Coal mine electric vehicle battery life prediction method based on big data

Abstract

The application belongs to the technical field of electric measurement, and discloses a coal mine electric vehicle battery life prediction method based on big data. The environmental temperature and humidity and electrical working condition data of the coal mine electric vehicle during operation are collected to extract multi-dimensional dynamic stress factors; the multi-dimensional dynamic stress factors are embedded into an electrochemical degradation model to adjust the chemical reaction activation energy parameter to control the solid electrolyte interface film growth rate to obtain physical degradation trajectory characteristics; a physical-data fusion prediction network is constructed to calculate the fluctuation variance of the multi-dimensional dynamic stress factors, and the weight proportion of the physical degradation trajectory characteristics and the data fitting characteristics is dynamically adjusted accordingly; and the weighted characteristics are fused to output the battery remaining life. The application couples the dynamic stress and the electrochemical degradation, and adaptively adjusts the physical and data weights according to the stress fluctuation, overcomes the battery life prediction distortion problem under the complex working conditions of the coal mine, and realizes the objective tracking of the nonlinear acceleration attenuation.

Description

Technical Field

[0001] This invention belongs to the field of electrical measurement technology and discloses a method for predicting the battery life of electric vehicles in coal mines based on big data. Background Technology

[0002] Current methods for predicting the battery life of electric vehicles in coal mines mostly employ pure data-driven models or static equivalent circuit models. Pure data-driven models directly input collected external operating data such as voltage, current, and temperature into a deep neural network, relying on network parameters to map battery life. Static equivalent circuit models use pre-set empirical formulas for fixed attenuation to estimate battery capacity loss. Both methods treat the battery as a black box or use a fixed degradation rate, failing to incorporate the dynamic stresses created by the interplay of temperature, humidity, and electrical conditions in the underground coal mine environment into the battery's internal electrochemical degradation process.

[0003] Existing technologies for predicting the lifespan of electric vehicle batteries in coal mines lack a mechanism to deeply couple dynamic stress in coal mines with the electrochemical degradation process and to adaptively adjust the weights of physical constraints and data fitting based on stress fluctuations, resulting in distorted battery lifespan predictions under complex working conditions in coal mines. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting the battery life of electric vehicles in coal mines based on big data, which can solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A big data-based method for predicting the battery life of electric vehicles in coal mines includes: collecting environmental temperature and humidity data and electrical condition data during the operation of the electric vehicle in the coal mine; performing time-series alignment and feature extraction on the environmental temperature and humidity data and the electrical condition data to obtain a multidimensional dynamic stress factor; embedding the multidimensional dynamic stress factor as an input variable into a pre-constructed electrochemical degradation model; adjusting the activation energy parameter of the chemical reaction in the electrochemical degradation model through the multidimensional dynamic stress factor to control the evolution process of the solid electrolyte interface film growth rate parameter and obtain physical degradation trajectory features; constructing a physical-data fusion prediction network; introducing a dynamic weight allocation mechanism into the data-driven network of the physical-data fusion prediction network; calculating the fluctuation variance of the multidimensional dynamic stress factor within a sliding time window; adjusting the weight ratio of the physical degradation trajectory features output by the electrochemical degradation model and the data fitting features output by the data-driven network in the final life prediction result according to the fluctuation variance; fusing the weighted physical degradation trajectory features and the data fitting features to output the remaining battery life.

[0006] Preferably, the process of temporally aligning and extracting features from the ambient temperature and humidity data and the electrical operating condition data to obtain a multidimensional dynamic stress factor includes: inputting the ambient temperature data from the ambient temperature and humidity data and the charge / discharge rate data from the electrical operating condition data into a preset thermo-electrical coupled stress field model to calculate the thermal distribution gradient data inside the battery cell; extracting the absolute humidity change rate data from the ambient temperature and humidity data, aligning the absolute humidity change rate data with the thermal distribution gradient data in time and space, and calculating the condensation probability index; and jointly encoding the thermal distribution gradient data, the condensation probability index, and the number of deep charge / discharge cycles from the electrical operating condition data to map them into the multidimensional dynamic stress factor.

[0007] Preferably, the method of adjusting the activation energy parameter of the chemical reaction in the electrochemical degradation model by the multidimensional dynamic stress factor to control the evolution process of the solid electrolyte interfacial membrane growth rate parameter and obtain the physical degradation trajectory characteristics includes: inputting the thermal stress component in the multidimensional dynamic stress factor into the exponential term of the Arrhenius equation to dynamically correct the activation energy parameter of the electrolyte decomposition reaction in the electrochemical degradation model; updating the diffusion coefficient in the solid electrolyte interfacial membrane growth rate parameter based on the corrected activation energy parameter, and performing attenuation iteration on the porosity parameter of the solid electrolyte interfacial membrane in combination with the mechanical stress component in the multidimensional dynamic stress factor; calculating the thickness growth variable of the solid electrolyte interfacial membrane according to the updated diffusion coefficient and the iterated porosity parameter, and determining the change sequence of the thickness growth variable with the cycle period as the physical degradation trajectory characteristics.

[0008] Preferably, calculating the fluctuation variance of the multidimensional dynamic stress factor within a sliding time window, and adjusting the weight ratio of the physical degradation trajectory features output by the electrochemical degradation model and the data fitting features output by the data-driven network in the final lifetime prediction result based on the fluctuation variance, includes: constructing a sliding time window along the time axis; calculating the covariance matrix of the multidimensional dynamic stress factor within the sliding time window; performing eigenvalue decomposition on the covariance matrix to obtain the main fluctuation component; inputting the main fluctuation component into a preset mapping function to calculate the dynamic adjustment coefficient at the current moment, wherein the mapping function is a monotonically increasing function; using the dynamic adjustment coefficient as the first weight coefficient of the data fitting features output by the data-driven network, and using the difference between the first and second weight coefficients as the second weight coefficient of the physical degradation trajectory features output by the electrochemical degradation model; and performing a weighted summation of the data fitting features and the physical degradation trajectory features based on the first and second weight coefficients.

[0009] Preferably, the process of collecting ambient temperature and humidity data and electrical condition data during the operation of the coal mine electric vehicle, performing time-series alignment and feature extraction on the ambient temperature and humidity data and the electrical condition data to obtain the multidimensional dynamic stress factor further includes: performing a short-time Fourier transform on the transient current change sequence in the electrical condition data to obtain a current frequency domain feature spectrum; extracting energy distribution data of high-frequency bands from the current frequency domain feature spectrum, and determining the energy distribution data of the high-frequency bands as the transient electric impact stress factor; performing feature cross-interaction between the transient electric impact stress factor and the cumulative humidity in the ambient temperature and humidity data to generate a humid-thermal-electric coupling stress feature vector, and incorporating the humid-thermal-electric coupling stress feature vector into the multidimensional dynamic stress factor.

[0010] Preferably, constructing the physical-data fusion prediction network and introducing a dynamic weight allocation mechanism in the data-driven network of the physical-data fusion prediction network further includes: using the physical degradation trajectory features output by the electrochemical degradation model as the state initialization constraint vector of the long short-term memory network in the data-driven network; during the forward propagation of the long short-term memory network, performing a Hadamard product operation on the rate of change of the derivative of the physical degradation trajectory features and the forget gate output inside the long short-term memory network to limit the degree to which the long short-term memory network retains features that deviate from the physical degradation trajectory features; and determining the constrained output sequence of the long short-term memory network as the data fitting feature.

