The Research Of Hybrid Intelligent Algorithms And Parallel Methods For Long-term Operation Of Cascaded Hydropower Stations

Over the past years, China has witnessed a very rapid hydroelectric development and the hydropower capacity is ranked the first in the world. Large quantities of cascaded hydropower systems, which are characterized with many reservoirs, huge installed capacity and wide transmission, have been and will put into operation. The optimization of large scale cascaded hydropower system has the characteristics of high dimension, nonlinearity, multistage, coupling constraints, and the extremely large number of cascaded reservoirs has aggravated curse of dimensionality, leading to greater difficulties in problem solving and lower computational efficiency. The results obtained from traditional methods can’t meet the fine demand in management of power grid and hydropower companies. Therefore, it’s very important and significative to explore reasonable and efficient solution algorithms for operation of hydropower systems. In the present paper, cascaded hydropower reservoirs on Lancang river and Hongshui river, which are located in southwestern China, are chosen as the engineering backgrounds. The research of hybrid intelligent algorithms has been carried out, and the recently popular parallel technique is adopted to improve the efficiency of dynamic programming and its improved ones. The main contents and conclusions are summarized as follows:In the view of the shortcomings of premature convergence which easily leads to falls into local optimum and stagnation in particle swarm optimization (PSO), a self-adaptive hybrid particle swarm optimization algorithm (AHPSO) is proposed to solve the long-term optimal operation model. AHPSO generates initial solutions with chaos to improve individual’s quality, and employs some variables including particle energy, particle similarity, the threshold of particle energy and the threshold of particle similarity for describing the self-adaptive changes and the evolving degree. In the meanwhile, the idea of genetic variation is introduced to improve global search ability in earlier stage and local refine ability in later stage to overcome premature convergence. A neighborhood based random greedy search strategy for overcoming the shortcoming of evolving slowly in the later running phase is also adopted. With the cascaded hydropower stations on Hongshui river as a study object, the result shows that AHPSO has better astringency and optimized solution than PSO, and is comparable with progressive optimization algorithm but less compute time. Therefore, AHPSO is a feasible method for solving the problem of large-scale cascaded hydropower stations operation.Aiming at weakness of premature convergence and inferior local search ability, a self-adaptive chaos whole annealing genetic algorithm (SCWAGA) based on simple genetic algorithm (SGA) is proposed to solve the long-term optimal operation problem for hydro cascaded system operation. In SCWAGA, chaos optimization is used for improving the quality of the initial solutions. Simultaneously, considering the influence of population diversity, the Boltzmann probability selection is applied for avoiding population precocity. In addition, crossover probability and mutation probability are adjusted constantly with self-adaptive function for improving local search ability and avoiding the local convergence. With hydropower station group on Hongshui River as a case study, the result shows that SCWAGA can obtain better global solution than SGA and POA and cost less computing time than POA. This method provides an efficient way for solving the optimal operation problem of large-scale hydropower system.With the curse of dimensionality appears when applying dynamic programming method to operation of large-scale hydropower systems, the computational time cost and low computational efficiency are great challenges. For this reason, parallel discrete differential dynamic programming (PDDDP) and parallel stochastic dynamic programming (PSDP), which are parallelized using Fork/Join parallel framework, are proposed in this paper to represent the normal way for parallelization of dynamic programming (DP) and its improved ones. On the basis of deep analysis on the parallelism of recursion equation of DP, all combinations of discrete variables in recursion equation at multistage and single stage are respectively designed as parent task for realization of PDDDP and PSDP. The proposed method was implemented to generation operation of cascaded reservoirs located on the lower stream of Lancang River, and two diverse configurations were used to test the parallel performance. The results show that the proposed parallel methods can obtain the same solution with serial methods, substantially cut down the execution time and take full advantage of multi-core resources. In addition, the proposed parallel design can provide guidance for other DP methods.For the extreme drought climate’s serious effect on hydropower generation and the security of power system, a multi-core parallel chance constrained dynamic programming (PCCDP) considering system reliability and extreme failure was proposed for long-term generation operation of cascaded reservoirs. In the proposed approach, the system reliability and extreme failure constraint were embedded in the recursion equation as Lagrangian term and penalty term respectively, and a multi-core parallel framework named Fork/Join was applied to parallelization of the proposed approach in which all combinations of variables for solution of recursion equation at any stage can be parallelized. By mean of continually updating Lagrangian multiplier and penalty coefficient, a variety of operating rules were generated, and then the simulative results were used to select the optimal operating rule. The proposed method was implemented to generation operation of cascaded reservoirs located on the lower stream of Lancang River, and two diverse configurations were used to test the parallel performance. The results show that the proposed PCCDP method can obtain the satisfactory operating rule responding to extreme drought climate which supplys a powerful technical support for the security of power system. Moreover, the execution time in parallel is substantially cut down, which provides a more efficient mean to research on the operating rules of cascaded reservoirs.Finally, a summary is made and some issues which need to be further studied are listed.