A safety early warning method and system for electric vehicle batteries

Abstract

This invention discloses a safety early warning method and system for electric vehicle batteries, relating to the technical field of battery safety. The method involves: acquiring an initial signal set and calculating sensitivity indices; calculating an average sensitivity index based on each sensitivity index and using it as the objective function; obtaining a preset parameter combination and optimizing it to obtain the optimal parameter combination; updating the parameter configuration of the preset algorithm based on the optimal parameter combination to obtain the target algorithm; and decomposing the acoustic signal to obtain the target mode function; performing time-frequency transformation on the target mode function to obtain a time-frequency spectrum and extracting fault features; and providing fault early warning based on the fault features. By optimizing the preset algorithm parameters, the efficiency and accuracy of signal decomposition are improved, extracting the acoustic signal of battery impact exhaust affected by external interference, accurately identifying fault features, and achieving early warning. This solves the problems of difficult signal extraction and feature recognition, improves the timeliness and accuracy of battery fault early warning, and avoids the risk of temperature runaway.

Description

Technical Field

[0001] This invention relates to the field of battery safety technology, and specifically to a safety warning method and system for electric vehicle batteries. Background Technology

[0002] Electric vehicle batteries are prone to safety hazards such as exhaust and thermal runaway when bumped or knocked. However, their exhaust acoustic signals are easily interfered with and fault characteristics are difficult to identify. Existing signal decomposition methods have problems with insufficient efficiency and accuracy. Publication No. CN115911623A discloses a method and system for diagnosing battery thermal runaway based on acoustic signals in an energy storage system. The method involves acquiring mixed acoustic signals from the battery energy storage system compartment, including exhaust acoustic signals, ambient noise, and other noise; performing spectral subtraction on the mixed acoustic signals to remove ambient noise; constructing an exhaust acoustic signal recognition model; and inputting the mixed acoustic signal with the ambient noise removed into the exhaust acoustic signal recognition model for identification, thereby identifying the exhaust acoustic signal and determining whether battery thermal runaway has occurred.

[0003] In existing signal decomposition methods, the parameters are fixed, which can achieve good results in solving specific signal decomposition problems. However, due to the different impact conditions on electric vehicle batteries, the frequencies of the generated acoustic signals are also different, resulting in poor acoustic signal decomposition and a decrease in the efficiency and accuracy of feature extraction. Summary of the Invention

[0004] The purpose of this invention is to address the problem mentioned in the background art that in existing signal decomposition methods, the parameters are fixed, and while they have good results in solving specific signal decomposition situations, the acoustic signal frequencies generated by electric vehicle batteries vary due to different impact conditions, resulting in poor acoustic signal decomposition effects and reduced efficiency and accuracy of feature extraction. Therefore, this invention proposes a safety warning method and system for electric vehicle batteries.

[0005] A first aspect of the present invention provides a safety warning method for electric vehicle batteries, the method comprising: An initial signal set is obtained by periodically acquiring acoustic signals within a preset time window, and a sensitivity index is calculated based on the target acoustic signal; the initial signal set contains multiple acoustic signals; the target acoustic signal is any one of the initial signal sets; The average sensitivity index is calculated based on each sensitivity index, and the average sensitivity index is used as the objective function. Obtain a preset parameter combination, and optimize the parameter combination based on the preset parameter combination and the objective function to obtain the optimal parameter combination; The target algorithm is obtained by updating the parameter configuration of the preset algorithm according to the optimal parameter combination, and the target mode function is obtained by decomposing the acoustic signal in the initial signal set according to the target algorithm. The target mode function is transformed by time-frequency to obtain a time-frequency spectrum. Fault features are extracted from the time-frequency spectrum, and fault warnings are given based on the fault features.

[0006] By periodically collecting acoustic signals from key locations of electric vehicle batteries, optimizing preset algorithm parameters with average sensitivity index as the objective function, and combining preprocessing, mode screening, and time-frequency transformation to accurately extract fault features, this method solves the problem that the acoustic signals from minor battery impacts are easily interfered with and difficult to extract. It improves signal decomposition efficiency and accuracy, as well as the accuracy of fault feature identification, enabling timely early warning of battery faults, reducing false alarm rates, and avoiding the risk of battery temperature runaway.

[0007] Optionally, a sensitivity index is calculated based on the target acoustic signal, the method comprising: The target acoustic signal is subjected to Hilbert transform to obtain an analytical signal, and the sum of squares of the real and imaginary parts of the analytical signal is taken to obtain an envelope signal. The envelope signal is processed to remove the DC component to obtain a DC envelope signal, and the DC envelope signal is subjected to preset operations to obtain the envelope power spectrum; the preset operations include: discrete Fourier transform operation, modulus squaring operation and normalization operation. The envelope power spectrum is summed across the entire frequency band to obtain the envelope spectral energy, and the original envelope spectral energy is calculated from the target acoustic signal. The energy proportion coefficient is calculated based on the envelope spectrum energy and the original envelope spectrum energy. The sensitivity index is obtained by performing a negative logarithmic multiplication on the energy proportion coefficient.

