A method for predicting battery thermal runaway in energy storage systems based on a hybrid model
By combining a battery model with an LSTM neural network, a battery data acquisition and electrothermal coupling model is constructed to predict the internal and surface temperature curves of lithium-ion batteries. This solves the problem of complex and slow response in the diagnosis of thermal runaway in lithium-ion batteries in existing technologies, and achieves fast and accurate early warning of thermal runaway.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ARMY ENG UNIV OF PLA
- Filing Date
- 2023-05-25
- Publication Date
- 2026-06-30
AI Technical Summary
Existing methods for diagnosing thermal runaway in lithium-ion batteries are complex and slow to respond, making it difficult to effectively predict thermal runaway accidents in energy storage systems.
A hybrid model-based approach is adopted, which combines the battery model with an LSTM neural network to construct a battery data acquisition, electrothermal coupling, and prediction model. The battery electrothermal coupling model accurately estimates the internal temperature and SOC, and the LSTM prediction model predicts the temperature curve. Combined with the thermal runaway judgment process, early warning is achieved.
It enables rapid and accurate prediction of thermal runaway in lithium-ion batteries, simplifies the fault diagnosis process, and improves the practicality and response speed of early warning.
Smart Images
Figure CN116626503B_ABST
Abstract
Description
Technical Field
[0001] This invention is applicable to the application of lithium-ion batteries in energy storage systems. It discloses a battery thermal runaway prediction method based on a hybrid model, aiming to reduce the occurrence of battery thermal runaway accidents in energy storage systems. Background Technology
[0002] Lithium-ion batteries are widely used in energy storage systems and electric vehicles due to their advantages such as high power density and energy density, long cycle life, low self-discharge rate, and moderate price. However, with their widespread adoption, a series of accidents have also occurred frequently. In particular, accidents such as spontaneous combustion and explosion caused by thermal runaway can have serious consequences.
[0003] Existing methods for diagnosing thermal runaway can be broadly categorized into two types: those based on battery characteristic information and those based on battery models. Characteristic-based methods primarily study the voltage and temperature evolution during thermal runaway, identifying useful characteristic information in the early stages of diagnosis, such as voltage, temperature, and impedance. Model-based methods require expertise in battery physics and chemical equations, involving complex mathematical modeling, observer design, and tedious parameter tuning processes. Summary of the Invention
[0004] This invention provides a method for predicting battery thermal runaway based on a hybrid model. It combines neural networks and battery models, and designs an algorithm based on the mechanism of thermal runaway in lithium-ion batteries to predict and determine abnormal internal and surface temperatures of the battery.
[0005] A hybrid model-based method for predicting thermal runaway in energy storage systems combines a battery model with an LSTM neural network model to construct the following model:
[0006] A battery data acquisition model used to collect battery-related parameters;
[0007] Used for accurate estimation of internal temperature T in and batteries SOC A battery electrothermal coupling model;
[0008] An LSTM prediction model is used to obtain the predicted internal temperature curve and the predicted surface temperature curve of the battery at various times.
[0009] A thermal runaway prediction model used to achieve early warning of thermal runaway.
[0010] Preferably, this invention combines a first-order equivalent circuit model and a lumped-parameter thermal model to form a battery electrothermal coupling model, which accurately estimates the battery's thermal performance. SOC and the internal temperature of the battery;
[0011] Parameter identification is performed on the first-order equivalent circuit model to identify the ideal voltage source. U oc, Ohmic internal resistance R 0, polarization resistance R d With polarization capacitor C d And use the ampere-hour integration method to perform SOC Estimation; based on this, coupling with the lumped-parameter thermal model, the first-order equivalent circuit model and the lumped-parameter thermal model are connected through ohmic internal resistance. R 0. Polarization internal resistance R d With internal temperature T in To establish a connection;
[0012] First, the battery's internal temperature is calculated using the load current and the battery's internal temperature. SOC ;
[0013] Secondly, according to SOC The relationship between temperature and internal resistance is used to determine the ohmic internal resistance. R 0. Polarization internal resistance R d The value of the resistance is obtained, and the heat generated by the battery is calculated based on the obtained resistance value;
[0014] Generating heat from lithium batteries Q j and ambient temperature T amb As input to the thermal model, the internal temperature of the lithium-ion battery is calculated. T in Then, the internal temperature T in The parameters are passed into the battery equivalent circuit model and, at the next moment, are related to the current. I Calculate the new battery SOC A loop is formed; this enables real-time and accurate estimation of the internal temperature. T in and batteries SOC Its function.
