Abstract: Power hub nodes are aggregation nodes that can be traded uniformly in the spot market, and they are also the cornerstone of the development and stable operation of the power financial market, which is of great significance to the construction of a unified power market system and the improvement of market functions. This paper analyzes the design of hub nodes from the perspectives of general attributes, node selection, weight design, and temporality and purposefully proposes the method of dividing hub areas based on spectral clustering and calculating weights based on a graph attention network. The core of the dual-scale metric-based spectral clustering partitioning strategy is to establish a dual-scale coupling matrix with key line power transmission distribution factor, multi-time nodes tariff, and nodes adjacency as metrics to prioritize the geographical characteristics of the hub area, the non-blocking characteristics within the area and nodes connectivity. Based on the graph attention network, a pricing nodes aggregation weight assignment model is established to achieve adaptive matching of node weights under different time slot nodes tariffs by using a self-attention mechanism to aggregate the neighboring nodes. Finally, the proposed design method is validated by IEEE 118 nodes arithmetic example, and the rationality and effectiveness of this paper's method is proved through cross-sectional comparison, which provides a reasonable basis for hub nodes setting in the actual power market.