Data Mining, Swarm Intelligence Algorithm And Composite Decision For Optimal Operation Of Hydropower Energy System

Water is the source of life and power is the engine of social development. Water is not only the indispensable basic needs to human life and production, but also is an important energy resource. On the whole, the water resource is very rich in our country and thirteen hydropower bases locate on the major rivers. Due to the cleanliness and renewable, hydropower has a very important role to reduce the dependence on fossil fuels and to improve the protection of natural environment. However, the amount of natural water is limited; it should be rational after meeting the water demand for life. From the broad perspective, the water resource is very scarce in our country. Therefore, research on the theory, method and technique for rational development and utilization of water resources and optimal allocation of hydroelectric energy, to exert the comprehensive benefits of water conservancy and hydropower, is of great practical significance and supporting role for the economic and social development.Hydropower energy system is a complex giant system; its optimal operation boundary is a comprehensive intersection of multiple needs on a variety of levels. It has been facing many scientific and engineering problems and urgently needs theoretical breakthrough and technical research. Rational use of and assigning the limited water resources is inseparable to the scientifically and accurately hydrological forecast, hydropower energy system optimal modeling and solution, multi-attribute multi-level decision making for multi-objective scheduling schemes. In response to these three problems, in this thesis, researches in the fields of data mining techniques for hydrology analysis and forecast, swarm intelligence solution methods for hydropower energy system optimization modeling, composite decision making theory and methodology with the consideration of multiple scheduling objectives, multiple attributes of scheduling schemes, multi-level of scheduling subjects, are carried out.(Ⅰ) Accurate runoff forecast is the precondition of rational use of water resources and scientific scheduling of cascade reservoirs. Accurate and timely runoff forecast is of great significance for reservoirs scheduling. For a long time in hydrological science, how to predict the reservoir inflow, especially the mid and long-term runoff forecast as accurately as possible is a difficult problem. Basin runoff process is a complex nonlinear dynamical process affected by many factors. Conventional linear methods cannot give a valid characterization. BP neural network is a non-linear data mining technique with extremely strong learning ability. But it needs to select appropriate input variables when used for hydrological forecast. According to the geographical and data features of the Jinsha river basin, the variables reflecting the interannual wet-dry characteristics of basin runoff are introduced into forecast factors to improve forecast accuracy; By refining the forecast scale, the forecast period is extended; The mechanisms of keeping-the-best and dynamic learning factor are introduced into BP neural network to improve the parameters learning efficiency and samples simulation accuracy. In the end, the case study for Jinsha River basin confirmed the feasibility and effectiveness of the proposed method.(II) The scheduling of hydroelectric energy and its interconnected power system is, in essence, a kind of nonlinear optimization problems. Many optimization methods have been used to solve these problems by previous scholars, but they have unique characteristics and difficulties. They mainly includes:i) the non-convexity caused by nonlinear objective function or constraints; ii) the high-dimensional caused by multiple decision variables; and iii) the strong constraints and narrow feasible region caused by complex constraints. Therefore, exploring the modeling theory and efficient solution methods is urgently needed. Three representative scheduling models are selected for studing effective swarm intelligence algorithm according to the characteristics of each problem. For the cascade reservoirs optimal generation problem, the Team-TLBO with advanced study algorithm is built to fullly use the global search ability of Teaching Learning based Optimization (TLBO) and avoid its lack of local optimization based on the concepts of "team teaching" and "advanced studies". For the difficulties of solving power system energy saving multi-objective dynamic scheduling problem and the shortage of differential eolution (DE) algorithm for solving high-dimensional and multi-objective problems, the extended double selection operator, and dynamic heuristic constraint handling and adaptive restart mechanism are proposed, and further the improved adaptive multi-objective differential eolution algorithm is built to enhance solution efficiency and accuracy. For the difficulties of solving hydrothermal power systems joint coordination optimization problem, according to the advantages and disadvantages of TLBO and DE, a hybrid algorithm in which TLBO is set as main algorithm and DE is set as local search operator is proposed to enhancing the overall optimization performance. The separately case studies of the three problems confirmed that the proposed methods can efficiently obtain the optimization schemes.(Ⅲ) Complex hydropower energy giant system is a multi-objective scheduling object of multi-level organizational structure decision-maker group. Hydropower energy system operation and dispatch must met many social and economic needs. Although, to some extent, building different mathematical model with different optimization goals maybe effectively take into account of various scheduling needs, hydro energy systems often involve basin and even inter-basin watershed. Water resource allocation is not only constrained by multiple scheduling body, but also guided and restricted by multiple levels decision groups. The decision of scheduling plan is a MCDM problem, and it is more of a multi-level group decision-making problem. Based on the operation process and logical relationship between decision bodies of the reservoirs regulation and management departments of our country, a multi-level multi-expert interactive decision theory and system meeting our national conditions and engineering project is proposed. By introducing the multi-level decision-making approach from social and economic field, a multi-objective multi-level interactive decision-making model with the consideration of organizational structure is established and its mathematical description and convergence proof are given. In the end, taking the multi-objective flood control of Three Gorges-Jinsha river cascade reservoirs as object, the scientific decision-making method that meets the requirements of reservoirs regulation and management departments is explored, providing theoretical basis and technical supporting for scientific decision-making of water resources optimal allocation in the future.