[0011] Preferably, jointly encoding the thermal distribution gradient data, the condensation probability index, and the number of deep charge-discharge cycles in the electrical condition data to map the multidimensional dynamic stress factor further includes: inputting the condensation probability index and the battery separator micropore size parameters into a preset wetting diffusion model to calculate the electrolyte abnormal wetting depth data; estimating the local micro-short circuit occurrence probability data based on the electrolyte abnormal wetting depth data and the local high temperature region data in the thermal distribution gradient data; and splicing the local micro-short circuit occurrence probability data as an additional dimension into the multidimensional dynamic stress factor to participate in subsequent evolution calculations.

[0012] Preferably, calculating the thickness growth variable of the solid electrolyte interface film based on the updated diffusion coefficient and the iterated porosity parameter, and determining the change sequence of the thickness growth variable with the cycle period as the physical degradation trajectory feature further includes: extracting lithium-ion consumption rate data corresponding to the updated diffusion coefficient, comparing the lithium-ion consumption rate data with the negative electrode lithium intercalation rate data; when the lithium-ion consumption rate data exceeds the negative electrode lithium intercalation rate data, activating the lithium plating side reaction calculation branch, calculating the dead lithium volume growth data based on the lithium plating side reaction calculation branch; superimposing and correcting the dead lithium volume growth data with the thickness growth variable, and updating the change sequence of the superimposed and corrected thickness growth variable with the cycle period as the physical degradation trajectory feature.

[0013] Preferably, inputting the main fluctuation component into a preset mapping function to calculate the dynamic adjustment coefficient at the current moment further includes: performing a differential operation on the main fluctuation component at adjacent time steps to obtain stress-induced impulse signal data; inputting the stress-induced impulse signal data into a preset exponential decay response function to calculate an impulse response decay coefficient, wherein the impulse response decay coefficient decays exponentially over time; multiplying the impulse response decay coefficient with the current amplitude of the main fluctuation component to obtain a corrected fluctuation component; and replacing the main fluctuation component with the corrected fluctuation component and inputting it into the mapping function to calculate the dynamic adjustment coefficient at the current moment.

[0014] Preferably, the process of performing feature cross-referencing between the transient electrical impact stress factor and the cumulative humidity in the environmental temperature and humidity data to generate a humid-thermal-electric coupling stress feature vector, and incorporating the humid-thermal-electric coupling stress feature vector into the multidimensional dynamic stress factor, further includes: calculating the information entropy of the energy distribution data in the high-frequency band of the current frequency domain feature spectrum to obtain the spectral disorder index; performing a nonlinear mapping between the spectral disorder index and the cumulative humidity to calculate the surface film repair blocking factor; performing vector concatenation between the surface film repair blocking factor and the transient electrical impact stress factor to generate the humid-thermal-electric coupling stress feature vector, and incorporating the humid-thermal-electric coupling stress feature vector into the multidimensional dynamic stress factor.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention embeds multidimensional dynamic stress factors extracted from environmental temperature and humidity and electrical operating conditions into an electrochemical degradation model, adjusts the activation energy parameters of chemical reactions and controls the evolution of the growth rate of the solid electrolyte interface film, and obtains physical degradation trajectory characteristics, so that the prediction process is constrained by electrochemical mechanisms. Based on the fluctuation variance of the multidimensional dynamic stress factors, the weight ratio of physical degradation trajectory characteristics and data fitting characteristics is dynamically adjusted in the physical-data fusion prediction network. The contribution of physical constraints and data fitting is adaptively allocated according to stress fluctuations, overcoming the prediction distortion problem of pure data-driven or static physical models under complex coal mine conditions, and realizing objective tracking of nonlinear accelerated battery degradation.

[0016] 2. This invention achieves comprehensive quantification of stress characteristics under the intertwined scenarios of damp heat and transient electrical shock in underground coal mines by jointly encoding the thermal distribution gradient and condensation probability index, and introducing the damp-heat-electric coupling stress generated by the cross-generation of high-frequency current energy distribution and humidity accumulation. It uses physical degradation trajectory characteristics as state initialization constraints for long short-term memory networks and performs Hadamard product operations with the forget gate output, limiting the data-driven network's feature outputs that deviate from the physical mechanism. Furthermore, it utilizes the comparison between lithium-ion consumption rate and lithium intercalation rate to activate the lithium plating side reaction branch, calculates the dead lithium volume to correct the thickness growth variable, and modifies the dynamic adjustment coefficient based on the stress abrupt shock signal using an exponential decay response function, enhancing the accuracy of characterizing the sudden drop in lifetime caused by transient shocks and lithium plating side reactions. Attached Figure Description

[0017] Figure 1 This is a flowchart of the overall method for predicting the battery life of electric vehicles in coal mines based on big data, as described in this invention. Figure 2 This is a flowchart of the multidimensional dynamic stress factor extraction and joint encoding process of the present invention; Figure 3 This is the process for solving the electrochemical degradation model and generating physical degradation trajectory features according to the present invention; Figure 4 This is a flowchart of the dynamic weight allocation and feature weighted fusion process of the present invention; Figure 5 This is a flowchart of the data fitting feature generation process for the physically constrained long short-term memory network of the present invention. Figure 6 This is a flowchart illustrating the refined generation process of the damp-heat-electric coupling stress characteristics of the present invention. Detailed Implementation

[0018] refer to Figure 1In one embodiment, ambient temperature and humidity data and electrical condition data of the electric vehicle operating in a coal mine are collected. The ambient temperature and humidity data and the electrical condition data are time-series aligned and feature extracted to obtain a multidimensional dynamic stress factor. This multidimensional dynamic stress factor is embedded as an input variable into a pre-constructed electrochemical degradation model. The activation energy parameters of the chemical reactions in the electrochemical degradation model are adjusted using the multidimensional dynamic stress factor to control the evolution of the solid electrolyte interface film growth rate parameter, thereby obtaining physical degradation trajectory features. A physical-data fusion prediction network is constructed. A dynamic weight allocation mechanism is introduced into the data-driven network of this physical-data fusion prediction network to calculate the fluctuation variance of the multidimensional dynamic stress factor within a sliding time window. Based on the fluctuation variance, the weight ratio of the physical degradation trajectory features output by the electrochemical degradation model and the data fitting features output by the data-driven network in the final lifetime prediction result is adjusted. The weighted physical degradation trajectory features and the data fitting features are fused together to output the remaining battery life.

[0019] refer to Figure 2 Specifically, ambient temperature and humidity data are collected using temperature and humidity sensors deployed inside and outside the battery box of the electric vehicle in the coal mine, with a sampling frequency of 1Hz. Electrical operating condition data is obtained through the battery management system, including parameters such as charging and discharging current, terminal voltage, charge and discharge rate, deep charge and discharge cycle count, and individual cell voltage difference, with a sampling frequency of 10Hz. The time alignment process uses linear interpolation to unify data from different sampling frequencies to the same time base. For any two adjacent sampling times... and At the midpoint The interpolated data values ​​at the specified points are calculated using the following formula:

[0020] in, Indicates time Interpolated data values ​​at the location, and Representing time respectively and The original sampled data value at that location.

