Distribution power system state estimation based on Gaussian mixture model-Neural network

Abstract

The main problem in the state estimation of distribution power system is that there are many nodes but few measuring points such that they are unobservable. With the construction and development of distribution network, most measuring devices of distribution power system have covered all nodes, but uploading their measured values to the power dispatching center in real time will take up a lot of communication resources, and once the data cannot be uploaded due to network congestion and other problems, the state estimation will be impossible to calculate. In this paper, the load Gaussian mixture model is established, and the load model under different scenarios is constructed. Obtain the load data from the smart meter, train the load model, and upload the model parameters to the power dispatching center where the neural network is trained with data such as node injection power generated by each node load model. Finally, the trained neural network is used to calculate the voltage and amplitude of each node. When some measure data is missing, the measure data generated by the compound model of the node stored in the power dispatching center is used as pseudo-measure for state estimation. The smart meter will update the training model regularly according to the change of node load, which helps to improve the robustness of the system. Compared with the traditional method of using power prediction as a pseudo-measurement, this method has the advantages of fast computing speed, high computing accuracy, small consumption of communication resources and strong robustness.