Power Lithium Battery Thermal Runaway Fault Classification and Risk Prediction Method and System

Abstract

The invention provides a power lithium battery thermal runaway fault classification and risk prediction method and system, including: a module-level power battery model fault injection method, a random fault generation and labeling method, and a power lithium-ion battery based on a deep learning method A multi-classification model for faults and a transfer learning method for applying the model to real vehicles. The invention can accurately express the real fault condition of the battery, and migrate to the specific actual vehicle working condition. The trained deep learning algorithm model can be successfully deployed in the real vehicle environment through mathematical processing and code conversion, and can diagnose faults in real time without adding additional calculations to the battery management system, while achieving high estimation accuracy.

Description

technical field [0001] The present invention relates to the technical field of transfer learning and electric vehicle battery management, in particular to a method and system for thermal runaway fault classification and risk prediction of power lithium batteries based on deep learning. Background technique [0002] In today's society, energy problems and environmental problems are becoming more and more serious, and new energy vehicles, especially pure electric vehicles, are gradually becoming the mainstream of the automotive industry. Lithium-ion batteries are an important core component of electric vehicles, and the battery management system (BMS) plays a role in ensuring the safe and stable operation of the battery. On-line monitoring and fault diagnosis of the power lithium-ion battery during the charging and discharging process of the electric vehicle power is the key point to ensure the stable and normal operation of the electric vehicle. [0003] Current fault diagno...

AI Technical Summary

Problems solved by technology

[0007] However, the above patents have the above two problems, one is the collection and labeling of fault data, and the other is the parallel computing and long-term dependence during model training

For the laboratory environment, it is easy to manually construct faults, such as constructing a short-term internal and external short circuit, setting the initial SOC, etc., and obtaining the fault label corresponding to the fault type, but for real vehicle data, especially for the cloud. quality data, apparently there is not enough fault data and their known labels to train deep learning algorithms

Therefore, satisfactory results cannot be obtained in the actual application process.

[0008] Secondly, for time series data, there is a problem of disappearing long-term dependence characteristics. For the general LSTM neural network, although a gating mechanism is introduced to suppress the problem of gradient disappearance and explosion, the information interaction distance between different time periods is in the time dimension. It is O(n), and parallel computing cannot be realized. When using massive data for prediction, the training performance is not good

Images