XGBoost-based error compensation method for lithium battery thermal runaway sound source positioning

Abstract

The application discloses a lithium battery thermal runaway sound source positioning error compensation method based on XGBoost, sound signals in the energy storage cabin are collected in a fixed microphone array, including time delay, mutual power spectrum data and sound source positioning coordinates calculated by a traditional geometric method, then, by comparing the geometric positioning coordinates with actual sound source position coordinates, positioning errors are calculated, and feature vectors are constructed. XGboost model is trained by using the feature vectors. In actual application, time delay, mutual power spectrum data and geometric positioning coordinates calculated in real time are input, error values obtained by the prediction model are used for correcting the traditional geometric sound source positioning coordinates, so that more accurate sound source positions are obtained. The application corrects errors generated by geometric positioning by introducing the XGBoost error compensation model, so that the position of the thermal runaway lithium battery can be accurately positioned.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of lithium battery safety, and particularly relates to a lithium battery thermal runaway sound source positioning error compensation method based on XGBoost. BACKGROUND

[0002] Lithium ion batteries have the characteristics of long discharge time, fast response speed, high conversion efficiency, not restricted by natural conditions, and easy to scale application, etc., and are the main development direction of electrochemical energy storage power stations. However, the organic electrolyte used by lithium ion batteries has a low boiling point and is flammable, and the material system has a high calorific value and currently lacks an effective fault early warning method. After a fault occurs in the battery body or electrical equipment, etc., the exothermic side reaction of the battery material is easily triggered, leading to battery thermal runaway, and further possibly evolving into a major safety accident such as a storage system combustion explosion.

[0003] In order to prevent the explosion of the sealed metal shell, a safety valve is now provided at the top of the lithium ion battery, which is a standard safety valve for each lithium ion battery and is the most important explosion-proof barrier. When the internal pressure of the battery is too large, the safety valve at the top will open to release pressure and prevent explosion. A specific sound signal will be generated when the safety valve is opened, and at this time the thermal runaway is not severe. Timely and effective identification and positioning of the sound signal will enable early warning of thermal runaway and make fault handling more targeted.

[0004] The traditional lithium battery thermal runaway positioning method mainly relies on spatial positioning algorithms. The sound source signal is collected by a microphone array, and the time delay difference of the signal received by each microphone is calculated to deduce the position of the sound source. The microphone array is usually configured in a linear or two-dimensional array to ensure that the sound source signal can be captured within a certain range. The position of each microphone is known, and the positioning accuracy of the sound source is optimized by arranging the spacing. Specifically, the sound waves emitted by the sound source propagate to each microphone at a fixed speed (about 343 m / s in air). The time difference (time delay difference) of the arrival of sound waves at different microphones can be calculated by dividing the distance difference between the sound source and the two microphones by the speed of sound. By knowing the positions of the microphones and the time delay difference between each pair of microphones, a system of equations can be established, and the three-dimensional coordinates of the sound source can be deduced using geometric methods. This usually involves measuring at least three microphones to ensure that the specific position of the sound source can be determined. However, in actual applications, the noise environment in the battery cabin is complex and diverse, including equipment operation noise and air flow noise, etc. These noises will interfere with the time delay measurement, thereby affecting the positioning accuracy. The accuracy of the spatial positioning method usually depends on the geometric structure of the microphone array and the ideal assumption of the speed of sound. Any irregular microphone arrangement, speed of sound variation, or microphone failure in the array can cause significant errors in positioning. In addition, in complex scenarios, the sound source propagation path may be affected by reflection, diffraction, and other nonlinear factors, further reducing the accuracy of positioning. SUMMARY

[0005] The purpose of the present application is to provide an XGBoost-based lithium battery thermal runaway sound source positioning error compensation method, which solves the technical problems of low positioning accuracy of lithium battery thermal runaway in the prior art due to irregular microphone arrangement, sound speed variation or failure of microphones in the array, and the influence of reflection, diffraction and other nonlinear factors on the sound source propagation path.

