An online intelligent fault diagnosis method for new energy vehicle batteries

Abstract

The application provides an online intelligent fault diagnosis method for a new energy automobile battery, early detection and early warning of thermal runaway are carried out by using a gas sensing technology, fire hazards are eliminated before the thermal runaway spreads, and great potential is achieved in the aspect of improving safety; when the battery is in thermal runaway, the combustible and explosive gas is discharged, real-time monitoring and timely alarm are carried out.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery safety, in particular to a new energy automobile battery online intelligent fault diagnosis method. BACKGROUND

[0002] At present, aiming at the characteristics that the current, voltage, internal resistance, internal pressure and surface temperature and other signals will change obviously in the process of battery thermal runaway, the battery state monitoring and thermal runaway early warning system mainly monitors and warns in real time based on the critical conditions of these signals. However, through monitoring these signals, there is limitation for early warning of lithium ion battery thermal runaway. The traditional temperature and voltage sensor detects the external temperature and terminal voltage of the battery, but these parameters change little in the early stage of thermal runaway, which is difficult to realize early warning of thermal runaway behavior. Emerging implantable temperature sensors and electrochemical impedance spectroscopy testing methods can provide the internal temperature of lithium ion battery cells, but their cost is high, and it is difficult to realize commercial application at present. Moreover, in large battery modules, the voltage and temperature of single battery change little, which is difficult to be detected before thermal runaway spreads to other batteries, and detecting the voltage and temperature of each single battery will weaken the energy density of the battery and increase the cost.

[0003] The behavior of gas release caused by internal chemical / electrochemical reactions of lithium ion batteries is an important feature in the operation process of the battery. Lithium ion batteries in different environments and operating states (such as low temperature / normal temperature / high temperature cycle, different charge and discharge cutoff voltage operation, high temperature storage, etc.) will affect the gas release concentration and gas production. In the early stage of battery thermal runaway, due to the gradual intensification of internal reactions in the battery, the concentration of characteristic gas produced by the reaction will increase from 0 to hundreds or even thousands of milligrams per cubic meter. Therefore, using gas detection technology to realize early warning of battery thermal runaway has great potential. Therefore, using gas sensing technology for early detection and early warning of thermal runaway can eliminate fire hazards before the spread of thermal runaway, and has great potential in improving the safety of lithium ion batteries. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a new energy automobile battery online intelligent fault diagnosis method, which comprises the following steps:

[0005] Step 1: measuring the time-varying data of each evolved gas composition of the battery with different positive electrode materials at different states of charge when kept at high temperature;

[0006] The evolved gas composition includes the concentration of at least one of CO2, CO, H2 and VOCs;

[0007] Step 2: classify the data measured in step 1 into time series of different gases evolved by batteries of different positive electrode materials at different states of charge, and divide the time series into input sequences and target sequences;

[0008] Step 3: train the LSTM neural network to obtain a prediction model of the composition of each evolved gas of the battery of the same positive electrode material at different states of charge;

[0009] Step 4: collect real-time data of the composition of evolved gas of each battery pack of the new energy vehicle;

[0010] Step 5: select a corresponding battery evolved gas composition prediction model according to the type of positive electrode material of the battery and the state of charge of the battery output by the battery management system of the new energy vehicle;

[0011] Step 6: using the sliding window method, extract the evolved gas composition data corresponding to the time steps in the sliding window range from the time steps of the real-time data extracted in step 4 to form a predicted time series;

[0012] Step 7: input the predicted time series of step 6 into the trained battery evolved gas composition prediction model selected in step 5, and output the result;

[0013] Step 8. Compare the output result with the threshold value, if greater than the threshold value, output a warning signal.

[0014] The evolved gas composition is the concentration of one of CO2, CO, H2 and VOCs.

[0015] The step 2 of dividing the time series into input sequences and target sequences comprises:

[0016] Using the sliding window method, extract the time series data corresponding to the time steps in the sliding window range from each time series data set;

[0017] The time series in a time window is taken as the input sequence, and the time series of the time steps after the input sequence is taken as the target sequence.

[0018] The step 3 comprises: initializing the parameters of the LSTM model, performing forward propagation on the input sequence through the LSTM model to obtain a prediction sequence, calculating a loss function according to the prediction sequence and the target sequence, updating the weight parameters of the LSTM model according to the calculated loss function gradient through a back propagation algorithm, repeating the training process until a preset number of training times is reached or until convergence is achieved.

