Artificial intelligence for personalized services in power marketing information systems

This paper presents an improved Transformer model with a dynamic gated attention mechanism that can predict power loads more accurately and computationally efficiently, especially in large-scale scenarios. To address the challenge of dynamic user behavior, we propose a heterogeneous graph neural network to simulate user interaction and consumption patterns to achieve accurate user clustering. Based on these clustering results, we developed a multilevel intelligent customer service system. The experimental results show that the framework improves the accuracy of user analysis, reduces the operating cost, reduces the manual workload, and improves the intelligence degree of the power marketing information system.