A lithium battery over-temperature fault risk assessment method and system based on dynamic impedance spectrum

Abstract

This invention discloses a method and system for assessing over-temperature fault risk in lithium batteries based on dynamic impedance spectroscopy. The method includes: simultaneously acquiring the dynamic impedance sequence and surface geometric center temperature sequence of the lithium battery under multiple operating conditions; extracting the time lag Δt between the abrupt change in dynamic impedance characteristics and the surface geometric center temperature response, advancing the surface geometric center temperature sequence by Δt time on the time axis to reconstruct the internal core temperature sequence of the lithium battery and use it as a training label; constructing a feature dataset and training an impedance-internal temperature soft measurement model; during online assessment, inputting the battery's dynamic impedance and external operating parameters in real time, and having the model directly output the predicted internal core temperature value at the current moment without delay, and executing graded over-temperature risk control accordingly. This invention achieves accurate soft measurement of internal temperature without the need for implanted internal sensors, completely eliminating the negative impact of surface temperature conduction lag on early warning.

Description

Technical Field

[0001] This invention belongs to the field of lithium battery testing and safety early warning technology, specifically a method and system for assessing the risk of lithium battery over-temperature faults based on dynamic impedance spectroscopy. Background Technology

[0002] With the large-scale application of lithium batteries in electric vehicles and energy storage systems, their thermal safety has become a focus of industry attention. If the heat generated during the charging and discharging process of lithium batteries cannot be dissipated in time, it will cause the internal temperature to rise sharply, leading to serious safety accidents such as capacity decay, shortened lifespan, and even thermal runaway.

[0003] Traditional lithium battery over-temperature monitoring primarily relies on temperature sensors placed on the battery casing surface. However, this monitoring method has significant limitations: First, due to the thermal conductivity limitations of the internal materials of lithium batteries, it takes a certain amount of time (denoted as time lag Δt) for internal heat to conduct to the battery surface. By the time the surface sensor detects a temperature anomaly, the core temperature inside the battery has usually already far exceeded the safety threshold. This physical lag results in a severe delay in the warning signal, causing the management system to miss the optimal intervention opportunity. Second, although implanted sensors (such as miniature thermocouples or fiber optic sensors) can acquire the actual internal temperature in real time, their installation process requires damaging the battery packaging structure and increases the battery's size, cost, and manufacturing complexity, making it difficult to promote and apply in large-scale mass production scenarios such as passenger vehicles. Finally, electrochemical impedance spectroscopy (EIS) is highly sensitive to temperature changes and is considered an ideal indicator for characterizing battery state. However, traditional EIS measurements usually require the battery to be in a steady state, and the testing cycle is long, making it difficult to adapt to frequently fluctuating dynamic operating conditions.

[0004] More importantly, existing soft measurement techniques fail to fully utilize the physical relationship between the immediacy of impedance response to internal temperature and the hysteresis of surface temperature response. If internal heating signs can be keenly detected through impedance characteristics, and the internal core temperature can be calculated in reverse by combining the surface temperature hysteresis Δt, then accurate early identification of internal overheating faults can be achieved without damaging the battery structure.

[0005] Therefore, developing a method for non-destructively reconstructing internal temperature tags using dynamic impedance spectroscopy and achieving delay-free risk assessment has significant application value. Summary of the Invention