[0019] The step 1, for the non-intrinsic gas CO, H2 and VOCs in the atmosphere, the evolution gas component refers to the change process of the volume concentration of the gas from nothing to something; for the intrinsic gas CO2 in the atmosphere, the evolution gas component refers to the change process after the volume concentration of the gas deviates from the average volume concentration of CO2 in the atmosphere.

[0020] The time sequence divided into input sequence and target sequence in the step 2 is obtained by the following steps:

[0021] According to the state of charge of the battery corresponding to each time sequence, the data is merged, if the state of charge of the battery belongs to the same state of charge distribution range, the time sequence of the evolution gas component is divided into the same group;

[0022] In each time sequence in the same state of charge distribution range, the evolution gas component data of the battery at different states of charge at the same time is averaged, and the average value of the evolution gas component at all times forms the time sequence of the evolution gas component of each state of charge distribution range.

[0023] The step 3 is used to obtain the evolution gas component prediction model of the battery with the same positive electrode material in different state of charge distribution ranges.

[0024] The state of charge distribution range includes SOC∈[0%, 5%], SOC∈(5%, 15%], SOC∈(15%, 25%], SOC∈(25%, 35%], SOC∈(35%, 45%], SOC∈(45%, 55%], SOC∈(65%, 75%], SOC∈(75%, 85%], SOC∈(85%, 95%], SOC∈(95%, 105%].

[0025] The evolution gas component is CO2. The present application has the beneficial effects that the potential thermal runaway risk of the power battery system of the vehicle can be quantitatively evaluated, the battery thermal runaway accident can be detected in time to reduce the battery thermal runaway accident, the safety performance of the lithium ion battery electric vehicle is improved, the safety hidden danger is reduced, and the property loss is reduced. The present application predicts the evolution gas concentration at the future time step according to the time sequence of the evolution gas component concentration detected in real time through the LSTM network, so that whether the evolution gas component will exceed the threshold value can be predicted in advance, the problem that the concentration of the evolution gas component of the battery with some kind of positive electrode material is low at some SOC state and is not conducive to detection is solved, or the problem that the CO2 evolved by the battery is not easy to distinguish from the CO2 in the air, and the change of the CO2 concentration evolved by the battery is not significant in the early evolution stage is solved. The present application also combines the detection data according to the SOC state of the battery to obtain the evolution gas component prediction model of the battery, and solves the problem that the SOC changes constantly during the operation of the new energy vehicle, and the battery thermal runaway cannot be predicted only according to one evolution gas component prediction model of the battery. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of the principle of LSTM neural network. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of the embodiments of the present disclosure more clear, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings of the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0028] Explanation of terms

[0029] SOC: State of Charge (SOC): This refers to the ratio of a battery's remaining capacity after a period of use or long-term storage to its fully charged capacity, usually expressed as a percentage. Its value range is 0-100. When SOC = 0, the battery is fully discharged, and when SOC = 100, the battery is fully charged.

[0030] Example 1

[0031] This embodiment proposes a new energy vehicle battery online intelligent fault diagnosis method, the steps are as follows:

[0032] S1. Training LSTM neural network

[0033] Experimental preparation:

[0034] Before the experiment, batteries containing different cathode materials were placed in a high-temperature chamber to heat the batteries. A gas detection system was installed in the chamber to monitor the evolution of CO2, CO, H2, and VOCs in real time at a measurement frequency of 1 Hz. For the non-inherent gases CO, H2, and VOCs in the atmosphere, gas evolution refers to the process by which the volume concentration of the gas changes from zero to a certain level. For the inherent gas CO2 in the atmosphere, gas evolution refers to the process by which the volume concentration of the gas deviates from the average volume concentration of CO2 in the atmosphere.

[0035] A battery testing system is set up in the high and low temperature chamber to charge and discharge the battery while heating it, and to measure the battery voltage, current and temperature. The measurement frequency is 10 Hz and the error is 0.1%.

[0036] Data Collection:

[0037] All experimental batteries were first charged to 3.65 V in a constant current-constant voltage mode (1 C charging and 0.01 C cutoff current, C represents the rate), and then multiple batteries of the same material were placed at different SOCs by controlling the discharge time. The initial temperature was (20±2) °C. Each battery was left for 1 h or more after being charged and discharged to different SOCs.

