A soft package battery fault detection method and system based on an adaptive neural network

Abstract

The application discloses a kind of soft package battery fault detection method and system based on adaptive neural network, constructs the battery temperature spatial distribution model considering thermal anomaly and sensor fault, utilizes auxiliary function decomposition method to make non-homogeneous boundary condition homogeneous, establishes ODE dimension reduction system in combination with Galerkin method;Collect battery surface temperature, current and terminal voltage data, design stable parameters through linear matrix inequality, estimate nonlinear heat generation item using neural network observer with adaptive weight matrix, and construct error system;Introduce lumped residual evaluation mechanism, generate dynamic fault threshold based on kernel density estimation technology, complete online detection by comparing residual evaluation value of output estimation error with threshold value in real time.Break through the limitation of traditional lumped parameter model, only a small amount of thermocouple is needed to realize the distributed thermal fault detection of two-dimensional soft package battery, while ensuring accuracy, significantly reduce hardware cost, provide a new technical path for power battery safety management.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of new energy vehicles, and particularly relates to a soft package battery fault detection method and system based on an adaptive neural network. BACKGROUND

[0002] Lithium-ion batteries have become the core energy storage device of electric vehicles due to their high energy density and long cycle life. However, the thermal safety problem during battery operation has always restricted its large-scale commercial application. Thermal effects not only significantly reduce battery performance, but also may trigger a chain reaction of thermal runaway under extreme conditions, causing severe safety accidents. Therefore, developing a high-reliability battery fault monitoring system is of great significance to ensure the safe operation of batteries. The existing battery fault detection methods have the following limitations: the detection method based on the lumped parameter model is difficult to accurately identify distributed thermal abnormalities due to the spatial homogenization assumption, and there is a risk of false positives and missed detection; and the detection method based on the partial differential equation model generally assumes that the model is completely known, and simplifies the heat generation process into a deterministic analytical model. However, the actual battery heat generation characteristics are affected by the nonlinear coupling of current, voltage, temperature and other factors, and traditional methods cannot accurately describe this complex characteristic. At present, there is still a lack of effective diagnostic means for two-dimensional battery thermal processes containing unknown nonlinear terms, especially under the condition of limited sensor configuration, achieving high-precision fault detection is still a technical problem to be solved. SUMMARY

[0003] In view of the above technical problems, the application provides a soft package battery fault detection method and system based on an adaptive neural network.

[0004] The technical solution adopted by the application to solve its technical problems is:

[0005] A soft package battery fault detection method based on an adaptive neural network, the method comprising the following steps:

[0006] S100: establishing a soft package battery temperature spatial distribution model considering distributed thermal abnormalities and sensor faults;

[0007] S200: constructing an auxiliary function to decompose the temperature distribution, converting the non-homogeneous boundary condition to a homogeneous boundary condition, obtaining an ODE reduced dimension infinite system based on the separation theorem and applying the Galerkin method;

[0008] S300: collecting battery surface temperature data, battery current and terminal voltage data, obtaining design parameters meeting the stability requirements of the neural network observer by solving linear matrix inequalities, estimating unknown nonlinear heat generation terms using a neural network, updating the neural network weight matrix according to the weight update rule, and constructing a neural network observer;

[0009] S400: Obtaining an error system based on the infinite dimensional system and the neural network observer;

[0010] S500: Introduce a lumped residual evaluation scheme, construct a residual evaluation function, and determine the fault detection threshold using kernel density estimation technology;

[0011] S600: Obtain an output estimation error according to the error system, substitute the output estimation error into the residual evaluation function to obtain a real-time evaluation value, compare the real-time evaluation value with the fault detection threshold, and implement online fault detection.

[0012] Preferably, S100 includes:

[0013] S110: Get the density of battery materials Specific heat capacity ,exist - Direction and Thermal conductivity in the - direction and ,according to 、 、 and Parameters, combined with the unknown heat generation of the battery and distributed thermal failures The two-dimensional heat conduction equation of the battery is obtained;

[0014] S120: Acquisition of measurement mechanism , combined with the battery temperature distribution and temperature sensor failure Get the measured output expression;

[0015] S130: Get convection coefficient and ambient temperature Parameters, combined with battery temperature distribution Get the boundary conditions and initial conditions;

[0016] S140: Establish a soft-pack battery temperature spatial distribution model by combining the heat conduction equation, measurement output expression, boundary conditions and initial conditions.

