Research On Dynamic Equivalent Modeling Of Distributed Photovoltaic Cluster Based On Clustering And Deep Learning Algorithm

With the introduction of the " carbon peak,carbon neutrality " policy,the country has accelerated the pace of building a new power system with clean energy as the main body.Distributed photovoltaic power stations have developed rapidly with their own flexibility advantages and the help of national policies.A large-scale distributed photovoltaic cluster has been formed in the current distribution network.The large number of power electronic devices contained in photovoltaic clusters and their inherent intermittency,randomness,and volatility have brought many challenges to the safe and stable operation of the distribution network.Constructing an accurate and efficient dynamic equivalent model of distributed photovoltaic clusters to achieve accurate control of the dynamic characteristics of photovoltaic power stations in the distribution network is an important basis for the current distribution network planning and design,operation control,system maintenance and other work.However,due to the large number of power stations in the distributed photovoltaic cluster and the large difference in dynamic characteristics,the traditional modeling method is difficult to solve the contradiction between model accuracy and simulation rapidity.In view of the above problems,this paper studies the dynamic equivalent modeling of distributed photovoltaic cluster in distribution network.The main work is as follows :(1)The grid-connected dynamic model of single-machine two-stage photovoltaic power station is established.The common grid-connected structure at this stage is analyzed,and a representative two-stage photovoltaic power station is selected as the research object.Modeling is carried out around several core devices such as photovoltaic array,DC / DC converter,DC / AC inverter and filter.The simplified mathematical model of photovoltaic array is introduced to further improve the practicability of the model in the engineering field while reducing the complexity of the model.Through simulation experiments,it is verified that the established dynamic model has a certain low voltage ride through capability,and based on this,the dynamic characteristics of photovoltaic gridconnected are analyzed.(2)A dynamic equivalent modeling method of distributed photovoltaic cluster based on Kmedoids clustering is proposed.Firstly,by analyzing the dynamic characteristics of grid-connected photovoltaic power stations,appropriate parameters are selected as clustering indicators,and Dynamic Time Warping(DTW)is introduced to dynamically transform the clustering features to reduce the adverse effects of high-dimensional data.Secondly,the equivalent treatment of parameters such as capacity,line impedance and load of photovoltaic power station after grouping is studied,and the dynamic equivalent model of the whole cluster is established.Finally,based on the improved IEEE33 distribution network system,the dynamic characteristics of the detailed model and the dynamic equivalent model under different working conditions are compared.The results show that the dynamic equivalent model greatly improves the simulation speed.(3)A distributed photovoltaic cluster fusion model based on clustering equivalence-error correction fusion is proposed.Firstly,a fusion model framework based on clustering equivalenceerror correction is built.The framework mainly includes a dynamic equivalent model based on clustering algorithm,an error correction model based on deep learning,and an error correction strategy combining the two.Secondly,an error correction model based on Gate Recurrent Unit(GRU)is proposed.Finally,the historical data of the detailed model and the dynamic equivalent model are collected to construct the data set to train the error correction model.The trained error correction model is combined with the dynamic equivalent model according to the proposed framework to achieve error correction.The proposed error correction strategy is verified in different typical scenarios.The results show that the proposed fusion model framework not only retains the advantages of high simulation efficiency of the clustering equivalent model,but also more accurately reflects the operating characteristics of the detailed model.