Power battery pack multi-fault diagnosis method and system based on adversarial transfer learning

Abstract

The invention provides a power battery pack multi-fault diagnosis method and system based on adversarial transfer learning, and relates to the technical field of power battery pack multi-fault diagnosis. The method comprises the following steps: establishing a simulation model of a target power battery; according to the fault type of the to-be-detected target power battery, coupling fault characteristics with the simulation model to obtain a fault generation model, and constructing a fault data set based on the fault generation model; and based on the fault data set, a power battery pack multi-fault diagnosis model is constructed, the power battery pack multi-fault diagnosis model processes real-time operation data of the power battery by adopting an adversarial transfer learning method, fault types are identified, and corresponding fault processing operations are executed according to different fault types. Multi-fault diagnosis of the power battery pack can be realized under different fault severity degrees, working conditions and temperatures based on a small amount of experimental data, extra hardware consumption is not needed, and the requirement of accurately diagnosing multiple faults under different working conditions, temperatures and fault severity degrees in practical application is met.

Description

Technical Field

[0001] The present invention belongs to the field of multi-fault diagnosis of power battery packs, and particularly relates to a multi-fault diagnosis method and system for power battery packs based on adversarial transfer learning. Background Technique

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] The fault diagnosis ability of the battery management system (BMS) is a core technical element to ensure the safety of power batteries. If the BMS fails to detect the faults existing in the power battery pack, the battery usage efficiency will become low, and even cascading safety risks may be triggered. For example, sensor faults will cause incorrect estimation of SOC and SOH, and then trigger incorrect control strategies; short-circuit faults will cause poor voltage consistency between battery cells; connection faults will lead to a sharp increase in contact resistance and cause overheating. If the faults are not detected and measures are not taken in time, it may even cause thermal runaway and endanger driving safety. Therefore, diagnosing multiple possible faults in the power battery pack is crucial for battery safety.

[0004] In the existing technical solutions, in order to diagnose multiple faults in the battery pack, the technical solutions for identifying different fault types are divided into two categories, namely, the hardware redundancy method and the signal redundancy method. However, the hardware redundancy method requires additional hardware losses such as sensors; the signal redundancy method identifies the fault type based on the characteristic differences between different faults. Since the types of battery data collected by the BMS are few (only voltage, current, and temperature), the types of diagnosed faults are limited. The emerging data-driven algorithms can improve the ability of signals to represent faults on the basis of signal redundancy, providing ideas for diagnosing more types of faults. However, the current data-driven methods do not consider the distribution shift of data between the training stage and the deployment stage, and the actual fault data is difficult to collect and the data volume is small, which may cause incorrect fault diagnosis results. In the data, factors such as fault severity, working conditions, and temperature will all cause the distribution inconsistency of the data collected by the BMS. Summary of the Invention

[0005] To overcome the above-mentioned deficiencies of the prior art, the present invention provides a multi-fault diagnosis method and system for power battery packs based on adversarial transfer learning, and designs a general neural network model that can accurately diagnose under different fault severities, working conditions, and temperatures based on a small amount of experimental data.

[0006] To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:

[0007] The first aspect of the present invention provides a multi-fault diagnosis method for power battery packs based on adversarial transfer learning;

[0008] A multi-fault diagnosis method for power battery packs based on adversarial transfer learning, including:

[0009] Establish a simulation model of the target power battery;

[0010] According to the fault types of the target power battery to be detected, couple the fault characteristics with the simulation model to obtain a fault generation model, and construct a fault data set based on the fault generation model;

[0011] Based on the fault data set, construct a multi-fault diagnosis model for the power battery pack. The multi-fault diagnosis model for the power battery pack uses the adversarial transfer learning method to process the real-time operation data of the power battery, identify the fault types, and perform corresponding fault handling operations according to different fault types.

[0012] As a further technical solution, the fault types of the target power battery to be detected include short-circuit faults, connection faults, and sensor faults of the power battery pack.

[0013] As a further technical solution, the simulation model of the target power battery is a P2D electrochemical model. By collecting the operation data of the power battery during offline testing and identifying the model parameters, a high-fidelity model of the target power battery can be obtained.

[0014] As a further technical solution, the step of coupling the fault characteristics with the simulation model according to the fault types of the target power battery to be detected to obtain a fault generation model is specifically: adding a fault characteristic formula corresponding to the fault type to the simulation model according to the fault types of the detected target power battery to realize the coupling of the fault and the simulation model, and obtain the fault generation model.