[0006] In order to solve the above technical problems, the present application adopts the following technical solutions:

[0007] The XGBoost-based lithium battery thermal runaway sound source positioning error compensation method specifically comprises the following steps:

[0008] Step S1: build and train a battery thermal runaway sound source positioning error compensation system

[0009] Step S1.1: build a lithium battery thermal runaway sound source positioning error compensation system; the system includes a sound collection component placed in the energy storage cabin where the lithium battery is placed, and a controller set at a remote end, wherein the controller includes a sound signal recognition model and an XGBoost model training module;

[0010] Step S1.2: collect training data; specifically comprising the following steps:

[0011] Step S1.2.1: collect real lithium battery safety valve opening audio in advance and store it in an audio player;

[0012] Step S1.2.2: place the audio player at the position of the lithium battery in the energy storage cabin as described in step S1 and play it to simulate the real lithium battery safety valve opening sound source position, and use the sound collection component to collect the audio signal of each play;

[0013] Step S1.2.3: perform the operation of step S1.2.2 on each position where the lithium battery is placed in the energy storage cabin, and repeat M times at each position, use the sound collection component to collect four audio signals at each play and send them to the controller; the four audio signals are recorded as a group;

[0014] Step S1.3: the controller processes the training data collected in step S2 and builds a lithium battery safety valve opening sound signal recognition model;

[0015] Step S1.4: the XGBoost model training module trains the XGBoost model using the lithium battery safety valve opening sound signal data identified by the sound signal recognition model, so that it can predict the positioning error;

[0016] Step S2: The sound sensor collects the sound information in the energy storage cabin in real time and transmits it to the controller, and the controller realizes real-time accurate positioning of the sound source of the lithium battery thermal runaway by using the trained error compensation system for positioning the sound source of the lithium battery thermal runaway.

[0017] Further optimization, in step S1.1, a three-dimensional coordinate system OXYZ is established in the energy storage cabin, and the real coordinates of the safety valve of each battery are measured in advance;

[0018] The sound collection module includes four sound sensors, which are respectively referred to as a first sound sensor, a second sound sensor, a third sound sensor and a fourth sound sensor; wherein the first sound sensor is arranged at the origin of the three-dimensional coordinate system, the second sound sensor is arranged on the X-axis of the three-dimensional coordinate system, and the distance to the origin is L1; the third sound sensor is arranged on the Y-axis of the three-dimensional coordinate system, and the distance to the origin is L2; the fourth sound sensor is arranged on the Z-axis of the three-dimensional coordinate system, and the distance to the origin is L3.

[0019] Further optimization, the sound sensor is a microphone.

[0020] Further optimization, in step S1.2.3, the audio player is placed at the safety valve position to play audio, and 100 times of audio are played at each safety valve position of the battery, and the audio duration is 3 seconds.

[0021] Further optimization, in step S1.3, the controller processes the training data collected in step S2 and constructs a sound signal recognition model for the opening of the lithium battery safety valve, which includes the following steps:

[0022] Step S1.3.1: Wavelet denoising is performed on the data collected by each sound sensor to remove background noise;

[0023] Step S1.3.2: The original audio signal is processed through a pre-emphasis filter to enhance the energy of high-frequency signals;

[0024] Step S1.3.3: Feature extraction is performed on the read audio information using Mel-frequency cepstral coefficients (MFCC), and the features extracted by each sound sensor are normalized;

[0025] Step S1.3.4: A preloaded SVM model is used to construct a feature vector, a data set based on MFCC parameters is constructed, and a sound signal recognition model for the opening of the lithium battery safety valve is established; the sound signal data recognized by the sound signal recognition model includes the relative time delay of the sound signal collected by different sensors, the mutual power spectrum information, the sound source positioning coordinates calculated by the geometric method, and the error value between the actual coordinates of the sound source and the geometric positioning coordinates.

[0026] Further optimization, each group of audio signal corresponding to 6 time delay difference and 18 mutual power spectrum, wherein, time delay difference includes T 12 、T 13 、T 14 、T 23 、T 24 、T 34 ; the real part of mutual power spectrum is R 12 、R 13 、R 14 、R 23 、R 24 、R 34 , the imaginary part is I 12 、I 13 、I 14 、I 23 、I 24 、I 34 , the modulus is M 12 、M 13 、M 14 、M 23 、M 24 、M 34 ; the real part, imaginary part and modulus of time delay mutual power spectrum and error value are normalized separately and the normalizer is saved for subsequent real-time detection of new data normalization input.

