The Calculation Of Static Voltage Stability Margin Based On Parallel Neural Networks

Power system voltage collapse is an important cause of voltage instability in the power system.The influence of voltage instability often leads to huge economic losses.The voltage margin index is a measure of the voltage collapse of the power system.It is an important indicator of voltage stability research.This thesis combines the existing methods of calculating the margin index with the neural network calculation method to achieve the goal of quickly obtaining a voltage stability margin.The work of this thesis mainly includes the following parts.The second chapter briefly summarizes the theories of static voltage stability,including the classification of static voltage stability indicators and its current research progress,the related theory of continuous power flow method and its application in power systems.In the third chapter,aiming at the problem that the calculation speed is slow and difficult to apply to large-scale power systems in the process of solving voltage stability margin based on PV curve,a parallel wavelet neural network is proposed,which replaces the traditional continuous power flow with the training and calculation process of neural network.By improving the learning rate of wavelet neural network and the adjustment of wavelet basis function,the calculation accuracy of wavelet neural network is improved.Finally,the simulation proves the parallel wavelet neural network proposed in this thesis.The effectiveness of the network is verified.Chapter 4 proposes a method which aims at the shortcomings of the extreme learning machine when the voltage stability margin calculation is not ideal.Since the calculation accuracy of the extreme learning machine is affected by the number of hidden layer nodes,and the continuous power flow method is difficult to meet the real-time requirements,therefore,a parallel extreme learning machine is introduced to find the hidden nodes of extreme learning machine,thus the accuracy is improved.The fifth chapter proposes a parallel sequential extreme learning machine.Although the learning time of the extreme learning machine is not long,but it is necessary to re-train the data whenever there is a large amount of data entering,and its real-time performance is still difficult to meet the requirements of online applications.Therefore,it is proposed to use a sequential extreme learning machine to predict the voltage stability margin.The simulation comparison shows that the calculation accuracy of the sequential extreme learning machine is greatly affected by the value of each newly added data,the length of the historical data and the number of hidden layer nodes.Therefore,this thesis proposes the method of parallel sequential extreme learning machine,it avoids the problem that the calculation speed is slow or even not converged due to improper selection of the step size in the calculation process of the continuous power flow method,finally the proposed method realize fast and accurate calculation.