Power System Transient Stability Assessment Based On Artificial Neural Network

Security and stability assessment is a key issue for power system operation. Thetraditional time domain simulation method and transient energy function method aredifficult to satisfy both the quickness and accuracy requirements in power systemtransient stability assessment (TSA). At the same time, they can’t provide operatorsenough information to help them to find proper control strategy and to make decisionquickly. TSA method based on artificial neural network (ANN) has many goodfeatures, such as self learning, fast speed and clear physical concept. Further, it canprovide operators some useful information between system failure data and operatingstate, so it can be another choice for power system online security assessment.For any TSA method based on neural network, how to select effective ANN’sinputs and how to optimize ANN’s structure are its two key points, since they havesignificant influence to the effectiveness and precision of this type method. In thisthesis, these two proplems are deeply studied and discussed. Main work of this thesisis as follows:1) A new method for power system TSA based on compound neural network isproposed. It combines probabilistic neural network (PNN) with radial basis functionnetwork (RBF) and uses PNN network to classify the sample data in advance, thenuses RBF network to predict the security margin of different classfications. Thetraining sample data in RBF sub-network takes false classification of PNN intoconsideration, and the approach adopts the overlap boundary classification method toreduce errors and revises the results to improve predicting accuracy. The givenmethod makes full use of the advantages of two networks and improves the transientstability judgment and predictive ability effectively.2) A method to optimally select ANN’s inputs based on sensitivity information isproposed. Sensitivity index between the input features and network output is defined.And, a numerical method based on perturbation is given to calculate such index.Sensitivity values with their gaps are used to determine the critical inputs. It can befound that the given method can remove the redundant inputs so as to improve theperformance of the given assessment method.3) IEEE-39nodes and IEEE-118nodes systems are selected to validate the correctness and effectiveness of the proposed method.