Soft Computing Methodology And Its Some Applications In Power System Fault Diagnosis

Electrical power system is a very complicated wide-area system with interconnected reticular networks, and often several kilometers extensions in length and breadth. Diverse types of equipments, such as switches, facilities, components, hardware and software pieces connect one another and interweave together within the gigantic system, thus a large-scale exoteric heterogeneous system is for that born. On the other hand, it is impossible for electrical energy to be stored during production and consumption, as at any time, the gross generated electrical energy is equal to the gross consumed one. Hence, electrical power system is a complex system of great trait. The open heterogeneous system is considered to be a complex one, and the solution under indeterminate and uncertain environment has long been a research core of related fields. In addition, with the complexities and opening increasingly, electrical power system is facing many new problems on all sides, e.g., equipment fault diagnosis and maintenance, reactive power control, load forecasting and electrical dispatch and so on. Conventional methods or some new emergent ones aren't very effective to these problems, and invariably expose many problems, more or less. However, soft computing (SC) methods but provide a quite effective support to resolve the puzzles, and can be applied to deal with those problems that it is impossible for conventional and new emergent ones to resolve. Hence, new ubiquitous ideals and algorithms suitable to special fielded problem-solving must be found in light of concrete problem of electrical power system.Based on methodology and theory of soft computing such as rough set, fuzzy set and neural network, and etc, in this thesis we proposed and discussd some soft computing methods and their combined models, and how to synergistically tackle practical problems in electrical power system. Moreover, we also investigated their innovative applications in some new fields and answer some cognitive puzzles that we could't understand them well before. This includes several aspects below.Considering the fuzziness of fault symptom information and complicated mapping relations between fault-masses and symptom-masses in process of transformer fault diagnosis, a new fault diagnosis method was proposed based on fuzzy set, rough set and evidence theory.Fuzzy set was embedded into Bayesian optimal classifier, another approach to resolve the above problem then was found. Vague set then was also embedded into Bayesian optimal classifier to generate such a Bayesian optimal classifier it could dispose the two facets of information, including positive or nagitive one, as well as indeterminate one, simultaneously. In the end, the likely maintenance strategies were investigated using probalistic rough model on Bayesian risk decision-making with the diagnostic results based.Based on rough set and other soft computing methods, substation fault information was hierarchically mined and diagnosed. Firstly, the overall substation fault area was divided into several independent sub-fault areas so as to reduce the scale of solution and improve reliability and response ability in real time as well as serviceability. Moreover, anti-interference capability was also particularly concerned. Meantime, transparency and explainable ability of diagnostic process were stressed too.The continuous attribute values of diagnosis decision table of stream turbine might be quantified based on Kohonen network, a simplified decision table then was generated by rough sets reduction, in the end, the ensemble neural network were applied to implement fault diagnosis.