Quantum Neural Network and Gaussian Process Framework for Lithium Battery State of Health Prediction

Abstract

Precise forecasting of lithium-ion battery State of Health (SOH) is crucial for effective prognostics and health management (PHM) to ensure safety, reliability, and optimal performance. The existing kernel functions in Gaussian process regression (GPR) exhibit limitations in capturing complex non-linear relationships within constrained datasets, resulting in suboptimal modeling of underlying data patterns and reduced forecasting efficiency. To address this gap, an innovative quantum neural network–Gaussian process regression (QNN–GPR) framework leverages quantum feature spaces to capture high-dimensional complexities and improve prediction accuracy in a three-step manner: (i) proposing three health indicators strongly correlated with SOH to characterize battery aging, (ii) designing QNN which leverages superposition and entanglement for processing multi-parametric battery information, yielding intermediate state estimations, and (iii) utilizing GPR as the subsequent stage to refine these intermediate predictions through probabilistic enhancement. Validation on four NASA battery datasets shows the QNN–GPR model attains an average 0.98% mean absolute error (MAE), indicating its superiority over conventional GPR for battery health management.