[0021] The feature extraction process is performed on time-aligned environmental temperature and humidity data and electrical operating condition data to extract feature parameters that reflect the external stresses experienced by the battery. The multidimensional dynamic stress factor is a vector containing multiple dimensions, each corresponding to a specific type of stress. This multidimensional dynamic stress factor is embedded as an input variable into a pre-constructed electrochemical degradation model. This model, based on the electrochemical mechanism of lithium-ion batteries, describes various electrochemical reaction processes occurring inside the battery and their impact on battery performance.

[0022] The activation energy parameters of chemical reactions in the electrochemical degradation model are adjusted by the multidimensional dynamic stress factor. The activation energy of a chemical reaction determines the ease with which the reaction proceeds; the lower the activation energy, the easier the reaction. The complex environmental conditions and electrical operating conditions in underground coal mines alter the internal temperature distribution and stress state of the battery, affecting the activation energy of the electrochemical reactions. The evolution of the solid electrolyte interfacial film growth rate parameter is controlled. The solid electrolyte interfacial film is a passivation film formed on the negative electrode surface during the first charge and discharge cycle of the battery, and its growth process directly affects the battery's cycle life. As charge and discharge cycles continue, the solid electrolyte interfacial film continuously thickens, leading to increased internal resistance and capacity decay.

[0023] Physical degradation trajectory features are obtained, reflecting the performance degradation trend of the battery under electrochemical constraints. A physical-data fusion prediction network is constructed, consisting of a physical model branch and a data-driven branch. The physical model branch outputs the physical degradation trajectory features, and the data-driven branch outputs the data fitting features. A dynamic weight allocation mechanism is introduced into the data-driven branch of the physical-data fusion prediction network. This mechanism adaptively adjusts the weight ratio of the physical degradation trajectory features and the data fitting features in the final lifetime prediction result based on the stress fluctuations experienced by the battery during operation.

[0024] The variance of the multidimensional dynamic stress factor within a sliding time window is calculated. This variance reflects the degree of stress variation over a certain time range. When stress fluctuations are small, the battery degradation process mainly follows an electrochemical mechanism, and the weight of the physical degradation trajectory features is relatively high. When stress fluctuations are large, the battery degradation process exhibits a nonlinear acceleration phenomenon, and the weight of the data fitting features is relatively high. The weighting of the physical degradation trajectory features output by the electrochemical degradation model and the data fitting features output by the data-driven network in the final lifetime prediction result is adjusted based on the variance of the variance. The weighted physical degradation trajectory features and the data fitting features are then fused to output the remaining battery lifetime.

[0025] In this embodiment, the composition and physical meaning of the multidimensional dynamic stress factor are shown in Table 1.

[0026] Table 1 Composition and Physical Significance of Multidimensional Dynamic Stress Factors ; In this embodiment, the multidimensional dynamic stress factor comprises five stress components, comprehensively quantifying the external stresses experienced by the battery from multiple aspects, including thermal, humidity, cycling, electrical shock, and their coupling effects. Each stress component corresponds to a specific degradation mechanism within the battery, accurately reflecting the impact of complex underground coal mine conditions on battery life. By embedding these multidimensional dynamic stress factors into the electrochemical degradation model, the model can more realistically simulate the battery degradation process under actual operating conditions.

[0027] In a preferred embodiment, the ambient temperature data from the ambient temperature and humidity data and the charge / discharge rate data from the electrical operating condition data are input into a preset thermo-electrical coupled stress field model to calculate the thermal distribution gradient data within the battery cell. The absolute humidity change rate data from the ambient temperature and humidity data is extracted, and the absolute humidity change rate data is spatiotemporally aligned with the thermal distribution gradient data to calculate the condensation probability index. The thermal distribution gradient data, the condensation probability index, and the number of deep charge / discharge cycles from the electrical operating condition data are jointly encoded and mapped to the multidimensional dynamic stress factor.

[0028] Specifically, the thermo-electric coupled stress field model is constructed based on Fourier's law of heat conduction and Joule's law of heat transfer, describing the heat generation and transfer processes inside the battery. The heat generation inside the battery mainly includes three parts: Joule heat, reaction heat, and polarization heat. Joule heat is the heat generated by the current passing through the battery's internal resistance; reaction heat is the heat generated by the entropy change during the electrochemical reaction; and polarization heat is the heat generated by electrode polarization. The governing equations of the thermo-electric coupled stress field model are as follows:

[0029] in, Indicates the density of battery materials. This indicates the specific heat capacity of the battery material. Indicates the internal temperature of the battery. Indicates time, Indicates the thermal conductivity of the battery material. Indicates the Joule heat production rate. Indicates the heat production rate of the reaction. This indicates the polarization heat production rate.

[0030] The Joule heat production rate is calculated using the following formula:

[0031] in, Indicates the charging and discharging current. This indicates the internal resistance of the battery.

[0032] The heat of reaction and the heat production rate are calculated using the following formula:

[0033] in, This indicates the open-circuit voltage of the battery. This represents the temperature coefficient of the open-circuit voltage.

[0034] The polarization heat generation rate is calculated using the following formula:

[0035] in, This indicates the battery's terminal voltage.

[0036] By solving the governing equations of the aforementioned thermo-electric coupled stress field model, the temperature value at any location within a single battery cell at any given time can be obtained. The thermal distribution gradient data, obtained by calculating the temperature difference between different locations within the battery, reflects the non-uniformity of the internal temperature. This non-uniformity leads to different electrochemical reaction rates at different locations, accelerating the battery degradation process.

[0037] Extract the absolute humidity change rate data from the environmental temperature and humidity data. Absolute humidity represents the mass of water vapor contained in a unit volume of air, and the unit is... The absolute humidity change rate is obtained by calculating the ratio of the difference in absolute humidity at adjacent moments to the time interval. The absolute humidity change rate data is spatiotemporally aligned with the heat distribution gradient data. This spatiotemporal alignment process ensures that the humidity data and temperature data correspond to each other in time and space.

[0038] The condensation probability index is calculated because condensation occurs when water vapor in the air condenses into liquid water on the battery surface when the battery surface temperature is lower than the dew point temperature of the surrounding air. Condensation can lead to a decrease in battery insulation performance and even cause short circuits. The condensation probability index is calculated using the following formula:

[0039] in, This represents the condensation probability index, with a value range of [value range missing]. , This indicates the dew point temperature of the air. Indicates the surface temperature of the battery. This represents the adjustment coefficient.

[0040] Dew point temperature Calculated using the following formula:

[0041] in, and For water, it is a constant. , , This represents the actual water vapor pressure in the air. This indicates the saturated water vapor pressure.

[0042] Saturated vapor pressure Calculated using the following formula:

[0043] in, Indicates air temperature, in units of .

[0044] The thermal distribution gradient data, the condensation probability index, and the number of deep charge-discharge cycles in the electrical operating condition data are jointly encoded and mapped to the multidimensional dynamic stress factor. The joint encoding process employs normalization and vector concatenation. First, the thermal distribution gradient data, the condensation probability index, and the number of deep charge-discharge cycles are each normalized to ensure their values ​​are within a uniform range. Within the interval. The normalization formula is as follows:

[0045] in, This represents the normalized data value. Represents the original data value. This represents the minimum value of this type of data. This indicates the maximum value of this type of data.