[0017] Preferably, S100 specifically includes:

[0018] ;

[0019] in, Indicates the temperature distribution of the soft pack battery, Indicates the density of the battery, represents the specific heat capacity of the battery, and Respectively expressed in - Direction and - direction thermal conductivity, and represents a spatial variable, Indicates time, Indicates unknown heat production, Represents the measurement output, represents the measurement mechanism, and They represent distributed thermal faults and temperature sensor faults occurring in the system, represents the convection coefficient, Indicates the ambient temperature.

[0020] Preferably, S200 specifically includes:

[0021] S210: Decomposing temperature distribution by constructing auxiliary functions , specifically:

[0022] ;

[0023] ;

[0024] in, Auxiliary function for construction, for The part corresponding to the homogeneous boundary conditions after decomposition;

[0025] S220: According to the decomposed A temperature spatial distribution model with homogeneous boundary conditions is established, specifically:

[0026] ;

[0027] in

[0028] ;

[0029] in, represents the differential operator, Indicates equality and Thermal conductivity in the direction, and It is an intermediate variable and has no actual meaning;

[0030] The spatial temperature distribution model with homogeneous boundary conditions includes the following boundary conditions:

[0031] ;

[0032] Among them, the parameters ,enter By terminal voltage and current composition, yes About current , terminal voltage and Nonlinear function of

[0033] S230: Based on the separation theorem and applying the Galerkin method, we can obtain an infinite dimensional system with ODE dimensionality reduction, specifically:

[0034] ;

[0035] The infinite-dimensional system has the following initial conditions:

[0036] ;

[0037] in

[0038] ;

[0039] Among them, the subscript and represent the slow subsystem and the fast subsystem respectively, and denote the time vectors corresponding to the slow subsystem and the fast subsystem respectively, and represent the unknown nonlinear heat generation terms after dimensionality reduction corresponding to the slow subsystem and the fast subsystem, and They represent the distributed thermal faults after dimensionality reduction corresponding to the slow subsystem and the fast subsystem, represents the measured output after conversion to homogeneous boundary conditions, and Represent the measured outputs after dimensionality reduction corresponding to the slow subsystem and the fast subsystem, represents the state variable corresponding to time zero, and Represent the two defined operators corresponding to the slow subsystem and the fast subsystem respectively, represents a spatial operator, represents the spatial basis function corresponding to the time vector, express The transpose of and denote the spatial basis functions of the slow subsystem and the fast subsystem, respectively. 、 、 and represent the coefficient matrices corresponding to the slow subsystem and the fast subsystem respectively, and represent the eigenvalues ​​and spatial basis function components corresponding to the slow subsystem, respectively. and represent the eigenvalues ​​and spatial basis function components corresponding to the fast subsystem, respectively. 、 and Respectively 、 and The transpose of .

[0040] Preferably, S300 includes:

[0041] S310: By solving the linear matrix inequality, the design parameter observer gain matrix that meets the stability requirements of the neural network observer is obtained and the auxiliary matrix ;

[0042] S320: Estimating unknown nonlinear heat generation terms using neural networks , update the neural network weight matrix according to the weight update rule;

[0043] S330: Get the slow subsystem time vector Estimated value of , combined coefficient matrix 、 and and the unknown nonlinear heat generation term , and get the estimated value Expressions for time derivatives and estimates of state quantities , and construct a neural network observer.

[0044] Preferably, S300 specifically includes:

[0045] S310: The design parameters that meet the stability requirements of the neural network observer are obtained by solving the linear matrix inequality problem. The linear matrix inequality problem is specifically:

[0046] ;

[0047] in, and is the defined matrix, is a symmetric positive definite matrix, is the identity matrix, is a block matrix derived from the symmetry, and is a matrix, , , For given parameters, and is an intermediate variable with no real meaning, and

[0048] ;

[0049] in, is a positive scalar, is a symmetric positive definite matrix, is the largest eigenvalue of the matrix, is the supremum of the Gaussian radial basis function vector elements;

[0050] S320: Using a neural network to estimate unknown nonlinear heat generation terms, specifically:

[0051] ;

[0052] in, represents the weight matrix of the neural network, is the vector of Gaussian radial basis functions, whose The specific form of the components is:

[0053] ;

[0054] The input vector is defined as , Middle The first element The center and width are respectively given by and Indicates that the number of nodes in the neural network is expressed as express;

[0055] By combining the output estimation error , proposed the weight update rule, and updated the neural network weight matrix according to the weight update rule, specifically:

[0056] ;

[0057] in, represents a constant coefficient, is a positive definite matrix representing the learning rate, represents the constant adjuvant matrix;

[0058] S330 specifically:

[0059] ;

[0060] in, Represents low-order state variables The estimated value of Indicates virtual output The estimated value of Represents an unknown nonlinear function The estimated value of represents the observation gain matrix.