[0027] Further optimization, the relative time delay is obtained by generalized cross correlation algorithm, after windowing processing of each group of four signals collected, the mutual power spectrum is calculated by using fast Fourier transform (FFT), and the phase normalization is carried out, so as to ensure that the influence of signal intensity on time delay estimation is minimum; the mutual power spectrum is subjected to inverse FFT to obtain cross correlation result, and the time delay estimation value is calculated, so as to provide accurate time difference data for subsequent positioning;

[0028] Wherein, the signal is windowed to reduce spectrum leakage, and the specific calculation formula is as follows:

[0029]

[0030] Wherein, w[k] is the Hanning window function, and K is the length of the signal; wherein, w[k] is the Hanning window function, and K is the length of the signal; that is, the total number of sampling points of the signal, k represents the position index of the current window function, and the range is k=0, 1,…, K-1.

[0031] The windowed signal: sig w = sig·w; refsig w = refsig·w;

[0032] w is the weight array of the window function, and the length is equal to the length of sig and refsig, so that the signal is zero at both ends, thereby reducing spectrum leakage and improving frequency energy distribution.

[0033] Fast Fourier Transform, FFT, of the sound signal:

[0034] SIG(f) = F(sig w ), RFSIG(f) = F(refsig w )

[0035] where F denotes the Fast Fourier Transform, SIG(f) and RFSIG(f) are the frequency domain representations of sig w and refsig w respectively;

[0036] R(f) = SIG(f) • RFSIG*(f)

[0037] where RFSIG*(f) is the complex conjugate of RFSIG(f);

[0038] Normalized cross-power spectrum:

[0039]

[0040] where |R(f)| is the amplitude of the cross-power spectrum and ∈ is a constant;

[0041]

[0042] where F -1 denotes the inverse Fourier Transform, r(τ) is the cross-correlation function

[0043] cc = [r(-τ max ),..., r(τ max )]

[0044]

[0045] where τ max is the maximum time delay, argmax(|cc|) is the position of the maximum of the cross-correlation, interp is the interpolation factor and fs is the sampling rate;

[0046] Cross-power spectrum extraction of the real and imaginary parts and the modulus as features:

[0047] Re(R(f)) = real(R(f))

[0048] Im(R(f)) = imag(R(f))

[0049]

[0050] Wherein, R(f) is the cross-power spectrum of the signal sig and the reference signal refsig, Re(R(f)) is the real part of the cross-power spectrum, Im(R(f)) is the imaginary part of the cross-power spectrum, |R(f)| is the modulus of the cross-power spectrum;

[0051] Further optimization, a nonlinear equation is constructed for each sound sensor, representing the relationship between the distance difference corresponding to the time delay difference of the sound source to the sound sensor and the distance difference;

[0052] First, the sound sensor located at the coordinate origin is set as the reference sound sensor, and the distance R0 from the reference sound sensor to the sound source is calculated:

[0053]

[0054] Then, the distance R from each i-th sound sensor to the sound source is calculated i :

[0055]

[0056] The distance difference corresponding to the time delay difference is calculated again:

[0057] d i =τ i ·c

[0058] Then a nonlinear equation is constructed:

[0059] R i -R0-d i =0

[0060] Optimization using nonlinear least squares, the goal is to minimize the residual of the above equation set:

[0061]

[0062] Wherein, the position coordinates of the sound source are (x, y, z); (x i ,y i ,z i ) is the position coordinates of the i-th sound sensor; τ i is the time delay difference of the i-th sound sensor; c is the sound speed; R0 is the distance from the sound source to the reference sound sensor; R i is the distance from the sound source to the i-th sound sensor; di is the distance difference corresponding to the time delay difference of the i-th sound sensor; N is the number of sound sensors; min (x,y,z) is the minimization objective function.

[0063] Further optimization, in the step S1.4, set the obtained sound source position coordinates of the geometric positioning calculation as (x, y, z), the error of the XGBoost prediction as (Delta x, Delta y, Delta z), and the accurate position coordinates of the corrected sound source position as (x+Delta x, y+Delta y, z+Delta z).