[0015] Preferably, the LSTM prediction model of the present invention uses the measured voltage parameters at each time step. U Current I , SOC Battery internal temperature T in Battery surface temperature T surf Using both as input matrices, and the battery predicted internal temperature and battery predicted surface temperature as output matrices, we obtain the battery predicted internal temperature curve and battery predicted surface temperature curve based on each time point.
[0016] Preferably, the thermal runaway prediction model of the present invention, combined with the temperature prediction model, proposes a battery thermal runaway judgment process. Based on the predicted temperature curve obtained by the prediction model, the predicted temperature is compared with the measured temperature to obtain the prediction result of battery thermal runaway and the cause of battery thermal runaway, thus completing the prediction and realizing thermal runaway early warning.
[0017] This invention proposes a hybrid model-based method for predicting thermal runaway in lithium batteries. It employs a hybrid model combining a battery electrothermal coupling model and an LSTM neural network. This hybrid model method integrates the thermal and electrical characteristics of the battery and applies a data-driven approach, providing a new perspective for predicting and diagnosing battery thermal runaway. Compared to other methods, this method has a simple fault judgment process, strong practicality, and fast response, and can be widely used in practical engineering due to these advantages. Attached Figure Description
[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0019] Figure 1 A model for predicting battery thermal runaway;
[0020] Figure 2 This is a first-order RC equivalent circuit model;
[0021] Figure 3 It is a lumped parameter thermal model;
[0022] Figure 4 It is an electrothermal coupling model;
[0023] Figure 5 This is a model for predicting the internal and surface temperatures of electrical measurements based on an LSTM neural network.
[0024] Figure 6 This is a schematic diagram of the thermal runaway prediction process;
[0025] Figure 7 This is data from an overcharging experiment;
[0026] Figure 8 This is a schematic diagram of the predicted internal temperature of the battery in Example 1;
[0027] Figure 9 This is a schematic diagram of the predicted external temperature of the battery in Example 1;
[0028] Figure 10 This is data from a thermal shock experiment;
[0029] Figure 11 This is a schematic diagram of the predicted internal temperature of the battery in Example 2;
[0030] Figure 12 This is a schematic diagram of the predicted surface temperature of the battery in Example 2. Detailed Implementation
[0031] The technical solution of the present invention will be described in further detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.
[0032] A hybrid model-based method for predicting thermal runaway in energy storage systems combines a battery model with an LSTM neural network model to construct the following model:
[0033] A battery data acquisition model used to collect battery-related parameters;
[0034] Used for accurate estimation of internal temperature T in and batteries SOC A battery electrothermal coupling model;
[0035] By combining the first-order equivalent circuit model and the lumped-parameter thermal model, a battery electrothermal coupling model is formed. This model is then used to accurately estimate the battery's thermal performance. SOC and the internal temperature of the battery;
[0036] Parameter identification is performed on the first-order equivalent circuit model to identify the ideal voltage source. U oc, Ohmic internal resistance R 0, polarization resistance R d With polarization capacitor C d And use the ampere-hour integration method to perform SOC Estimation; based on this, coupling with the lumped-parameter thermal model, the first-order equivalent circuit model and the lumped-parameter thermal model are connected through ohmic internal resistance. R 0. Polarization internal resistance R d With internal temperature T in To establish a connection;
[0037] First, the battery's internal temperature is calculated using the load current and the battery's internal temperature. SOC ;
[0038] Secondly, according to SOC The relationship between temperature and internal resistance is used to determine the ohmic internal resistance. R 0. Polarization internal resistance R d The value of the resistance is obtained, and the heat generated by the battery is calculated based on the obtained resistance value;
[0039] Generating heat from lithium batteries Q j and ambient temperature Tamb As input to the thermal model, the internal temperature of the lithium-ion battery is calculated. T in Then, the internal temperature T in The parameters are passed into the battery equivalent circuit model and, at the next moment, are related to the current. I Calculate the new battery SOC A loop is formed; this enables real-time and accurate estimation of the internal temperature. T in and batteries SOC Its function.
