Research And Its Application Of Multilayer Nested Parallel Multi-dimensional Dynamic Programming

With the rapid propulsion of cascade hydropower development in our country, the joint operation optimization for large-scale cascade hydropower stations is getting more and more attention of people.There are some limitations for the traditional analysis theories and optimization methods in solving this kind of problems, which leads the water resources unable to be fully used. So, in order to give full play to the mutual compensation of cascade reservoirs, and improve the utilization efficiency of water resources, it has very important theoretical and practical significance to carry out the operation optimization research on theory and method for large cascade hydropower stations.In view of this, this paper takes the cascade hydropower stations of Li Xianjiang River as the research object, and carrys out the joint operation optimization research from the aspects of theoretical study and practical application, based on the generalizations of current hydropower station operation optimization theory. The thorough research has been carried out from the aspects of modeling of multi-layer nested multi-dimensional dynamic programming and its parallel computing, application of discriminant criterion method, total output operation chart optimization of cascade reservoirs, modeling of the optimal output distribution, extraction of operation rules and instance application, respectively. The main research contents and results are as follows:(1) In order to alleviate the "curse of dimensionality" of Multi-dimensional Dynamic Programming (MDP) in cascade reservoirs operation optimization, this paper proposes a new MDP algorithm named Multi-layer Nested multi-dimensional Dynamic Programming (MNDP) based on multi-layer nested structure, and takes the cascade reservoirs of Li Xianjiang River in China as an application instance, the global optimality of MNDP is validated by the calculation results of MDP. Contrast and analysis of MNDP and MDP from the aspects of run-time, memory usage and computational complexity by the calculation results with the same discretized points are provided. Finally, the advantages and disadvantages and applicability of these two algorithms are shown.(2) Aimed at the defect of MNDP on run-time, this paper proposes the Multilayer Nested Parallel multi-dimensional Dynamic Programming Algorithm (MNPDPA) by introducing the parallel processing technology, and realizes the parallel computing of MNPDPA in a computer with multiple cores. Finally, analyzes and evaluates the parallel performance of MNPDPA by the calculation results.(3) In order to explore and mining the possible room to promote the efficiency of parallel algorithm, this paper proposes three parallel modes for MDP by the parallel analysis, i.e. the stages parallel mode, state combinations parallel mode and hybrid parallel mode, which are used to study the effect of different modes on parallel computation efficiency. Finally, analyzes and evaluates the parallel performance of these three parallel modes from the aspects of run-time and memory usage.(4) By taking the cascade reservoirs of Li Xianjiang River as the research background, the following work has been done based on the long series of optimal results by MNDP. First of all, summarized and extracted the general operation laws for each reservoir in each period. Secondly, divided the whole operation period into four special phases according to the water level variation law of each reservoir, and developed the operation rules for each reservoir over the whole operation period based on the four special phases. Finally, implemented the simulation operation calculation by the obtained operation rules, compared and analyzed the simulation results with that of MNDP and conventional method.(5) Considering the mutual compensation effect among cascade reservoirs, this paper sets up the maximum guarantee output model and maximum power generation model based on the discriminant criterion method and the principle of compiling cascade energy storage operation chart, and in order to optimize the initial cascade energy storage operation chart, the Progressive Optimality Algorithm has been embedded in the solving process. From the aspects of power generation, guarantee output and assurance rate, the contrast and analysis of cascade energy storage operation charts wtih the present cascade conventional joint operation method are provided in the final.(6) In view of the three shortcomings of discriminant criterion method in the application of cascade reservoirs operation, with the goal of maximizing power generation over the entire planning horizon, this paper establishes a new double nested model which is used to optimize the total output operation chart and determine the optimal output distribution. Moreover, mean value method and stepwise regression method is respectively used to extract the output distribution ratio based on the long series of optimal simulation operation data. Finally, through the simulation calculation results, comparison and analysis of the obtained operation rules with the discriminant criterion method and conventional method are shown.