[0064] Further optimization, in the step S2, the controller uses the trained battery thermal runaway sound source positioning error compensation system to process the received sound signal as follows, specifically comprising:

[0065] Step S2.1: normalize the extracted features using the method described in step S1.3, and use the pre-trained SVM model and normalizer for prediction to determine whether the sound signal is the sound of the lithium battery safety valve opening, and if so, set the lithium battery as the target lithium battery;

[0066] Step S2.2: real-time acquisition of the sound signal of the target lithium battery identified in the energy storage cabin, calculation of the time delay, cross-power spectrum data and sound source positioning coordinates calculated by the traditional geometric method, and input into the normalizer, and use the pre-trained normalizer to normalize each feature data frame to ensure that different features are on the same scale; horizontally stack the normalized feature array to form a comprehensive feature matrix, which contains all the features used for model prediction, and use the loaded model to predict the combined features and output the error value;

[0067] Step S2.3: correct the sound source position coordinates output by the geometric positioning method with the error value predicted by the model, and finally locate the accurate position of the target lithium battery.

[0068] Compared with the prior art, the present application has the following beneficial effects:

[0069] The present application collects the sound signal in the energy storage cabin through the sound sensor, then compares the geometric positioning coordinates with the actual sound source position coordinates, calculates the positioning error, and constructs the feature vector. The XGboost model is trained using these feature vectors. In actual application, the time delay, cross-power spectrum data and geometric positioning coordinates calculated in real time are input, and the error value obtained by the prediction model is used to correct the traditional geometric sound source positioning coordinates, realizing real-time accurate positioning of the lithium battery thermal runaway sound source. The problem of low positioning accuracy of lithium battery thermal runaway in the prior art is solved. BRIEF DESCRIPTION OF DRAWINGS

[0070] Figure 1 is a schematic diagram of the battery thermal runaway sound source positioning error compensation system described in the present application;

[0071] Figure 2 is a block diagram of the target lithium battery positioning system;

[0072] Figure 3 is an overall diagram of the energy storage cabin constructed in equal proportions;

[0073] Figure 4 is a schematic diagram of the energy storage cabin door opening;

[0074] Figure 5 is a schematic diagram of the microphone layout. DETAILED DESCRIPTION

[0075] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the present application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0076] The XGBoost-based lithium battery thermal runaway sound source positioning error compensation method specifically includes the following steps: step S1: building and training a battery thermal runaway sound source positioning error compensation system, as shown in Figure 1 .

[0077] Step S1.1: build a lithium battery thermal runaway sound source positioning error compensation system, as shown in Figure 2 . The system includes an energy storage cabin where lithium batteries are placed, a sound collection component installed in the energy storage cabin, and a controller set at a remote end, which includes a sound signal recognition model and an XGBoost model training module.

[0078] In this embodiment, an energy storage cabin equipment model is designed in equal proportions, with a length of 2 meters, a width of 1 meter, and a height of 1.2 meters, as shown in Figure 3 .

[0079] The sound sensor is a microphone.

[0080] A three-dimensional coordinate system OXYZ is established in the energy storage cabin. The channel one microphone is located at the coordinate origin, with coordinates (0, 0, 0). The channel two microphone is located 1 meter away from the origin in the x direction, defined as the width direction of the energy storage cabin, with coordinates (100, 0, 0). The channel three microphone is located 2 meters away from the origin in the y direction, defined as the length direction of the energy storage cabin, with coordinates (0, 200, 0). The channel four microphone is located 1.2 meters away from the origin in the z direction, defined as the height direction of the energy storage cabin, with coordinates (0, 0, 120). The specific layout form is shown in Figure 4 . A support is placed in the energy storage cabin for placing a box to simulate a battery module, as shown in Figure 4 . At the same time, since the coordinate system is established, as shown in Figure 5 ​As shown, the three-dimensional coordinates of the safety valve of each box need to be measured as the true value for subsequent calculation of error value. The error values of the safety valve positions of each module are collected to train the XGBoost error compensation model. At the same time, the known true value coordinates of the box safety valve are used to calibrate the error through multiple experiments and data collection. The microphone position and array configuration are adjusted to reduce the positioning error.

