Research On Reliability Assessment Models And Algorithms Of Power Systems Based On Soft Computing Theory

With the forming of electricity market and the increasing constantly of demand for electric power, people put forward the higher request to the electric power quantity. At the same time, the increase in the risk of transmission congestion caused by the uncertainty of demand for electric power and decision-making dispersedly of IPP make it difficult to know the operation conditions of power systems in advance, and these systems will be close to the condition under which some components in the systems carry the limit load. These bring operation securityof power systems in danger. Thus, the research on reliability assessment and load prediction of bulk power systems has great significance of both theory and practice. From internal and external information and following closely advanced technology, reliability assessment theories and procedures of bulk power systems are systematically studied and the results show that the "computation catastrophe or calamity"of reliability assessment is synthetically caused by a few complicated calculation processes, in hard computing methods. Therefore, to alleviate the "computation catastrophe" soft computing methods from each link which causes the "computation catastrophe" should be designed. For this reason, we have studied in depth the basic logic, and principles of rough sets---the newest soft computing theories, artificial neural network and genetic algorithms. Soft computing models and algorithms of reliability assessment and load prediction for bulk power systems are presented by means of integrating rough set, artificial neural network and genetic algorithm organically. The numerical experiments for a practical system, and the results tested on the RBTS and IEEE-RTS79 show the correctness, feasibility and usefulness of the proposed soft computing models and algorithms. The main results in detail include the following:1. In terms of analyzing the dependency of decision attributes on condition attributes, the thesis proves the conclusions as follows:(1) The reducing of condition attribute set doesn't make the positive region become increscent ;( 2) The monotone increasing characteristic of the dependency degree of decision attribute: The reducing of condition attribute set doesn't make the dependency degree of decision attribute increscent. (3)The minimum condition attribute set which keeps the dependency degree changeless is just the minimum reducing set of condition attribute set. 2. A valid reducing method of condition attribute set----Depth Approach Reducing Method (DARM) has been put forward so as to provide technology support for the application of rough set theories in power system reliability assessment and load forecasting.3. Some new concepts-----rough classification and rough event classes of stochastic events have been presented to establish soft computing models used to identify stochastic events pattern.4. Rough sets theories are originally employed to study power system reliability assessment in this thesis. By the strong and weak coupling, a rough neural network integrated model (RNNIM) applying to stochastic events pattern identification is presented. A genetic algorithm is designed to learn the network parameters with a nonlinear error function. A new method is offered to decrease the times of power flow calculation of identified stochastic events in bulk power systems reliability evaluation and to identify whether a stochastic event belongs to a specified contingency pattern quickly and effectively.5. Soft computing models of load forecasting of power systems are studies through rough sets theories to benefit evaluating reliability of power systems considering load model1) The concept of condition attribute sensitivity based on information entropy is given. A rough composite forecasting method (RCFM) is proposed on the foundation of DARM. An effective way is supplied for choosing prediction models and composite coefficients under data driving.2) For multifactor nonlinear correlation load forecasting problems, making use of rough s...