CN-121995235-A - Physical information transducer-based lithium ion battery health state estimation method

CN121995235ACN 121995235 ACN121995235 ACN 121995235ACN-121995235-A

Abstract

The invention discloses a physical information transducer-based lithium ion battery health state estimation method which comprises the steps of obtaining voltage and current time sequence data of a lithium ion battery under a pulse charging condition, extracting characteristic parameters from the time sequence data, constructing a training data set containing the characteristic parameters and corresponding health state labels, constructing a physical information neural network model based on a transducer architecture, inputting the training data set into the physical information neural network model for training, processing the pulse charging data of the lithium ion battery to be tested by using the physical information neural network model after training, and outputting a battery health state estimation result. The invention makes an important contribution to building the battery health prediction model with higher interpretability, better data efficiency and stronger physical credibility by building a framework with strong principle and popularization.

Inventors

LIN LIN
LI XIAOXIA
WANG SHUNLI
YANG XIAO

Assignees

西南科技大学

Dates

Publication Date: 20260508
Application Date: 20260313

Claims (8)

1. The lithium ion battery health state estimation method based on the physical information transducer is characterized by comprising the following steps of: acquiring voltage and current time sequence data of the lithium ion battery under the pulse charging condition; extracting characteristic parameters from the time sequence data, and constructing a training data set containing the characteristic parameters and corresponding health state labels; And constructing a physical information neural network model based on a transducer architecture, inputting the training data set into the physical information neural network model for training, processing pulse charging data of the lithium ion battery to be tested by using the trained physical information neural network model, and outputting a battery health state estimation result.
2. The method for estimating a state of health of a lithium ion battery based on a physical information transducer according to claim 1, wherein extracting characteristic parameters from the time series data comprises: Calculating an ohmic polarization voltage jump value according to the voltage transient response at the current step moment, and acquiring ohmic polarization characteristics according to the ohmic polarization voltage jump value; fitting a relaxation phase voltage recovery process by adopting a third-order exponential relaxation model to extract a relaxation time constant, and acquiring relaxation dynamics characteristics according to the relaxation time constant; Constructing a differential capacity curve through numerical differentiation and Savitzky-Golay filtering, and acquiring differential capacity characteristics according to the differential capacity curve; Calculating charging energy demand characteristics based on voltage envelope integration, and acquiring envelope energy characteristics according to the charging energy demand characteristics; And obtaining the characteristic parameters based on the ohmic polarization characteristic, the relaxation dynamics characteristic, the differential capacity characteristic and the envelope energy characteristic.
3. The method for estimating the state of health of the lithium ion battery based on the physical information transducer according to claim 1, wherein the physical information neural network model comprises a feature embedding and position encoding module, a physical information attention encoding module, a physical constraint fusion module and an SOH regression decoding module; The feature embedding and position coding module is used for mapping input features to a high-dimensional space, and capturing sequence information and actual time interval information simultaneously by adopting bimodal position coding; the physical information attention coding module is used for processing according to the output characteristics of the characteristic embedding and position coding module to acquire data driving characteristics; the physical constraint fusion module is used for realizing the deep interactive fusion of the physical constraint characteristics and the data driving characteristics; The SOH regression decoding module is used for mapping the decrypted time sequence characteristics into a single SOH value.
4. The method for estimating a state of health of a lithium ion battery according to claim 3, wherein the physical information attention encoding module comprises a multi-head attention unit and a physical constraint feedforward unit; the multi-head attention unit introduces a physically directed attention bias matrix in a standard self-attention mechanism; the physical constraint feedforward unit adds a physical regularization term based on electrochemical constraint in a feedforward network.
5. The method for estimating a state of health of a lithium-ion battery based on a physical information transducer of claim 4, wherein introducing a physically directed attention bias matrix in a standard self-attention mechanism comprises: constructing a first bias term according to the time adjacency, constructing a second bias term based on the monotonous attenuation constraint of the battery health state, constructing a third bias term according to the characteristic similarity, fusing the three biases to form an attention bias matrix, adding the attention bias matrix with the attention scores of the query key value pairs, and inputting the attention bias matrix into a Softmax function.
6. The method for estimating a state of health of a lithium ion battery based on a physical information transducer according to claim 3, wherein implementing the deep interaction fusion of the physical constraint feature and the data-driven feature comprises: Calculating physical constraint characteristics based on the SEI film growth model, the lithium inventory loss model and the active material loss model; and splicing the data driving features and the physical constraint features, and realizing the deep interaction fusion of the physical constraint features and the data driving features through a multi-layer perceptron network.
7. The method for estimating a state of health of a lithium ion battery based on a physical information transducer of claim 1, wherein training the physical information neural network model further comprises: Performing auxiliary training by using a composite loss function, wherein the composite loss function comprises a main task loss, a physical consistency loss and a regularization loss, the main task loss adopts a mean square error to measure the difference between a predicted value and a label value, the physical consistency loss comprises a health state monotonically decreasing constraint term and a degradation rate matching constraint term, and a AdamW optimizer is adopted to update model parameters.
8. The method for estimating a state of health of a lithium ion battery based on a physical information transducer according to claim 7, wherein said updating model parameters with AdamW optimizer comprises: And executing a two-stage training strategy, wherein the first stage uses the synthetic data generated based on the electrochemical model to pretrain the physical constraint fusion module, and the second stage uses the real experimental data to perform end-to-end fine tuning on the whole model, and adopts an adaptive constraint intensity mechanism to dynamically adjust the weight coefficient of the physical constraint in the training process.