[0008] Hilbert transform can extract the envelope of the target acoustic signal, focus on the impact characteristics of battery exhaust faults and remove high-frequency interference. Removing the DC component directly eliminates its negative impact on the baseline shift of the spectrum analysis. The combination of discrete Fourier transform, modulus square and normalization can transform the time-domain envelope signal into the frequency-domain envelope power spectrum, realizing the frequency-domain quantization of fault characteristics. Then, by calculating the energy proportion coefficient of the envelope spectrum, the energy contribution of the fault signal relative to the original signal is locked. Combined with the negative logarithmic multiplication operation, this contribution is transformed into a quantifiable sensitivity index, which not only highlights the fault impact component, but also accurately reflects the richness of fault-related information in the signal.

[0009] Optionally, an optimal parameter combination is obtained by optimizing the preset parameter combination and the objective function. The preset parameter combination includes: filter length, number of modes, and number of segmented frequency bands. The method includes: Step 1: Determine the population size, maximum number of iterations, and search space. Use the objective function as the fitness function. Randomly generate multiple individuals based on the preset parameter combination and population size. Combine the individuals to obtain the population. Step 2: Obtain the chaotic mapping formula, generate a chaotic sequence according to the chaotic mapping formula, and map the chaotic sequence to the search space to obtain the initial position of the population; the initial position of the population includes the initial positions of multiple individuals; Step 3: Substitute each individual into the preset algorithm to calculate the fitness index, and select the individual with the highest fitness value as the optimal individual. Step 4: Calculate the strategy selection probability according to the strategy selection formula and generate a random number. If the strategy selection probability is less than the random number, execute the reverse learning strategy; otherwise, execute the random mutation strategy. Step 5: If the current iteration count reaches the maximum iteration count, or the fitness of the best individual changes less than the threshold over multiple consecutive iterations, then stop the iteration; otherwise, return to step 4 and re-execute the population update operation. The population update operation includes a reverse learning strategy and a random mutation strategy.

[0010] By generating ergodic initial positions of the population through chaotic mapping, and combining dynamic strategy selection probability to achieve adaptive switching between reverse learning and random mutation, while setting iteration termination conditions to avoid invalid iterations, the system efficiently searches for the optimal combination of filter length, number of modes, and number of segmented frequency bands, taking into account both global exploration and local optimization capabilities, thereby improving the efficiency and accuracy of parameter optimization.

[0011] Optionally, the reverse learning strategy includes: The reverse solution for the current optimal individual position is generated by the first preset formula, and the population position is updated by the second preset formula. First preset formula: in, Let X be the current optimal individual position at the t-th iteration. best (t) represents the inverse solution, where t is the current iteration number, ub is the upper bound of the search space, lb is the lower bound of the search space, rand(·) generates a uniformly distributed random number in the interval (0,1), and X best (t) represents the position of the individual with the best fitness value in the population at the t-th iteration; Second preset formula: in, This represents the new population position updated using the back-learning strategy at the (t+1)th iteration, where t is the current iteration number. Let X be the current optimal individual position at the t-th iteration. best The inverse solution corresponding to (t), where b1 is the adjustment coefficient. Let be the current optimal individual position in the t-th iteration.

[0012] By generating a reverse solution for the current optimal individual position using a preset formula, and combining this with the update formula of the dynamic adjustment coefficient, the population position is updated. This approach can both expand the global search range by covering the other end of the search space with the reverse solution and improve the rationality of population updates by leveraging the dynamic changes of the adjustment coefficient with the iteration process. In the early stages of iteration, the search is expanded by favoring the reverse solution, and in the later stages, it is refined by favoring the current optimal solution. This enhances the global exploration capability and local optimization accuracy of the parameter optimization algorithm.

[0013] Optionally, the random mutation strategy includes: Individuals are randomly selected from the population to obtain individuals to be mutated. A mutation factor is generated, and the individuals to be mutated are mutated according to the mutation factor to obtain mutated individuals. The fitness of the mutated individuals is determined. If the fitness is not greater than the fitness of the individuals to be mutated, the individuals to be mutated are retained; otherwise, the individuals to be mutated are replaced with mutated individuals.

[0014] By performing controllable random mutations on individuals within the parameter search space, combining the mutation coefficient to control the perturbation intensity and independent dimension mutation to enhance randomness, and using boundary correction to ensure the validity of parameters, a fitness-based optimal replacement mechanism is adopted to retain only the better mutated individuals. This approach can perform fine-grained searches near high-quality solutions, avoid jumping out of the optimal region, and appropriately jump out of local optima. While ensuring the stability of the algorithm, it enhances the optimization capability and further improves the accuracy and efficiency of parameter optimization.

[0015] A second aspect of the present invention provides a safety warning system for electric vehicle batteries, the system comprising: The signal acquisition module is used to periodically acquire acoustic signals within a preset time window to obtain an initial signal set, and to calculate a sensitivity index based on the target acoustic signal; the initial signal set contains multiple acoustic signals; the target acoustic signal is any one of the initial signal sets; The index acquisition module is used to calculate the average sensitivity index based on each sensitivity index and use the average sensitivity index as the objective function. The parameter optimization module is used to obtain a preset parameter combination and optimize it according to the preset parameter combination and the objective function to obtain the optimal parameter combination. The signal decomposition module is used to update the parameter configuration of the preset algorithm according to the optimal parameter combination to obtain the target algorithm, and to decompose the acoustic signals in the initial signal set according to the target algorithm to obtain the target mode function; The fault early warning module is used to perform time-frequency transformation on the target mode function to obtain a time-frequency spectrum, extract fault features from the time-frequency spectrum to obtain fault features, and provide fault early warning based on the fault features.

