Feeder loss estimation of transformer in long-short memory network, based on FCM clustering

Abstract In order to improve the accuracy of estimating the feeder line loss rate in distribution networks and make it more effective for line maintenance management, a feeder line loss estimation method based on the fuzzy C-means clustering long short-term memory network Transformer model is proposed. Firstly, based on the two dimensions of data parameter availability and line loss correlation, a three-dimensional evaluation index for the feeder line loss rate of the distribution system was constructed. Fuzzy clustering technology was used to effectively classify the feeders, identify the benchmark feeders of each category, and preprocess the original data. Secondly, a line loss prediction model with a dual layer structure is introduced, in which the first layer adopts a gate mechanism of long short-term memory network, aiming to capture the dependency characteristics in the data sequence related to feeder line loss in the distribution network. The second layer integrates the multi head self attention mechanism of the Transformer model, and obtains prediction data by combining it with the characteristic data of distribution network feeder line loss, which can ensure the efficiency and accuracy of short-term distribution network feeder line loss prediction. Finally, to verify the effectiveness and practicality of the proposed method, an application analysis was conducted using the distribution network feeder of a power supply enterprise in a city in Guangdong Province as an actual case.