[0061] Preferably, S400 includes:

[0062] According to the neural network observer, infinite dimensional system and weight update law, the simultaneous equations are simplified to obtain The estimated error The expression for the time derivative is, The estimated error The expression of , and the estimated error of the ideal constant neural network weight matrix The expression of the derivative with respect to time, and then the error system, is obtained as follows:

[0063] ;

[0064] in, is a low-order state variable The estimation error of is the estimation error of the ideal constant neural network weight matrix, For unknown functions The estimation error.

[0065] Preferably, S500 specifically includes:

[0066] ;

[0067] in, represents the length of the evaluation time window, represents the residual signal, represents the residual evaluation function, Indicates the fault detection threshold.

[0068] Preferably, S600 specifically includes:

[0069] .

[0070] A soft-pack battery fault detection system based on an adaptive neural network includes a soft-pack battery temperature spatial distribution model establishment module, an infinite-dimensional system determination module, a neural network observer construction module, an error system determination module, a fault detection threshold determination module, and an online fault detection module;

[0071] A module for establishing a soft-pack battery temperature spatial distribution model, which is used to establish a soft-pack battery temperature spatial distribution model that takes into account distributed thermal anomalies and sensor failures;

[0072] Infinite-dimensional system determination module, used to construct auxiliary functions to decompose temperature distribution, transform non-homogeneous boundary conditions into homogeneous boundary conditions, and obtain infinite-dimensional systems with ODE dimensionality reduction based on the separation theorem and the application of the Galerkin method;

[0073] The neural network observer construction module is used to collect battery surface temperature data, battery current and terminal voltage data. By solving linear matrix inequalities, the design parameters that meet the stability requirements of the neural network observer are obtained. The neural network is used to estimate the unknown nonlinear heat generation terms. The neural network weight matrix is ​​updated according to the weight update rule to construct the neural network observer.

[0074] An error system determination module is used to obtain the error system based on the infinite dimensional system and the neural network observer;

[0075] The fault detection threshold determination module is used to introduce the lumped residual evaluation scheme, construct the residual evaluation function, and determine the fault detection threshold through the kernel density estimation technology;

[0076] The online fault detection module is used to obtain the output estimation error according to the error system, substitute the output estimation error into the residual evaluation function to obtain a real-time evaluation value, compare the real-time evaluation value with the fault detection threshold, and realize online fault detection.

[0077] The above-mentioned soft-pack battery fault detection method and system based on an adaptive neural network achieves efficient detection by integrating distributed temperature modeling and neural network observation technology. First, a battery temperature spatial distribution model that takes into account thermal anomalies and sensor failures is constructed. The inhomogeneous boundary conditions are homogenized using the auxiliary function decomposition method, and an ODE dimensionality reduction system is established in combination with the Galerkin method. Battery surface temperature, current, and terminal voltage data are collected, and stability parameters are designed through linear matrix inequalities. A neural network observer with an adaptive weight matrix is ​​used to estimate the nonlinear heat generation term, and an error system is constructed. An innovative lumped residual evaluation mechanism is introduced, and a dynamic fault threshold is generated based on kernel density estimation technology. Online detection is completed by comparing the residual evaluation value of the output estimation error with the threshold in real time. This method breaks through the limitations of traditional lumped parameter models and can achieve distributed thermal fault detection of two-dimensional soft-pack batteries with only a small number of thermocouples. While ensuring accuracy, it significantly reduces hardware costs and provides a new technical path for power battery safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 This is a flow chart of a method for detecting soft-pack lithium-ion battery faults based on an adaptive neural network in one embodiment of the present invention;

[0079] Figure 2 The estimated error distribution at a certain moment before a fault occurs provided by an embodiment of the present invention;

[0080] Figure 3 The estimated error distribution at a certain moment after a fault occurs provided by an embodiment of the present invention;

[0081] Figure 4This is a fault detection result provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0082] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below with reference to the accompanying drawings.