[0038] The batteries of different positive electrode materials at different SOCs were placed in different high and low temperature boxes, and each high and low temperature box was continuously heated to 150 °C at a speed of 5 °C / min based on the initial temperature and was kept. Each battery was left in the high and low temperature box for 1 h or more, and the gas detection system monitored the evolution of CO2, CO, H2 and VOC in real time. The results are shown in Table 1.

[0039] Table 1

[0040]

[0041] The time sequence of the evolved gas composition of the battery of different positive electrode materials at different SOCs was obtained, including the time sequence of the concentration of CO2, CO, H2 and VOC evolved.

[0042] The time sequence of the evolved gas composition was classified according to the positive electrode material of the battery, and then classified into multiple time sequence data sets according to the SOC.

[0043] It should be noted that the gas detection system can only monitor the concentration of one of CO2, CO, H2 and VOC in real time during data collection, or can monitor the concentration of several or all gases.

[0044] The corresponding time sequence of the evolved gas composition is the time sequence of the concentration of one of CO2, CO, H2 and VOC, or the time sequence of the concentration of several or all gases.

[0045] If the gas detection system monitors the concentration of several or all gases in real time during data collection, the further time sequence data set can be further classified into time sequence data sets of different gases according to the gas composition.

[0046] Each time sequence data set is trained into a battery evolved gas composition prediction model for batteries of different positive electrode materials at different SOCs using an LSTM neural network.

[0047] If the time sequence data set can be further classified into time sequence data sets of different gases according to the gas composition, each time sequence data set is trained using an LSTM neural network, and is trained into a battery evolved gas composition prediction model for batteries of different positive electrode materials at different SOCs.

[0048] Dividing the dataset: dividing each time series dataset into an input sequence and a target sequence;

[0049] That is, using the sliding window method, the time step corresponding to the time series data in the sliding window range is extracted from each time series dataset;

[0050] The time series in a time window is taken as the input sequence, and the time series of the next time step of the input sequence is taken as the target sequence;

[0051] LSTM neural network training: initialize the LSTM model parameters, pass the input sequence through the LSTM model for forward propagation, obtain the predicted sequence, calculate the loss function according to the predicted sequence and the target sequence, update the weight parameters of the LSTM model according to the calculated loss function gradient through the back propagation algorithm, repeat the training process until the preset training times are reached or until convergence.

[0052] S2. Collecting real-time data of the evolved gas components of each battery pack of the new energy vehicle, the evolved gas components including the concentrations of CO2, CO, H2 and VOC;

[0053] It should be noted that the evolved gas components can be the concentration of only one of CO2, CO, H2 and VOC, or the concentration of multiple gases.

[0054] S3. Selecting a corresponding battery evolved gas component prediction model according to the battery positive electrode material type and the SOC of the battery output by the battery management system of the new energy vehicle;

[0055] Corresponding to the above technical solution, the battery evolved gas component prediction model is a prediction model for a certain gas component evolved by the battery when the battery with different positive electrode materials is in different SOC conditions, or a combined prediction model for multiple gas components.

[0056] S4. Using the sliding window method, extracting the evolved gas component data corresponding to the time step in the sliding window range from the real-time data extracted in step S2 to form a to-be-predicted time series, wherein the time step in the sliding window range includes the time step of the real-time data extracted in S2;

[0057] S5. Inputting the to-be-predicted time series of S4 into the input trained battery evolved gas component prediction model selected in step S3 to output a result;

[0058] S6. Comparing the output result with a threshold value, if greater than the threshold value, outputting a warning signal.

[0059] Embodiment 2

[0060] A new energy vehicle battery online intelligent fault diagnosis method, which is different from embodiment 1 only in that:

[0061] In step S1

[0062] During data collection, the gas detection system only monitors the CO2 evolution in real time, and obtains the time sequence of CO2 evolution of the battery with different positive electrode materials at different SOC;

[0063] After training the LSTM neural network, the prediction model of the battery CO2 evolution of the battery with different positive electrode materials at different SOC is obtained.

[0064] In step S2, the real-time data of CO2 evolution of each battery pack of the new energy vehicle is collected.

[0065] In step S3, the corresponding prediction model of the battery CO2 evolution is selected according to the battery positive electrode material type and the SOC of the battery output by the battery management system of the new energy vehicle.

[0066] In step S4, the sliding window method is used to extract the data of CO2 evolution corresponding to the time step in the sliding window range from the real-time data extracted in step S2 to form a predicted time sequence, and the time step in the sliding window range includes the time step of the real-time data extracted in S2.

