Equivalent Modeling of Distributed Photovoltaic Power Station Clusters Based on Deep Belief Network

摘要

This paper presents a novel equivalent modeling method for distributed photovoltaic (PV) power station clusters. Deep Learning (DL) is proposed to PV power station clusters modeling and the training algorithm is Deep Belief Network (DBN) made up of multiple layers of restricted Boltzmann machines (RBM). A DBN algorithm is applied to dynamic equivalent modeling of PV clusters after first-step clustering using the improved K-means algorithm. The input variables of the neural network are irradiance variation, voltage fluctuation, reactive power reference of the dual-loop controller, the output active and reactive power of PV clusters. The output variables are the output active and reactive power of PV clusters. The datasets are obtained through 6560 experiments. Then a layer-by-layer unsupervised learning method is used to pre-train the network followed by fine-tuning the parameters using a supervised back-propagation (BP) method. Finally, the equivalent model of PV clusters based on DBN is built and applied to a typical distribution network in Anhui Province. The validity and accuracy of the proposed model are verified in three different disturbance cases. At the same time, the computational complexity and the simulation time are reduced using the proposed model significantly.