[0006] To address the core shortcomings of existing technologies: First, traditional surface temperature monitoring suffers from thermal conduction lag, meaning that by the time a surface temperature alarm is triggered, the internal temperature of the battery has already far exceeded the safety threshold, missing the optimal intervention opportunity; second, implanted internal temperature sensors require damage to the battery packaging, making them unsuitable for mass production scenarios; third, traditional electrochemical impedance spectroscopy requires the battery to be in a static steady state, making it unsuitable for dynamic charge and discharge conditions; and fourth, existing internal temperature soft measurement technologies lack real internal temperature monitoring labels, resulting in low model prediction accuracy and poor generalization ability. This invention provides a lithium battery overheating fault risk assessment method based on dynamic impedance spectroscopy, which can achieve early and accurate identification of internal overheating faults without damaging the battery structure.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for assessing the risk of over-temperature faults in lithium batteries based on dynamic impedance spectroscopy, comprising: S1. In offline testing, acquire dynamic impedance sequence data of lithium battery under multiple operating conditions, and simultaneously acquire the surface geometric center temperature sequence, charge / discharge rate sequence and ambient temperature sequence of lithium battery in real time; extract the time lag Δt between dynamic impedance feature mutation and center temperature response; advance the surface geometric center temperature sequence by Δt time on the time axis to non-destructively reconstruct the core temperature sequence inside the lithium battery and use it as a training label. S2. Align the dynamic impedance sequence, surface geometric center temperature sequence, charge / discharge rate sequence, and ambient temperature sequence according to timestamps, and combine them with training labels to construct a multi-dimensional time-series feature dataset; S3. Based on the multi-dimensional time-series feature dataset, train a deep learning network to establish a nonlinear mapping relationship between dynamic impedance, surface geometric center temperature, charge / discharge rate, ambient temperature and internal core temperature, and obtain an impedance-internal temperature soft measurement model. S4. During actual operation of the lithium battery, the dynamic impedance, surface geometric center temperature, charge / discharge rate, and ambient temperature within a specified time window are acquired in real time and input into the impedance-internal temperature soft measurement model. The model outputs the real-time predicted value T of the lithium battery's internal core temperature at the current moment. 内温 According to T 内温 Conduct real-time over-temperature risk assessment.

[0008] This invention cleverly utilizes the hysteresis time Δt to reconstruct the internal temperature tag in reverse, achieving accurate soft measurement of internal temperature without the need for implanting internal sensors, and completely eliminating the negative impact of surface temperature conduction hysteresis on early warning.

[0009] Furthermore, the real-time predicted internal core temperature T output by the model is... 内温 Compared with the preset battery internal over-temperature safety threshold T 阈值 Compare the data, classify the risk levels of overheating, and implement tiered control strategies.

[0010] Furthermore, when the real-time predicted internal core temperature T 内温 At: T 阈值 -10℃<T 内温 ≤T 阈值 -5℃ is determined to be a low-risk over-temperature condition. The system triggers a minor warning and increases the sampling frequency of the dynamic impedance spectrum.

[0011] Furthermore, when the real-time predicted internal core temperature T 阈值 -5℃<T 内温 ≤T 阈值 If the temperature is deemed to be at medium risk due to overheating, the battery management system will be immediately adjusted to reduce the charging and discharging power.

[0012] Furthermore, when the real-time predicted internal core temperature T 内温 >T 阈值 If the temperature is deemed to be at high risk of overheating, the main battery circuit will be immediately cut off, and the liquid cooling system will be forcibly activated.

[0013] Furthermore, in step S1, the dynamic impedance value of the lithium battery is obtained by the following method: firstly, the response voltage of the battery is simultaneously acquired using a four-probe method. U r With excitation current i e The sampling voltage formed on the precision sampling resistor U s This eliminates the interference of line voltage drop and lead impedance on the measurement results in principle; among which the excitation current i e Through a 10mΩ precision sampling resistor R s Linear conversion to sampled voltage U s , and response voltage U r For a typical strong DC and weak AC signal, the DC component is first removed by a voltage follower unit to avoid amplification clipping distortion and phase shift error caused by residual DC voltage; subsequently... U s and U r Both signals are converted to single-ended signals by an instrumentation amplifier with the same architecture, and then amplified by an independent programmable amplifier with identical parameters to achieve adjustable gain of 10 / 100 / 1000 times. This ensures that the frequency response, temperature drift, and phase drift of the two signals are consistent to the greatest extent, eliminating phase deviation introduced by the amplification stage. The amplified signals are then converted to digital using a synchronous sampling ADC (analog-to-digital converter) to ensure the time synchronization of the voltage and current sequences and avoid phase errors caused by asynchronous conversion. Finally, the amplitude conversion of the converted data is performed to obtain the complete signal. i e andU r The sequence is used to extract the voltage and current components at the target frequency through Fourier transform, and the battery impedance value at the corresponding frequency is calculated by vector division.

[0014] Furthermore, a high-precision K-type thermocouple sensor is arranged at the geometric center of the lithium battery surface to independently collect temperature data in real time, and the sensor accuracy is ±0.2℃.