[0015] As a further technical solution, the step of constructing a fault data set based on the fault generation model is specifically: constructing a fault data set according to the fault severity range, different working conditions, and different temperatures of the target power battery.

[0016] As a further technical solution, the multi-fault diagnosis model for the power battery pack is a multi-layer convolutional neural network model with classification ability, including a convolutional layer, a pooling layer, and a fully connected layer.

[0017] As a further technical solution, the multi-fault diagnosis model for the power battery pack uses the adversarial transfer learning method to process the real-time operation data of the power battery and identify the fault types, specifically:

[0018] Use the convolutional layer to extract the sample features of the real-time operation data of the power battery;

[0019] Construct a classifier cross-entropy loss and a domain discriminator cross-entropy loss based on the sample features, train the classifier and the discriminator based on the obtained classifier cross-entropy loss and domain discriminator cross-entropy loss, output fault information through the classifier, and output operating condition and temperature information through the discriminator; the discriminator and the classifier are composed of fully connected layers.

[0020] The second aspect of the present invention provides a multi-fault diagnosis system for power battery packs based on adversarial transfer learning.

[0021] A multi-fault diagnosis system for power battery packs based on adversarial transfer learning, comprising:

[0022] A battery model construction module, which is configured to: establish a simulation model of the target power battery;

[0023] A fault data set construction module, which is configured to: couple the fault characteristics with the simulation model according to the fault types of the target power battery to be detected to obtain a fault generation model, and construct a fault data set based on the fault generation model;

[0024] A fault diagnosis module, which is configured to: based on the fault data set, construct a multi-fault diagnosis model for the power battery pack, and the multi-fault diagnosis model for the power battery pack uses an adversarial transfer learning method to process the real-time operation data of the power battery, identify the fault types, and perform corresponding fault handling operations according to different fault types.

[0025] The third aspect of the present invention provides a computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, the steps in the multi-fault diagnosis method for power battery packs based on adversarial transfer learning as described in the first aspect of the present invention are implemented.

[0026] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored on the memory and executable on the processor, and when the processor executes the program, the steps in the multi-fault diagnosis method for power battery packs based on adversarial transfer learning as described in the first aspect of the present invention are implemented.

[0027] The above one or more technical solutions have the following beneficial effects:

[0028] The present invention obtains a dense fault data set for constructing a multi-fault diagnosis model for power battery packs by establishing a high-fidelity power battery fault model. The model further improves the ability of the convolutional neural network to extract domain-invariant features through the adversarial transfer learning method. Therefore, it can realize multi-fault diagnosis of power battery packs under different fault severities, operating conditions and temperatures based on a small amount of experimental data, without additional hardware consumption. The established multi-fault diagnosis model for power battery packs has excellent generalization ability and meets the requirements of accurately diagnosing multiple faults under different operating conditions, temperatures and fault severities in practical applications.

[0029] Advantages of additional aspects of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings forming a part of this specification are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0031] Figure 1 It is a flowchart of the method for the first embodiment.

[0032] Figure 2 It is a schematic structural diagram of the multi-fault diagnosis model for the first embodiment.

[0033] Figure 3 It is a system structure diagram for the second embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0035] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention.

[0036] In the case of no conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0037] Embodiment 1

[0038] This embodiment discloses a multi-fault diagnosis method for power battery packs based on adversarial transfer learning;

[0039] As Figure 1 shown, the multi-fault diagnosis method for power battery packs based on adversarial transfer learning includes:

[0040] S1: Establish a simulation model of the target power battery;

[0041] S2: According to the fault types of the target power battery to be detected, couple the fault characteristics with the simulation model to obtain a fault generation model, and construct a fault data set based on the fault generation model;

[0042] S3: Based on the fault data set, construct a multi-fault diagnosis model for the power battery pack. The multi-fault diagnosis model for the power battery pack uses the adversarial transfer learning method to process the real-time operation data of the power battery, identify the fault types, and perform corresponding fault handling operations according to different fault types.

[0043] In step S1, the constructed simulation model of the target power battery is a P2D electrochemical model. The P2D model is based on electrochemical principles and can accurately describe the physical and chemical processes inside the battery, including the diffusion of lithium ions in the electrode material, mass transfer in the electrolyte, and electrode reaction kinetics. Compared with simplified models (such as equivalent circuit models), the P2D model can more accurately simulate the dynamic behavior of the battery, especially its performance under different working conditions, temperatures, and fault conditions. The P2D electrochemical model is constructed by measuring, identifying, and other means to obtain the electrochemical parameters of the target power battery.