[0067] In step S5, the predicted time sequence of S4 is input into the prediction model of the battery CO2 evolution selected in step S3, and the output result is output.

[0068] In step S6, the output result is compared with the threshold value, and if it is greater than the threshold value, an early warning signal is output.

[0069] When the battery with different positive electrode materials occurs thermal runaway at different SOC values, the main gas of the battery thermal gas is CO2, CO, H2, C x H y , C x H y O z , CH3F, C2H5F, POF3, HF. According to the chemical properties C x H y (except CH4) and C x H y O zAs VOCs, photoionization detection sensors can accurately monitor VOC levels. CO and H2 are monitored using electrochemical sensors, while CO2, CH4, and C2H5F can be accurately monitored using infrared sensors. CO2 accounts for a significant volume percentage in each of these sensors. Using CO2 as a thermal runaway warning signal offers a faster response time than using battery temperature and voltage as warning signals, providing more immediate battery safety monitoring. However, due to the high concentration of CO2 in air, significant concentration changes require a longer time to be observed. Therefore, existing technologies have not yet utilized CO2 as a thermal runaway warning signal. Existing technologies primarily use CO, H2, and alkanes and alkenes as monitoring signals. When the battery is at a low SOC, VOCs, CO, and H2 are released late and at low concentrations, making them difficult to apply to thermal runaway warnings for batteries in all conditions. Sensors for detecting CO and H2 need to be embedded in optical fibers and embedded within the battery pack to monitor their concentrations. Because optical fiber sensors cross-react to temperature and pressure changes, additional design is required to decouple these two signals.

[0070] The LSTM neural network in this embodiment is a special type of RNN (recurrent neural network). The LSTM network structure inputs the CO2 concentration at time t into the network in chronological order. In the LSTM unit, the output of the current state needs to be calculated based on the state of the memory cell and the input. The updated memory cell information is passed to the next hidden layer unit, and finally the CO2 concentration at time t+1 is output.

[0071] When training a prediction model for battery CO2 emissions, it's necessary to use previous CO2 data to predict future long-term battery CO2 emissions. This requires continuously looping the LSTM network structure, using the output at time t+1 as the input at time t+2. This cycle completes the prediction of long-term battery CO2 emissions. Experiments were conducted using battery CO2 data of various step sizes as network input. During network training, the network evaluates and optimizes the output predictions based on known battery CO2 emissions data. The average difference between the predicted and actual values ​​is calculated as the training target. The mean squared error between the predicted and actual values ​​is used as the loss function. A backpropagation algorithm is used to update the LSTM model's weight parameters based on the calculated loss function gradient. The training process is repeated until the preset number of training cycles is reached or convergence is achieved.

[0072] Therefore, the battery CO2 evolution prediction model of this embodiment can predict the subsequent CO2 evolution concentration of batteries based on existing real-time data collected from each battery pack of a new energy vehicle. This overcomes the problem of the existing technology of slow observation of significant changes in CO2 concentration. Batteries with different positive electrode materials at different SOCs all produce CO2 earlier and at higher concentrations, thus compensating for the shortcomings of using other gases as a thermal runaway warning signal.

[0073] If the precipitated gas component of this embodiment is replaced by other gas components instead of CO2, the time for generating the thermal runaway warning signal can also be shortened.

[0074] In step S6 of this embodiment, the threshold value corresponds to different values ​​according to the type of positive electrode material of the battery and the SOC state value of the battery. For example, when the battery SOC is 70%, the CO2 concentration threshold value is 500×10 -6 .

[0075] Example 3

[0076] This embodiment proposes a new energy vehicle battery online intelligent fault diagnosis method, the details of which are as follows:

[0077] S1. Training LSTM neural network

[0078] Experimental preparation:

[0079] Before the experiment, batteries containing different cathode materials were placed in a high-temperature chamber to heat the batteries. A gas detection system was installed in the chamber to monitor the evolution of CO2, CO, H2, and VOCs in real time at a measurement frequency of 1 Hz. For the non-inherent gases CO, H2, and VOCs in the atmosphere, gas evolution refers to the process by which the volume concentration of the gas changes from zero to a certain level. For the inherent gas CO2 in the atmosphere, gas evolution refers to the process by which the volume concentration of the gas deviates from the average volume concentration of CO2 in the atmosphere.

[0080] A battery testing system is set up in the high and low temperature chamber to charge and discharge the battery while heating it, and to measure the battery voltage, current and temperature. The measurement frequency is 10 Hz and the error is 0.1%.

