考虑流域站点水位特性的水库群防洪调度研究

Research on the Reservoir Group Flood Control Operation Considering Water Level Characteristics of Basin Flood Control Stations

针对常规水库防洪调度对防洪控制站点水位安全需求考虑不足的问题,研究提出了一种耦合流域防洪站点水位计算的水库群防洪优化调度方法。首先,通过多站点水位流量关系解析站点水位特性,辨识站点水位关键影响因子,进而根据各影响因子与水位之间的相关关系筛选用于水位计算的输入特征,并引入多层前馈型反向传播神经网络(Back Propagation neural network,BP神经网络),用于拟合输入特征与实测水位之间的复杂非线性关系,从而实现站点水位精确计算;其次,提出一种表征防洪控制站点水位安全程度的指标——安全裕度,以梯级下游防洪站点的整体安全裕度最大为目标,构建了耦合BP神经网络的梯级水库群防洪优化调度模型;最后,综合逐步逼近动态规划(Dynamic Programming Successive Approximation,DPSA)、逐次优化算法(Progressive Optimization Algorithm,POA)和粒子群算法(Particle Swarm Optimization,PSO)的优势,提出一种混合优化算法(DPSA-POA-PSO)对模型进行求解。以金沙江下游-三峡梯级水库群为对象开展实例研究,结果表明:考虑站点水位特性显著提升了BP神经网络的水位计算精度,对典型洪水的洪峰水位拟合偏差均不超过0.05m;与逐步逼近动态规划和粒子群优化算法相比,在梯级三库联合调度时,混合优化算法使得下游站点整体安全裕度分别提高了0.88%和2.58%,梯级五库联合调度时,下游整体安全裕度分别提高了0.85%和1.87%,并均能够同时满足各站点的保证水位要求,为改善流域水库群防洪安全水平提供了可靠支撑。

To address the insufficient consideration on water level safety requirements at flood control stations in conventional reservoir flood control operation,this study proposes an optimal flood control operation method for reservoir group coupled with water level calculation at ba⁃sin flood control stations.First,the station water level characteristics are analyzed through the multi-site water level and flow relationship to identify the key factors affecting the water level.The input features for water level calculation are screened according to the correlation between the influencing factors and the water level. The multi-layer feed-forward back propagation neural network (BP neural network) is in⁃troduced to fit the complex non-linear relationship between the input features and the observed water level, to realize the accurate calcula⁃tion of water level at the station. Then, the safety margin index is proposed to quantify the station water level safety. With the objective of max⁃imizing the overall safety margin of the downstream flood control stations, the optimal flood control operation model for cascade reservoirs,coupled with BP neural network, is constructed. Finally, a hybrid optimization algorithm (DPSA-POA-PSO) is developed to solve the mod⁃el by combining the advantages of the Dynamic Programming Successive Approximation (DPSA), Progressive Optimization Algorithm(POA) and Particle Swarm Optimization( PSO). The results show that the water level calculation accuracy of the BP neural network is signifi⁃cantly improved by considering the station water level characteristics, and the flood level fitting deviation for typical floods is limited within0.05m. Compared with the DPSA and PSO algorithms, the downstream overall safety margin from the hybrid optimization algorithm is im⁃proved by 0.88% and 2.58% in the joint operation of three cascade reservoirs, and is improved by 0.85% and 1.87% in the joint operation offive cascade reservoirs. Meanwhile, the derived operation schemes satisfies all the guaranteed water level requirements, which provides reli⁃able support for improving the basin flood control safety.