| Distribution network is situated in the end of power system,it is an essential part for connecting the transmission system to users.The observability and controllability of distribution system are the cornerstones for providing sustainable energy to users.The observability and controllability of distribution systems depend critically on accurately knowing its topology and line parameters.During the operation of distribution network,the topology and line parameters are evolved due to distribution network reconfiguration,fault isolation,etc.However,because of management factors,these changes are poorly recorded in some parts of the distribution system,which has a low degree of automation.Directly obtaining these dynamic topologies and parameters through manual calculation and verification requires high costs,so distribution system urgently needs a data-driven approach to accurately recognize the dynamic topology and estimate line parameters.In this paper,the medium-voltage distribution network is taken as the research object.Based on the distribution network operation data provided by smart meters,data-driven distribution network topology recognition and line parameter estimation models are researched:Firstly,to recognize the real-time topology of distribution system,this paper converts the topology recognition to interconnection recognition of distribution transformers.A data-driven topology recognition model based on ensemble multi-layer perceptron(MLP)is proposed.Considering the smart meter data contains noise,the two-dimension wavelet threshold de-noising method is implemented on historical distribution transformer measurements.After that,this paper analyzes the influence factors of the interconnection of distribution transformers.Based on the knowledge of power systems and statistics,an informative and non-redundant feature extraction method is proposed,which generates the input of MLP.To improve the accuracy of MLP classifier,the grid search method and ensemble learning strategy are implemented to improve the generalization ability and efficiency of MLP.Finally,experiments based on a real-world distribution network topology are implemented on the Tensor Flow platform and verified the proposed ensemble MLP model.Toaccurately estimate the line parametersof distribution system,this paper proposes a Dynamic Bayesian Network(DBN)based distribution line parameters estimation model.The causal relationship between influencing factors and line parameters was determined according to Joules law and power flow formula.The influencing factor data of the line parameters come from two different data sources: smart meters and weather stations.The multi-source raw data are preprocessed on the Python platform and generate the training set.Considering that there may be missing values in the training set,the expectation maximization(EM)algorithm is used to train the dynamic Bayesian model.The belief propagation(BP)algorithm is used to infer distribution line parameters with the DBN model.The accuracy of the proposed method is investigated by carrying out experiments on a real-world medium voltage distribution network using pgmpy.Finally,this paper researches the application of distribution network topology recognition and line parameter estimation.Targeted on the scenario of locating distributed generators,a dynamic multi-objective optimization model is constructed with the objective functions of voltage deviation rate,voltage fluctuation rate,and circuit net loss.Considering that the position cannot be changed after the PV is connected,the probabilistic power flow based on the Montelaro simulation is used to convert the dynamic problem into multiple static problems to solve.Latin hypercube sampling is used to optimize the selection of testing samples.Given that the objective function is not large,the vector evaluated genetic algorithm(VEGA)is used to obtain the Pareto solution set of the multi-objective optimization model.Finally,under the condition of dynamic distribution network topology and line parameters,this paper implements experimets on a medium-voltage distribution network using pandapower and geatpy modules.By comparing the adaptability of the access schemes,the rationality of the photovoltaic power access optimization algorithm is verified. |
