EV battery degradation forecasting based on high dimensional data
Examensarbete för masterexamen
This study is made to investigate the benefit of using high-dimensional data from sensors in electric vehicles (EVs) in State-of-Health (SOH) forecasting. By the term "High dimensional data", it means the data has a considerably large dimension that the computation on the data is very difficult or very time consuming without a proper handling. EV is one of the biggest trends in the automobile industry in the past few years. One of the biggest concerns about the EVs is their lithium-ion battery. The lithium-ion batteries will degrade over time based on several factors such as calendar ageing, accumulated charge, temperature, and etc. These factors can lead the Li-ion batteries to degrade, which can lead to several problems for the EVs. The study is made on the Diagnostic Read Out (DRO) dataset from real-world Volvo Cars’ customers. Several machine learning models were used to predict and analyzed to find the significant causes of battery degradation. Principal Component Analysis (PCA) was used to reduce the dimension of data before fitting it into the prediction models. For the results, this study found that Long Short-Term Memory (LSTM) is the most suitable machine learning model for SOH forecasting from all machine learning models considered in this research. Moreover, applying PCA to the data significantly improves the models’ performances than using the whole data without dimensionality reduction. Furthermore, adding a suitable number of lag features to the input also increases the models’ performance considerably. Last but not least, SOC and the temperature while cranking and starting the engine of the hybrid EVs are considered to be important to the battery degradation problem.
Lithium-ion Battery degradation , Electric Vehicle , State-of-Health forecasting , Machine learning , Linear Regression , Recurrent Neural Network , Random Forest Regression , Long Short-Term Memory , Principal Component Analysis