[0044] The P2D electrochemical model includes five equations: liquid-phase diffusion, solid-phase diffusion, liquid-phase Ohm's law, solid-phase Ohm's law, and kinetic equation.

[0045] Among them, the liquid-phase diffusion equation is:

[0046]

[0047] In the formula, ε e,i is the liquid-phase volume fraction, c e is the liquid-phase lithium-ion concentration, is the effective liquid-phase diffusion coefficient, j r,i is the ionic pore wall flux, F is the Faraday constant, is the charge transfer number; p is the positive electrode of the battery; n is the negative electrode of the battery.

[0048] The solid-phase diffusion equation is:

[0049]

[0050] Among them, c s is the solid-phase lithium-ion concentration, D s,i is the solid-phase diffusion coefficient, and r is the radius of the solid-phase particles.

[0051] Liquid-phase Ohm's law:

[0052]

[0053] is the effective reaction rate constant, Φ e is the liquid-phase electric potential, R is the gas constant, and T is the battery temperature.

[0054] The solid-phase Ohm's law is:

[0055]

[0056] Φ s is the solid-phase electric potential, and is the solid-phase effective ionic conductivity.

[0057] The kinetic equation is:

[0058]

[0059] where k i is the kinetic reaction rate constant, c e is the liquid-phase lithium-ion concentration, c s,max is the maximum lithium-ion concentration in the solid phase, c s,surf is the lithium-ion concentration on the solid-phase surface, and η is the overpotential.

[0060] The battery terminal voltage V can be expressed as:

[0061] V = U oc + η + R0I

[0062] where U oc represents the battery open-circuit voltage, η represents the overpotential, R0 is the ohmic internal resistance, and I is the applied current.

[0063] By collecting the operating data of the power battery during off-line testing and identifying the model parameters, a high-fidelity model of the target power battery can be obtained. Among them, the model parameters can be divided into fixed parameters and non-fixed parameters. Among them, fixed parameters such as the Faraday constant and the gas constant directly adopt fixed values; non-fixed parameters such as the liquid-phase volume fraction, the solid-phase particle radius, the effective liquid-phase diffusion coefficient, the solid-phase diffusion coefficient, the effective reaction rate constant, the solid-phase effective ionic conductivity, the kinetic reaction rate constant, the maximum lithium-ion concentration in the solid phase, and the ohmic internal resistance can obtain the value range of the parameters and determine the initial values by referring to relevant literature and the battery specification sheet. Among them, parameters such as the solid-phase particle radius that need to be measured can directly use the literature values or be measured using measuring equipment in the model, and the remaining parameters need to be identified based on the pulse charge and discharge experiments of the target battery using optimization algorithms such as particle swarm optimization and genetic algorithms to obtain the identified values of the parameters.

[0064] Furthermore, in step S2, the fault types of the power battery to be detected include power battery pack short-circuit faults, connection faults, and sensor faults. Specifically:

[0065] S21: According to the fault type to be detected, add the corresponding fault characteristic formula in the P2D model to realize the coupling of the fault and the P2D model, and obtain the fault generation model.

[0066] Among them, the short-circuit fault can be simulated by connecting a short-circuit resistor in parallel at both ends of the battery model:

[0067]

[0068] I' = I + I SC

[0069] In the formula, I SC is the short - circuit current, I' is the operating current of the battery after the short - circuit fault, V is the voltage; R sc is the short - circuit resistance.

[0070] The connection fault is simulated by increasing the contact resistance between the batteries:

[0071] R’ con = R con + R con,fault

[0072] Among them, R con is the normal contact resistance of the battery, R con,fault is the faulty contact battery, and R’ con is the contact resistance after the fault.

[0073] The sensor fault is simulated by superimposing the corresponding fault information on the measured values of the simulation model:

[0074] V = V + V fault

[0075] Among them, V fault is the sensor fault voltage signal, such as the error increase value, measurement offset value, measurement drift value, etc.