[0081] Data Collection:

[0082] All experimental cells were first charged to 3.65 V using a constant current and constant voltage method (1 C charge and 0.01 C cutoff current, where C represents the rate). Multiple cells made of the same material were then discharged at different SOCs using controlled discharge durations. The initial temperature was (20 ± 2)°C. After charging and discharging to different SOCs, each cell was allowed to rest for at least 1 hour.

[0083] The batteries of the same positive electrode material at different SOC are placed in different high and low temperature boxes, each high and low temperature box is continuously heated to 150 ℃ at a speed of 5 ℃ / min based on the initial temperature and is kept, each battery stays in the high and low temperature box for more than 1h, and the gas detection system monitors the release of CO2, CO, H2 and VOC in real time.

[0084] The time sequence of the released gas components of the batteries of different positive electrode materials at different SOC is obtained, including the time sequence of the concentration of CO2, CO, H2 and VOC released.

[0085] The time sequence of the released gas components is classified according to the positive electrode material of the battery, and then classified into multiple time sequence data sets according to SOC.

[0086] It should be noted that the gas detection system can only monitor the concentration of CO2 in real time during data collection. The corresponding time sequence of the released gas components is the time sequence of the concentration of CO2.

[0087] Before the time sequence data set is used to train the LSTM neural network, the data merging step is also included:

[0088] Since the batteries of the same positive electrode material change similarly when the SOC value is close, the time sequence of the released gas components is divided into several groups according to the SOC distribution range (state of charge distribution range) divided into multiple continuous SOC distribution ranges, each group of time sequence of the released gas components corresponds to the SOC in the divided SOC distribution range, and the SOC distribution range includes SOC∈[0%, 5%], SOC∈(5%, 15%], SOC∈(15%, 25%], SOC∈(25%, 35%], SOC∈(35%, 45%], SOC∈(45%, 55%], SOC∈(65%, 75%], SOC∈(75%, 85%], SOC∈(85%, 95%], SOC∈(95%, 105%].

[0089] The battery of the new energy vehicle changes from one boundary of the SOC distribution range to another boundary during the charging and discharging process, which often takes tens of minutes, and the chemical composition of the battery is relatively stable during this process. The same model can be used to predict the change of the gas component.

[0090] The time sequence of the released gas components in each group is summed to obtain the average time sequence of the released gas components, that is, the released gas component data of the battery at different SOC values at the same time in each time sequence in the same SOC distribution range is averaged, and the average value of the released gas component at all times is grouped to obtain the time sequence of the released gas component of each SOC distribution range.

[0091] The time series data set of the evolved gas composition of each SOC distribution range is trained into a battery evolved gas composition prediction model for batteries with different positive electrode materials in different SOC distribution ranges using an LSTM neural network.

[0092] The time series data corresponding to the time steps in the sliding window range is extracted from each time series data set using a sliding window method to divide the data set;

[0093] The time series in a time window is taken as an input sequence, and the time series of the next time step of the input sequence is taken as a target sequence;

[0094] After dividing the time series of the evolved gas composition of each SOC distribution range into input sequences and target sequences;

[0095] LSTM neural network training: initialize the LSTM model parameters, pass the input sequence through the LSTM model for forward propagation to obtain a predicted sequence, calculate the loss function according to the predicted sequence and the target sequence, update the weight parameters of the LSTM model according to the calculated loss function gradient through the backpropagation algorithm, repeat the training process until the preset number of training times is reached or until convergence is achieved.

[0096] S2. Collecting real-time data of the evolved gas composition of each battery pack of the new energy vehicle, the evolved gas composition including the concentrations of CO2, CO, H2, and VOC;

[0097] It should be noted that the evolved gas composition is only the concentration of CO2.

[0098] S3. Selecting the corresponding battery evolved gas composition prediction model according to the battery positive electrode material type and the SOC of the battery output by the battery management system of the new energy vehicle;

[0099] Corresponding to the above technical solution, the battery evolved gas composition prediction model is a prediction model for the CO2 evolved by the battery when the battery with different positive electrode materials is in different SOC distribution ranges.

[0100] S4. Extracting the evolved gas composition data corresponding to the time steps in the sliding window range from the real-time data extracted in step S2 to form a predicted time series, the time steps in the sliding window range including the time steps of the real-time data extracted in S2;

[0101] S5. Inputting the predicted time series of S4 into the input trained battery evolved gas composition prediction model selected in step S3 to output a result;

[0102] S6. Comparing the output result with a threshold value, if it is greater than the threshold value, outputting a warning signal.

