Model And Algorithm Of Approximate Dynamic Programming For Large-Scale Power System Unit Commitment Problems

With the construction and development of intelligent power dispatching system, department of power grid operation have higher and higher require-ments on the accuracy and computation speed of unit commitment. At the same time, the interconnection scale expansion of grid makes the unit com-mitment problem increasingly complex and difficult to solve. Therefore, re-searching and exploring theory and methods suitable for solving the unit com-bination problem of large-scale power systems has important practical signifi-cance and economic value for improving generation efficiency and saving en-ergy and resources.This dissertation focused on the model and algorithm of approximate dy-namic programming and their applications in the unit commitment problem of a large-scale power system. The approximate dynamic programming theory, which is the latest achievements in mathematical programming, is applied to the unit commitment problems of power systems for the first time. The disser-tation has conducted the thorough careful research to the approximate dynamic programming-based method for day-ahead conventional unit commitment problems and security-constrained unit commitment problems. This study has enriched and perfected the theory of approximate dynamic programming. The key contributions and results are as follows:The mathematics of solving the "curses of dimensionality" by approxi-mate dynamic programming is discussed first. Four kinds of models of ap-proximate value functions and existing methods of function approximation are analysis. And then the principles of two approximate iterative algorithms are introduced, which are called approximate value iteration and approximate pol-icy iteration. Combining the characteristics of the unit commitment problems, seven basic concepts of approximate dynamic programming are defined. And on this basis, the formulations of decision function, contribution and approxi- mate value function are given by the introduction of an observing cost function. An approximate value function of the unit commitment problem using post-decision state variables is constructed.By using the proposed approximate value function, the approximate value iteration and approximate policy iteration are implemented with MAT-LAB-2013a on an IBM-compatible PC.10 to 100-unit systems over 24 time periods are calculated. The efficiency of the two approximate iterative algo-rithms for unit commitment problems are compared for the first time, and then suitability of the approximate policy iteration is pointed out.According to the characters of power system operation, a policy itera-tion-approximate dynamic programming method is presented for the large-scale unit commitment problems, since the computation time of original algorithm is too long. Firstly, the original approximate policy iteration is im-proved according to unit commitment problems. An algorithm suitable for unit commitment problems is proposed to update the constructed approximate value function. Detailed flow and implementation of this algorithm are dis-cussed. And the update to the approximate value functions of adjacent periods, the problems of ramping constraints as well as the shortage of traditional trun-cation technique without considering the minimum up/down time characteris-tics are studied. The concrete plans of updating the value of post-decision states by only one pre-decision state once a time, using dynamic limits of ac-tive power and an extended sequential truncation technique are put forward. The results of 10 to 1000-unit systems over 96 time periods verify the correct-ness and practicability of the method for solving large-scale unit commitment problems.Combined with the study of approximate dynamic programming theory and its application in solving day-ahead conventional unit commitment prob-lems, the proposed policy iteration-approximate dynamic programming method is extended to security-constrained unit commitment problems. Con-straints such as the power flow equations, transmission flow constraints, and limitations of the bus voltage make the difficulties and calculations of solving the problem increase significantly. A method of handling the power flow con-straints by adjusting the calculation of the contribution without changing the approximate policy iteration is proposed for this problem. Results of the IEEE 30 to 300-bus systems and the Polish 2737-bus system verify the correctness and practicability of the method for solving security-constrained unit commit-ment problems.This new research is the core product of the 5th subject of National 973 Program of China (2013CB228205) "collaborative optimization regulation of energy complementary between power supplies with different characteristics and load". The research work is jointly supported by the National 973 Program and the National Natural Science Foundation of China (51167001). The find-ings of the research offer reliable solutions and technical support on how to make generation schedule of a large-scale power system for energy saving. It not only has important application value in the direction of intelligent optimi-zation scheduling and control, but also expands a new field for the develop-ment of the approximate dynamic planning theory.