[0015] Furthermore, the multi-dimensional temporal feature dataset undergoes the following preprocessing steps before model construction: Outlier removal: Abnormal deviations caused by sensor noise are removed using the 3σ principle to ensure the continuity and smoothness of time-series feature data; Normalization: The min-max normalization method is used to uniformly map multi-dimensional time-series feature data and training labels to the [0,1] interval, eliminating the negative impact of different physical dimensions on the gradient descent of deep learning networks.

[0016] Furthermore, the structure of the impedance-internal temperature soft measurement model is as follows: a bidirectional long short-term memory network (BiLSTM) combined with an attention mechanism is used as the deep learning network; the input layer receives the dynamic impedance sequence, the surface geometric center temperature sequence, the charge / discharge rate sequence, and the ambient temperature sequence; the feature extraction layer processes the input layer data simultaneously through two forward and backward LSTM layers to mine the physical mapping law between the dynamic impedance features and the internal core temperature; the attention layer assigns weights to the input layer data through a multi-head attention mechanism, giving higher weights to moments of drastic change in the dynamic impedance sequence; the fully connected layer uses the ReLU activation function to deeply integrate the features output by the attention layer; and the output layer outputs the real-time predicted value of the target lithium battery's internal core temperature at the current moment.

[0017] This invention also provides a lithium battery over-temperature fault risk assessment system based on dynamic impedance spectroscopy, used to implement the above-mentioned lithium battery over-temperature fault risk assessment method, comprising: Offline data acquisition unit: In offline testing mode, acquires dynamic impedance sequence data of lithium battery under multiple operating conditions, and simultaneously acquires the surface geometric center temperature sequence, charge / discharge rate sequence and ambient temperature sequence of lithium battery in real time; Internal core temperature sequence reconstruction unit: Extracts the time lag Δt between the abrupt change in dynamic impedance characteristics and the temperature response at the surface geometric center; advances the surface geometric center temperature sequence by Δt time on the time axis to non-destructively reconstruct the internal core temperature sequence of the lithium battery and use it as a training label. Dataset construction unit: used to align the dynamic impedance sequence, surface geometric center temperature sequence, charge / discharge rate sequence and ambient temperature sequence by timestamp, and combine them with training labels to construct a multi-dimensional time-series feature dataset; Impedance-internal temperature soft measurement model construction unit: Based on the multi-dimensional time-series feature dataset, a deep learning network is trained to establish a nonlinear mapping relationship between dynamic impedance, surface geometric center temperature, charge / discharge rate, ambient temperature and internal core temperature, and thus obtains the impedance-internal temperature soft measurement model. Online risk assessment unit: During actual operation of the lithium battery, the dynamic impedance, surface geometric center temperature, charge / discharge rate, and ambient temperature are acquired in real time within a specified time window and input into the impedance-internal temperature soft measurement model. The model outputs the real-time predicted value T of the lithium battery's internal core temperature at the current moment. 内温 According to T 内温 Conduct real-time over-temperature risk assessment.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Breaking through the core bottleneck of internal temperature soft measurement, this invention enables non-destructive reconstruction of realistic internal temperature tags. Utilizing the instantaneous response characteristics of dynamic impedance to internal temperature changes, this invention extracts the time lag Δt between impedance characteristic abrupt changes and the surface temperature response. Without implanting internal sensors or damaging the battery packaging structure, it can reverse-engineer high-precision internal core temperature sequence tags, completely solving the industry pain point of existing internal temperature soft measurement technologies lacking realistic supervisory tags, and providing a reliable physical benchmark for model training.

[0019] 2. Completely eliminates thermal conduction hysteresis, achieving zero-delay early warning of over-temperature faults. This invention constructs an impedance-internal temperature soft measurement model based on dynamic impedance characteristics. During online operation, it can directly output the current core temperature of the battery, without waiting for internal heat to be conducted to the battery surface. This fundamentally solves the fatal flaw of traditional surface temperature monitoring and early warning hysteresis, providing a golden window for thermal runaway intervention and significantly improving the timeliness of early warning.

[0020] 3. Adaptable to all dynamic operating conditions and with strong generalization ability. The dynamic impedance spectroscopy online measurement technology adopted in this invention breaks through the limitation of traditional electrochemical impedance spectroscopy requiring the battery to be in a static steady state. It can complete the real-time acquisition of impedance data under all dynamic operating conditions of battery charging and discharging. At the same time, based on the deep learning network of BiLSTM + attention mechanism, it can deeply explore the nonlinear mapping law between impedance characteristics and internal temperature under different ambient temperatures and different charge and discharge rates, and has excellent internal temperature prediction accuracy under all operating conditions.