[0081] Step S1.2: Collecting training data, specifically including the following steps:

[0082] Step S1.2.1: Collecting real audio of lithium battery safety valve opening in advance, each audio is three seconds long, and is stored in an audio player;

[0083] Step S1.2.2: Place the audio player at the position of the lithium battery in the energy storage cabin as described in step S1, and play it to simulate the real sound source position of the lithium battery safety valve opening, and use the sound collection component to collect the audio signal of each play;

[0084] Step S1.2.3: Perform the operation of step S1.2.2 on each position of the lithium battery in the energy storage cabin, and repeat M times at each position. Use the sound collection component to collect four audio signals during each play and send them to the controller. The four audio signals are recorded as a group.

[0085] In this embodiment, 100 times of audio playing are performed at each lithium battery position, and the microphone array is used to collect the audio signal of each play and send it to the controller.

[0086] Step S1.3: The controller processes the training data collected in step S2 and builds a lithium battery safety valve opening sound signal recognition model. Specifically, it includes the following steps:

[0087] Step S1.3.1: Wavelet denoising is performed on the data collected by each sound sensor to remove background noise;

[0088] Step S1.3.2: The original audio signal is processed through a pre-emphasis filter to enhance the energy of high-frequency signals;

[0089] Step S1.3.3: Use Mel Frequency Cepstral Coefficient (MFCC) to extract features from the read audio information, and normalize the extracted features corresponding to each sound sensor;

[0090] Step S1.3.4: Constructing a feature vector using a pre-loaded SVM model, completing the construction of the MFCC parameter-based data set, and establishing a lithium battery safety valve opening sound signal recognition model; the sound signal recognition model-recognized lithium battery safety valve opening sound signal data includes the relative time delay of the sound signals collected by different sensors, mutual power spectrum information, sound source positioning coordinates calculated by a geometric method, and error values between the actual coordinates of the sound source and the geometric positioning coordinates.

[0091] The 6 time delay features include T12, T13, T14, T23, T24, and T34, representing the time delay between each pair of microphones. The 18 mutual power spectrum features include the real parts R 12 , R 13 , R 14 , R 23 , R 24 , R 34 , the imaginary parts I 12 , I 13 , I 14 , I 23 , I 24 , I 34 , and the module values M

[0092] M 12 , M 13 , M 14 , M 23 , M 24 , M 34 . The geometric positioning coordinates (x, y, z) represent the positioning coordinates obtained by the spherical interpolation method. The error values (Δx, Δy, Δz) represent the deviation between the geometric positioning coordinates and the known true value coordinates.

[0093] Each experimental result forms a feature vector, marked as a.csv file, and the file name includes the feature name. 100 data points are collected for each lithium battery position to form a complete data set. A total of 20 sets of position data are collected, with 100 data in each set, for a total of 2000 groups. Important features can be added according to the model prediction error of the position to optimize the model prediction stability.

[0094] Step S1.4: The XGBoost model training module trains the XGBoost model using the lithium battery safety valve opening sound signal data recognized by the sound signal recognition model, so that it can predict the positioning error.

[0095] Before model training, data normalization is required. Initialize the XGBoost regression model xgb.XGBRegressor, and set the objective function as the mean square error reg:squarederror. Define a hyperparameter grid param_grid containing multiple hyperparameter options, including the maximum depth of the decision tree max_depth (which controls the depth of the tree to prevent overfitting), the learning rate learning_rate (which controls the step size of weight updates in each iteration of the model), the number of base learners (trees) to be trained n_estimators, the proportion of random sampling for each tree subsample, the proportion of randomly selected features for each tree colsample_bytree (similar to subsample, controlling the random sampling of features can improve the robustness of the model), the minimum sample weight in the sub-tree min_child_weight, and the regularization weight reg_alpha, reg_lambda. The specific values of the hyperparameters are shown in Table 1.

[0096] Table 1 Specific values of hyperparameters

[0097] Parameter name Optimal value colsample_bytree 0.6 learning_rate 0.01 max_depth 5 min_child_weight 1 n_estimators 50 subsample 0.7

[0098] Use GridSearchCV to perform grid search to find the best combination of hyperparameters, and use 5-fold cross-validation (cv=5) to evaluate the performance of the model. Finally, train the model using the best combination of hyperparameters and save it. Use the trained model to predict the test set and calculate the evaluation metrics, including mean square error mse, root mean square error rmse, and R 2 (r2) score.

