Online Prediction of Electric Vehicle Battery Failure Using LSTM Network

Xuemei Li, Hao Chang, Ruichao Wei, Shenshi Huang, Shaozhang Chen, Zhiwei He, Dongxu OuyangEnergies 2023 · DOI

TL;DR

This paper introduces an online prediction method for electric vehicle (EV) battery failures using Long Short-Term Memory (LSTM) networks. By analyzing real-world vehicle operation data, the model can predict low State of Charge (SOC) and insulation faults with over 85% accuracy, enhancing EV safety and monitoring capabilities.

What Problem Does This Paper Address?

The rapid growth of the electric vehicle industry, while crucial for energy transition, brings increased attention to power battery safety. Battery failures, particularly spontaneous combustions, pose a significant risk, with power batteries being the primary cause. Existing fault prediction methods are often experimental, rely on offline data, and lack real-time prediction capabilities. This paper aims to bridge this gap by developing an online, real-time prediction system for critical EV battery faults, specifically low SOC and insulation alarms, to improve safety and operational reliability.

Methods

The core of this research lies in leveraging Long Short-Term Memory (LSTM) networks for fault prediction. LSTMs are a type of recurrent neural network (RNN) adept at handling sequential data and overcoming the vanishing gradient problem inherent in traditional RNNs, making them suitable for time-series data like vehicle operation logs.

The methodology involves several key steps:

  1. Data Pre-processing: Raw vehicle operation data is collected, cleaned (outliers removed, missing values interpolated), and then dimensionless processed to ensure consistency.
  2. Relevance Analysis: Pearson correlation coefficients are calculated to identify battery parameters that are strongly correlated with specific faults. This helps in selecting the most relevant features for the prediction model.
  3. LSTM Model Training: The LSTM model is trained using the pre-processed and feature-selected data. The process involves optimizing hyperparameters such as window size, batch size, and epochs to achieve the lowest loss function.
  4. Fault Prediction:
    • Low SOC Alarm: The problem is reframed as predicting the future SOC value. The LSTM model predicts SOC based on correlated parameters (total voltage, max/min cell voltage). A low SOC alarm is then triggered if the predicted SOC falls below a defined threshold.
  5. Insulation Alarm: The insulation resistance value and the occurrence of a fault alarm are used as model inputs to predict the probability of a fault.
  6. Validation and Evaluation: The trained models are validated using test sets and real vehicle verification. Evaluation metrics like Max Error (ME), Mean Square Error (MSE), Root Mean Square Error (RMSE) for SOC prediction, and Accuracy (ACC), True Positive Rate (TPR), and False Positive Rate (FPR) for insulation alarm prediction are used.

Datasets

The study utilizes a large dataset from the Guangzhou new energy intelligent vehicle big data monitoring platform. This dataset comprises 6,568,070 data points collected between April and October 2022. The data includes 55 characteristics, such as battery operation data, drive motor data, and fault codes. The data was collected at a sampling frequency of 20 Hz during normal operation and 2 Hz during fault events. For the specific faults studied:

  • Low SOC Alarm: 1,060,962 data samples were used for SOC prediction.
  • Insulation Alarm: 7,681 total samples, with 87 fault data samples, were used.

The data was split into training and validation sets, typically with an 8:2 ratio.

Key Insights and Ideas

  • LSTM's Efficacy for EV Battery Faults: The study strongly demonstrates that LSTMs are highly effective for predicting EV battery failures, outperforming previous experimental and offline methods.
  • Reframing Fault Prediction: The paper cleverly reframes the low SOC alarm prediction as a direct SOC value prediction problem, simplifying the approach and leveraging LSTM's time-series forecasting capabilities.
  • Correlation is Key: Identifying strongly correlated parameters (like total voltage, max/min cell voltage for SOC) is crucial for building an accurate and efficient prediction model.
  • Handling Data Nuances: The research highlights the importance of data pre-processing, including outlier removal and interpolation, and even addresses subtle issues like data rounding in BMS that can affect fault detection.
  • Real-time Prediction Potential: The proposed method is designed for online prediction, meaning it can continuously monitor and alert drivers or maintenance systems before a critical failure occurs, offering a significant safety advantage.

Key Results

  • The LSTM-based model achieved an accuracy of over 85% for predicting both low SOC and insulation alarms.
  • For low SOC prediction, the model demonstrated low prediction errors, with ME, MSE, and RMSE values significantly below 0.4 for one vehicle type and below 0.2 for another. In one case, after rounding, 43 out of 45 predicted data points matched actual data points.
  • For insulation alarm prediction, the model showed a high True Positive Rate (TPR) of 87.50% and a low False Positive Rate (FPR) of 1.56%, indicating reliable fault detection.
  • The model can provide fault warnings approximately 150 seconds in advance, allowing drivers or systems to take preventative action.

Limitations and Future Work

  • Historical Data Dependency: The current study relies on historical vehicle operation data. The authors acknowledge that a crucial next step is to embed the model into an online supervision platform for real-time data processing and fault prediction.
  • Limited Fault Types: The study focused on two specific faults (low SOC and insulation). Future work should aim to expand the range of predictable faults and integrate other diagnostic methods to improve robustness.
  • Data Availability: While the dataset is large, it's from a specific region and platform. Generalizing the model to diverse EV models and operating conditions might require more extensive and varied datasets.
  • Real-time Implementation: The paper proposes an online prediction method, but the actual implementation and integration into a live EV monitoring system is a future endeavor.

Summary updated

Abstract

The electric vehicle industry is developing rapidly as part of the global energy structure transformation, which has increased the importance of overcoming power battery safety issues. In this paper, first, we study the relationship between different types of vehicle faults and battery data based on the actual vehicle operation data in the big data supervisory platform of new energy vehicles. Second, we propose a method to realize the online prediction of electric vehicle battery faults, based on a Long Short-Term Memory (LSTM). Third, we carry out prediction research for two kinds of faults: low State of Charge (SOC) alarm and insulation alarm. Last, we show via experimental results that the model based on the LSTM network can effectively predict battery faults with an accuracy of more than 85%. Through this research, it is possible to complete online pre-processing of vehicle operation data and fault prediction of power batteries, improve vehicle monitoring capabilities and ensure the safety of electric vehicle use.
BibTeX
@inproceedings{li2023online,
  author = {Xuemei Li and Hao Chang and Ruichao Wei and Shenshi Huang and Shaozhang Chen and Zhiwei He and Dongxu Ouyang},
  title = {{Online Prediction of Electric Vehicle Battery Failure Using LSTM Network}},
  booktitle = {Energies},
  year = {2023},
  doi = {10.3390/en16124733},
}
Figure