Research On Methods For Fault Diagnosis Of Power Grids Based On Computational Intelligence

When a fault occurs in a power grid, a large number of operated protective relays (PRs) and circuit breakers (CBs)(referred to as alarm messages) with uncertainties will flood into the dispatching control center through the SCADA system or the fault information system. Using these alarm messages to quickly and accurately diagnose the faulted sections is of significance in restoring the power supply and reducing the power cut loss. Fault diagnosis of power grids has experienced decades of development and gained delightful achievements. With the increasing development of modern power system, the growing scale and the increasing complexity of the structure put forward higher requirements on the operation level of power system. In this context, existing fault diagnosis methods are confronted with serious challenges in the terms of application scope, diagnostic accuracy and precision. In light of this, this thesis is dedicated to the study of fault diagnosis methods and the emphasis is placed on the use of computational intelligence (CI). Additionally, since one of the CI methods, i.e., biogeography-based optimization (BBO) shows high performance on the fault diagnosis of power grids, it is also used to solve the economic dispatch problems. The major works and achievements of this thesis are as follows:A new method for fault diagnosis of power grids using fuzzy reasoning spiking neural P systems is proposed. This method is able to not only simulate the cause-effect behaviors between PRs and CBs by graphical approach, but also perform the fuzzy reasoning among candidate faulted section, PRs, and CBs by means of a simple matrix operation, and thus help dispatchers analyse the fault clearance process. The simulations show that this method has characteristics of capability in handling uncertainties and rapidity in reasoning.A divisional fault diagnosis method for power grids based on radial basis function (RBF) neural network and fuzzy integral is presented. It aims at effectively combating the "curse of dimensionality" when neural networks are applied to large-scale power grids. An overlapping network division method is proposed to divide a power grid into a number of eligible sub-networks. When a fault occurs, local RBF neural network diagnostic modules will be selectively triggered according to the local alarm information. If it involves a tie line, a fuzzy integral fusion module will be then triggered to fuse the diagnostic outputs derived from connected sub-networks about the tie line. The diagnostic results demonstrate that this method, based on the "divide and conquer" idea, is efficient in diagnosing various faults involving tie lines with high diagnostic accuracy and adapts to large-scale power grids well. A method based on RBF neural network for fuzzy cellular fault diagnosis of power grids is proposed. It aims at solving the transportability problem of accommodating the network topology changes when applying neural networks to fault diagnosis of power grids.This method directly takes all the associated PRs and CBs used to protect the section as inputs to establish generalized cellular neural network diagnostic model. And the fuzzy reasoning rules containing uncertainties, which are extracted from a sagittal diagram, are used to train the RBF neural network. The results show that the method accommodats the network topology changes well through a simple modification of the diagnostic model. It can solve different complex faults with good fault tolerance and transportability.An improved analytic model for fault diagnosis of power grids and a self-adaptive BBO (SaBBO) are proposed. After analysing the reason of multi-solution problem within the existing analytic model, a corresponding improved analytic model is proposed. Moreover, an efficient method based on SaBBO is presented to solve it. The simulations reflect that the diagnostic results of the improved model are unique and more reasonable, and SaBBO is characterized by less iterations, high reliability, and good performance.A multi-strategy ensemble BBO (MsEBBO) for the static economic dispatch (SED) problem of power system is proposed. Multiple strategies are employed to improve BBO’s migration model, migration operator, and mutation operator, to balance its global search ability and local search ability. In addition, an effective constraint handling technique without penalty factor settings is developed to handle various complicated constraints of the SED problem. The experimental results confirm that these improved components can significantly enhance the performance of MsEBBO by mutual beneficial cooperation.An enhanced BBO based on polyphyletic migration operator and orthogonal learning (OL) strategy is proposed to solve the dynamic economic dispatch (DED) problem. On one hand, the polyphyletic migration operator can not only generate new features from broader areas in the search space, but also effectively increase the population diversity. On the other hand, instead of the previous blind search behaviour, the OL strategy based on orthogonal experimental design can directionally guide the algorithm toward the global optimum. In addition, an effective constraint handling technique without penalty factor settings is also developed. The results show that this algorithm is able to skip poor local optima and converge to more promising area, and thus achieve better economic dispatch scheme.