Quantum-Enhanced Anomaly Detection for Supply Chain Disruption Risk in Renewable Energy Component Procurement: An Empirical Comparison of Variational Quantum Circuits and Classical Machine Learning Baselines

Abstract

Renewable energy supply chains — dependent on geographically concentrated raw materials such as lithium, cobalt, rare earths, and polysilicon — are increasingly exposed to procurement disruption risk from geopolitical shocks, commodity price volatility, and demand shifts tied to climate policy. This study develops and empirically tests a hybrid quantum-classical anomaly detection framework using a Variational Quantum Circuit (VQC) to flag early-warning disruption signals in renewable energy procurement data, and benchmarks it against three classical baselines: Isolation Forest, One-Class SVM, and an LSTM Autoencoder. A monthly panel (2008–2025, n = 216) combining the Caldara–Iacoviello Geopolitical Risk Index, World Bank Pink Sheet commodity prices, and IRENA renewable capacity data was engineered into 147 rolling-window features and reduced via PCA to six components for a 6-qubit angle-encoded circuit, implemented independently in PennyLane and Qiskit. Using three documented macro-disruption windows (2008–09 Global Financial Crisis, COVID-19 onset, and the 2022 Russia-Ukraine invasion) as ground truth, the VQC was evaluated against the classical baselines on precision, recall, F1-score, false-positive rate, per-typology recall, noise robustness, and computational cost. Contrary to the study's initial hypotheses, Isolation Forest achieved the highest F1-score (0.593) among all five models, modestly outperforming both VQC implementations (F1 = 0.556). The VQC showed no measurable robustness advantage over Isolation Forest at any tested noise level (0–50% of feature standard deviation) and required approximately 1,230 times longer to train than Isolation Forest for a comparable or weaker result. All models, including the VQC, failed to detect the COVID-19-era logistics disruption window. These findings suggest that, on this dataset and circuit design, near-term quantum machine learning does not yet offer a measurable advantage over well-tuned classical anomaly detectors for renewable energy procurement risk monitoring, and the paper discusses the practical and methodological implications of this null result for procurement risk management and future QML research.