[0040] An LSTM prediction model is used to obtain the predicted internal temperature curve and the predicted surface temperature curve of the battery at various times.
[0041] The LSTM prediction model will measure the voltage parameters at each time step. U Current I , SOC Battery internal temperature T in Battery surface temperature T surf Using both as input matrices, and the battery predicted internal temperature and battery predicted surface temperature as output matrices, we obtain the battery predicted internal temperature curve and battery predicted surface temperature curve based on each time point.
[0042] A thermal runaway prediction model used to achieve early warning of thermal runaway.
[0043] Based on the temperature prediction model, a battery thermal runaway judgment process is proposed. According to the predicted temperature curve obtained by the prediction model, the predicted temperature is compared with the measured temperature to obtain the prediction result of battery thermal runaway and the cause of battery thermal runaway, thus completing the prediction and realizing thermal runaway early warning.
[0044] A method for predicting battery thermal runaway in energy storage systems based on a hybrid model consists of four modules: a data acquisition module, a battery electrothermal coupling model, an LSTM prediction model, and a prediction model. The input parameters of the LSTM are obtained through the data acquisition module and the battery electrothermal coupling model. The output of the LSTM prediction model is used as the basis for the judgment of the prediction module. Finally, the prediction model is used to obtain the prediction of battery thermal runaway and the determination of the cause.
[0045] Battery voltage, battery current, and battery surface temperature are the easiest data to collect during battery operation. They can be used as key parameters in the battery electrothermal model to estimate the battery's internal temperature and SOC. In the prediction model, they are correlated with the desired results T'in and T'surf, and therefore serve as inputs to the model.
[0046] Most existing battery fault diagnosis models employ a single battery model and are primarily based on the battery's electrical characteristics. An ideal battery diagnosis method should simultaneously study both the battery's electrical characteristics and thermal effects. This would shorten the alarm time and improve the accuracy of fault monitoring. Therefore, this paper employs an electrothermal coupling model of the battery, which plays a crucial role in accurately estimating Tin and State of Charge (SOC) during the prediction process.
[0047] The prediction model chosen is LSTM, a recurrent neural network (RNN) with memory capabilities, first proposed by Hochreiter et al., specifically designed for processing time-series data. LSTM not only solves the "information loss" problem in traditional BP (backpropagation) and CNN algorithms that rely solely on current input data to determine the output state, but also addresses the shortcomings of traditional RNNs in processing long-term data series, which are prone to "vanishing gradients" and "exploding gradients" due to long-term dependency mechanisms. The prediction results output by this model are used in the diagnostic module.
[0048] In the prediction module, based on the predicted results T'in and T'surf, the temperature rise relationship between the battery's interior and surface is obtained according to different thermal runaway triggers. Based on this, a thermal runaway diagnosis process is proposed, where ωi and ωs are diagnostic thresholds. Thermal runaway determination is achieved by comparing the difference γin(t) with the threshold ωi to determine if thermal runaway has occurred. When the difference is less than the threshold, it can be determined that the battery does not have a thermal fault; when the difference is greater than the threshold, it can be determined that the battery has experienced thermal runaway and a warning signal is generated. The time t at this point is the alarm time. The second step is to detect the thermal runaway trigger by comparing the difference γsurf(t) with the threshold ωs to determine the external triggers that caused the thermal runaway.
