Cuckoo Search Based Relevance Vector Machine with Hybrid Kernel for Battery Remaining Useful Life Prediction

The study of lithium-ion battery management has always been a hot issue in energy field. Predicting the remaining useful life (RUL) helps to know the time of maintenance and replacement in advance so that users can be avoided from unexpected battery fault. In this paper, a RUL prediction method on the basis of a sparse probability model in the Bayesian framework which is Relevance Vector Machine (RVM) is proposed. Hybrid kernel function is introduced to ensure the prediction accuracy. A newly heuristic algorithm Cuckoo Search (CS) is adopted to intelligently determine all the kernel parameters along with the weight of each kernel function. Contrast experiments show that our proposed approach based on hybrid kernel RVM with CS (CS-HKRVM) outperforms the basic machine learning algorithms including single kernel RVM, Back Propagation Neural Network and Support Vector Machine. The hybrid kernel RVM with particle swarm optimization (PSO) is also conducted in order to demonstrate that CS is more effective and accurate than PSO in searching for the optimal solution of prediction model construction.