[0046] Then, the normalized thermal distribution gradient data, condensation probability exponent, and deep charge-discharge cycle count are concatenated into vectors to form a three-dimensional initial stress vector. This initial stress vector is then nonlinearly mapped through a fully connected layer to obtain the final multidimensional dynamic stress factor. The mapping relationship of the fully connected layer is as follows:

[0047] in, Represents the multidimensional dynamic stress factor. The activation function is represented by the ReLU function. Represents the weight matrix. Represents the initial stress vector. This represents the bias vector.

[0048] refer to Figure 3In this embodiment, the thermal stress component of the multidimensional dynamic stress factor is input into the exponential term of the Arrhenius equation to dynamically correct the activation energy parameter of the electrolyte decomposition reaction in the electrochemical degradation model. Based on the corrected activation energy parameter, the diffusion coefficient in the solid electrolyte interface film growth rate parameter is updated, and the porosity parameter of the solid electrolyte interface film is iteratively decayed using the mechanical stress component of the multidimensional dynamic stress factor. According to the updated diffusion coefficient and the iterated porosity parameter, the thickness growth variable of the solid electrolyte interface film is calculated, and the sequence of changes in the thickness growth variable with the cycle period is determined as the physical degradation trajectory feature.

[0049] Specifically, the Arrhenius equation describes the relationship between chemical reaction rate and temperature, and its expression is as follows:

[0050] in, Represents the reaction rate constant. Indicates pre-exponential factor, Indicates the activation energy of the reaction. Represents the gas constant. This indicates absolute temperature.

[0051] The thermal stress component of the multidimensional dynamic stress factor is input into the exponential term of the Arrhenius equation to dynamically correct the activation energy parameter of the electrolyte decomposition reaction. The corrected activation energy parameter is calculated using the following formula:

[0052] in, This represents the corrected activation energy parameter. This represents the activation energy parameter under standard conditions. Indicates the influence coefficient of thermal stress. This represents the thermal stress component in the multidimensional dynamic stress factor.

[0053] The diffusion coefficient in the solid electrolyte interface film growth rate parameter is updated based on the corrected activation energy parameter. The growth process of the solid electrolyte interface film is mainly controlled by the diffusion process of lithium ions in the film, and the relationship between the diffusion coefficient and temperature and activation energy also follows the Arrhenius equation. The updated diffusion coefficient is calculated using the following formula:

[0054] in, This represents the updated diffusion coefficient. This represents the diffusion coefficient under standard conditions.

[0055] The porosity parameter of the solid electrolyte interface membrane is iteratively decayed by incorporating the mechanical stress component of the multidimensional dynamic stress factor. Mechanical stress causes microcracks in the solid electrolyte interface membrane, increasing its porosity. With charge-discharge cycles, the porosity gradually decreases. The decay iterative formula for the porosity parameter is as follows:

[0056] in, Indicates the first Porosity after one cycle Indicates the first Porosity after one cycle Indicates the influence coefficient of mechanical stress. This represents the mechanical stress component in the multidimensional dynamic stress factor.

[0057] The thickness growth variable of the solid electrolyte interface film is calculated based on the updated diffusion coefficient and the iterative porosity parameter. The relationship between the thickness growth of the solid electrolyte interface film and the diffusion coefficient and porosity is as follows:

[0058] in, The variable representing the thickness increase of the solid electrolyte interface film. Indicates the charge / discharge cycle time. The tortuosity factor represents the degree of tortuosity of the diffusion path of lithium ions in the solid electrolyte interface film.

[0059] The sequence of changes in the thickness growth variable over the cycle period is defined as the physical degradation trajectory feature. The physical degradation trajectory feature is a time series, with each element corresponding to the thickness growth of the solid electrolyte interface film within one charge-discharge cycle. This feature reflects the performance degradation trend of the battery under electrochemical constraints.

[0060] refer to Figure 4 In this embodiment, a sliding time window is constructed along the time axis, and the covariance matrix of the multidimensional dynamic stress factor within the sliding time window is calculated. Eigenvalue decomposition is performed on the covariance matrix to obtain the principal fluctuation component. The principal fluctuation component is input into a preset mapping function to calculate the dynamic adjustment coefficient at the current moment. The mapping function is a monotonically increasing function. The dynamic adjustment coefficient is used as the first weighting coefficient of the data fitting features output by the data-driven network, and the difference between the dynamic adjustment coefficient and the first weighting coefficient is used as the second weighting coefficient of the physical degradation trajectory features output by the electrochemical degradation model. The data fitting features and the physical degradation trajectory features are then weighted and summed based on the first and second weighting coefficients.

[0061] Specifically, a sliding time window is constructed along the time axis, with the window size set to... A time step. For the current moment... The sliding time window contains time from time At the time The multidimensional dynamic stress factor data is obtained. The covariance matrix of the multidimensional dynamic stress factor within the sliding time window is calculated. The elements of the covariance matrix are calculated using the following formula:

[0062] in, Represents the first element in the covariance matrix. Line number Column elements, Indicates the first The first time step The value of the stress factor, Indicates the first The average value of the stress factor within the sliding time window. Indicates the first The first time step The value of the stress factor, Indicates the first The average value of the stress factor within the sliding time window.

[0063] The covariance matrix is ​​subjected to eigenvalue decomposition to obtain the principal fluctuation components. The expression for eigenvalue decomposition is as follows:

[0064] in, This represents the eigenvector matrix, where each column corresponds to an eigenvector. This represents an eigenvalue diagonal matrix, where the elements on the diagonal correspond to the eigenvalues ​​of each eigenvector.

[0065] The eigenvalues ​​are arranged in descending order, and the eigenvector corresponding to the largest eigenvalue is taken as the principal fluctuation component. The principal fluctuation component reflects the main fluctuation direction and degree of the multidimensional dynamic stress factor within the sliding time window.

[0066] The main fluctuation component is input into a preset mapping function to calculate the dynamic adjustment coefficient at the current moment. The mapping function uses the sigmoid function, and its expression is as follows:

[0067] in, This represents the dynamic adjustment coefficient, with a value range of [value missing]. , This represents the adjustment coefficient. This represents the amplitude of the main fluctuation component.

[0068] The dynamic adjustment coefficient is used as the first weighting coefficient of the data fitting features output by the data-driven network, and the difference between the dynamic adjustment coefficient and the data fitting coefficient is used as the second weighting coefficient of the physical degradation trajectory features output by the electrochemical degradation model. The data fitting features and the physical degradation trajectory features are then weighted and summed based on the first and second weighting coefficients to obtain the final battery remaining life prediction result. The weighted summation formula is as follows:

[0069] in, This indicates the final predicted battery remaining life. This represents the remaining lifetime corresponding to the data-fitted features output by the data-driven network. This represents the remaining lifetime corresponding to the physical degradation trajectory characteristics output by the electrochemical degradation model.

[0070] In this embodiment, the parameter update process and parameter meanings of the electrochemical degradation model are shown in Table 2.