[0083] In one embodiment, Figure 1 As shown, a soft pack battery fault detection method based on an adaptive neural network comprises the following steps:

[0084] S100: Establish a spatial temperature distribution model for soft-pack batteries that takes into account distributed thermal anomalies and sensor failures;

[0085] S200: Construct auxiliary functions to decompose the temperature distribution, transform the non-homogeneous boundary conditions into homogeneous boundary conditions, and obtain an infinite-dimensional system of ODE dimensionality reduction based on the separation theorem and the application of the Galerkin method;

[0086] S300: Collect battery surface temperature data, battery current, and terminal voltage data, solve linear matrix inequalities to obtain design parameters that meet the stability requirements of the neural network observer, use the neural network to estimate the unknown nonlinear heat generation term, update the neural network weight matrix according to the weight update rule, and construct the neural network observer;

[0087] S400: Obtaining an error system based on the infinite dimensional system and the neural network observer;

[0088] S500: Introduce a lumped residual evaluation scheme, construct a residual evaluation function, and determine the fault detection threshold using kernel density estimation technology;

[0089] S600: Obtain an output estimation error according to the error system, substitute the output estimation error into the residual evaluation function to obtain a real-time evaluation value, compare the real-time evaluation value with the fault detection threshold, and implement online fault detection.

[0090] Specifically, the infinite dimensional system of ODE dimensionality reduction obtained in S200 is for constructing a neural network observer in step S300; the temperature spatial distribution model takes into account the existence of thermal anomalies and sensor failures, and the design parameters that meet the stability requirements of the neural network observer are obtained by solving linear matrix inequalities. The performance of the adaptive neural network observer is verified based on experimental results and the parameters are adjusted. The neural network weight matrix is ​​updated according to the weight update law. The purpose of constructing the neural network observer is to estimate the unknown nonlinear part. The error system is obtained by combining the infinite dimensional system, and a residual evaluation function is constructed. The fault detection threshold is determined by the kernel density estimation technology, and the output estimation error is substituted into the residual evaluation function to obtain a real-time evaluation value. The real-time evaluation value is compared with the preset threshold. When the evaluation value is greater than the threshold, it is judged that a battery fault occurs at this time, and online fault detection is realized; further, when the evaluation value is less than or equal to the threshold, it indicates that there is no fault in the battery at this time.

[0091] In one embodiment, S100 includes:

[0092] S110: Get the density of battery materials Specific heat capacity ,exist - Direction and Thermal conductivity in the - direction and ,according to 、 、 and Parameters, combined with the unknown heat generation of the battery and distributed thermal failures The two-dimensional heat conduction equation of the battery is obtained;

[0093] S120: Acquisition of measurement mechanism , combined with the battery temperature distribution and temperature sensor failure Get the measured output expression;

[0094] S130: Get convection coefficient and ambient temperature Parameters, combined with battery temperature distribution Get the boundary conditions and initial conditions;

[0095] S140: Establish a soft-pack battery temperature spatial distribution model by combining the heat conduction equation, measurement output expression, boundary conditions and initial conditions.

[0096] In one embodiment, S100 specifically includes:

[0097] ;

[0098] in, Indicates the temperature distribution of the soft pack battery, Indicates the density of the battery, represents the specific heat capacity of the battery, and Respectively expressed in - Direction and - direction thermal conductivity, and represents a spatial variable, Indicates time, Indicates unknown heat production, Represents the measurement output, represents the measurement mechanism, and They represent distributed thermal faults and temperature sensor faults occurring in the system, represents the convection coefficient, Indicates the ambient temperature.

[0099] In one embodiment, S200 specifically includes:

[0100] S210: Decomposing temperature distribution by constructing auxiliary functions , specifically:

[0101] ;

[0102] ;

[0103] in, Auxiliary function for construction, for The part corresponding to the homogeneous boundary conditions after decomposition;

[0104] S220: According to the decomposed A temperature spatial distribution model with homogeneous boundary conditions is established, specifically:

[0105] ;

[0106] in

[0107] ;

[0108] in, represents the differential operator, Indicates equality and Thermal conductivity in the direction, and It is an intermediate variable and has no actual meaning;

[0109] The spatial temperature distribution model with homogeneous boundary conditions includes the following boundary conditions:

[0110] ;

[0111] Among them, the parameters ,enter By terminal voltage and current composition, yes About current , terminal voltage and Nonlinear function of