[0016] Optionally, the indicator acquisition module includes: The signal analysis module is used to perform Hilbert transform on the target acoustic signal to obtain an analytical signal, and to perform square root operation on the sum of squares of the real and imaginary parts of the analytical signal to obtain an envelope signal; The signal splitting module is used to remove the DC component from the envelope signal to obtain a DC envelope signal, and to perform preset operations on the DC envelope signal to obtain an envelope power spectrum; the preset operations include: discrete Fourier transform operation, modulus squaring operation, and normalization operation. The energy extraction module is used to sum the envelope power spectrum across the entire frequency band to obtain the envelope spectrum energy, and to calculate the original envelope spectrum energy of the target acoustic signal. The index calculation module is used to calculate the energy proportion coefficient based on the envelope spectrum energy and the original envelope spectrum energy, and to perform a negative logarithmic multiplication operation on the energy proportion coefficient to obtain the sensitivity index.

[0017] Optionally, the parameter optimization module includes: The first operation module is used to determine the population size, maximum number of iterations and search space, use the objective function as the fitness function, randomly generate multiple individuals according to the preset parameter combination and population size, and combine the individuals to obtain the population. The second operation module is used to obtain a chaotic mapping formula, generate a chaotic sequence according to the chaotic mapping formula, and map the chaotic sequence to the search space to obtain the initial position of the population; the initial position of the population includes the initial position of multiple individuals; The third operation module is used to calculate the fitness by substituting each individual into the preset algorithm and then using the sensitivity index to obtain the fitness, and the individual with the largest fitness value is taken as the optimal individual. The fourth operation module is used to calculate the strategy selection probability according to the strategy selection formula and generate a random number. If the strategy selection probability is less than the random number, the reverse learning strategy is executed; otherwise, the random mutation strategy is executed. The fifth operation module is used to stop iteration if the current iteration count reaches the maximum iteration count, or if the fitness of the best individual changes less than a threshold over multiple consecutive iterations; otherwise, it returns to the fourth operation module and re-executes the population update operation. The population update operation includes a reverse learning strategy and a random mutation strategy.

[0018] Optionally, the fifth operation module is used to: generate a reverse solution of the current optimal individual position through a first preset formula, and then update the population position through a second preset formula; First preset formula: in, Let X be the current optimal individual position at the t-th iteration. best (t) represents the inverse solution, where t is the current iteration number, ub is the upper bound of the search space, lb is the lower bound of the search space, rand(·) generates a uniformly distributed random number in the interval (0,1), and X best (t) represents the position of the individual with the best fitness value in the population at the t-th iteration; Second preset formula: in, This represents the new population position updated using the back-learning strategy at the (t+1)th iteration, where t is the current iteration number. Let X be the current optimal individual position at the t-th iteration. best The inverse solution corresponding to (t), where b1 is the adjustment coefficient. Let be the current optimal individual position in the t-th iteration.

[0019] Optionally, the fifth operation module is further configured to: randomly select individuals from the population to obtain individuals to be mutated, generate a mutation factor, mutate the individuals to be mutated according to the mutation factor to obtain mutated individuals, determine the fitness of the mutated individuals, and retain the individuals to be mutated if the fitness is not greater than the fitness of the individuals to be mutated, otherwise replace the individuals to be mutated with mutated individuals.

[0020] The beneficial effects of this invention are: This invention proposes a safety early warning method for electric vehicle batteries. It involves periodically collecting acoustic signals and calculating sensitivity indices. The optimal parameter combination of a preset algorithm is obtained by optimizing the average sensitivity index as the objective function. The updated preset algorithm then decomposes the signal to obtain the target mode function. Finally, fault features are extracted through time-frequency transformation to achieve fault early warning. This method effectively solves the problem that exhaust acoustic signals from minor impacts to electric vehicle batteries are easily interfered with by external interference and are difficult to extract effectively. The optimized preset algorithm improves the efficiency and accuracy of signal decomposition, enabling accurate identification of fault features in the exhaust acoustic signals. This enhances the timeliness and accuracy of battery fault early warning, effectively capturing exhaust signals caused by minor impacts to the battery and providing timely warnings, thus avoiding the risk of battery temperature runaway due to difficulties in signal extraction and feature recognition. Attached Figure Description

[0021] Figure 1A flowchart of a safety warning method for electric vehicle batteries provided in an embodiment of the present invention; Figure 2 This is an algorithm flowchart for a safety warning method for electric vehicle batteries provided in an embodiment of the present invention. Detailed Implementation

[0022] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0023] This invention provides a safety warning method for electric vehicle batteries. See also... Figure 1 , Figure 1 A flowchart illustrating a safety warning method for electric vehicle batteries provided in an embodiment of the present invention. The method includes the following steps: S101, periodically collect acoustic signals within a preset time window to obtain an initial signal set, and calculate the sensitivity index based on the target acoustic signal; S102, the average sensitivity index is calculated based on each sensitivity index, and the average sensitivity index is used as the objective function. S103, obtain the preset parameter combination, and optimize the optimal parameter combination based on the preset parameter combination and the objective function; S104, Update the parameter configuration of the preset algorithm according to the optimal parameter combination to obtain the target algorithm, and decompose the acoustic signal in the initial signal set according to the target algorithm to obtain the target mode function; S105, perform time-frequency transformation on the target modal function to obtain a time-frequency spectrum, extract features from the time-frequency spectrum to obtain fault features, and provide fault warning based on the fault features.