[0099] Visualize the feature importance and display it on the graphical interface for subsequent testing of inaccurate places to increase important features. Use the logging library to record important information during training, including the best parameters, the best cross-validation score, evaluation metrics, etc. for subsequent analysis and debugging. Finally, save the data normalizer and the optimal error compensation model for real-time detection calls later.

[0100] Step S2: The sound sensor collects sound information in the energy storage cabin in real time and transmits it to the controller. The controller uses the trained battery thermal runaway sound source positioning error compensation system to realize real-time accurate positioning of the lithium battery thermal runaway sound source.

[0101] In this embodiment, detection is carried out in a real scene, the microphone array in the energy storage compartment remains unchanged, the audio player is accurately placed at the position of the simulated battery safety valve for playing, the signals of multiple microphones are collected in real time to obtain complete sound field information, and wavelet transform is used for multi-resolution analysis of the collected signals to remove background noise. Through the steps of decomposition, threshold processing and reconstruction, clearer signals are extracted. A pre-emphasis filter is applied to enhance the signal components and improve the effect of subsequent feature extraction. Time delay features and cross-power spectrum features are extracted from the denoised and pre-emphasized signals. The generalized cross-correlation phase transform (GCC-PHAT) algorithm is used to estimate the time delay between each pair of microphones, forming 6 groups of time delay features. The positions and time delays between the microphones are combined with the spherical interpolation method to calculate the coordinates in the geometric positioning aspect. 18 groups of cross-power spectrum features are calculated to describe the frequency domain characteristics and phase information of the signals.

[0102] To eliminate the scale differences between different features, the time delay features, cross-power spectrum features, and geometric positioning coordinates are normalized to ensure consistent input into the error compensation model. The normalized data is input into the pre-trained error compensation model, and the predicted geometric positioning error is represented in the form of (x, y, z) coordinates. The offset predicted by the error model is applied to the geometric positioning coordinates to obtain the corrected sound source position. This process significantly reduces the positioning error fluctuations caused by environmental instability (such as reflection, temperature change, and equipment noise).

[0103] Through repeated field tests and data collection, the layout of the microphone array and the signal processing algorithm are optimized to adapt to different energy storage compartment designs and environmental conditions. At the same time, the error compensation model is continuously updated and trained to improve its adaptability to environmental changes.

[0104] The audio player is accurately placed at different positions of the simulated battery safety valve for repeated tests. Several sets of experimental data are shown in Table 2, including the true value coordinates of the sound source position, the geometric positioning coordinates of the sound source, and the corrected sound source position.

[0105] Table 2 Simulation experimental data (unit: cm)

[0106] True coordinates of sound source position Geometrically located coordinates of sound source Corrected coordinates of sound source position (63,118.2,16.5) (64,118,15) (64,118.1,15.9) (88.2,94.3,16.5) (89,95,16) (88.6,95,16.5) (63,70.5,81) (52,73,80) (60.1,72,80.2) (63,70.5,57.4) (64,71,57) (63.5,70.7,57.4)

[0107] From the data in Table 2, it can be clearly seen that the XGboost model is trained using the sound feature vector, the error value obtained through the prediction model is used to correct the traditional geometric sound source positioning coordinates, the corrected sound source position coordinates have very small errors with the true value, and real-time accurate positioning of the lithium battery thermal runaway sound source is achieved.

[0108] With the above ideal embodiments according to the present application as the inspiration, through the above description, relevant staff can make various changes and modifications without deviating from the technical idea of the present application. The technical scope of the present application is not limited to the content of the specification, and must be determined according to the scope of the claims.