[0112] S230: Based on the separation theorem and applying the Galerkin method, we can obtain an infinite dimensional system with ODE dimensionality reduction, specifically:

[0113] ;

[0114] The infinite-dimensional system has the following initial conditions:

[0115] ;

[0116] in

[0117] ;

[0118] Among them, the subscript and represent the slow subsystem and the fast subsystem respectively, and denote the time vectors corresponding to the slow subsystem and the fast subsystem respectively, and represent the unknown nonlinear heat generation terms after dimensionality reduction corresponding to the slow subsystem and the fast subsystem, and They represent the distributed thermal faults after dimensionality reduction corresponding to the slow subsystem and the fast subsystem, represents the measured output after conversion to homogeneous boundary conditions, and Represent the measured outputs after dimensionality reduction corresponding to the slow subsystem and the fast subsystem, represents the state variable corresponding to time zero, and Represent the two defined operators corresponding to the slow subsystem and the fast subsystem respectively, represents a spatial operator, represents the spatial basis function corresponding to the time vector, express The transpose of and denote the spatial basis functions of the slow subsystem and the fast subsystem, respectively. 、 、 and represent the coefficient matrices corresponding to the slow subsystem and the fast subsystem respectively, and represent the eigenvalues ​​and spatial basis function components corresponding to the slow subsystem, respectively. and represent the eigenvalues ​​and spatial basis function components corresponding to the fast subsystem, respectively. 、 and Respectively 、 and The transpose of .

[0119] In one embodiment, S300 includes:

[0120] S310: By solving the linear matrix inequality, the design parameter observer gain matrix that meets the stability requirements of the neural network observer is obtained and the auxiliary matrix ;

[0121] S320: Estimating unknown nonlinear heat generation terms using neural networks , update the neural network weight matrix according to the weight update rule;

[0122] S330: Get the slow subsystem time vector Estimated value of , combined coefficient matrix 、 and and the unknown nonlinear heat generation term , and get the estimated value Expressions for time derivatives and estimates of state quantities , and construct a neural network observer.

[0123] In one embodiment, S300 specifically includes:

[0124] S310: The design parameters that meet the stability requirements of the neural network observer are obtained by solving the linear matrix inequality problem. The linear matrix inequality problem is specifically:

[0125] ;

[0126] in, and is the defined matrix, is a symmetric positive definite matrix, is the identity matrix, is a block matrix derived from the symmetry, and is a matrix, , , For given parameters, and is an intermediate variable with no real meaning, and

[0127] ;

[0128] in, is a positive scalar, is a symmetric positive definite matrix, is the largest eigenvalue of the matrix, is the supremum of the Gaussian radial basis function vector elements;

[0129] S320: Using a neural network to estimate unknown nonlinear heat generation terms, specifically:

[0130] ;

[0131] in, represents the weight matrix of the neural network, is the vector of Gaussian radial basis functions, whose The specific form of the components is:

[0132] ;

[0133] The input vector is defined as , Middle The first element The center and width are respectively given by and Indicates that the number of nodes in the neural network is expressed as express;

[0134] By combining the output estimation error , proposed the weight update rule, and updated the neural network weight matrix according to the weight update rule, specifically:

[0135] ;

[0136] in, represents a constant coefficient, is a positive definite matrix representing the learning rate, represents the constant adjuvant matrix;

[0137] S330 specifically:

[0138] ;

[0139] wherein, denotes an estimate of the low-order state variable , denotes an estimate of the virtual output , denotes an estimate of the unknown nonlinear function , denotes an estimate of the observation gain matrix.

[0140] In particular, the output estimation error is defined as .

[0141] In one embodiment, S400 comprises:

[0142] According to the neural network observer, the infinite-dimensional system and the weight update law, the simultaneous equations are solved, and the expression of the estimation error of , the derivative of the estimation error of , the estimation error of the derivative of the estimation error of the ideal constant neural network weight matrix, the error system is obtained, in particular:

[0143] ;

[0144] wherein, is the estimation error of the low-order state variable , is the estimation error of the ideal constant neural network weight matrix, is the estimation error of the unknown function ;

[0145] In particular, the state estimation error is defined as , the neural network weight estimation error is defined as , and the unknown function estimation error is defined as: . Wherein, the ideal constant matrix satisfies the following relationship:

[0146] ,

[0147] Wherein, the neural network approximation error is bounded under the peak norm, satisfying .

