A gaussian process model with indirect health indicators for battery prognosis

Yinjia Li, Sara Kohtz, Pingfeng Wang

Research output: Chapter in Book/Report/Conference proceedingConference contribution


This paper focuses on predicting lithium battery capacity and state of health based on pre-recorded datasets. Data-driven techniques, including machine learning approaches have become prevalent. However, data is often inputted into these methods without much thought. Therefore, data pre-processing before utilizing machine learning techniques can optimize the performance and efficiency of these models. Since the battery degradation process is highly nonlinear and influenced by multiple factors, including both manufacturing aspects and operating conditions, which complicate the battery capacity prognostics, we need some tool that can keep track on useful data and produce reliable predictions in the end [1]. Specifically, the main methodology of this study includes defining indirect health indicators as inputs into two machine learning techniques: Gaussian Process Regression (GPR) and Feed Forward Neural Network. To demonstrate this method, the NASA battery dataset [2] is utilized; in particular, the indirect health indicators are extracted from the battery's temperature and charging state (the voltage and current change process), then a machine learning model is utilized to optimize prediction of capacity degradation.The first method for predicting the State of Health (SOH) of a battery is Gaussian Process Regression, which modifies the isotropic squared exponential kernel with an automatic relevance determination structure. The entire process extracts the highly relevant input features for capacity predictions [1]. Utilizing this method enables quick extraction of relevant data and provides fast predictions of capacity.The second strategy is Feedforward Neural Networking (FFNN). Overall, FFNN is a widely used machine learning method with outstanding performance in nonlinear modeling. It first intake one input as the starting layer, and each layer except the output layer is fully connected to the next layer, so the final prediction of FFNN is tightly connected with the previous data [3].However, the inputs to these advanced models need to be considered, as irrelevant inputs can lead to inaccurate predictions (see figure 1). Therefore, indirect health indicator (IHI) is applied as a form of feature extraction as an input to the model. Indirect health indicator is what affects the capacity of a battery during the charging or discharging state. During those processes, the performance of lithium-ion batteries will deteriorate with capacity decreasing and impedance increasing, which will cause equipment and system failures or even catastrophic loss [4]. Therefore, it is necessary to consider the IHI before predicting the model.For this study, indirect health indicators are determined with data pre-processing and are subsequently inputted into the GPR and FFNN. The results show success for capacity estimation for lithium ion batteries.

Original languageEnglish (US)
Title of host publication2023 Annual Reliability and Maintainability Symposium, RAMS 2023
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781665460538
StatePublished - 2023
Event2023 Annual Reliability and Maintainability Symposium, RAMS 2023 - Orlando, United States
Duration: Jan 23 2023Jan 26 2023

Publication series

NameProceedings - Annual Reliability and Maintainability Symposium
ISSN (Print)0149-144X


Conference2023 Annual Reliability and Maintainability Symposium, RAMS 2023
Country/TerritoryUnited States


  • Gaussian Process Regression
  • Machine Learning
  • Neural Network
  • SOH Estimation

ASJC Scopus subject areas

  • Safety, Risk, Reliability and Quality
  • General Mathematics
  • Computer Science Applications


Dive into the research topics of 'A gaussian process model with indirect health indicators for battery prognosis'. Together they form a unique fingerprint.

Cite this