[0071] Table 2. Update process and meaning of parameters in the electrochemical degradation model. ; In this embodiment, the parameters of the electrochemical degradation model are dynamically updated based on the multidimensional dynamic stress factors experienced by the battery during operation, enabling the model to more accurately simulate the degradation process of the battery under actual operating conditions. By incorporating the thermal stress component into the correction of the activation energy parameter and the mechanical stress component into the decay iteration of the porosity parameter, the influence of various external stresses on the internal electrochemical degradation process of the battery can be comprehensively considered.

[0072] refer to Figure 6 In a preferred embodiment, a short-time Fourier transform is performed on the transient current change sequence in the electrical operating condition data to obtain a current frequency domain feature spectrum. Energy distribution data in the high-frequency band is extracted from the current frequency domain feature spectrum and identified as the transient electrical impact stress factor. The transient electrical impact stress factor is then cross-referenced with the cumulative humidity in the environmental temperature and humidity data to generate a humid-thermal-electric coupling stress feature vector, which is then incorporated into the multidimensional dynamic stress factor.

[0073] Specifically, a short-time Fourier transform (SFT) is performed on the transient current change sequence in the electrical operating condition data. The SFT is a time-frequency analysis method that can simultaneously provide information about the signal in both the time and frequency domains. The expression for the SFT is as follows:

[0074] in, Indicates time ,frequency The short-time Fourier transform value at that point, Represents a sequence of transient current changes. The window function is represented, using the Hanning window.

[0075] The time-frequency distribution of the current signal, i.e., the current frequency domain characteristic spectrum, can be obtained through short-time Fourier transform. Energy distribution data in the high-frequency band is extracted from this spectrum. The high-frequency band is set to be above 100Hz, as the energy in this band mainly reflects the transient changes in the current. The energy distribution data in the high-frequency band is obtained by calculating the sum of squares of the amplitudes at each frequency point within this band. This high-frequency band energy distribution data is then determined as the transient electric shock stress factor.

[0076] The transient electrical shock stress factor is cross-referenced with the cumulative humidity from the ambient temperature and humidity data. The cumulative humidity is calculated by integrating the ambient humidity over a period of time, reflecting the total degree of battery exposure in a humid environment. The feature cross-reference process employs a polynomial feature expansion method to generate a new feature that incorporates the interaction between the two features. The expression for the feature cross-reference is as follows:

[0077] in, This represents the vector after feature crossing. Indicates the transient electrical impact stress factor. This indicates the cumulative humidity level.

[0078] A hygrothermal-electrical coupling stress eigenvector is generated and incorporated into the multidimensional dynamic stress factor. The hygrothermal-electrical coupling stress eigenvector reflects the damage to the battery caused by the combined effects of a humid environment and transient electrical shocks. In underground coal mines, the humid environment accelerates the corrosion of battery metal components, while transient electrical shocks cause localized high temperatures inside the battery; the combined effect of these two factors significantly accelerates the battery degradation process.

[0079] refer to Figure 5 In this embodiment, the physical degradation trajectory features output by the electrochemical degradation model are used as the state initialization constraint vector of the Long Short-Term Memory (LSTM) network in the data-driven network. During the forward propagation of the LSTM network, the rate of change of the derivative of the physical degradation trajectory features is multiplied by the Hadamard product of the forget gate output within the LSTM network to limit the degree to which the LSTM network retains features that deviate from the physical degradation trajectory features. The constrained output sequence of the LSTM network is then determined as the data fitting feature.

[0080] Specifically, the data-driven network adopts a Long Short-Term Memory (LSTM) network structure. LSTM is a special type of recurrent neural network that can effectively process long sequences of data and solves the gradient vanishing and gradient exploding problems existing in traditional recurrent neural networks. The basic unit of LSTM consists of four parts: input gate, forget gate, output gate, and cell state.

[0081] The physical degradation trajectory features output by the electrochemical degradation model are used as the state initialization constraint vector for the Long Short-Term Memory (LSTM) network in the data-driven network. The initial cell states and initial hidden states of the LSTM are obtained by initializing the physical degradation trajectory features. The initialization formula is as follows:

[0082]

[0083] in, Indicates the initial cell state. Indicates the initial hidden state. and Represents the weight matrix. Indicates the characteristics of physical degradation trajectory. and This represents the bias vector.

[0084] During the forward propagation of the Long Short-Term Memory (LSTM) network, the rate of change of the derivative of the physical degradation trajectory feature is multiplied by the Hadamard product of the forget gate output within the LSTM network. The rate of change of the derivative of the physical degradation trajectory feature reflects the change in the battery degradation rate. The forget gate controls which information in the cell state needs to be forgotten. By performing the Hadamard product operation between the rate of change of the derivative of the physical degradation trajectory feature and the forget gate output, the degree to which the LSTM retains features deviating from the physical degradation trajectory feature can be limited, making the output of the data-driven network more consistent with the electrochemical mechanism.

[0085] The expression for the Hadamard product operation is as follows:

[0086] in, This indicates the corrected output of the forget gate. This represents the original forget gate output. This represents the Hadamard product operation. The rate of change of the derivative represents the characteristic of the physical degradation trajectory.

[0087] The constrained output sequence of the Long Short-Term Memory (LSTM) network is determined as the data fitting feature. This feature reflects the performance degradation trend of the battery under actual operational data-driven conditions. By using the physical degradation trajectory features as the state initialization constraint for the LSTM and correcting the forget gate output, the data-driven network can learn statistical regularities from the data while being constrained by electrochemical mechanisms, thus avoiding predictions that do not conform to physical laws.

[0088] In a preferred embodiment, the condensation probability index and the battery separator micropore size parameters are input into a preset wetting diffusion model to calculate the abnormal electrolyte wetting depth data. Based on the abnormal electrolyte wetting depth data and the local high-temperature region data in the thermal distribution gradient data, the probability data of local micro-short circuit occurrence is estimated. The probability data of local micro-short circuit occurrence is then added as an additional dimension to the multidimensional dynamic stress factor for subsequent evolution calculations.

[0089] Specifically, the wetting diffusion model, based on Fick's diffusion law, describes the wetting process of the electrolyte in the battery separator. Abnormal electrolyte wetting refers to the phenomenon where condensation increases the water content in the electrolyte, causing changes in its wetting properties. The abnormal electrolyte wetting depth data is calculated using the following formula:

[0090] in, Indicates the abnormal wetting depth of the electrolyte. This represents the diffusion coefficient of the electrolyte. Indicates the soaking time. Indicates the condensation effect coefficient. This represents the probability index of condensation. The tortuosity factor of the diaphragm.

[0091] Based on the abnormal electrolyte wetting depth data and the local high-temperature region data in the thermal distribution gradient data, the probability of local micro-short circuit occurrence is estimated. A local micro-short circuit refers to a tiny short circuit between the positive and negative electrodes inside the battery caused by separator damage or conductive impurities. Local micro-short circuits can lead to a localized increase in battery temperature and even thermal runaway. The probability of local micro-short circuit occurrence is calculated using the following formula:

[0092] in, This represents the probability of a local micro-short circuit occurring, with a value range of [value missing]. , This represents the adjustment coefficient. This indicates the threshold for abnormal electrolyte wetting depth. This represents the temperature gradient value of a local high-temperature region in the thermal distribution gradient data.