[0049] like Figure 1 As shown, in the specific implementation of the thermal runaway prediction method, it is first necessary to collect real-time parameters during battery operation, including battery voltage. U Battery current I Battery surface temperature T surf Ambient temperature T amb As the basis for prediction, the battery electrothermal coupling model and LSTM prediction model described in this method are input to obtain the battery internal temperature prediction curve and the battery surface temperature prediction curve. The prediction model then determines the occurrence of thermal runaway in the battery and, in the event of thermal runaway, obtains the alarm time. t And the causes of failure due to thermal runaway.
[0050] Specifically, the battery electrothermal coupling model consists of an equivalent circuit model and a lumped heat model, as shown in the equivalent circuit model. Figure 2The Thevenin model, which considers the polarization effect during battery charging and discharging, can better describe the dynamic characteristics of the battery. The parameters in the equivalent circuit include an ideal voltage source. U oc Ohmic internal resistance R 0, polarization resistance R d With polarization capacitor C d Using these parameters and SOC Parameter estimation is performed based on the correlation with temperature, and then the model is applied. SOC Real-time estimation. The lumped heat model, as shown in Figure 3, simulates the relationship between the internal temperature of the battery and the external temperature, aiming to estimate the internal temperature of the battery as an auxiliary indicator for judging battery thermal runaway. To reduce the complexity of the model, the method only considers radial thermal behavior with convective heat transfer boundary conditions and assumes that the internal temperature of the battery is uniform and that all internal heat generation comes from the internal resistance of the battery. The battery equivalent circuit model is coupled with the lumped heat model, as shown in Figure 3. Figure 4 As shown, the combination of the electrical model after parameter identification and the thermal model after thermal calculation yields the battery at each time step. SOC and battery internal temperature T in .
[0051] After the battery parameters are prepared, the internal temperature and surface temperature of the battery are predicted, such as... Figure 5 As shown, the battery voltage U Battery current I Battery surface temperature T surf ,Battery SOC and battery internal temperature T in The input matrix is divided into training and test data according to time intervals. The training data is fed into the training model to train the LSTM neural network. The test data is input to obtain the prediction data, and the result is iteratively fed into the LSTM neural network. Finally, the prediction results are plotted as a temperature prediction curve with time as the horizontal axis.
[0052] After obtaining the temperature prediction curve, the system proceeds to the thermal runaway prediction module. Figure 6 This is a thermal runaway prediction process; the prediction process first calculates the difference between the actual temperature and the predicted temperature:
[0053] (1)
[0054] (2)
[0055] in T in It is the internal temperature of the battery.T ’ in This is the predicted internal temperature of the battery. T surf It is the surface temperature of the battery. T ’ surf This is the predicted battery surface temperature.
[0056] Specifically, the first step in the process is to determine whether thermal runaway has occurred. Thermal runaway is determined by comparing differences. γ in(t) and threshold ω i To determine whether thermal runaway has occurred, when the difference... γ in(t) Less than the threshold ω i This indicates that the battery does not have a thermal fault. When the difference is... γ in(t) Greater than the threshold ω i When the battery experiences thermal runaway, a warning signal can be generated. t This is the alarm time.
[0057] The second step of the process is to detect the cause of thermal runaway by comparing the differences. γ surf(t) and threshold ω s To determine the external triggers that cause thermal runaway. When the difference γ surf(t) Greater than the threshold ω s This allows us to determine that the thermal runaway was caused by thermal shock. When the difference... γ surf(t) Less than the threshold ω s This allows us to determine that the thermal runaway was caused by overcharging and discharging.
[0058] The present invention will be further explained below with reference to specific embodiments.
