Load Prediction Based Remaining Discharge Energy Estimation Using a Combined Online and Offline Prediction Framework

Remaining discharge energy (RDE) indicates how much useful energy can be extracted from a battery before reaching the discharge limit. Future current loading on vehicle battery systems can be predicted to increase the accuracy of RDE estimations. This is done by using clustering techniques to group load measurements into states, and then using a probability-based framework, along with real-world data, to calculate the transitional probabilities between states. Here, an adapted K-means clustering method is used to cluster load profile data. Markov modelling is used to produce state transition probabilities. Two methods for load prediction are used, which are referred to as the offline-training method and the moving window method, where the offline-training method has not been implemented for this application before. Additional control logic is implemented to combine the proposed load prediction methods to produce a new hybrid load prediction method. This hybrid method shows improved RDE accuracy for a generalised load case. The robustness of the proposed technique is assessed in the presence of model errors, still showing good accuracy when compared to state-of-charge based calculations.