[0093] The probability data of local micro-short circuit occurrence is added as an additional dimension to the multidimensional dynamic stress factor for subsequent evolution calculations. Local micro-short circuits are one of the important causes of sudden battery failure in electric vehicles in coal mines. Introducing the probability of local micro-short circuit occurrence into the multidimensional dynamic stress factor enables the prediction model to more accurately capture the risk of sudden battery failure.

[0094] In a preferred embodiment, the lithium-ion consumption rate data corresponding to the updated diffusion coefficient is extracted and compared with the negative electrode lithium intercalation rate data. When the lithium-ion consumption rate data exceeds the negative electrode lithium intercalation rate data, the lithium plating side reaction calculation branch is activated, and the dead lithium volume growth data is calculated based on the lithium plating side reaction calculation branch. The dead lithium volume growth data is superimposed and corrected with the thickness growth variable, and the superimposed and corrected thickness growth variable change sequence with cycle period is updated to the physical degradation trajectory feature.

[0095] Specifically, the updated lithium-ion consumption rate data corresponding to the diffusion coefficient is extracted. The lithium-ion consumption rate refers to the number of lithium ions consumed per unit time due to electrolyte decomposition and solid electrolyte interfacial film growth. The lithium-ion consumption rate data is calculated using the following formula:

[0096] in, This indicates the rate of lithium-ion consumption. This indicates the lithium ion concentration in the electrolyte. This indicates the surface area of ​​the electrode.

[0097] The lithium intercalation rate of the negative electrode refers to the number of lithium ions that can be intercalated into the negative electrode per unit time. The lithium intercalation rate of the negative electrode is calculated using the following formula:

[0098] in, Indicates the lithium intercalation rate at the negative electrode. Indicates the charging and discharging current density. This indicates the number of charges carried by each lithium ion. This represents the Faraday constant.

[0099] The lithium-ion consumption rate data is compared with the negative electrode lithium intercalation rate data. When the lithium-ion consumption rate data exceeds the negative electrode lithium intercalation rate data, it indicates that excess lithium ions cannot intercalate into the negative electrode and will deposit metallic lithium on the negative electrode surface, i.e., a lithium plating side reaction occurs. The lithium plating side reaction calculation branch is activated, and the dead lithium volume growth data is calculated based on this branch. Dead lithium refers to the portion of the deposited metallic lithium that can no longer participate in electrochemical reactions. The dead lithium volume growth data is calculated using the following formula:

[0100] in, This indicates the volume increase of dead lithium. Indicates the molar mass of lithium. This indicates the density of metallic lithium.

[0101] The data on the volume increase of dead lithium is combined with the data on the thickness increase variable for correction. The deposition of dead lithium increases the effective thickness of the solid electrolyte interface film, accelerating the increase in internal resistance and capacity decay of the battery. The corrected thickness increase variable is calculated using the following formula:

[0102] in, This represents the thickness growth variable after superposition and correction.

[0103] The sequence of changes in the thickness growth variable over the cycle period after superposition correction is updated to reflect the physical degradation trajectory characteristics. By introducing a calculation branch for lithium plating side reactions, the electrochemical degradation model can more accurately simulate the degradation process of batteries under high-current charge-discharge and low-temperature conditions, improving the accuracy of predicting sudden drops in battery life.

[0104] In a preferred embodiment, the main fluctuation component is differentially processed at adjacent time steps to obtain stress-induced impulse signal data. The stress-induced impulse signal data is input into a preset exponentially decaying response function to calculate the impulse response decay coefficient, which decays exponentially over time. The impulse response decay coefficient is multiplied by the current amplitude of the main fluctuation component to obtain a corrected fluctuation component. This corrected fluctuation component replaces the main fluctuation component and is input into the mapping function to calculate the dynamic adjustment coefficient at the current moment.

[0105] Specifically, the main fluctuation component is differentially analyzed at adjacent time steps to obtain stress-induced shock signal data. The expression for the differential operation is as follows:

[0106] in, Indicates time Stress abrupt change impulse signal data at the location, Indicates time The amplitude of the main fluctuation component at that location. Indicates time The amplitude of the main fluctuation component at that location.

[0107] The stress-induced impulse signal data is input into a preset exponentially decaying response function to calculate the impulse response decay coefficient. The expression for the exponentially decaying response function is as follows:

[0108] in, This represents the impulse response attenuation coefficient. Represents the decay time constant. This represents the unit step function.

[0109] The impulse response decay coefficient decays exponentially over time, reflecting that the impact of stress mutations on battery life prediction gradually weakens over time. The corrected fluctuation component is obtained by multiplying the impulse response decay coefficient by the current amplitude of the main fluctuation component. The expression for the corrected fluctuation component is as follows:

[0110] in, Indicates time The corrected fluctuation component at that point.

[0111] The corrected fluctuation component is used to replace the main fluctuation component in the mapping function to calculate the dynamic adjustment coefficient at the current moment. By introducing the exponentially decaying response of the stress mutation impulse signal, the dynamic weight allocation mechanism can more accurately reflect the impact of stress mutation on the battery degradation process, avoid drastic fluctuations in the weight coefficients caused by stress mutation, and improve the stability of the lifetime prediction results.

[0112] In a preferred embodiment, information entropy is calculated on the energy distribution data of the high-frequency band in the current frequency domain feature spectrum to obtain a spectral disorder index. The spectral disorder index is then nonlinearly mapped to the accumulated humidity to calculate a surface film repair blocking factor. The surface film repair blocking factor is then vector-concatenated with the transient electro-impact stress factor to generate a hygrothermal-electric coupling stress feature vector, which is then incorporated into the multidimensional dynamic stress factor.

[0113] Specifically, information entropy is calculated on the energy distribution data of the high-frequency band in the current frequency domain feature spectrum. Information entropy is an indicator that measures the degree of uncertainty and disorder in a signal. The spectral disorder index is calculated using the following formula:

[0114] in, This represents the spectral disorder index. Indicates the number of frequency points within the high-frequency band. Indicates the first The proportion of energy at each frequency point to the total energy of the high-frequency band.

[0115] The surface film repair blocking factor is calculated by performing a nonlinear mapping between the spectral disorder index and the cumulative humidity. The surface film repair blocking factor reflects the degree to which the humid and hot environment and electrical shock hinder the repair process of the solid electrolyte interface film. The nonlinear mapping uses a hyperbolic tangent function, and its expression is as follows:

[0116] in, This represents the surface film repair blocking factor, with a value range of [value missing]. , This represents the adjustment coefficient.

[0117] The surface film repair blocking factor and the transient electrical shock stress factor are vector-concatenated to generate the hygrothermal-electrical coupling stress feature vector. This hygrothermal-electrical coupling stress feature vector is then incorporated into the multidimensional dynamic stress factor. During battery operation, the solid electrolyte interface film continuously deteriorates and repairs. When the surface film repair process is hindered, the damage accumulates, accelerating battery degradation. By introducing the surface film repair blocking factor, the impact of hygrothermal-electrical coupling stress on battery life can be more comprehensively quantified.

[0118] In this embodiment, the parameter settings for each layer of the physical-data fusion prediction network are shown in Table 3.