[0024] The initial signal set contains multiple acoustic signals; the target acoustic signal is any one of the signals in the initial signal set. This invention provides a safety early warning method for electric vehicle batteries. It involves periodically collecting acoustic signals and calculating sensitivity indices. The optimal parameter combination of a preset algorithm is obtained by optimizing the average sensitivity index as the objective function. The updated preset algorithm then decomposes the signal to obtain the target mode function. Finally, fault features are extracted through time-frequency transformation to achieve fault early warning. This method effectively solves the problem that exhaust acoustic signals from minor impacts to electric vehicle batteries are easily interfered with by external interference and are difficult to extract effectively. The optimized preset algorithm improves the efficiency and accuracy of signal decomposition, enabling accurate identification of fault features in the exhaust acoustic signals. This enhances the timeliness and accuracy of battery fault early warning, effectively capturing exhaust signals caused by minor impacts to the battery and providing timely warnings, thus avoiding the risk of battery temperature runaway due to difficulties in signal extraction and feature recognition.

[0025] In one implementation, if the battery of an electric vehicle is bumped or knocked during driving, a short circuit may easily occur inside the battery, leading to temperature runaway. The electrolyte inside the battery will vaporize due to the increased temperature, and the exhaust valve will discharge the excess gas to prevent the battery from bulging. Therefore, by monitoring the acoustic signal of the exhaust valve, the electric vehicle can be given a fault warning (battery fault prediction and health management, pre-diagnosis and health management).

[0026] In one implementation, high-sensitivity acoustic sensors (with a pickup range of 20Hz~20kHz, adapted to the wideband acoustic characteristics of exhaust airflow) are deployed at key locations such as the exhaust valve and housing seams of the electric vehicle battery pack. Acoustic signals are periodically acquired with a time window of 100ms and a acquisition step size of 10ms. The initial signal set contains 10 sets of continuous acoustic signals within a single acquisition cycle (covering the complete stage of exhaust signal from weak to strong). At the same time, vibration and temperature data of the battery pack are acquired synchronously as auxiliary references to eliminate interference signals such as road noise and motor operating noise.

[0027] In one implementation, a parameter optimization strategy adapted to the characteristics of battery exhaust signals improves the decomposition effect of the preset algorithm (FMD algorithm): The scenario-based range of the preset parameter combination is set according to the characteristics of the battery exhaust acoustic signal (main frequency 500Hz~5kHz, strong non-stationarity). The preset parameter range of the FMD algorithm is set as follows: filter length L∈[50,200] (adapting to the time domain length of the exhaust signal), number of modes N∈[3,8] (covering the main modes of the exhaust signal), and number of segmented frequency bands K∈[2,6] (matching the frequency band distribution of exhaust characteristics), so as to avoid the optimization efficiency being low due to the parameter range being too large.

[0028] In one implementation, in addition to updating the filter length, number of modes, and number of segmented frequency bands, a mode selection threshold parameter (set to 80% of the sensitivity index) is set to address low-frequency noise interference in the battery exhaust signal. After decomposition, only mode functions with a sensitivity index ≥ the threshold are retained as target mode functions (BLIMF), eliminating noise modes without exhaust characteristics and reducing subsequent computational load. Initial signal preprocessing optimization: Before decomposition, the acoustic signals in the initial signal set are preprocessed: first, background noise such as road surface and motor noise is removed using wavelet threshold denoising, and then the signal is normalized (amplitude normalized to [-1,1]) to improve the decomposition accuracy of the FMD algorithm for exhaust signals. Simultaneously, the acquired acoustic signals are resampled (sampling rate uniformly set to 48kHz) to ensure signal consistency across different acquisition periods. After obtaining the target mode functions, the exhaust feature matching degree of each mode is calculated (compared to the modal features of the standard battery exhaust signal), retaining only mode functions with a matching degree ≥ 70% to ensure the targeted nature of subsequent feature extraction and avoid invalid mode interference in fault diagnosis.

[0029] In one implementation, to address the non-stationary characteristics of battery exhaust signals, the Hilbert-Huang Transform (HHT) is used instead of the traditional Fourier Transform to perform time-frequency analysis on the target mode function: first, the mode function is decomposed into intrinsic mode functions (IMF) through empirical mode decomposition (EMD), and then the IMF is subjected to Hilbert transform to obtain the time-frequency spectrum, accurately capturing the time-varying characteristics of the exhaust signal (such as the change of exhaust intensity over time). Three core fault features were extracted from the time-frequency spectrum: frequency domain features: energy proportion of the exhaust characteristic frequency band (500Hz~5kHz) and main frequency offset (compared to the reference main frequency when the normal battery has no exhaust); time domain features: signal peak value, root mean square value, and impulse factor (reflecting the intensity of the exhaust airflow); time-frequency domain features: marginal spectral entropy and variance of instantaneous frequency (reflecting the stability of the exhaust signal), forming a 12-dimensional fault feature vector. A three-level early warning system is constructed based on fault feature vectors: Level 1 warning (minor impact): The energy percentage of the exhaust characteristic frequency band is ≥10% and <30%, triggering the vehicle terminal to remind: battery pack abnormality, and it is recommended to drive at low speed to the repair shop; Level 2 warning (moderate impact): The energy percentage is ≥30% and <50%, triggering the vehicle to actively limit power (down to 50%), and pushing the warning information to the owner and the repair platform; Level 3 warning (severe impact): The energy percentage is ≥50%, triggering the vehicle to stop urgently and cut off the high-voltage circuit, and at the same time linking the rescue system; The warning thresholds are determined through training with a large amount of battery impact test data and are updated and optimized regularly.

