Abstract: To meet the research needs for the evaluation index system of software-defined command and control systems capable of diverse tasks, a prediction method is proposed for the correlation relationships of the dynamic evaluation index network based on Transformer. By establishing a dynamic index network based on the interrelationships of each index, the predictions is generated for future index networks based on the evolutionary patterns of dynamic index networks, supporting further downstream research on the dynamic index networks, including the mining of key indices, etc. The simulation experiment results on two kinds of different data sets show that the algorithm can maintain high accuracy and stability under both the transductive setting with fixed index quantity and inductive setting with
the index quantity increase with the system evolution, compared with the traditional graph neural network, demonstrating the wide applicability of the algorithm.