Description

Physical information transducer-based lithium ion battery health state estimation method Technical Field The invention belongs to the technical field of battery health state estimation, and particularly relates to a lithium ion battery health state estimation method based on physical information converters. Background The accurate prediction of the health state of the lithium ion battery is important to ensure the safety, reliability and economy of wide technologies from portable electronic products, electric automobiles to power grid-level energy storage systems and the like. With the growing global demand for efficient, robust energy storage solutions, developing robust and accurate SOH estimation methods has become the focus of research in academia and industry. Conventional SOH monitoring methods can be broadly divided into experimental-based and model-based techniques. While electrochemical models, such as the open-ended Doyle-Fuller-Newman model, can provide profound physical insight into the internal dynamics of the battery, their practical use in real-time battery management systems is often hindered by great computational complexity and challenges in identifying numerous model parameters under varying operating conditions. The advent of data-driven methods brings about a paradigm shift that uses machine learning algorithms to learn directly from complex nonlinear relationships between operational data and battery degradation. Techniques ranging from support vector machines to complex deep neural networks have demonstrated excellent capabilities in capturing aging patterns from historical cyclic data. However, there are two fundamental limitations to these purely data-driven approaches. First, they typically operate as a "black box" model, providing predictions that lack physical interpretability, which is critical to diagnostic and predictive decisions in safety critical applications. Second, their performance is severely dependent on acquiring massive, high-fidelity aging datasets covering different stress factors and battery lots. This craving for data severely constrains its generalization ability and places a significant hurdle to model deployment, especially for new battery chemistry systems or when the operational history is limited. Meanwhile, pulse current charging (PCC-Pulse Current Charging) is not only widely studied for its potential to extend battery cycle life by inhibiting detrimental aging mechanisms such as lithium precipitation, but it is also of great interest as a rich dynamic data source that is highly sensitive to internal electrochemical states. The characteristic voltage transients and relaxation phases in the PCC process encode valuable information regarding kinetic limitations, concentration polarization, and internal impedance evolution. This makes PCC not just a charging strategy, but an effective diagnostic tool, providing fertile soil for feature extraction intended for State of Health (SOH-State of Health) estimation. To address the limitations of existing methods, a hybrid paradigm of collaborative fusion data-driven learning and domain-specific physical knowledge has become a new front with broad prospects. The physical information neural network framework originally proposed by Raissi et al has shown significant success in embedding the physical system control law, typically described by partial differential equations, into the loss function of a deep learning model. However, the application of such deep learning of physical information to analysis of long sequences of battery data, which is complex to operate, especially in the study of PCC conditions, remains essentially blank. Furthermore, while the Transformer architecture with self-attention mechanisms has drastically changed sequence modeling in the field of natural language processing, etc., its potential to interpret time dependencies in battery cycle data under physical constraints has not been fully exploited. Disclosure of Invention In order to solve the technical problems, the invention provides a lithium ion battery health state estimation method based on physical information converters, which makes an important contribution to building a battery health prediction model with higher interpretability, better data efficiency and higher physical credibility by building a framework with strong principle and popularization. In order to achieve the above object, the present invention provides a method for estimating the state of health of a lithium ion battery based on a physical information transducer, comprising: acquiring voltage and current time sequence data of the lithium ion battery under the pulse charging condition; extracting characteristic parameters from the time sequence data, and constructing a training data set containing the characteristic parameters and corresponding health state labels; And constructing a physical information neural network model based on a transducer architecture, inputting