[0119] Table 3 Parameter settings for each layer of the physical-data fusion prediction network ; In this embodiment, the data-driven branch of the physical-data fusion prediction network employs a two-layer LSTM structure and a two-layer fully connected layer structure. The input layer receives 5-dimensional multidimensional dynamic stress factor data. The first LSTM layer maps the input data to a 64-dimensional hidden state space. The second LSTM layer further extracts sequence features, reducing the dimension of the hidden state space to 32 dimensions. The first fully connected layer maps the 32-dimensional sequence features to a 16-dimensional feature space. The second fully connected layer maps the 16-dimensional features to a 1-dimensional output, i.e., the remaining lifetime corresponding to the data-fitted features.

[0120] In this embodiment, ambient temperature and humidity data and electrical condition data are collected during the operation of the coal mine electric vehicle. The ambient temperature and humidity data includes ambient temperature, relative humidity, absolute humidity, absolute humidity change rate, and cumulative humidity. The electrical condition data includes charging and discharging current, terminal voltage, charge and discharge rate, deep charge and discharge cycle count, current transient change sequence, and individual battery voltage difference. The ambient temperature and humidity data and the electrical condition data are time-aligned, and linear interpolation is used to unify the 1Hz sampling frequency temperature and humidity data and the 10Hz sampling frequency electrical condition data to a 1Hz time base.

[0121] Feature extraction is performed on the time-aligned data. Ambient temperature data and charge / discharge rate data are input into the thermo-electric coupled stress field model to calculate the thermal distribution gradient data inside the battery cell. The absolute humidity change rate data is extracted and spatiotemporally aligned with the thermal distribution gradient data to calculate the condensation probability index. The thermal distribution gradient data, condensation probability index, and deep charge / discharge cycle count are jointly encoded and mapped to an initial multidimensional dynamic stress factor.

[0122] A short-time Fourier transform is performed on the transient current change sequence to obtain the current frequency domain feature spectrum. Energy distribution data above 100Hz is extracted from the current frequency domain feature spectrum and identified as the transient electro-impact stress factor. Information entropy is calculated from the energy distribution data in the high-frequency band to obtain the spectral disorder index. The spectral disorder index is nonlinearly mapped to the accumulated humidity to calculate the surface film repair blocking factor. The surface film repair blocking factor and the transient electro-impact stress factor are vector-concatenated to generate a hygrothermal-electric coupling stress feature vector, which is then incorporated into the multidimensional dynamic stress factor.

[0123] The condensation probability index and the battery separator micropore size parameters are input into the wetting and diffusion model to calculate the abnormal electrolyte wetting depth data. Based on the abnormal electrolyte wetting depth data and the local high-temperature region data in the thermal distribution gradient data, the probability data of local micro-short circuit occurrence is estimated. The local micro-short circuit occurrence probability data is then added as an additional dimension to the multidimensional dynamic stress factor.

[0124] A multidimensional dynamic stress factor is embedded as an input variable into a pre-constructed electrochemical degradation model. The thermal stress component of the multidimensional dynamic stress factor is input into the exponential term of the Arrhenius equation to dynamically correct the activation energy parameter of the electrolyte decomposition reaction. The diffusion coefficient in the growth rate parameter of the solid electrolyte interfacial film is updated based on the corrected activation energy parameter. The porosity parameter of the solid electrolyte interfacial film is then iteratively decayed by incorporating the mechanical stress component of the multidimensional dynamic stress factor.

[0125] Based on the updated diffusion coefficient and iterated porosity parameters, the thickness growth variable of the solid electrolyte interface film is calculated. The lithium-ion consumption rate data corresponding to the updated diffusion coefficient is extracted and compared with the negative electrode lithium intercalation rate data. When the lithium-ion consumption rate data exceeds the negative electrode lithium intercalation rate data, the lithium plating side reaction calculation branch is activated to calculate the dead lithium volume growth data. The dead lithium volume growth data is superimposed and corrected with the thickness growth variable, and the sequence of changes in the superimposed and corrected thickness growth variable with the cycle period is determined as the physical degradation trajectory characteristic.

[0126] A physical-data fusion prediction network is constructed, using physical degradation trajectory features as the state initialization constraint vector for the Long Short-Term Memory (LSTM) network within the data-driven network. During the forward propagation of the LSTM network, the rate of change of the derivative of the physical degradation trajectory features is multiplied by the Hadamard product of the forget gate output within the LSTM network to limit the LSTM network's retention of features deviating from the physical degradation trajectory features. The constrained output sequence of the LSTM network is then determined as the data fitting feature.

[0127] A dynamic weight allocation mechanism is introduced into the data-driven network of the physics-data fusion prediction network. A sliding time window of size 100 time steps is constructed along the time axis, and the covariance matrix of the multidimensional dynamic stress factor within the sliding time window is calculated. Eigenvalue decomposition is performed on the covariance matrix to obtain the main fluctuation component. The main fluctuation component is differentially analyzed at adjacent time steps to obtain the stress mutation impulse signal data. The stress mutation impulse signal data is input into the exponentially decaying response function to calculate the impulse response decay coefficient. The impulse response decay coefficient is multiplied by the current amplitude of the main fluctuation component to obtain the corrected fluctuation component.

[0128] The corrected fluctuation component is input into the sigmoid mapping function to calculate the dynamic adjustment coefficient at the current moment. This dynamic adjustment coefficient is used as the first weighting coefficient for the data fitting features output by the data-driven network, and the difference between the dynamic adjustment coefficient and the dynamic adjustment coefficient is used as the second weighting coefficient for the physical degradation trajectory features output by the electrochemical degradation model. Based on the first and second weighting coefficients, the data fitting features and the physical degradation trajectory features are weighted and summed to output the remaining battery life.

Claims

1. A method for predicting the battery life of electric vehicles in coal mines based on big data, characterized in that, include: Collect ambient temperature and humidity data and electrical condition data during the operation of electric vehicles in coal mines, perform time-series alignment and feature extraction on the ambient temperature and humidity data and the electrical condition data, and obtain multi-dimensional dynamic stress factors; The multidimensional dynamic stress factor is embedded as an input variable into a pre-constructed electrochemical degradation model. The activation energy parameters of the chemical reactions in the electrochemical degradation model are adjusted by the multidimensional dynamic stress factor to control the evolution process of the solid electrolyte interface film growth rate parameters and obtain the physical degradation trajectory characteristics. A physical-data fusion prediction network is constructed, and a dynamic weight allocation mechanism is introduced into the data-driven network of the physical-data fusion prediction network. The fluctuation variance of the multidimensional dynamic stress factor within the sliding time window is calculated, and the weight ratio of the physical degradation trajectory features output by the electrochemical degradation model and the data fitting features output by the data-driven network in the final lifetime prediction result is adjusted according to the fluctuation variance. The remaining battery life is output by fusing the weighted physical degradation trajectory features with the data fitting features.

2. The method for predicting the battery life of electric vehicles in coal mines based on big data according to claim 1, characterized in that, The process of performing time-series alignment and feature extraction on the ambient temperature and humidity data and the electrical operating condition data to obtain multidimensional dynamic stress factors includes: inputting the ambient temperature data in the ambient temperature and humidity data and the charge / discharge rate data in the electrical operating condition data into a preset thermo-electrical coupling stress field model to calculate the thermal distribution gradient data inside the battery cell; Extract the absolute humidity change rate data from the environmental temperature and humidity data, align the absolute humidity change rate data with the heat distribution gradient data in time and space, and calculate the condensation probability index. The thermal distribution gradient data, the condensation probability index, and the number of deep charge-discharge cycles in the electrical operating condition data are jointly encoded and mapped to the multidimensional dynamic stress factor.

