iTransformer-based Blockchain Supply Chain Anomaly Detection and Forecasting System

This paper proposes a blockchain-based supply chain anomaly detection and forecasting system built upon the iTransformer architecture, which achieves precise prediction and anomaly identification of supply chain time series data through a reverse attention mechanism. The system leverages blockchain technology to enhance supply chain transparency and integrates deep learning models for intelligent analytics. Experimental results show that the iTransformer achieves a MAPE value of 7.923 in time series forecasting tasks, significantly outperforming the traditional ARIMA model’s 12.513. Using LIME interpretability analysis, key factors such as contract status, delivery time, and environmental temperature and humidity were identified as having a significant impact on prediction outcomes. The system demonstrates substantial practical benefits with 45% reduction in false positive rates and real-time anomaly detection capabilities, reducing response time from traditional 24-48 hours to minutes. The proposed system provides an effective solution for anomaly detection and forecasting in supply chain management, with considerable practical value.