[0030] The extracted fault features are cross-validated by combining other data from the battery management system (BMS) (such as battery voltage, temperature, and casing strain). If the acoustic features trigger an alarm and the BMS data is synchronously abnormal, the alarm is confirmed to be valid. If only the acoustic features are abnormal, it is judged as a suspected alarm and further signal collection is required for verification to reduce the false alarm rate.

[0031] In one embodiment, a sensitivity index is calculated based on the target acoustic signal, the method comprising: The target acoustic signal is subjected to Hilbert transform to obtain an analytic signal, and the envelope signal is obtained by taking the square root of the sum of the squares of the real and imaginary parts of the analytic signal. The envelope signal is processed to remove the DC component to obtain the DC envelope signal, and the DC envelope signal is then subjected to preset operations to obtain the envelope power spectrum; the preset operations include: discrete Fourier transform operation, modulus squaring operation and normalization operation; The envelope power spectrum is summed across the entire frequency band to obtain the envelope spectral energy, and the original envelope spectral energy is calculated from the target acoustic signal. The energy proportion coefficient is calculated based on the energy of the envelope spectrum and the energy of the original envelope spectrum. The sensitivity index is obtained by performing a negative logarithmic multiplication on the energy proportion coefficient.

[0032] In one implementation, the square root of the sum of the squares of the real and imaginary parts of the analytic signal is taken, i.e., the modulus of the analytic signal is calculated. The result is the envelope signal of the original signal. The envelope signal extracts the slowly changing amplitude information in the original signal. In battery fault diagnosis, it can highlight the periodic impact component caused by exhaust, thereby effectively suppressing high-frequency carrier interference and highlighting fault characteristics.

[0033] In one implementation, the mean of the envelope signal is subtracted to obtain a zero-mean envelope signal, called the de-DC envelope signal. The DC component corresponds to a constant offset in the signal, which appears as a spectral line at zero frequency in the spectrum, interfering with subsequent analysis of fault characteristic frequencies. Removing the DC component eliminates the negative impact of the zero-frequency component on spectrum calculations, allowing spectrum analysis to focus more on the AC components.

[0034] In one implementation, the complex spectrum obtained from the discrete Fourier transform is first modulo (amplitude) and then squared to obtain the power spectrum (or energy spectrum). Modulo squaring eliminates phase information, retaining only the intensity (energy) of each frequency component, reflecting the distribution of signal power with frequency. The physical meaning of the power spectrum is clear, facilitating subsequent summation or comparison of the energy of frequency components.

[0035] In one implementation, the envelope power spectrum is summed across the entire frequency band to obtain the envelope spectral energy: the values ​​of the normalized envelope power spectrum at all frequency points are summed to obtain the envelope spectral energy E1. This energy reflects the total intensity of the envelope signal at all frequency components and is a scalar indicator measuring the overall contribution of fault impact in the signal. The higher the energy, the richer the fault-related information contained in the signal.

[0036] In one implementation, the energy percentage coefficient is calculated based on the envelope spectrum energy and the original envelope spectrum energy. For example, if the envelope spectrum energy is denoted by E1, the original envelope spectrum energy by E0, and the energy percentage coefficient by p, then... The sensitivity index is obtained by performing a negative logarithmic operation on the energy proportion coefficient p, and by taking the natural logarithm of the energy proportion coefficient p and multiplying it by a negative sign.

[0037] In one embodiment, an optimal parameter combination is obtained by optimization based on a preset parameter combination and an objective function. The preset parameter combination includes: filter length, number of modes, and number of segmented frequency bands. The method includes: Step 1: Determine the population size, maximum number of iterations, and search space. Use the objective function as the fitness function. Randomly generate multiple individuals based on the preset parameter combination and population size. Combine the individuals to obtain the population. Step 2: Obtain the chaos mapping formula, generate a chaotic sequence according to the chaos mapping formula, and map the chaotic sequence to the search space to obtain the initial position of the population; the initial position of the population contains the initial positions of multiple individuals; Step 3: Substitute each individual into the preset algorithm to calculate the fitness index, and select the individual with the highest fitness value as the optimal individual. Step 4: Calculate the strategy selection probability according to the strategy selection formula and generate a random number. If the strategy selection probability is less than the random number, execute the reverse learning strategy; otherwise, execute the random mutation strategy. Step 5: If the current iteration count reaches the maximum iteration count, or the fitness of the best individual changes less than the threshold over multiple consecutive iterations, then stop the iteration; otherwise, return to step 4 and re-execute the population update operation. The population update operation includes a reverse learning strategy and a random mutation strategy.

[0038] In one implementation, the population size, maximum number of iterations, and search space are determined. For example, the population size is the total number of individuals, and the search space is the upper and lower bounds of the parameters to be optimized (where ub represents the upper bound of the search space and lb represents the lower bound of the search space). The optimization objective of the algorithm is to find the parameter combination that minimizes the sensitivity index.

