The analysis of intelligent university performance cost accounting driven by big data

The traditional performance cost accounting model of higher education shows obvious lag in dealing with the dynamic allocation of resources and multidimensional performance evaluation, so it is necessary to integrate data intelligence to improve accounting accuracy and management efficiency. Firstly, based on the data set related to education finance published by the Education Economics Laboratory, a heterogeneous graph modeling mechanism is designed to reconstruct the relationship between multiple node types, including teachers, courses, funds and outputs, and form a dynamic knowledge graph sequence. Secondly, by combining the Graph Attention Network (GAT) with the transformer architecture, a cross-time and cross-type performance cost representation model is constructed to support the teaching expenditure prediction and performance fitting analysis. Finally, an anomaly detection mechanism combining Autoencoder and Local Outlier Factor (AE + LOF) is introduced, and a closed-loop system is formed by optimizing budget allocation based on reinforcement learning strategy network. The results show that the prediction accuracy of the GAT model is better than that of the four baseline methods. In the annual scenario, the Root Mean Square Error (RMSE) and the mean absolute percentage error (MAPE) are reduced by 1.52% and 1.63% respectively, which has stronger error stability. In anomaly detection, the average F1 score of AE + LOF framework reaches 0.907, the false positive rate is controlled within 4.3%, and the sample fluctuation of adaptive institutions is less than 3.6%. In budget optimization experiments, the reinforcement learning policy achieved a ΔPerf/ΔCost of 0.178, with a performance improvement rate of 21.5% under high-performance preference conditions. System tests showed end-to-end latency below 200 ms and self-learning convergence within 18 epochs, with the highest user satisfaction rating in “interactivity.” Under the constraints of the data and simulation conditions of this study, graph-structured modeling and intelligent embedding mechanisms hold the potential to enhance the level of intelligence in performance-based cost accounting.