[0148] In one embodiment, S500 is specifically:

[0149] ;

[0150] wherein, denotes the length of the evaluation time window, represents a residual signal, represents a residual evaluation function, represents a fault detection threshold.

[0151] In one embodiment, S600 is specifically:

[0152] .

[0153] In one detailed embodiment, the present application builds an experimental platform including a battery test cabinet, an incubator and a battery management system to estimate unknown nonlinear heat generation terms. The experimental object is a LiFePO4 / graphite soft pack lithium ion battery with a length of 0.24 m, a width of 0.18 m and a thickness of 7.83*10 -3 m. Figure 2 is the estimated error distribution at a certain moment before the fault occurs, and a typical sensor fault type is investigated: the temperature sensor deviation fault of Nos. 10-15 starts to inject from . Figure 3 The estimated error distribution at a certain moment after the fault occurs is shown. Figure 4 The corresponding fault detection result is shown, which can accurately detect the temperature sensor fault. By comparing Figure 2 and Figure 3 , it can be found that the estimated error distribution changes obviously before and after the fault occurs. Figure 4 The value of the residual evaluation function changes obviously after the fault occurs, and the changed value exceeds the detection threshold, so the method can accurately detect the fault in this embodiment.

[0154] The present invention proposes a method for fault detection of soft-pack lithium-ion batteries based on an adaptive neural network, which has achieved a significant breakthrough in the technical path. Compared with traditional fault detection schemes, this method constructs a two-dimensional temperature spatial distribution model, combines the coupling mechanism of distributed thermal anomalies and sensor faults, and avoids the use of approximate lumped parameter models. It only needs to deploy a limited number of temperature sensors to complete high-precision detection. Its innovation is reflected in three aspects: First, a neural network observer with an adaptive weight update mechanism is used to dynamically estimate the unknown nonlinear part of the battery thermal model, effectively solving the problem of modeling uncertainty under complex working conditions; second, by integrating the Galerkin dimensionality reduction method and linear matrix inequality parameter design, distributed state reconstruction based on local temperature data is realized, overcoming the traditional method's reliance on full-state measurement; third, the innovative introduction of an infinite-dimensional system analysis framework in the observer design stage completely avoids the observation spillover effect caused by dimensionality reduction in traditional lumped parameter modeling, and significantly improves the accuracy of model analysis. This method achieves online diagnosis of multiple faults such as thermal runaway and sensor failure through lumped residual evaluation and kernel density estimation dynamic threshold technology, demonstrating higher engineering practical value in the field of power battery safety management and providing an innovative solution for reducing hardware deployment costs and improving real-time detection.

[0155] In one embodiment, a soft-pack battery fault detection system based on an adaptive neural network is also provided, comprising a soft-pack battery temperature spatial distribution model establishment module, an infinite-dimensional system determination module, a neural network observer construction module, an error system determination module, a fault detection threshold determination module, and an online fault detection module;

[0156] A module for establishing a soft-pack battery temperature spatial distribution model, which is used to establish a soft-pack battery temperature spatial distribution model that takes into account distributed thermal anomalies and sensor failures;

[0157] Infinite-dimensional system determination module, used to construct auxiliary functions to decompose temperature distribution, transform non-homogeneous boundary conditions into homogeneous boundary conditions, and obtain infinite-dimensional systems with ODE dimensionality reduction based on the separation theorem and the application of the Galerkin method;

[0158] The neural network observer construction module is used to collect battery surface temperature data, battery current and terminal voltage data. By solving linear matrix inequalities, the design parameters that meet the stability requirements of the neural network observer are obtained. The neural network is used to estimate the unknown nonlinear heat generation terms. The neural network weight matrix is ​​updated according to the weight update rule to construct the neural network observer.

[0159] An error system determination module is used to obtain the error system based on the infinite dimensional system and the neural network observer;

[0160] The fault detection threshold determination module is used to introduce the lumped residual evaluation scheme, construct the residual evaluation function, and determine the fault detection threshold through the kernel density estimation technology;

[0161] The online fault detection module is used to obtain the output estimation error according to the error system, substitute the output estimation error into the residual evaluation function to obtain a real-time evaluation value, compare the real-time evaluation value with the fault detection threshold, and realize online fault detection.