3. The method for predicting the battery life of electric vehicles in coal mines based on big data according to claim 1, characterized in that, By adjusting the activation energy parameters of the chemical reactions in the electrochemical degradation model using the multidimensional dynamic stress factor, the evolution process of the solid electrolyte interface film growth rate parameters is controlled, and physical degradation trajectory characteristics are obtained. This includes: inputting the thermal stress component in the multidimensional dynamic stress factor into the exponential term of the Arrhenius equation to dynamically correct the activation energy parameters of the electrolyte decomposition reaction in the electrochemical degradation model. The diffusion coefficient in the growth rate parameter of the solid electrolyte interface film is updated based on the corrected activation energy parameter, and the porosity parameter of the solid electrolyte interface film is attenuated iteratively by combining the mechanical stress component in the multidimensional dynamic stress factor. Based on the updated diffusion coefficient and the iterated porosity parameter, the thickness growth variable of the solid electrolyte interface film is calculated, and the change sequence of the thickness growth variable with the cycle period is determined as the physical degradation trajectory feature.

4. The method for predicting the battery life of electric vehicles in coal mines based on big data according to claim 1, characterized in that, Calculating the fluctuation variance of the multidimensional dynamic stress factor within a sliding time window, and adjusting the weight ratio of the physical degradation trajectory features output by the electrochemical degradation model and the data fitting features output by the data-driven network in the final lifetime prediction result based on the fluctuation variance, includes: constructing a sliding time window along the time axis, calculating the covariance matrix of the multidimensional dynamic stress factor within the sliding time window, performing eigenvalue decomposition on the covariance matrix, and obtaining the main fluctuation component; The main fluctuation component is input into a preset mapping function to calculate the dynamic adjustment coefficient at the current moment. The mapping function is a monotonically increasing function. The dynamic adjustment coefficient is used as the first weighting coefficient of the data fitting feature output by the data-driven network, and the difference between the dynamic adjustment coefficient and the data fitting feature output by the electrochemical degradation model is used as the second weighting coefficient. The data fitting feature and the physical degradation trajectory feature are weighted and summed based on the first weighting coefficient and the second weighting coefficient.

5. The method for predicting the battery life of electric vehicles in coal mines based on big data according to claim 1, characterized in that, Collecting ambient temperature and humidity data and electrical condition data during the operation of electric vehicles in coal mines, performing time-series alignment and feature extraction on the ambient temperature and humidity data and the electrical condition data, and obtaining multi-dimensional dynamic stress factors also includes: performing short-time Fourier transform on the current transient change sequence in the electrical condition data to obtain the current frequency domain feature spectrum. Energy distribution data of high-frequency bands are extracted from the current frequency domain feature spectrum, and the energy distribution data of the high-frequency bands are determined as transient electric shock stress factors; The transient electrical impact stress factor is cross-referenced with the cumulative humidity in the environmental temperature and humidity data to generate a humid-thermal-electrical coupled stress feature vector, which is then incorporated into the multidimensional dynamic stress factor.

6. The method for predicting the battery life of electric vehicles in coal mines based on big data according to claim 1, characterized in that, The construction of the physical-data fusion prediction network, and the introduction of a dynamic weight allocation mechanism in the data-driven network of the physical-data fusion prediction network, further includes: using the physical degradation trajectory features output by the electrochemical degradation model as the state initialization constraint vector of the long short-term memory network in the data-driven network; During the forward propagation of the Long Short-Term Memory (LSTM) network, the rate of change of the derivative of the physical degradation trajectory feature is multiplied by the Hadamard product of the forget gate output within the LTM network to limit the degree to which the LTM network retains features that deviate from the physical degradation trajectory feature. The constrained output sequence of the Long Short-Term Memory Network is determined as the data fitting feature.

7. The method for predicting the battery life of electric vehicles in coal mines based on big data according to claim 2, characterized in that, The method of jointly encoding the thermal distribution gradient data, the condensation probability index, and the number of deep charge-discharge cycles in the electrical condition data and mapping them to the multidimensional dynamic stress factor also includes: inputting the condensation probability index and the battery separator micropore size parameters into a preset wetting diffusion model to calculate the abnormal wetting depth data of the electrolyte. Based on the abnormal wetting depth data of the electrolyte and the local high temperature region data in the thermal distribution gradient data, the probability data of local micro short circuit occurrence is estimated. The probability data of local micro-short circuit occurrence is added as an additional dimension to the multidimensional dynamic stress factor for subsequent evolution calculation.

8. The method for predicting the battery life of electric vehicles in coal mines based on big data according to claim 3, characterized in that, Based on the updated diffusion coefficient and the iterated porosity parameter, the thickness growth variable of the solid electrolyte interface film is calculated. The change sequence of the thickness growth variable with the cycle period is determined as the physical degradation trajectory feature. The method further includes: extracting the lithium ion consumption rate data corresponding to the updated diffusion coefficient and comparing the lithium ion consumption rate data with the negative electrode lithium intercalation rate data. When the lithium-ion consumption rate data exceeds the negative electrode lithium intercalation rate data, the lithium plating side reaction calculation branch is activated, and the dead lithium volume growth data is calculated based on the lithium plating side reaction calculation branch. The data on the increase in the volume of dead lithium is superimposed and corrected with the data on the increase in thickness. The sequence of changes in the thickness increase variable with the cycle period after superposition and correction is then updated to the physical degradation trajectory feature.

9. The method for predicting the battery life of electric vehicles in coal mines based on big data according to claim 4, characterized in that, The calculation of the dynamic adjustment coefficient at the current moment by inputting the main fluctuation component into a preset mapping function further includes: performing differential operation on the main fluctuation component at adjacent time steps to obtain stress sudden impulse signal data; The stress-induced impulse signal data is input into a preset exponential decay response function to calculate the impulse response decay coefficient, which decays exponentially over time. The impulse response attenuation coefficient is multiplied by the current amplitude of the main fluctuation component to obtain the corrected fluctuation component. The corrected fluctuation component is then used to replace the main fluctuation component and input into the mapping function to calculate the dynamic adjustment coefficient at the current moment.

10. The method for predicting the battery life of electric vehicles in coal mines based on big data according to claim 5, characterized in that, The transient electrical impact stress factor is cross-referenced with the cumulative humidity in the environmental temperature and humidity data to generate a humid-thermal-electric coupling stress feature vector. The humid-thermal-electric coupling stress feature vector is then incorporated into the multidimensional dynamic stress factor. The method further includes: calculating the information entropy of the energy distribution data of the high-frequency band in the current frequency domain feature spectrum to obtain the spectral disorder index. The surface film repair blocking factor is calculated by performing a nonlinear mapping between the spectral disorder index and the cumulative humidity. The surface film repair blocking factor and the transient electrical shock stress factor are vector-concatenated to generate the hygrothermal-electrical coupling stress feature vector, and the hygrothermal-electrical coupling stress feature vector is incorporated into the multidimensional dynamic stress factor.

Images