Battery (electricity)

Quantum mechanics

Engineering

Lithium (medication)

Reliability engineering

Statistics

Support vector machine

Endocrinology

Power (physics)

Computer science

Relevance vector machine

Physics

Nuclear engineering

Organic chemistry

Randomness

Mathematics

Environmental science

Artificial intelligence

Medicine

Chemistry

Artificial neural network

Machine learning

Energy

Lithium-ion batteries are currently being widely used. Accurately predicting their remaining useful life (RUL) is essential for battery management systems (BMS) and rationally planning the battery usage. There exist problems such as battery capacity regeneration and randomness caused by single time prediction and parameter settings. This paper proposes a hybrid algorithm that combines the broad learning system (BLS) with the relevance vector machine (RVM). First, use the empirical mode decomposition (EMD) to extract the features of the used data. Then input the training data into the BLS network and set different prediction starting points, and the corresponding prediction data is output. All prediction data is formed into a matrix to train the RVM. The RVM is used as the prediction layer of the hybrid model. Eventually, the RVM's output is the RUL prediction of the hybrid model. In this paper, the proposed method is experimentally validated using Li-ion battery experimental data from three sources, and its accuracy is compared with several common machine learning algorithms. Experimental results show that BLS-RVM has higher prediction accuracy, stronger long-term prediction, and generalization capabilities, and its root mean square error is about 0.01. The algorithm proposed in this paper for multiple training and prediction followed by fusion of the results broadens the research horizon of lithium-ion battery life hybrid methods for prediction.

Lithium-ion batteries remaining useful life prediction based on BLS-RVM