[0021] 4. Hierarchical closed-loop control with strong engineering applicability. This invention sets up a three-level over-temperature risk hierarchical control strategy based on the internal core temperature prediction value, realizing closed-loop control of the entire process from early warning and power regulation to emergency shutdown; at the same time, no modification is required to the battery structure or production line process, which can be achieved only through surface sensors and impedance measurement modules, making the cost controllable. It can be directly integrated into the existing battery management system and is perfectly adapted to large-scale mass production application scenarios such as electric vehicles and energy storage power stations. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart of a lithium battery over-temperature fault risk assessment method based on dynamic impedance spectroscopy according to the present invention. Figure 2 This is a schematic diagram of the real-time lithium battery surface temperature acquisition system of the present invention; Figure 3 This is a detailed flowchart of a lithium battery over-temperature fault risk assessment method based on dynamic impedance spectroscopy according to the present invention. Figure 4 This is a schematic diagram of a lithium battery over-temperature fault risk assessment system based on dynamic impedance spectroscopy according to the present invention. Detailed Implementation

[0024] Specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings. While specific embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0025] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings, and the drawings do not constitute a limitation on the embodiments of the present invention.

[0026] Example 1 This embodiment presents a method for assessing the risk of over-temperature faults in lithium batteries based on dynamic impedance spectroscopy. Figure 1 As shown, it includes: S1. In offline testing, acquire dynamic impedance sequence data of the lithium battery under multiple operating conditions, and simultaneously acquire the surface geometric center temperature sequence, charge / discharge rate sequence, and ambient temperature sequence of the lithium battery in real time; extract the time lag Δt between the abrupt change in dynamic impedance characteristics and the surface geometric center temperature response; advance the surface geometric center temperature sequence by Δt on the time axis to non-destructively reconstruct the core temperature sequence inside the lithium battery and use it as a training label. The specific details are as follows: During the offline testing phase, a multi-condition simulation test platform was built. High and low temperature constant temperature chambers were used to simulate the ambient temperature of the entire scenario, and the ambient temperature gradients were set to -10℃, 0℃, 10℃, 25℃, 40℃, and 55℃. A battery charge and discharge test system was used to simulate the full-condition charge and discharge scenarios of the vehicle, with charge and discharge rates set to 0.3C, 0.5C, 1C, 2C, and 3C, covering all scenarios such as normal driving, fast charging, and high-rate discharge, and a multi-condition cross-test matrix was constructed.

[0027] During the test, two synchronous data acquisition systems were used, such as Figure 2 As shown, time synchronization acquisition of multi-source data is completed: The first system is a dynamic impedance acquisition system (i.e.) Figure 2 The impedance meter in the system employs a high-frequency synchronous excitation-response measurement architecture. The specific acquisition process is as follows: First, the response voltage of the battery is synchronously acquired using a four-probe method. U r With excitation current i e The sampling voltage formed on the precision sampling resistor U s This eliminates the interference of line voltage drop and lead impedance on the measurement results in principle; among which the excitation current i e Through a 10mΩ precision sampling resistor R s Linear conversion to sampled voltage U s , and response voltage U r For a typical strong DC and weak AC signal, the DC component is first removed by a voltage follower unit to avoid amplification clipping distortion and phase shift error caused by residual DC voltage; subsequently... U s and U rBoth signals are converted to single-ended signals by an instrumentation amplifier with the same architecture, and then amplified by an independent programmable amplifier with identical parameters to achieve adjustable gain of 10 / 100 / 1000 times. This ensures that the frequency response, temperature drift, and phase drift of the two signals are consistent to the greatest extent, eliminating phase deviation introduced by the amplification stage. The amplified signals are then converted to digital using a synchronous sampling ADC (analog-to-digital converter) to ensure the time synchronization of the voltage and current sequences and avoid phase errors caused by asynchronous conversion. Finally, the amplitude conversion of the converted data is performed to obtain the complete signal. i e and U r The sequence is used to extract the voltage and current components at the target frequency through Fourier transform, and the battery impedance value at the corresponding frequency is calculated by vector division.