[0039] In one implementation, participants Figure 2 , Figure 2 This invention provides an algorithm flowchart for a safety warning method for electric vehicle batteries; obtaining a chaotic mapping formula and generating a chaotic sequence based on the chaotic mapping formula; the chaotic mapping formula is: , where Y n+1 The updated chaos value, β is a constant, Y n The nth chaotic value; β takes the value of 4, which puts the system in a completely chaotic state and generates a sequence with optimal ergodicity; Y n Let ∈(0,1) be the nth chaotic value. The initial value Y0 is a random number that is neither 0 nor 1 within (0,1), such as Y0=0.3. Map the chaotic sequence to the search space to obtain the initial position of the population. The chaotic sequence (that is, the chaotic value, symbolically represented as Y) is then used to determine the initial position of the population. ij The initial position of the individual is obtained by converting the original chaotic value corresponding to the j-th parameter of the i-th individual into the actual value within the search space of the corresponding parameter using a linear scaling formula: , where X ij Let lb be the actual value of the j-th parameter of the i-th individual. j ub is the lower bound for the search of the j-th parameter. j Y is the upper limit of the search for the j-th parameter. ijThis represents the original chaotic value corresponding to the j-th parameter of the i-th individual (the original chaotic value is randomly generated); The fitness is obtained by substituting each individual into the preset algorithm and calculating the sensitivity index: the initial parameter value of each individual is substituted into the FMD algorithm to decompose the signal, and the sensitivity index of each BLIMF component after decomposition is calculated as the fitness value of that individual.

[0040] In one implementation, the strategy selection formula is: t is the current iteration number, T max ρ represents the maximum number of iterations; it is used to generate a random number rand between 0 and 1 for subsequent strategy selection; the dynamic probability ρ enables adaptive switching between two population update strategies. In the early stages of iteration, ρ is relatively large, so reverse learning is preferred to expand the search range; in the later stages of iteration, ρ is relatively small, so random mutation is preferred to refine the search for the optimal solution, taking into account both the algorithm's global search and local optimization capabilities.

[0041] In one implementation, if the current iteration number t reaches the maximum iteration number Tmax, or the found optimal fitness value tends to stabilize (without significant decrease), the iteration is terminated; continue iterating: if the termination condition is not met, the current iteration number t+1 is incremented, and the process returns to step 4 to continue the population update operation; control the iteration process of the algorithm to avoid invalid iterations, terminate the algorithm in a timely manner when it converges or reaches the maximum computational load, and output the optimal individual.

[0042] In one embodiment, the reverse learning strategy includes: The reverse solution for the current optimal individual position is generated by the first preset formula, and the population position is updated by the second preset formula. First preset formula: in, Let X be the current optimal individual position at the t-th iteration. best (t) represents the inverse solution, where t is the current iteration number, ub is the upper bound of the search space, lb is the lower bound of the search space, rand(·) generates a uniformly distributed random number in the interval (0,1), and X best (t) represents the position of the individual with the best fitness value in the population at the t-th iteration; Second preset formula: in, This represents the new population position updated using the back-learning strategy at the (t+1)th iteration, where t is the current iteration number. Let X be the current optimal individual position at the t-th iteration. best The inverse solution corresponding to (t), where b1 is the adjustment coefficient. Let be the current optimal individual position in the t-th iteration.

[0043] In one implementation, the essence of the reverse solution is to generate a solution X that is similar to the current optimal solution X within the parameter search space [lb, ub], using lb as a reference. best (t) A new symmetric solution. For example: if lb=10, ub=200, X best Given that (t) = 48 and rand(·) = 0.2, the inverse solution X is... best (t)=200+0.2×(10-48)=200-7.6=192.4, that is, the reverse solution of the current optimal solution 48 is 192.4, covering the other end of the search space.

[0044] In one implementation, b1 is the adjustment coefficient: T max The maximum number of iterations (preset value, such as T) max =100, used to control the dynamic change trend of b1), t is the current iteration number; in the early stage of iteration (t is small): the value of b1 is large, and the new position ζ(t+1) is more biased towards the reverse solution X. best (t), expand the search range to find the global optimal solution; in the later stage of iteration (t is close to T) max As the value of b1 gradually decreases, the new position ζ(t+1) is more biased towards the current optimal solution X. best (t), focusing on local fine-grained search to ensure the accuracy of the solution.

[0045] In one embodiment, the random mutation strategy includes: Randomly select individuals from the population to obtain individuals to be mutated, generate mutation factors, mutate the individuals to be mutated according to the mutation factors to obtain mutated individuals, determine the fitness of the mutated individuals, if the fitness is not greater than the fitness of the individuals to be mutated, retain the individuals to be mutated, otherwise replace the individuals to be mutated with the mutated individuals.

[0046] In one implementation, mutations are performed only within the preset search space [lb, ub] of the parameters and will not exceed the boundary, thus ensuring the physical validity of the parameters; the perturbation intensity is controlled within a small range to avoid excessive mutations that could cause the algorithm to jump out of the current high-quality search area and lose the information of the optimal solution already found; mutations can be performed on individuals with poor fitness or on all individuals in the population.