[0103] The embodiment groups the training data according to the SOC distribution range, trains the LSTM neural network respectively, selects the prediction model according to the positive electrode material and SOC of the battery pack, and uses the prediction model to predict the real-time data of the battery pack gas composition. The SOC in the dynamic process of the new energy vehicle lasts for tens of minutes in the same SOC distribution range, and the selected prediction model has good stability for predicting the battery pack gas composition, and does not need to frequently replace the prediction model.

Claims

1. A new energy vehicle battery online intelligent fault diagnosis method, characterized in that: The method comprises the following steps: Step 1: measuring the temporal variation data of various evolved gas components when batteries with different positive electrode materials are kept at a high temperature at different states of charge; The components of the precipitated gas include: the concentration of at least one of CO2, CO, H2 and VOCs; Step 2: Classify the data measured in step 1 into time series of different gases released by batteries with different cathode materials at different states of charge, and divide the time series into input series and target series; Step 3: Perform LSTM neural network training to obtain prediction models for the components of evolved gases when batteries with the same cathode material are at different states of charge. Step 4: Collect real-time data on the composition of the gas released from each battery pack of the new energy vehicle; Step 5: Select the corresponding battery gas composition prediction model based on the type of battery positive electrode material and battery state of charge output by the new energy vehicle battery management system; Step 6: Using the sliding window method, extract the gas composition data corresponding to the time steps within the sliding window range from the time steps of the real-time data extracted in step 4 to form the time series to be predicted; Step 7: Input the time series to be predicted in step 6 into the trained battery gas composition prediction model selected in step 5, and output the result; Step 8. Compare the output result with the threshold. If it is greater than the threshold, output a warning signal. The step 2 of dividing the time series into an input sequence and a target sequence includes: The sliding window method is used to extract the time series data corresponding to the time steps within the sliding window range from each time series data set; Take the time series sequence within a time window as the input sequence, and the time series sequence of the time step after the input sequence as the target sequence; The time series sequence divided into the input sequence and the target sequence in step 2 is obtained by processing the following steps: Data is merged based on the state of charge of the batteries corresponding to each time series. If the state of charge of the batteries belongs to the same state of charge distribution range, the time series of the gas components are divided into the same group; In each time series within the same state of charge distribution range, the average value of the evolved gas composition data of batteries at different states of charge at the same time is calculated, and the average value data of the evolved gas composition at all times constitutes the time series of the evolved gas composition in each state of charge distribution range.

2. The method according to claim 1, characterized in that The precipitated gas component is the concentration of one of CO2, CO, H2 and VOCs.

3. The method according to claim 1, characterized in that The step 3 includes: initializing the LSTM model parameters, forward propagating the input sequence through the LSTM model to obtain a predicted sequence, calculating the loss function based on the predicted sequence and the target sequence, updating the weight parameters of the LSTM model according to the calculated loss function gradient through the backpropagation algorithm, and repeating the training process until a preset number of training times is reached or until convergence.

4. The method according to claim 1, wherein In step 1, for the non-inherent gases CO, H2 and VOCs in the atmosphere, the precipitated gas composition refers to the process in which the volume concentration of the gas changes from zero to something; for the inherent gas CO2 in the atmosphere, the precipitated gas composition refers to the process in which the volume concentration of the gas deviates from the average volume concentration of CO2 in the atmosphere.

5. The method according to claim 1, wherein The step 3 is used to obtain prediction models for various precipitated gas components when batteries of the same positive electrode material are in different state of charge distribution ranges.

6. The method according to claim 5, characterized in that The battery precipitated gas composition prediction model selected in step 5 is a battery precipitated gas composition prediction model when batteries with different positive electrode materials are in different state of charge distribution ranges.

7. The method according to claim 6, characterized in that The state of charge distribution range includes SOC∈[0%, 5%], SOC∈(5%, 15%], SOC∈(15%, 25%], SOC∈(25%, 35%], SOC∈(35%, 45%], SOC∈(45%, 55%], SOC∈(65%, 75%], SOC∈(75%, 85%], SOC∈(85%, 95%], SOC∈(95%, 105%]).

8. The method according to any one of claims 1 to 7, characterized in that The component of the precipitated gas is CO2.

Images