[0028] The second system is the operating condition and temperature acquisition system, located at the geometric center of the surface of the lithium battery under test (i.e., Figure 2 A high-precision K-type thermocouple sensor is attached to the central temperature sampling point of the battery. The sensor accuracy reaches ±0.2℃. The temperature recorder collects the temperature sequence of the geometric center of the battery surface in real time. At the same time, the charge and discharge rate sequence is collected synchronously through the charge and discharge test system, and the ambient temperature sequence is collected synchronously through the constant temperature chamber. All data are marked with a unified timestamp to ensure complete synchronization with the timing of the dynamic impedance sequence.

[0029] After completing the full-process data acquisition under multiple operating conditions, the internal core temperature label is reconstructed. The specific process is as follows: Extraction of time lag Δt: Using cross-correlation analysis, the moment when the dynamic impedance characteristics change abruptly is located. t 1, and the time when the surface geometric center temperature shows a corresponding response. t 2. The time lag Δt is calculated as follows: t 2- t 1; At the same time, the variation law of Δt under different charge and discharge rates and different ambient temperatures was obtained by fitting the single-factor variable method to ensure the full-condition adaptability of Δt.

[0030] Internal core temperature sequence reconstruction: The surface geometric center temperature sequence is shifted forward by Δt time on the time axis. The reconstruction formula is:

[0031] T 内温 T represents the real-time predicted internal core temperature of a lithium battery. 表面 This is the temperature at the geometric center of the lithium battery surface.

[0032] This non-destructively reconstructs the core temperature sequence inside the lithium battery and uses it as training labels.

[0033] S2. Align the dynamic impedance sequence, surface geometric center temperature sequence, charge / discharge rate sequence, and ambient temperature sequence according to timestamps, and combine them with training labels to construct a multi-dimensional time-series feature dataset.

[0034] After the multi-dimensional time-series feature dataset is constructed, two preprocessing steps are performed sequentially: Outlier removal: The 3σ principle is used to identify and remove abnormal deviations caused by sensor noise and signal interference. The missing values ​​after removal are filled in by linear interpolation to ensure the continuity and smoothness of time series data. Normalization: The min-max normalization method is used to map all feature data and internal core temperature labels to the [0,1] interval. The normalization formula is: x'=(xx min ) / (x max -x min Where x is the original data, x' is the normalized data, and x' is the normalized data. min x max These are the minimum and maximum values ​​of the feature dimension, respectively, to eliminate the negative impact of different physical dimensions on model training and improve the model's convergence speed and accuracy.

[0035] After preprocessing, the dataset is randomly divided into training, validation, and test sets in a ratio of 7:1.5:1.5. The training set is used for model parameter learning, the validation set is used for hyperparameter optimization and overfitting control, and the test set is used for final model performance verification.

[0036] S3. Based on the aforementioned multi-dimensional time-series feature dataset, a deep learning network is trained to establish a nonlinear mapping relationship between dynamic impedance, surface geometric center temperature, charge / discharge rate, ambient temperature, and internal core temperature, thereby obtaining an impedance-internal temperature soft measurement model. The specific network structure and training process are as follows: We employ a bidirectional long short-term memory network (BiLSTM) combined with an attention mechanism as a deep learning network.

[0037] Input layer: Receives a multivariate sliding time window sequence with a length of 100 time steps. The input feature dimensions include dynamic impedance sequence, surface geometric center temperature sequence, charge / discharge rate sequence, and ambient temperature sequence. The sliding step size is set to 1 time step.

[0038] Feature extraction layer: Two BiLSTM networks are set up to process time series data (i.e., input layer data) simultaneously through two LSTM units in the forward and backward directions, and to explore the nonlinear mapping law between impedance features and internal core temperature from both forward and backward time series directions; the first BiLSTM layer has 128 hidden units and the second BiLSTM layer has 64 hidden units, both with a dropout=0.2 random deactivation strategy to avoid model overfitting.

[0039] Attention layer: An 8-head multi-head attention mechanism is set up to adaptively allocate the weights of the time-series features output by BiLSTM, giving higher feature weights to the early signs of overheating when the dynamic impedance sequence changes drastically, thereby enhancing the model's ability to capture fault precursors and suppressing interference from irrelevant noise.

[0040] Fully connected layer: Two fully connected network layers are set up. The first layer has 64 neurons and uses the ReLU activation function, and the second layer has 32 neurons and uses the ReLU activation function to deeply integrate the features output by the attention layer.