[0047] In one implementation, the mutated object is the parameter position X of all individuals in the population after the t-th iteration. i (t) (i=1,2,...,M, M is the population size, X i(t) represents the parameter vector of the i-th individual in the t-th iteration; Mutation dimension: consistent with the dimension of the parameters to be optimized (this scheme is 3-dimensional, corresponding to filter length (L), number of modes (N), and number of segmented frequency bands (K)), each dimension mutates independently to ensure the randomness of parameter perturbation. Generating random mutation factor: For each parameter dimension of each individual, a uniformly random number in the interval (0,1) is generated as the mutation factor, denoted as ξ. ij (t) (i is the individual ID, j is the parameter dimension, j=1 / 2 / 3 corresponds to L / N / K), and satisfies the variation magnitude constraint ξ. ij (t)∈[0,α] (α is the coefficient of variation, ranging from 0.1 to 0.3. When α=0.2, it can guarantee small perturbations); the coefficient of variation α is the core adjustment parameter: if α is too small, the perturbation is insufficient and it is impossible to effectively escape the local optimum; if α is too large, it is easy for the algorithm to fall into meaningless random search and lose the ability to find local optimization.

[0048] In one implementation, the mathematical expression for the mutation operation is: ,in, This represents the new position of the i-th individual after the j-th parameter mutation. This represents the original position of the i-th individual before the j-th parameter mutation. is the mutation factor of the j-th parameter of the i-th individual, 2rand-1: generates random numbers in the interval (-1,1) to control the direction of mutation (positive / negative), so that the perturbation can be carried out in the direction of increasing or decreasing the parameter, ensuring the comprehensiveness of the search; New location after mutation It is possible that the random disturbance slightly exceeds the preset search upper and lower limits [lb] j ,ub j [Boundary correction is required to ensure that the parameters conform to physical meaning: if] <lb j Then let =lb j ;like >ub j Then let =ub j If in [lb] j, ub j Within the range, retain directly. .

[0049] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should still fall within the scope of the claims of the present invention.

Claims

1. A safety warning method for electric vehicle batteries, characterized in that, The method includes: An initial signal set is obtained by periodically acquiring acoustic signals within a preset time window, and a sensitivity index is calculated based on the target acoustic signal; the initial signal set contains multiple acoustic signals; the target acoustic signal is any one of the initial signal sets; The average sensitivity index is calculated based on each sensitivity index, and the average sensitivity index is used as the objective function. Obtain a preset parameter combination, and optimize the parameter combination based on the preset parameter combination and the objective function to obtain the optimal parameter combination; The target algorithm is obtained by updating the parameter configuration of the preset algorithm according to the optimal parameter combination, and the target mode function is obtained by decomposing the acoustic signal in the initial signal set according to the target algorithm. The target mode function is transformed by time-frequency to obtain a time-frequency spectrum. Fault features are extracted from the time-frequency spectrum, and fault warnings are given based on the fault features.

2. The safety warning method for electric vehicle batteries according to claim 1, characterized in that, The method for calculating a sensitivity index based on the target acoustic signal includes: The target acoustic signal is subjected to Hilbert transform to obtain an analytical signal, and the sum of squares of the real and imaginary parts of the analytical signal is taken to obtain an envelope signal. The envelope signal is processed to remove the DC component to obtain a DC envelope signal, and the DC envelope signal is subjected to preset operations to obtain the envelope power spectrum; the preset operations include: discrete Fourier transform operation, modulus squaring operation and normalization operation. The envelope power spectrum is summed across the entire frequency band to obtain the envelope spectral energy, and the original envelope spectral energy is calculated from the target acoustic signal. The energy proportion coefficient is calculated based on the envelope spectrum energy and the original envelope spectrum energy. The sensitivity index is obtained by performing a negative logarithmic multiplication on the energy proportion coefficient.

3. The safety warning method for electric vehicle batteries according to claim 1, characterized in that, The optimal parameter combination is obtained by optimizing the preset parameter combination and the objective function. The preset parameter combination includes: filter length, number of modes, and number of segmented frequency bands. The method includes: Step 1: Determine the population size, maximum number of iterations, and search space. Use the objective function as the fitness function. Randomly generate multiple individuals based on the preset parameter combination and population size. Combine the individuals to obtain the population. Step 2: Obtain the chaotic mapping formula, generate a chaotic sequence according to the chaotic mapping formula, and map the chaotic sequence to the search space to obtain the initial position of the population; the initial position of the population includes the initial positions of multiple individuals; Step 3: Substitute each individual into the preset algorithm to calculate the fitness index, and select the individual with the highest fitness value as the optimal individual. Step 4: Calculate the strategy selection probability according to the strategy selection formula and generate a random number. If the strategy selection probability is less than the random number, execute the reverse learning strategy; otherwise, execute the random mutation strategy. Step 5: If the current iteration count reaches the maximum iteration count, or the fitness of the best individual changes less than the threshold over multiple consecutive iterations, then stop the iteration; otherwise, return to step 4 and re-execute the population update operation. The population update operation includes a reverse learning strategy and a random mutation strategy.

4. A safety warning method for electric vehicle batteries according to claim 3, characterized in that, The reverse learning strategy includes: The reverse solution for the current optimal individual position is generated by the first preset formula, and the population position is updated by the second preset formula. First preset formula: in, Let X be the current optimal individual position at the t-th iteration. best (t) represents the inverse solution, where t is the current iteration number, ub is the upper bound of the search space, lb is the lower bound of the search space, rand(·) generates a uniformly distributed random number in the interval (0,1), and X best (t) represents the position of the individual with the best fitness value in the population at the t-th iteration; Second preset formula: in, This represents the new population position updated using the back-learning strategy at the (t+1)th iteration, where t is the current iteration number. Let X be the current optimal individual position at the t-th iteration. best The inverse solution corresponding to (t), where b1 is the adjustment coefficient. Let be the current optimal individual position in the t-th iteration.

