searching random coefficients of characteristic terms is given to efficiently solve optimal operations of large-scale hydropower plants. The presented method is verified by long-term operations of a provincial hydropower system with more than 100 plants in Yunnan. The validity,efficiency,and sensitivity of this method are respectively demonstrated by several case studies. A comparison with conventional dynam-ic programming and its modifications shows that the method contributes a substantial improvement of compu-tational efficiency while ensuring basically the same solution accuracy.