Research On Dimensionality Reduction Optimization Methods For Mid-long Term Hydropower System Operation

Since the implementation of west-east power transmission, China has been witnessing the rapid development of hydropower industry in extra-large river basin, provincial and regional power grid. In the past few years, the hydropower system have changed greatly, including hydro unit scale, installment capacity, generating head and the number of hydropower stations. The computation complexity of hydropower system operation increases exponentially due to the explosion of system scale. Based on the China’s southwestern regions with rich hydropower resources, in order to overcome the curse of dimensionality, the paper focuses on the theories of dimension reduction optimization methods, including four different aspects:knowledge rule, parallel computing, experimental design and swarm intelligence. The main research contents of this paper are summarized as follows:(1) Traditional methods cannot dynamically identify the feasible search space in most cases, which has direct impact on the algorithm performances. Thus, the knowledge rule dimensional reduction approach is presented to overcome the abovementioned problem. The set operations theory is first used to change the preset interval of different constraints into the specified one, and then the feasible region of multi-reservoir system is obtained by the dynamic coordination between all the reservoirs, leading to the reduction of computational measure for conventional optimization methods. The results show that the proposed method can identify the feasible zone effectively, and help traditional methods avoid the superfluous computation so that the computational efficiency is enhanced remarkably, which provide new means to alleviate the problem of dimension disaster.(2) When applied to solve the complex computation-intensive tasks of large-scale hydropower system operation problem, classic methods, like dynamic programming (DP) and progressive optimality algorithm (POA), usually take too much time to reach fine satisfactory solution. To overcome the problem, the potential parallel mechanisms of DP and POA are first thoroughly analyzed, then the parallel strategy for different methods are designed:the discrete state combinations of the POA sub-problem has good parallel computing capacity, while DP has parallelism in four different layers which are combination-layer, phase-layer, state-layer and decision-layer. Using the Fork/Join parallel framework, parallel methods are implemented to enhance the computation efficiency. The simulation results indicate that parallel methods can make the utmost of multi-core parallel computing resource and greatly improve the computational efficiency while obtaining the same results with traditional serial methods, which provide new technique to alleviate the problem of dimension disaster.(3) With the rapid expansion of hydropower system, the computing scale of traditional methods is undergoing an explosive growth due to the comprehensive combination at each stage. Thus, to alleviate the problem, three novel optimization approaches combining classical methods and experiment design are proposed, which are orthogonal progressive optimization algorithm, orthogonal discrete differentiation and dynamic programming and uniform dynamic programming. The proposed methods starts from an initial feasible solution, and takes advantage of the specific design table to choose some small but representative state vectors instead of all the vectors at each stage, and then the recursion formula is employed to obtain better solution. Moreover, these methods will continue the iteration calculation until the termination condition is satisfied. The theoretical analysis and simulation results indicate that three methods has polynomial-growth computational complexity, which makes a significant reduction in computing time and memory usage, and provide new ideas to alleviate the problem of dimension disaster.(4) In general, swarm intelligence algorithms use the complexity of evolutionary iteration rather than combination computation, which helps them handle with large-scale hydropower system operation problem. However, due to the mechanism-based defects, these methods have the premature convergence problem. Therefore, the quantum-behaved particle swarm optimization (QPSO) and social spider optimization (SSO) are introduced to solve the hydropower operation problem. Moreover, based on the initialization mechanism, evolution pattern and mutation search strategy, the improved quantum-behaved particle swarm optimization (IQPSO) and elite-gather social spider optimization (ESSO) are developed to alleviate the existing defects of original methods. In the way, both the local exploration ability and global searching ability of two methods can be enhanced simultaneously. The results demonstrate the effectiveness and practicality of the proposed methods, which provide new methods to alleviate the problem of dimension disaster.