[0162] Regarding the specific definition of a soft-pack battery fault detection system based on an adaptive neural network, please refer to the definition of a soft-pack battery fault detection method based on an adaptive neural network above, which will not be repeated here. The various modules in the above-mentioned soft-pack battery fault detection system based on an adaptive neural network can be implemented in whole or in part through software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0163] The above is a detailed introduction to the soft-pack battery fault detection method and system based on an adaptive neural network provided by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the core idea of ​​the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A soft pack battery fault detection method based on adaptive neural network, characterized in that: The method comprises the following steps: S100: Establish a spatial temperature distribution model for soft-pack batteries that takes into account distributed thermal anomalies and sensor failures; S200: Construct auxiliary functions to decompose the temperature distribution, transform the non-homogeneous boundary conditions into homogeneous boundary conditions, and obtain an infinite-dimensional system of ODE dimensionality reduction based on the separation theorem and the application of the Galerkin method; S300: Collect battery surface temperature data, battery current, and terminal voltage data, solve linear matrix inequalities to obtain design parameters that meet the stability requirements of the neural network observer, use the neural network to estimate the unknown nonlinear heat generation term, update the neural network weight matrix according to the weight update rule, and construct the neural network observer; S400: Obtaining an error system based on the infinite dimensional system and the neural network observer; S500: Introduce a lumped residual evaluation scheme, construct a residual evaluation function, and determine the fault detection threshold using kernel density estimation technology; S600: Obtain an output estimation error according to the error system, substitute the output estimation error into the residual evaluation function to obtain a real-time evaluation value, compare the real-time evaluation value with the fault detection threshold, and implement online fault detection.

2. The method according to claim 1, characterized in that S100 includes: S110: Obtaining the density of battery materials Specific heat capacity ,exist - Direction and Thermal conductivity in the - direction and ,according to 、 、 and Parameters, combined with the unknown heat generation of the battery and distributed thermal failures The two-dimensional heat conduction equation of the battery is obtained; S120: Acquisition of measurement mechanism , combined with the battery temperature distribution and temperature sensor failure Get the measured output expression; S130: Get convection coefficient and ambient temperature Parameters, combined with battery temperature distribution Get the boundary conditions and initial conditions; S140: Establish a soft-pack battery temperature spatial distribution model by combining the heat conduction equation, measurement output expression, boundary conditions and initial conditions.

3. The method according to claim 2, characterized in that S100 is specifically: ； in, Indicates the temperature distribution of the soft pack battery, Indicates the density of the battery, represents the specific heat capacity of the battery, and Respectively expressed in - Direction and - direction thermal conductivity, and represents a spatial variable, Indicates time, Indicates unknown heat production, Represents the measurement output, represents the measurement mechanism, and They represent distributed thermal faults and temperature sensor faults occurring in the system, represents the convection coefficient, Indicates the ambient temperature.

4. The method according to claim 3, characterized in that S200 is specifically: S210: Decomposing temperature distribution by constructing auxiliary functions , specifically: ； ； in, Auxiliary function for construction, for The part corresponding to the homogeneous boundary conditions after decomposition; S220: According to the decomposed A temperature spatial distribution model with homogeneous boundary conditions is established, specifically: ； in: ； in, represents the differential operator, Indicates equality and Thermal conductivity in the direction, and It is an intermediate variable and has no actual meaning; The spatial temperature distribution model with homogeneous boundary conditions includes the following boundary conditions: ； Among them, the parameters ,enter By terminal voltage and current composition, yes About current , terminal voltage and Nonlinear function of S230: Based on the separation theorem and applying the Galerkin method, we can obtain an infinite dimensional system with ODE dimensionality reduction, specifically: ； The infinite-dimensional system has the following initial conditions: ； in: ； Among them, the subscript and represent the slow subsystem and the fast subsystem respectively, and denote the time vectors corresponding to the slow subsystem and the fast subsystem respectively, and represent the unknown nonlinear heat generation terms after dimensionality reduction corresponding to the slow subsystem and the fast subsystem, and They represent the distributed thermal faults after dimensionality reduction corresponding to the slow subsystem and the fast subsystem, represents the measured output after conversion to homogeneous boundary conditions, and Represent the measured outputs after dimensionality reduction corresponding to the slow subsystem and the fast subsystem, represents the state variable corresponding to time zero, and Represent the two defined operators corresponding to the slow subsystem and the fast subsystem respectively, represents a spatial operator, represents the spatial basis function corresponding to the time vector, express The transpose of and denote the spatial basis functions of the slow subsystem and the fast subsystem, respectively. 、 、 and represent the coefficient matrices corresponding to the slow subsystem and the fast subsystem respectively, and represent the eigenvalues ​​and spatial basis function components corresponding to the slow subsystem, respectively. and represent the eigenvalues ​​and spatial basis function components corresponding to the fast subsystem, respectively. 、 and Respectively 、 and The transpose of .