[0041] Output layer: Set up 1 neuron, use a linear activation function, and output the real-time predicted value of the core internal temperature of the target lithium battery at the current moment.

[0042] During model training, mean squared error was used as the loss function, and the Adam optimizer was used for iterative parameter updates. The initial learning rate was set to 0.001, the batch size was set to 64, and the maximum number of training epochs was set to 200. During training, the model's performance on the validation set was verified every 5 epochs. If the validation set loss did not decrease for 10 consecutive epochs, an early stopping strategy was triggered to stop training and avoid overfitting. After training, hyperparameters such as the number of hidden units in the BiLSTM, the number of attention heads, the learning rate, and the dropout value were optimized using a grid search method. The final solidified model had an average absolute error of ≤1℃ and a root mean square error of ≤1.5℃ for internal temperature prediction on the test set, meeting the accuracy requirements for engineering applications.

[0043] S4. During actual operation of the lithium battery, the dynamic impedance, surface geometric center temperature, charge / discharge rate, and ambient temperature within a specified time window are acquired in real time and input into the impedance-internal temperature soft measurement model. The model outputs the real-time predicted value T of the lithium battery's internal core temperature at the current moment. 内温 According to T 内温 The specific process for conducting a real-time over-temperature risk assessment is as follows: Real-time data synchronous acquisition: Following the same measurement method as the offline stage, the battery dynamic impedance data, surface geometric center temperature data, current charge / discharge rate and ambient temperature data are acquired in real time within the sliding time window. After completing the preprocessing operation consistent with the offline stage, the data is input into the impedance-internal temperature soft measurement model in real time. No-delay prediction of internal core temperature: Based on the input real-time feature data, the model directly outputs the predicted value T of the internal core temperature of the lithium battery at the current moment. 内温 The response delay is ≤100ms, completely eliminating the time lag caused by surface temperature heat conduction; The process of risk assessment and closed-loop management for overheating is as follows: Figure 3 As shown: The real-time predicted internal core temperature T 内温 Compared with the preset over-temperature safety threshold T 阈值 The risk levels were compared and divided into three categories, and corresponding tiered control strategies were implemented accordingly: Low risk of overheating: When 50℃ <T 内温 When the temperature is ≤55℃, the BMS triggers a slight warning and sends a warning to the driver through the vehicle terminal. At the same time, the sampling frequency of the dynamic impedance spectrum is increased to enhance the monitoring density of battery thermal status. Medium risk of overheating: when it reaches 55℃ <T 内温 When the temperature is ≤60℃, the BMS immediately implements the charge and discharge power regulation strategy, limits the maximum charge and discharge rate of the battery to 0.5C, reduces the battery heat generation rate, and starts the vehicle liquid cooling system to enter half power cooling mode. High risk of overheating: When T 内温 When the temperature exceeds 60℃, the BMS immediately triggers the high-voltage safety protection, cuts off the battery main circuit, and forcibly starts the cooling thermal management system to enter full-power cooling mode, and issues an emergency fault warning to the driver to avoid thermal runaway accidents.

[0044] Example 2 This embodiment provides a lithium battery over-temperature fault risk assessment system based on dynamic impedance spectroscopy, such as... Figure 4 As shown, the lithium battery over-temperature fault risk assessment method described in Example 1 consists of an offline data acquisition unit, an internal core temperature sequence reconstruction unit, a dataset construction unit, an impedance-internal temperature soft measurement model construction unit, and an online risk assessment unit.

[0045] The offline data acquisition unit acquires dynamic impedance sequence data of the lithium battery under multiple operating conditions in offline testing mode, and simultaneously acquires the surface geometric center temperature sequence, charge / discharge rate sequence, and ambient temperature sequence of the lithium battery in real time.

[0046] The internal core temperature sequence reconstruction unit extracts the time lag Δt between the abrupt change in dynamic impedance characteristics and the surface geometric center temperature response; advances the surface geometric center temperature sequence by Δt time on the time axis, thereby non-destructively reconstructing the internal core temperature sequence of the lithium battery and using it as a training label.

[0047] The dataset construction unit is used to align the dynamic impedance sequence, surface geometric center temperature sequence, charge / discharge rate sequence, and ambient temperature sequence by timestamp, and combine them with training labels to construct a multi-dimensional time-series feature dataset.

