Improvement Of Smart Grid Security Based On Artificial Intelligence

Smart grid is a next-generation electrical power system including renewable energy,intelligent transmission technology, communications and information technology. The smartgrid is more stable and economical than the current power system. However, it also haspotential security problems. For example, the transmission lines may be broken down easilydue to the increase of the length and transmission voltage. The power consumption, whichaffected by the dynamic price of the electricity, may be fluctuated dramatically in a shortperiod of time. The large fluctuation causes the transmission unstable. As a result, the smartgrid security is a crucial problem. This thesis focuses on two major problems of the securityissues in the smart grid, which are the fault location of the transmission line, and theprediction of the price and the load values.Transmission line is one of the most important components in power system. It has ahigh fault rate due to the outdoor exposure. Even worse, locating the fault position is adifficult task because of the length of the transmission line. Consequently, the accurate faultposition play an important role in improving the security of the smart grid. The currentmethods using machine learning techniques estimates the fault position by using the signalcollected from one or two terminals of a transmission line. However, suffer from that thecollected signal less accurate when the distance between the terminal and the fault location islonger. Hence, the accuracy of the estimation is suffered from the inaccurate collected signal.This study proposes a fault location method which firstly determine which terminal is closestto the fault position. Then, the signal collected by the closest terminal is used to estimate thefault position. The experimental results are obtained by using the multi-terminals powersystem network simulated in Matlab.11different types of the faults are generated. Theextreme learning machine (ELM) is applied as the estimator. The relative error of proposedmethod has more significantly improvement than the existing methods.The second contribution of this thesis is the prediction of the price and the load. Theproblems of the price and the load prediction contain the huge amount of historical data.Choosing suitable subset of the features can improve the performance of prediction. Thisthesis propose the multivariable mutual information(MMI) and random forest based ensemblesystem based on feature selection for electricity price prediction and load forecasting are used.PJM, NYISIO and AEMO electricity market price and load dataset are used. Experimentalresults show that the proposed method is better and more stable.