[0059] Example 1
[0060] The method was validated using experimental data on thermal runaway caused by battery overcharging. An APR18650 lithium-ion battery was subjected to overcharge cycles. With increasing cycle number, the maximum surface temperature of the battery showed an upward trend, and thermal runaway occurred during the 18th overcharge cycle. The experimental data were processed to obtain the battery voltage, surface temperature, and internal temperature. Figure 7 As shown, the battery's internal temperature rose sharply around 4600 seconds, while the surface temperature began to rise after 4600 seconds, indicating a prolonged period of thermal runaway. Some data records are shown in Table 1.
[0061] Table 1 Data near temperature abrupt change nodes
[0062] Time (s) Battery voltage (V) Battery current (A) Battery surface temperature (°C) 4580 4.454 2.42 34.3 4590 4.462 2.42 35.4 4600 4.463 2.43 35.3 4610 4.462 2.42 37.5 4620 4.469 2.42 38.4
[0063] The battery operating data is input into the thermal runaway prediction model, and the prediction results are as follows: Figure 8 , Figure 9 As shown. Among them. Figure 8 This is the predicted internal temperature of the battery in Example 1. Figure 9 The external temperature prediction result of the battery in Example 1 is shown. According to the thermal runaway diagnosis process, the internal temperature is first compared with the prediction result. At 4600s, the difference γ between the actual temperature and the predicted temperature is calculated. in Exceeding the threshold ω i This allows us to determine that the battery experienced thermal runaway at 4600s and generate an alarm; then, we compare the surface temperature with the predicted result, and the difference γ between the actual and predicted temperatures is calculated. surf The threshold ω was not exceeded. s According to the diagnostic process, the cause of thermal runaway was determined to be internally caused electrical abuse, namely, overcharging and over-discharging.
[0064] The diagnosis was: thermal runaway occurred, alarm time. t =4600s, the thermal runaway was caused by battery overcharge and discharge. The diagnostic results are consistent with the experimental results, verifying the effectiveness of the proposed diagnostic method.
[0065] Example 2
[0066] This method was validated using experimental data on thermal runaway caused by battery thermal shock. Before the thermal shock experiment, the lithium-ion battery cells were treated as follows: using a cyclic charge-discharge apparatus, the cells were fully charged. At room temperature, they were first discharged at a 0.5C rate to a termination voltage of 3.0V, then allowed to rest for 60 minutes. Next, they were charged at a 0.5C rate to a charging termination voltage of 4.2V, then switched to constant voltage charging. After stopping charging, the cells were allowed to rest for 60 minutes. After charge-discharge cycles according to QC / T743 standard, the experimental batteries were placed in an adiabatic accelerated calorimeter for stable discharge. The heating equipment was used to raise the temperature to 150℃ at a rate of 2℃ / min, then heating was stopped, and experimental data were recorded. The battery caught fire and burned in the later stages of heating, and the casing ruptured. The experimental data were processed to obtain the battery voltage, battery surface temperature, and battery internal temperature as shown below. Figure 10 As shown, the battery's internal temperature rose sharply around 4400s, and the surface temperature began to rise after 4400s. The battery experienced thermal runaway, with the combustion temperature reaching a maximum of 900℃. After combustion, the temperature gradually decreased, and the thermal runaway process was relatively long. Some experimental data are recorded in Table 2:
[0067] Table 2 Data near temperature abrupt change nodes
[0068] Time (s) Battery voltage (V) Battery current (A) Battery surface temperature (°C) 4380 3.944 0.72 153.3 4440 3.945 0.72 155.8 4500 3.942 0.73 362.2 4560 3.942 0.73 479.5 4720 3.941 0.93 588.1
[0069] The battery operating data is input into the thermal runaway prediction model, and the prediction results are as follows: Figure 11 , Figure 12 As shown. Among them. Figure 11 This is the predicted internal temperature of the battery in Example 2. Figure 12 This is the predicted surface temperature result of the battery in Example 2. According to the thermal runaway diagnosis process, the internal temperature is first compared with the predicted result. At 4440s, the difference between the actual temperature and the predicted temperature is... γ in Exceeding the threshold ω i This allows us to determine that the battery experienced thermal runaway at 4440s and generate an alarm; then, we compare the surface temperature with the predicted result, and the difference between the actual temperature and the predicted temperature. γ surf Greater than the threshold ω s According to the diagnostic process, the cause of thermal runaway was determined to be thermal abuse caused by thermal shock.