5. A safety warning method for electric vehicle batteries according to claim 1, characterized in that, The random mutation strategy includes: Individuals are randomly selected from the population to obtain individuals to be mutated. A mutation factor is generated, and the individuals to be mutated are mutated according to the mutation factor to obtain mutated individuals. The fitness of the mutated individuals is determined. If the fitness is not greater than the fitness of the individuals to be mutated, the individuals to be mutated are retained; otherwise, the individuals to be mutated are replaced with mutated individuals.

6. A safety warning system for electric vehicle batteries, characterized in that, The system includes: The signal acquisition module is used to periodically acquire acoustic signals within a preset time window to obtain an initial signal set, and to calculate a sensitivity index based on the target acoustic signal; the initial signal set contains multiple acoustic signals; the target acoustic signal is any one of the initial signal sets; The index acquisition module is used to calculate the average sensitivity index based on each sensitivity index and use the average sensitivity index as the objective function. The parameter optimization module is used to obtain a preset parameter combination and optimize it according to the preset parameter combination and the objective function to obtain the optimal parameter combination. The signal decomposition module is used to update the parameter configuration of the preset algorithm according to the optimal parameter combination to obtain the target algorithm, and to decompose the acoustic signals in the initial signal set according to the target algorithm to obtain the target mode function; The fault early warning module is used to perform time-frequency transformation on the target mode function to obtain a time-frequency spectrum, extract fault features from the time-frequency spectrum to obtain fault features, and provide fault early warning based on the fault features.

7. A safety warning system for electric vehicle batteries according to claim 6, characterized in that, The indicator acquisition module includes: The signal analysis module is used to perform Hilbert transform on the target acoustic signal to obtain an analytical signal, and to perform square root operation on the sum of squares of the real and imaginary parts of the analytical signal to obtain an envelope signal; The signal splitting module is used to remove the DC component from the envelope signal to obtain a DC envelope signal, and to perform preset operations on the DC envelope signal to obtain an envelope power spectrum; the preset operations include: discrete Fourier transform operation, modulus squaring operation, and normalization operation. The energy extraction module is used to sum the envelope power spectrum across the entire frequency band to obtain the envelope spectrum energy, and to calculate the original envelope spectrum energy of the target acoustic signal. The index calculation module is used to calculate the energy proportion coefficient based on the envelope spectrum energy and the original envelope spectrum energy, and to perform a negative logarithmic multiplication operation on the energy proportion coefficient to obtain the sensitivity index.

8. A safety warning system for electric vehicle batteries according to claim 6, characterized in that, The parameter optimization module includes: The first operation module is used to determine the population size, maximum number of iterations and search space, use the objective function as the fitness function, randomly generate multiple individuals according to the preset parameter combination and population size, and combine the individuals to obtain the population. The second operation module is used to obtain a chaotic mapping formula, generate a chaotic sequence according to the chaotic mapping formula, and map the chaotic sequence to the search space to obtain the initial position of the population; the initial position of the population includes the initial position of multiple individuals; The third operation module is used to calculate the fitness by substituting each individual into the preset algorithm and then using the sensitivity index to obtain the fitness, and the individual with the largest fitness value is taken as the optimal individual. The fourth operation module is used to calculate the strategy selection probability according to the strategy selection formula and generate a random number. If the strategy selection probability is less than the random number, the reverse learning strategy is executed; otherwise, the random mutation strategy is executed. The fifth operation module is used to stop iteration if the current iteration count reaches the maximum iteration count, or if the fitness of the best individual changes less than a threshold over multiple consecutive iterations; otherwise, it returns to the fourth operation module and re-executes the population update operation. The population update operation includes a reverse learning strategy and a random mutation strategy.

9. A safety warning system for electric vehicle batteries according to claim 8, characterized in that, The fifth operation module is used to: generate a reverse solution for the current optimal individual position using a first preset formula, and then update the population position using a second preset formula; First preset formula: in, Let X be the current optimal individual position at the t-th iteration. best (t) represents the inverse solution, where t is the current iteration number, ub is the upper bound of the search space, lb is the lower bound of the search space, rand(·) generates a uniformly distributed random number in the interval (0,1), and X best (t) represents the position of the individual with the best fitness value in the population at the t-th iteration; Second preset formula: in, This represents the new population position updated using the back-learning strategy at the (t+1)th iteration, where t is the current iteration number. Let X be the current optimal individual position at the t-th iteration. best The inverse solution corresponding to (t), where b1 is the adjustment coefficient. Let be the current optimal individual position in the t-th iteration.

10. A safety warning system for electric vehicle batteries according to claim 8, characterized in that, The fifth operation module is further configured to: randomly select individuals from the population to obtain individuals to be mutated, generate a mutation factor, mutate the individuals to be mutated according to the mutation factor to obtain mutated individuals, determine the fitness of the mutated individuals, and retain the individuals to be mutated if the fitness is not greater than the fitness of the individuals to be mutated, otherwise replace the individuals to be mutated with mutated individuals.

Images