Claims

1. A lithium battery thermal runaway sound source localization error compensation method based on XGBoost, characterized by: The specific steps include: Step S1: Build and train the battery thermal runaway sound source localization error compensation system Step S1.1: Build a lithium battery thermal runaway sound source localization error compensation system; the system includes a sound collection component installed in the energy storage compartment where the lithium battery is placed, and a controller located remotely, the controller including an acoustic signal recognition model and an XGBoost model training module; Step S1.2: Collect training data: Specifically includes the following steps: Step S1.2.1: Pre-collect the audio of the actual lithium battery safety valve opening and store it in an audio player; Step S1.2.2: Place the audio player in the position of the lithium battery in the energy storage compartment described in step S1. Set and play to simulate the real sound source position of the lithium battery safety valve opening, and use the sound collection component to collect the audio signal of each playback; Step S1.2.3: Perform step S1.2.2 for each lithium battery location in the energy storage compartment, repeating M times for each location. Use the sound collection component to collect four audio signals during each playback and send them to the controller; the four audio signals are recorded as a group; Step S1.3: The controller processes the training data collected in step S2 and constructs a lithium battery The safety valve opening acoustic signal recognition model specifically includes the following steps: Step S1.3.1: Perform wavelet denoising on the data collected by each sound sensor to remove background noise; Step S1.3.2: Process the original audio signal through the pre-emphasis filter to enhance the high frequency signal. The energy of the number; Step S1.3.3: Use Mel-Frequency Cepstral Coefficients (MFCCs) to extract features from the read audio information, and normalize the features extracted by each sound sensor. Step S1.3.4: Use the pre-loaded SVM model to construct the feature vector and complete the MFCC-based A parameter data set is constructed to establish a lithium battery safety valve opening acoustic signal recognition model. The lithium battery safety valve opening acoustic signal data identified by the acoustic signal recognition model includes the relative time delay of the sound signals collected by different sensors, cross-power spectrum information, the sound source location coordinates calculated by geometric methods, and the error value between the actual coordinates of the sound source and the geometric location coordinates. Step S1.4: The XGBoost model training module uses the lithium battery safety valve opening acoustic signal data identified by the acoustic signal recognition model to train the XGBoost model so that it can predict the positioning error; Step S2: The sound sensor collects sound information in the energy storage compartment in real time and transmits it to the controller. The controller uses the trained battery thermal runaway sound source positioning error compensation system to achieve real-time and accurate positioning of the lithium battery thermal runaway sound source.

2. The XGBoost-based lithium battery thermal runaway sound source localization error compensation method according to claim 1, characterized in that: In step S1.1, a three-dimensional coordinate system OXYZ is established in the energy storage cabin, and the real coordinates of the safety valve of each battery are measured in advance; The sound collection module includes four sound sensors, which are respectively recorded as the first sound sensor, the second sound sensor, the third sound sensor and the fourth sound sensor; among them, the first sound sensor is set at the origin of the three-dimensional coordinate system, the second sound sensor is set on the X-axis of the three-dimensional coordinate system, and the distance to the origin is L1; the third sound sensor is set on the Y-axis of the three-dimensional coordinate system, and the distance to the origin is L2; ​​the fourth sound sensor is set on the Z-axis of the three-dimensional coordinate system, and the distance to the origin is L3.

3. The XGBoost-based lithium battery thermal runaway sound source localization error compensation method according to claim 2, characterized in that: The sound sensor is a microphone.

4. The XGBoost-based lithium battery thermal runaway sound source localization error compensation method according to claim 2, characterized in that: In step S1.2.3, the audio player is placed at the safety valve position to play the audio. Each battery safety valve position is played 100 times, and the audio duration is 3 seconds.

5. The XGBoost-based lithium battery thermal runaway sound source localization error compensation method according to claim 4 is characterized in that: Each group of audio signals corresponds to 6 time delay differences and 18 cross-power spectra, where the time delay difference includes T 12 、T 13 、T 14 、T 23 、T 24 、T 34 ; The real part of the cross power spectrum is R 12 、R 13 、R 14 、R 23 、R 24 、R 34 , the imaginary part is I 12 , I 13 , I 14 , I 23 , I 24 , I 34 , the modulus is M 12 、M 13 、M 14 、M 23 、M 24 、M 34 ; The real part, imaginary part, modulus and error value of the time delay cross power spectrum are normalized separately and the normalizer is saved for subsequent real-time detection and comparison with the normalized input of new data.