5. The method according to claim 4, characterized in that S300 includes: S310: By solving the linear matrix inequality, the design parameter observer gain matrix that meets the stability requirements of the neural network observer is obtained and the auxiliary matrix ; S320: Estimating unknown nonlinear heat generation terms using neural networks , update the neural network weight matrix according to the weight update rule; S330: Get the slow subsystem time vector Estimated value of , combined coefficient matrix 、 and and the unknown nonlinear heat generation term , and get the estimated value Expressions for time derivatives and estimates of state quantities , and construct a neural network observer.

6. The method according to claim 5, characterized in that S300 is specifically: S310: The design parameters that meet the stability requirements of the neural network observer are obtained by solving the linear matrix inequality problem. The linear matrix inequality problem is specifically: ； in, and is the defined matrix, is a symmetric positive definite matrix, is the identity matrix, is a block matrix derived from the symmetry, and is a matrix, , , For given parameters, and is an intermediate variable with no real meaning, and ; in, is a positive scalar, is a symmetric positive definite matrix, is the largest eigenvalue of the matrix, is the supremum of the Gaussian radial basis function vector elements; S320: Using a neural network to estimate unknown nonlinear heat generation terms, specifically: ； in, represents the weight matrix of the neural network, is the vector of Gaussian radial basis functions, whose The specific form of the components is: ； The input vector is defined as , Middle The first element The center and width are respectively given by and Indicates that the number of nodes in the neural network is expressed as express; By combining the output estimation error , proposed the weight update rule, and updated the neural network weight matrix according to the weight update rule, specifically: ； in, represents a constant coefficient, is a positive definite matrix representing the learning rate, represents the constant adjuvant matrix; S330 specifically: ； in, Represents low-order state variables The estimated value of Indicates virtual output The estimated value of Represents an unknown nonlinear function The estimated value of represents the observation gain matrix.

7. The method according to claim 6, characterized in that S400 includes: According to the neural network observer, infinite dimensional system and weight update law, the simultaneous equations are simplified to obtain The estimated error The expression for the time derivative is, The estimated error The expression of , and the estimated error of the ideal constant neural network weight matrix The expression of the derivative with respect to time, and then the error system, is obtained as follows: ； in, is a low-order state variable The estimation error of is the estimation error of the ideal constant neural network weight matrix, For unknown functions The estimation error.

8. The method according to claim 7, characterized in that S500 is specifically: ； in, represents the length of the evaluation time window, represents the residual signal, represents the residual evaluation function, Indicates the fault detection threshold.

9. The method according to claim 8, characterized in that S600 specifically: 。 10. A soft pack battery fault detection system based on adaptive neural network, characterized in that: It includes a soft-pack battery temperature spatial distribution model building module, an infinite-dimensional system determination module, a neural network observer construction module, an error system determination module, a fault detection threshold determination module, and an online fault detection module; A module for establishing a soft-pack battery temperature spatial distribution model, which is used to establish a soft-pack battery temperature spatial distribution model that takes into account distributed thermal anomalies and sensor failures; Infinite-dimensional system determination module, used to construct auxiliary functions to decompose temperature distribution, transform non-homogeneous boundary conditions into homogeneous boundary conditions, and obtain infinite-dimensional systems with ODE dimensionality reduction based on the separation theorem and the application of the Galerkin method; The neural network observer construction module is used to collect battery surface temperature data, battery current and terminal voltage data, obtain design parameters that meet the stability requirements of the neural network observer by solving linear matrix inequalities, use the neural network to estimate the unknown nonlinear heat generation terms, and update the neural network weight matrix according to the weight update rule to build the neural network observer; An error system determination module is used to obtain the error system based on the infinite dimensional system and the neural network observer; The fault detection threshold determination module is used to introduce the lumped residual evaluation scheme, construct the residual evaluation function, and determine the fault detection threshold through the kernel density estimation technology; The online fault detection module is used to obtain the output estimation error according to the error system, substitute the output estimation error into the residual evaluation function to obtain a real-time evaluation value, compare the real-time evaluation value with the fault detection threshold, and realize online fault detection.

Images