[0070] The diagnosis was: thermal runaway occurred, alarm time. t =4440s, the cause of thermal runaway was external thermal shock. The diagnostic results are consistent with the experimental results, verifying the effectiveness of the proposed diagnostic method.
Claims
1. A method for predicting battery thermal runaway in an energy storage system based on a hybrid model, characterized in that... By combining the battery model with the LSTM neural network model, the following model is constructed: A battery data acquisition module for collecting battery-related parameters; collecting battery voltage. U Battery current I Battery surface temperature T surf and battery ambient temperature T amb ; Used for accurate estimation of internal temperature T in and batteries SOC A battery electrothermal coupling model; An LSTM prediction model is used to obtain the predicted internal temperature curve and the predicted surface temperature curve of the battery at various times. A thermal runaway prediction model used to achieve early warning of thermal runaway; By combining the first-order equivalent circuit model and the lumped-parameter thermal model, a battery electrothermal coupling model is formed. This model is then used to accurately estimate the battery's thermal performance. SOC and the internal temperature of the battery T in ; Parameter identification is performed on the first-order equivalent circuit model to identify the ideal voltage source. U oc, Ohmic internal resistance R 0, polarization resistance R d With polarization capacitor C d And use the ampere-hour integration method to perform SOC Estimation; based on this, coupling with the lumped-parameter thermal model, the first-order equivalent circuit model and the lumped-parameter thermal model are connected through ohmic internal resistance. R 0. Polarization internal resistance R d With internal temperature T in To establish a connection; First, the battery's internal temperature is calculated using the load current and the battery's internal temperature. SOC ; Secondly, according to SOC The relationship between temperature and internal resistance is used to determine the ohmic internal resistance. R 0. Polarization internal resistance R d The value of the resistance is obtained, and the heat generated by the battery is calculated based on the obtained resistance value; Generating heat from lithium-ion batteries Q j and ambient temperature T amb As input to the thermal model, the internal temperature of the lithium-ion battery is calculated. T in Then, the internal temperature T in The parameter is passed into the first-order equivalent circuit model, and at the next moment, it is related to the current. I Calculate the new battery SOC A loop is formed; this enables real-time and accurate estimation of the internal temperature. T in and batteries SOC Its function; The above LSTM prediction model will measure the voltage parameter at each time step. U Current I , SOC Battery internal temperature T in Battery surface temperature T surf Together, they form the input matrix, and the battery predicted internal temperature and battery predicted surface temperature form the output matrix, resulting in the battery predicted internal temperature curve and battery predicted surface temperature curve based on each time point. The above thermal runaway prediction model, combined with the LSTM prediction model, proposes a battery thermal runaway judgment process. Based on the predicted temperature curve obtained by the LSTM prediction model, the predicted temperature is compared with the measured temperature to obtain the prediction result of battery thermal runaway and the cause of battery thermal runaway, thus completing the prediction and realizing thermal runaway early warning. The specific determination process is as follows: First, calculate the measured internal temperature of the battery. T in With battery predicted internal temperature T ’ in The difference γ in ( t ), calculate the measured surface temperature of the battery T surf With battery predicted surface temperature T ’ surf The difference γ surf ( t When the difference γ in ( t Less than the threshold ω i The battery is determined not to have thermal runaway; when the difference is... γ in ( t ) greater than the threshold ω i At that time, it is determined that the battery has experienced thermal runaway; At this time, when the difference γ surf ( t ) greater than the threshold ω s The cause of thermal runaway was determined to be thermal shock, when the difference... γ surf ( t Less than the threshold ω s The cause of thermal runaway was determined to be overcharge and discharge.