6. The XGBoost-based lithium battery thermal runaway sound source localization error compensation method according to claim 5, characterized in that: The relative time delay is obtained through a generalized cross-correlation algorithm. After windowing each group of four collected signals, the cross-power spectrum is calculated using a fast Fourier transform (FFT) and phase normalized to minimize the impact of signal strength on the time delay estimate. The cross-power spectrum is then subjected to an inverse FFT to obtain a cross-correlation result, and the time delay estimate is calculated to provide accurate time difference data for subsequent positioning. Among them, the signal is windowed to reduce spectrum leakage. The specific calculation formula is as follows: Where w[k] is the Hanning window function, K is the length of the signal, that is, the total number of sampling points of the signal, and k represents the position index of the current window function, ranging from k = 0, 1, ..., K-1. Windowed signal: sig w =sig·w;refsig w =refsig·w; w is the weight array of the window function, whose length is equal to the length of sig and refsig, so that the transition between the two ends of the signal is zero, thereby reducing spectrum leakage and improving the frequency domain energy distribution; Perform Fast Fourier Transform FFT calculation on the sound signal: SIG(f)=F(sig w ),REFSIG(f)=F(refsig w ) Where F stands for fast Fourier transform, SIG(f) and REFSIG(f) are sig w and refsig w Frequency domain representation of ; R(f)=SIG(f)·REFSIG*(f) where REFSIG*(f) is the complex conjugate of REFSIG(f); Normalized cross power spectrum: Where |R(f)| is the amplitude of the cross-power spectrum and ∈ is a constant; Among them, F -1 represents the inverse Fourier transform, r(τ) is the cross-correlation function cc=[r(-τ max ),……,r(τ max )] Among them, τ max is the maximum delay, argmax(|cc|) is the position of the maximum cross-correlation value, interp is the interpolation factor, and fs is the sampling rate; The cross power spectrum extracts the real and imaginary parts as well as the modulus as features: Re(R(f))=real(R(f)) Im(R(f))=imag(R(f)) Wherein, R(f) is the cross power spectrum of the signal sig and the reference signal refsig, Re(R(f)) is the real part of the cross power spectrum, Im(R(f)) is the imaginary part of the cross power spectrum, and |R(f)| is the modulus of the cross power spectrum.

7. The XGBoost-based lithium battery thermal runaway sound source localization error compensation method according to claim 6, characterized in that: Construct a nonlinear equation for each sound sensor to express the relationship between the distance difference between the sound source and the sound sensor and the distance difference corresponding to the time delay difference; First, set the sound sensor at the coordinate origin as the reference sound sensor and calculate the distance R0 from the reference sound sensor to the sound source: Then, calculate the distance R from each i-th sound sensor to the sound source i : Then calculate the distance difference corresponding to the delay difference: d i =t i ·c Then construct the nonlinear equation: R i -R0-d i =0 Using nonlinear least squares optimization, the goal is to minimize the residual of the above system of equations: Among them, the position coordinates of the sound source are (x, y, z); (x i ,y i ,z i ) is the position coordinate of the i-th sound sensor; τ i is the time delay difference of the i-th sound sensor; c is the speed of sound; R0 is the distance from the sound source to the reference sound sensor; R i is the distance from the sound source to the i-th sound sensor; di is the distance difference corresponding to the time delay difference of the i-th sound sensor; N is the number of sound sensors; min (x,y,z) To minimize the objective function.

8. The XGBoost-based lithium battery thermal runaway sound source localization error compensation method according to claim 7, characterized in that: In step S1.4, the sound source position coordinates obtained by geometric positioning calculation are set to (x, y, z), and the error predicted by XGBoost is (Δx, Δy, Δz). Then, the precise position coordinates of the corrected sound source position are (x+Δx, y+Δy, z+Δz).

9. The XGBoost-based lithium battery thermal runaway sound source localization error compensation method according to claim 8, characterized in that: In step S2, the controller uses the trained battery thermal runaway sound source localization error compensation system to process the received sound signal as follows: Step S2.1: Normalize the extracted features using the method described in step S1.3, and use the pre-trained SVM model and normalizer to perform predictions to determine whether the sound signal is the sound of a lithium battery safety valve opening. If so, set the lithium battery as the target lithium battery; Step S2.2: Acquire the acoustic signals of the target lithium battery identified in the energy storage compartment in real time, calculate the time delay, cross-power spectrum data, and the sound source location coordinates calculated by traditional geometric methods, and input them into the normalizer. Use the pre-trained normalizer to normalize each feature data frame to ensure that different features are on the same scale; stack the normalized feature arrays horizontally to form a comprehensive feature matrix, which contains all the features used for model prediction. Use the loaded model to predict the combined features and output the error value; Step S2.3: Use the error value predicted by the model to correct the sound source position coordinates output by the geometric positioning method, and finally locate the precise position of the target lithium battery.

Images