Research And Application Of Parallel Methods For Solving Optimal Generation Operation Of Hydropower Station Group

Many large-scale groups of hydropower stations cross basins and provinces have been laid out in China after decades of construction, especially the last decade. Joint optimal regulation of hydropower station group, which is supposed to maximizing the compensation benefits from aspects of hydrology, reservoir storage and power generation, has drawn growing attention as an important approach to achieve more economic benefits as well as secure flood control, power supply, energy saving and emission reduction.Research on optimal operation of station group is mainly focused on model establishment and algorithm design. In previous studies, the rapid expansion of hydropower stations’size and the corresponding increasing complexity of models have highlighted the problems of limited calculation speed and being easily stuck in local optimal solutions. This thesis applies parallel computing techniques in the joint optimal operation of large-scale hydropower station group in upper reach of Yangtze River with various parallel strategies involved with multiple traditional algorithms and intelligent algorithms. Results of parallel calculation are validated through the case study in Yangtze River as well as standard testing function. Main contents and conclusions of the study are as follows:(1) Parallel dynamic programming algorithms based on the independence of state point and stage reconstruction are proposed against the curse of dimensionality. First, calculation process of dynamic programming is divided into two technical steps including calculation in single stage and recursion among stages by analyzing the principle of dynamic programming recursion. Then parallel dynamic programming based on the independence of state space is suggested in light of the independence among state points in the state space, which is corresponding to the step of calculation in single stage. The second parallel dynamic programming is carried out by adjusting the calculation orders of different stages and reconstructing them into new sub-stages, which is corresponding to the step of recursion among stages. The above two methods are applied to the optimal operation for power generation in a cascade station group in Yalongjiang River, and the effects of parameter control on the performance of parallel dynamic programming are analyzed and summarized. Simulated results reveal the superiority of parallel dynamic programming over serial mode.(2) Parallel progressive optimization algorithms based on multiple initial solutions and the independence of stages are proposed against the weakness of traditional progressive optimal algorithm. Through the analysis of the structure of progressive optimization algorithm, the calculation process of progressive optimization algorithm is divided into two technical steps:generation of initial solutions and two-stage recursion optimization. In the first step, we implement iterative optimization in multiple processes with different initial solutions instead of single initial solution in the traditional method, which is suggested as parallel progressive optimization algorithm based on multiple initial solutions. For the second step, parallel progressive optimization algorithm based on the independence of stages is suggested with a concern of the independence between different two-stages. These two methods are then applied to the optimal operation for power generation in a cascade station group in Yalongjiang River, and the effects of parameter control on the performance of parallel progressive optimization are analyzed. Simulated results indicate that parallel algorithms are significantly better than serial algorithm.(3) Parallel differential evolution algorithms based on sub-group division and adaptive immigration are respectively presented for the degrading performance of intelligent algorithms along with the increase of dimension in solving complex optimal issues such as joint operation of cascade reservoirs. The algorithm based on sub-group division is proposed according to the natural parallelism of populations in the structure of differential evolution algorithm while the dynamic adaptive immigration principle is suggested through an analysis of information exchange mode among different sub-groups. Performance of the algorithms are then tested and compared through test function and the case of joint operation of cascade reservoirs after calibrating the selection range of control parameters. Results also suggest that parallel differential evolution algorithms can outperform the serial counterpart.(4) A parallel optimization algorithm based on decomposition of hydropower stations is proposed for short-term joint optimal scheduling of large-scale hydropower station group. Stations are decomposed to several basic calculation units from three aspects:spatial distribution of stations, propagation characteristics of flow among stations and regulation performance of reservoirs. Then the units are allocated to different parallel processes in order to improve the computing speed. The algorithm is applied in the short-term joint scheduling of large-scale hydropower station group in upper reach of Yangtze River, and results reveal that the suggested algorithm can meet the requirement of timeliness with higher efficiency.