An intelligent load-feeder matching method of distribution networks based on an improved SAX algorithm and Bayesian hyperparameter optimization

Abstract

As the operation mode of distribution networks is adjusted more and more frequently in the new-type power system， the load-feeder matching of distribution networks meets with difficulties such as high-dimensional heterogeneity and low value density of sampled data， high dependence of existing matching algorithms on physical characteristics of load， and poor flexibility of parameter settings， etc. To this end， an intelligent load-feeder matching method of distribution networks based on an improved SAX （symbolic aggregation approximation） algorithm and Bayesian hyperparameter optimization is proposed. Firstly， a data value enhancement model for discrete symbolic time data series is established to approximate the high-dimensional heterogeneous data into low-dimensional uniform symbols to correct and fill the abnormal data and blank data. Secondly， an improved CNN-LSTM （convolutional neural network-long short-term memory） hybrid neural network is constructed to train the load data as per the matching， and the multi-headed attention is employed to explore the potential mathematical relationships between load and data to curtail dependency on physical features of load. Thirdly， a Bayesian hyperparameter optimization algorithm is introduced to update the neural network training parameters one by one to improve the flexibility and adaptability of the neural network model in the case of feeder topology changes. Finally， the proposed method is experimentally validated by matching between load and 100 feeders in a region. The results prove that the proposed method is superior to the traditional method in matching accuracy.

Keywords

• improved sax
• bayesian hyperparameter optimization
• multi-head attention
• improved cnn-lstm
• load-feeder matching