[0076] After adding the fault, the output of the terminal voltage of the battery model is updated to:

[0077] V = U - η - RI' - R'I' + V' con I' + V' fault

[0078] S22: According to the requirements of density, set the simulation step size of the fault severity. For example, the step size for the short - circuit fault range of 5 - 500 ohms is 1 ohm, the step size for the connection fault range of 5 - 50 milliohms is 1 milliohm, and set different application working conditions and temperatures;

[0079] S23: According to the simulation step size and working condition temperature settings in S22, generate multiple groups of battery simulation operation data based on the established fault generation model, and construct a fault data set, where the data sets containing different temperatures, working conditions, and fault severities are divided into different domains.

[0080] Step S3: Based on the fault data set, construct a multi-fault diagnosis model for the power battery pack. The multi-fault diagnosis model for the power battery pack is a multi-layer convolutional neural network model with classification ability, including a convolutional layer, a pooling layer, and a fully connected layer. Divide the fault data set into a training set and a validation set according to a set ratio, such as 80% for the training set and 20% for the validation set, and train the multi-fault diagnosis model for the power battery pack.

[0081] Combined with Figure 2 , the multi-fault diagnosis model for the power battery pack consists of n convolutional layers and m fully connected layers. Among them, a pooling layer is connected in series after every q convolutional layers, so there are n / q pooling layers. m, n, and q are all hyperparameters to be optimized. The output of the last fully connected layer is the extracted sample features. Therefore, the convolutional layer and the pooling layer together constitute the feature extractor;

[0082] The extracted sample features are used on the one hand to construct the cross-entropy loss L of the classifier c , that is, the training of the fault classification ability. On the other hand, they are used to construct the cross-entropy loss L of the domain discriminator d , that is, the domain discrimination ability. The classifier and the domain discriminator are respectively composed of r and s fully connected layers. The output label of the classifier is 0-3, which is used to represent whether there is a fault and the fault type. The output label of the domain discriminator is 0-t, which is used to represent which working condition and temperature the data domain comes from. Here, t is the number of domains divided according to the types of working conditions and temperatures in the training set;

[0083] The feature extractor is used to generate sample features, enabling the classifier to identify the fault type as much as possible and making the domain discriminator unable to distinguish which domain the features come from. That is, the established multi-fault diagnosis model for the power battery pack can identify the fault type represented by the data but cannot distinguish the domain source of the data, thereby improving the ability of the neural network to obtain domain-invariant features and diagnose and classify faults. In addition, the objective loss function in the model training process is calculated as follows:

[0084]

[0085] In the formula, L is the objective loss function, n is the number of training samples, D is the training sample set, x i , c i , d i represent the model input, the classifier output, and the domain discriminator output respectively. θ f , θ c , θ d are the network parameters to be updated in the feature extractor, the classifier, and the domain discriminator respectively. Their update principles are as follows. They are the updated parameters respectively:

[0086]

[0087] Finally, fault information is output through the classifier, and operating condition and temperature information are output through the discriminator; corresponding fault handling operations are performed according to different fault types. For example, when a battery short - circuit fault is detected, a short - circuit fault alarm is issued, the battery pack forced cooling system is started, and immediate repair suggestions are provided. If conditions permit, power can be cut off immediately for inspection, the short - circuit point can be repaired or the short - circuited cell can be replaced; when a connection fault is detected, a connection fault alarm is issued, and suggestions for checking the connecting wires at the fault location are provided, and the connecting wires are strengthened or replaced; when a sensor fault is detected, a sensor fault alarm is issued, and suggestions for the sensor fault location and sensor replacement are provided.

[0088] Embodiment Two

[0089] This embodiment discloses a multi - fault diagnosis system for power battery packs based on adversarial transfer learning;

[0090] As Figure 3 shown, the multi - fault diagnosis system for power battery packs based on adversarial transfer learning includes:

[0091] A battery model construction module, which is configured to: establish a simulation model of the target power battery;

[0092] A fault data set construction module, which is configured to: couple the fault characteristics with the simulation model according to the fault types of the target power battery to be detected to obtain a fault generation model, and construct a fault data set based on the fault generation model;

[0093] A fault diagnosis module, which is configured to: based on the fault data set, construct a fault diagnosis model for the power battery pack. The multi - fault diagnosis model for the power battery pack uses the adversarial transfer learning method to process the real - time operation data of the power battery, identify the fault types, and perform corresponding fault handling operations according to different fault types.

[0094] Embodiment Three

[0095] The purpose of this embodiment is to provide a computer - readable storage medium.

