Resource Optimization Method Based on Spatio-Temporal Modeling in a Complex Cluster Environment for Electric Vehicle Charging Scenarios

In intelligent cluster systems, the spatio-temporal complexity of agent data collection and resource allocation, as well as the problems in collaborative organizations, present substantial challenges to efficient resource distribution. To address this, a novel self-organizing prediction method for spatio-temporal resource allocation is proposed. In the spatio-temporal modeling part, dilated convolution is applied for time modeling. Its dilation rate grows exponentially with the layer depth, allowing it to effectively capture the time trends of graph nodes and handle long time series data. For spatial modeling, an innovative dual-view dynamic graph convolutional network architecture is utilized to accurately explore the static and dynamic correlation information of the spatial layout of charging piles. Meanwhile, a composite self-organizing mechanism integrating a trust model is put forward. The trust model assists agents in choosing partners, and the Q-learning algorithm of the intelligent cluster realizes the independent evaluation of rewards and the optimization of relationship adaptation. In the experimental scenario of electric vehicle charging, considering charging piles as agents, under the home charging mode, the self-organizing charging scheduling can reduce the total load range by up to 90.37%. It effectively shifts the load demand from peak periods to valley periods, minimizes the total peak–valley load difference, and significantly improves the security and reliability of the microgrid, thus providing a practical solution for resource allocation in intelligent clusters.