Power System Load Clustering Integrating Attention Mechanism and Contrastive Learning

To address the limitations of traditional clustering methods in power load analysis, such as inaccurate classification, complex parameter settings, and poor generalization capabilities, this study proposes a novel power load clustering approach integrating enhanced attention mechanisms and contrastive learning (CL-IAM). Initially, a residual neural network (ResNet) combined with an attention mechanism is utilized to extract features from power load data. Subsequently, these features are mapped into a contrastive loss space using a multi-layer perceptron (MLP) network, with cosine similarity employed to measure the similarity between data points, thereby accomplishing the clustering task. Experimental results demonstrate that this method significantly improves the precision and efficiency of power load clustering, providing robust support for subsequent power load analysis.