[0096] A computer - readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps in the multi - fault diagnosis method for power battery packs based on adversarial transfer learning as described in Embodiment 1.

[0097] Embodiment Four

[0098] The purpose of this embodiment is to provide an electronic device.

[0099] An electronic device includes a memory, a processor, and a program stored on the memory and executable on the processor. When the processor executes the program, the steps in the multi-fault diagnosis method for a power battery pack based on adversarial transfer learning as described in Embodiment 1 are implemented.

[0100] The steps involved in the devices of Embodiments Two, Three, and Four above correspond to those in Method Embodiment One. For specific implementation manners, reference may be made to the relevant description part of Embodiment One. The term "computer-readable storage medium" should be understood to include a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0101] Those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0102] Although the specific implementation manners of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation to the protection scope of the present invention. Those skilled in the art should understand that, based on the technical solutions of the present invention, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the protection scope of the present invention.

Claims

1. A multi-fault diagnosis method for power battery packs based on adversarial transfer learning, characterized in that: include: Establish a simulation model of the target power battery; According to the fault type of the target power battery to be detected, coupling the fault characteristics with the simulation model to obtain a fault generation model, and constructing a fault data set based on the fault generation model; Based on the fault data set, a multi-fault diagnosis model for a power battery pack is constructed. The multi-fault diagnosis model for a power battery pack uses an adversarial transfer learning method to process the real-time operation data of the power battery, identify the fault type, and perform corresponding fault handling operations according to different fault types.

2. The power battery pack multi-fault diagnosis method based on adversarial transfer learning as claimed in claim 1, characterized in that: The fault types of the target power battery to be detected include power battery pack short circuit fault, connection fault, and sensor fault.

3. The power battery pack multi-fault diagnosis method based on adversarial transfer learning as claimed in claim 1, characterized in that: The simulation model of the target power battery is a P2D electrochemical model, which acquires a high-fidelity model of the target power battery by collecting operating data of an offline test of the power battery and identifying model parameters.

4. The power battery pack multi-fault diagnosis method based on adversarial transfer learning as claimed in claim 1, characterized in that: According to the fault type of the target power battery to be detected, the fault characteristics are coupled with the simulation model to obtain the fault generation model, specifically: according to the fault type of the target power battery to be detected, the fault characteristic formula corresponding to the fault type is added to the simulation model to achieve the coupling of the fault and the simulation model to obtain the fault generation model.

5. The power battery pack multi-fault diagnosis method based on adversarial transfer learning as claimed in claim 1, characterized in that: The constructing of the fault data set based on the fault generation model specifically includes: constructing the fault data set according to the fault severity range, different operating conditions, and different temperatures of the target power battery.

6. The power battery pack multi-fault diagnosis method based on adversarial transfer learning as claimed in claim 1, characterized in that: The power battery pack multi-fault diagnosis model is a multi-layer convolutional neural network model with classification capability, including a convolutional layer, a pooling layer and a fully connected layer.

7. The power battery pack multi-fault diagnosis method based on adversarial transfer learning as claimed in claim 1, characterized in that: The power battery pack multi-fault diagnosis model uses an adversarial transfer learning method to process the real-time operation data of the power battery and identify the fault type, specifically: Use the convolutional layer to extract sample features of the real-time operation data of the power battery; A classifier cross entropy loss and a domain discriminator cross entropy loss are constructed based on the sample features, and the classifier and the discriminator are trained based on the obtained classifier cross entropy loss and the domain discriminator cross entropy loss. The classifier outputs fault information, and the discriminator outputs operating condition and temperature information; the discriminator and the classifier are composed of a fully connected layer.

8. A power battery pack multi-fault diagnosis system based on adversarial transfer learning, characterized by: include: A battery model building module is configured to: establish a simulation model of a target power battery; A fault data set construction module is configured to: according to the fault type of the target power battery to be detected, couple the fault characteristics with the simulation model to obtain a fault generation model, and construct a fault data set based on the fault generation model; A fault diagnosis module is configured to: construct a power battery pack fault diagnosis model based on the fault data set, wherein the power battery pack multi-fault diagnosis model uses an adversarial transfer learning method to process the real-time operation data of the power battery, identify the fault type, and perform corresponding fault handling operations according to different fault types.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the power battery pack multi-fault diagnosis method based on adversarial transfer learning as described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the power battery pack multi-fault diagnosis method based on adversarial transfer learning as described in any one of claims 1 to 7 are implemented.

Images