[0048] The impedance-internal temperature soft measurement model construction unit: based on the multi-dimensional time-series feature dataset, a deep learning network is trained to establish a nonlinear mapping relationship between dynamic impedance, surface geometric center temperature, charge / discharge rate, ambient temperature and internal core temperature, thereby obtaining the impedance-internal temperature soft measurement model.

[0049] The online risk assessment unit, during actual operation of the lithium battery, acquires dynamic impedance, surface geometric center temperature, charge / discharge rate, and ambient temperature in real time within a specified time window and inputs them into the impedance-internal temperature soft measurement model. The model then outputs the real-time predicted value T of the lithium battery's internal core temperature at the current moment. 内温 According to T 内温 Conduct real-time over-temperature risk assessment.

[0050] It should be noted that each unit in the aforementioned lithium battery over-temperature fault risk assessment system based on dynamic impedance spectroscopy can be implemented entirely or partially through software, hardware, or a combination thereof. These units can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each unit. For specific limitations regarding the lithium battery over-temperature fault risk assessment system based on dynamic impedance spectroscopy, please refer to the limitations of the lithium battery over-temperature fault risk assessment method based on dynamic impedance spectroscopy (i.e., Example 1) above; both have the same function and role, and will not be repeated here.

[0051] The above description of the embodiments is provided to enable those skilled in the art to understand and apply the present invention. It will be apparent to those skilled in the art that various modifications can be made to the above embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. A method for assessing the risk of over-temperature failure in lithium batteries based on dynamic impedance spectroscopy, characterized in that, Including the following steps: S1. In offline testing, acquire dynamic impedance sequence data of lithium battery under multiple operating conditions, and simultaneously acquire lithium battery surface geometric center temperature sequence, charge / discharge rate sequence and ambient temperature sequence in real time; extract the time lag Δt between dynamic impedance feature mutation and surface geometric center temperature response; advance the surface geometric center temperature sequence by Δt time on the time axis to non-destructively reconstruct the core temperature sequence inside the lithium battery and use it as training label; S2. Align the dynamic impedance sequence, surface geometric center temperature sequence, charge / discharge rate sequence, and ambient temperature sequence according to timestamps, and combine them with training labels to construct a multi-dimensional time-series feature dataset; S3. Based on the multi-dimensional time-series feature dataset, train a deep learning network to establish a nonlinear mapping relationship between dynamic impedance, surface geometric center temperature, charge / discharge rate, ambient temperature and internal core temperature, and obtain an impedance-internal temperature soft measurement model. S4. During actual operation of the lithium battery, the dynamic impedance, surface geometric center temperature, charge / discharge rate, and ambient temperature within a specified time window are acquired in real time and input into the impedance-internal temperature soft measurement model. The model outputs the real-time predicted value T of the lithium battery's internal core temperature at the current moment. 内温 According to T 内温 Conduct real-time over-temperature risk assessment.

2. The lithium battery over-temperature fault risk assessment method based on dynamic impedance spectroscopy according to claim 1, characterized in that, The real-time predicted internal core temperature T output by the model 内温 Compared with the preset battery internal over-temperature safety threshold T 阈值 Compare the data, classify the risk levels of overheating, and implement tiered control strategies.

3. The lithium battery over-temperature fault risk assessment method based on dynamic impedance spectroscopy according to claim 2, characterized in that, When the real-time internal core temperature prediction value T 内温 At: T 阈值 -10℃<T 内温 ≤T 阈值 -5℃ is determined to be a low-risk over-temperature condition. The system triggers a minor warning and increases the sampling frequency of the dynamic impedance spectrum.

4. The lithium battery over-temperature fault risk assessment method based on dynamic impedance spectroscopy according to claim 2, characterized in that, When the real-time internal core temperature prediction value T 阈值 -5℃<T 内温 ≤T 阈值 If the temperature is deemed to be at medium risk due to overheating, the battery management system will be immediately adjusted to reduce the charging and discharging power.

5. The lithium battery over-temperature fault risk assessment method based on dynamic impedance spectroscopy according to claim 2, characterized in that, When the real-time internal core temperature prediction value T 内温 >T 阈值 If the temperature is deemed to be at high risk of overheating, the main battery circuit will be immediately cut off, and the liquid cooling system will be forcibly activated.

