| With the rapid development of new energy and micro grid technology,more and more non-linear loads are connected to micro grid,which increases the difficulty of power quality management.At the same time,the increasing high-precision instruments put forward higher requirements for power quality.The Unified Power Quality Controller(UPQC)in the microgrid can comprehensively improve various power quality problems and increase the penetration rate of new energy.First,based on the analysis of the structure,mathematical model and power quality management principle of the UPQC system,a corresponding simulation model was built.Then several conventional methods of command signal detection were introduced,and through simulation,it is found that the instruction signal detection method based on ip-iq method has high detection accuracy and load following ability,but due to the need for multiple coordinate transformations and other reasons,this method has time-delay problems in the detection process,which affects the governance effect.Aimed at the problem of compensation delay,a harmonic current extraction scheme combining radial basis neural network(RBF)and ip-iq method was proposed.The radial basis function neural network was used to fit the current compensation deviation caused by the delay in the process of detection by ip-iq.Training and generating a time-delay harmonic compensation network,which worked together with the ip-iq method to compensate the command current output by the ip-iq method in real time and improved the extraction accuracy of the command current signal.The simulation results show that the proposed detection scheme can effectively reduce the lag effect of ip-iq method.It is found that the structure of RBF neural network has a direct impact on the compensation effect.Obtaining the best number of hidden layer neurons can reduce the complexity of the algorithm while ensure the current compensation accuracy.The structure of traditional neural network is selected by human experience.In this paper,a random configuration neural network which can automatically adjust the number of neurons according to the fitting deviation was proposed to realize the autonomous optimization of the time-delay harmonic compensation neural network.Simulation results show that,compared with the ip-iq method based on RBF neural network,the ip-iq method based on random configuration neural network can quickly and accurately obtain the optimal neural network structure,with higher precision of instruction signal extraction,which can offset the lag of ip-iq harmonic extraction and reduce the harmonic content of power grid current to a greater extent. |
