PREDICTING THE REMAINING USEFUL LIFETIME OF LITHIUM BATTERIES USING MACHINE LEARNING – PAUL KITCHING

For this project, two machine learning regression models were used to create two web services, which are consumed by an application. The first web service uses linear regression and the second web service uses decision forest regression. The application sends battery data to the web services, which returns predictions on the remaining useful lifetime of the batteries.

The accuracy of the predictions was calculated to be 78% for linear regression and 75% for decision forest regression. It was observed that, as a general rule, batteries that had a smaller remaining useful lifetime would have more accurate predictions.

Regression was found to be a suitable approach to predicting and expressing the remaining lifetimes of batteries. Through analysing the weights of the parameters available in the dataset, a number of parameters were able to be removed without significantly impacting the accuracy of the predictions. This allows a streamlined model and is especially useful for batch requests, where the amount of information being sent is significantly reduced.

Skip to content