6. The lithium battery over-temperature fault risk assessment method based on dynamic impedance spectroscopy according to claim 1, characterized in that, In step S1, the dynamic impedance value of the lithium battery is obtained using the following method: First, the response voltage of the battery is simultaneously acquired using a four-probe method. U r With excitation current i e The sampling voltage formed on the precision sampling resistor U s ; then U s and U r Both signals are converted to single-ended signals by an instrumentation amplifier with the same architecture, and then amplified by an independent programmable amplifier with identical parameters to achieve adjustable gain of 10 / 100 / 1000 times. The amplified signals are then converted to digital signals using a synchronous sampling ADC. Finally, the amplitude of the converted data is converted to obtain the complete signal. i e and U r The sequence is subjected to Fourier transform to extract the voltage and current components at the target frequency, and the dynamic impedance value at the corresponding frequency is calculated by vector division.

7. The lithium battery over-temperature fault risk assessment method based on dynamic impedance spectroscopy according to claim 1, characterized in that, A high-precision K-type thermocouple sensor is placed at the geometric center of the lithium battery surface to independently collect temperature data in real time, and the sensor accuracy is ±0.2℃.

8. The lithium battery over-temperature fault risk assessment method based on dynamic impedance spectroscopy according to claim 1, characterized in that, Before building the model, the multi-dimensional time-series feature dataset undergoes the following preprocessing steps: Outlier removal: Abnormal deviations caused by sensor noise are removed using the 3σ principle to ensure the continuity and smoothness of time-series feature data; Normalization: The min-max normalization method is used to uniformly map multi-dimensional time-series feature data and training labels to the [0,1] interval, eliminating the negative impact of different physical dimensions on the gradient descent of deep learning networks.

9. The method for assessing the risk of over-temperature faults in lithium batteries based on dynamic impedance spectroscopy according to claim 1, characterized in that, The structure of the impedance-internal temperature soft measurement model is as follows: a bidirectional long short-term memory network (BiLSTM) combined with an attention mechanism is used as a deep learning network. Input layer: Receives dynamic impedance sequence, surface geometric center temperature sequence, charge / discharge rate sequence, and ambient temperature sequence; Feature extraction layer: Processes input layer data simultaneously through two LSTM layers (forward and backward) to uncover the physical mapping between dynamic impedance features and internal core temperature; Attention layer: Weights input layer data are assigned using a multi-head attention mechanism, giving higher weights to moments of drastic change in the dynamic impedance sequence; Fully connected layer: Employs the ReLU activation function to deeply integrate the features output by the attention layer; Output layer: Outputs the real-time predicted internal core temperature of the target lithium battery at the current moment.

10. A lithium battery over-temperature fault risk assessment system based on dynamic impedance spectroscopy, used to implement the lithium battery over-temperature fault risk assessment method according to any one of claims 1-9, characterized in that, include: Offline data acquisition unit: In offline testing mode, acquires dynamic impedance sequence data of lithium battery under multiple operating conditions, and simultaneously acquires the surface geometric center temperature sequence, charge / discharge rate sequence and ambient temperature sequence of lithium battery in real time; Internal core temperature sequence reconstruction unit: Extracts the time lag Δt between the abrupt change in dynamic impedance characteristics and the temperature response at the surface geometric center; advances the surface geometric center temperature sequence by Δt time on the time axis to non-destructively reconstruct the internal core temperature sequence of the lithium battery and use it as a training label. Dataset construction unit: used to align the dynamic impedance sequence, surface geometric center temperature sequence, charge / discharge rate sequence and ambient temperature sequence by timestamp, and combine them with training labels to construct a multi-dimensional time-series feature dataset; Impedance-internal temperature soft measurement model construction unit: Based on the multi-dimensional time-series feature dataset, a deep learning network is trained to establish a nonlinear mapping relationship between dynamic impedance, surface geometric center temperature, charge / discharge rate, ambient temperature and internal core temperature, and thus obtains the impedance-internal temperature soft measurement model. Online risk assessment unit: During actual operation of the lithium battery, the dynamic impedance, surface geometric center temperature, charge / discharge rate, and ambient temperature are acquired in real time within a specified time window and input into the impedance-internal temperature soft measurement model. The model outputs the real-time predicted value T of the lithium battery's internal core temperature at the current moment. 内温 According to T 内温 Conduct real-time over-temperature risk assessment.

Images