Battery pack protection method and device, vehicle and readable storage medium

Abstract

The invention provides a battery pack protection method and device, a vehicle and a readable storage medium, and is applied to the technical field of batteries. The method comprises the steps of determining impact characteristics of impact on a battery pack in a vehicle, determining response characteristics of each structure node in the battery pack under the impact effect, and outputting control parameters of a protection mechanism of the battery pack through a pre-trained parameter prediction model according to the impact characteristics and the response characteristics, and controlling the protection mechanism according to the control parameters. According to the method provided by the invention, dynamic adjustment of the control parameters of the protection mechanism can be realized, the control parameters of the protection mechanism are enabled to be matched with the impact on the battery pack, and the control parameters of the protection mechanism are enabled to be matched with the response of each structure node to the impact, so that the impact resistance of the battery pack under the current impact working condition can be improved; and the safety of the battery pack can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of batteries, and more particularly, to a battery pack protection method, device, vehicle and readable storage medium in the technical field of batteries. BACKGROUND

[0002] With the continuous development of battery technology, electric vehicles (hereinafter collectively referred to as vehicles) are becoming more and more popular. As the power source of vehicles, the safety of the battery pack (also known as the battery pack) is crucial to the safety of the vehicle. Therefore, there is an urgent need for a battery pack protection method to improve the safety of the battery pack. SUMMARY

[0003] The present application provides a battery pack protection method, device, vehicle and readable storage medium, which can improve the safety of the battery pack.

[0004] In a first aspect, a battery pack protection method is provided, the method comprising: determining an impact feature of an impact on a battery pack in a vehicle, and determining a response feature of each structural node in the battery pack under the impact; outputting, according to the impact feature and the response feature, a control parameter of a protection mechanism of the battery pack by a pre-trained parameter prediction model; controlling the protection mechanism according to the control parameter to protect the battery.

[0005] In the present application, during the driving of the vehicle, the impact feature of the impact on the battery pack in the vehicle is determined, and the response feature of each structural node in the battery pack under the impact is determined. According to the impact feature and the response feature, the control parameter of the protection mechanism of the battery pack is output by the pre-trained parameter prediction model, and the protection mechanism is controlled according to the control parameter. The impact feature can represent the impact type and intensity of the impact currently suffered by the battery pack, and the response feature can represent the state of each structural node in the battery pack when the battery pack suffers an impact. When the control parameter is determined according to the impact feature and the response feature and the protection mechanism is controlled, the dynamic adjustment of the control parameter can be realized, so that the control parameter of the protection mechanism matches the impact on the battery pack, and the control parameter of the protection mechanism matches the response of each structural node to the impact. Therefore, the anti-impact performance of the battery pack under the current impact condition can be improved, and the safety of the battery pack can be improved.

[0006] Optionally, the determination of the impact feature of the impact on the battery pack in the vehicle comprises: acquiring an impact pulse signal and a motion state parameter of the battery pack under the impact, and a strain parameter of each structural node under the impact; The impact feature is output by a pre-trained impact identification model based on the impact pulse signal, the motion state parameter, and the strain parameter.

[0007] In the embodiments of the present application, the impact pulse signal can represent the mechanical response characteristics of the battery pack when subjected to impact, the motion state parameter can represent the state change of the battery pack when subjected to impact, and the strain parameter can represent the response of the battery pack to impact. Therefore, the impact pulse signal, the motion state parameter, and the strain parameter can reflect the impact working condition of the battery pack accurately, and thus a more accurate impact feature can be obtained based on the impact pulse signal, the motion state parameter, and the strain parameter.

[0008] Optionally, the control parameter of the protection mechanism of the battery pack is output by a pre-trained parameter prediction model based on the impact feature and the response feature, including: The impact intensity of the battery pack under impact is predicted based on the impact feature. In a case where the impact intensity is not lower than a preset intensity threshold, the control parameter is output by the parameter prediction model based on the impact feature and the response feature.

[0009] In the embodiments of the present application, in a case where the impact intensity is not lower than a preset intensity threshold, the control parameter is determined based on the impact feature and the response feature to control the protection mechanism, and in a case where the impact intensity is lower than the preset intensity threshold, the protection mechanism is not controlled, which can reduce unnecessary control of the protection mechanism and prolong the service life of the protection mechanism.

[0010] Optionally, the impact pulse signal and the motion state parameter of the battery pack under impact and the strain parameter of each structural node under impact are obtained, including: A plurality of continuous accelerations and a plurality of continuous angular velocities of the battery pack in a historical time period before the current time are obtained to obtain the motion state parameter. The impact pulse signal of the battery pack in the historical time period is obtained. A plurality of continuous strain parameters of each structural node in the historical time period are obtained.

[0011] In the embodiments of the present application, the impact feature is output by a pre-trained impact identification model based on a plurality of continuous sensor data of each sensor in a historical time period before the current time. Since the plurality of sensor data of each sensor in the historical time can accurately reflect the dynamic change of the impact suffered by the battery pack, a more accurate impact feature can be obtained based on the plurality of continuous sensor data in the historical time period.

[0012] Optionally, the response feature of each structural node of the battery pack under impact is determined, including: obtain a topology of the structural nodes; output, by a trained response identification model, a heat map based on the topology and the impact feature, the heat map including the response feature of each of the structural nodes.

[0013] In the embodiments of the present application, the response identification model outputs the response feature of each structural node based on the encoding vector, and the impact identification model does not need to process the original sensor data, so that the data processing amount of the response identification model can be reduced. Moreover, the encoding vector can accurately represent the impact feature of the impact suffered by the battery pack, so that the response feature output by the response identification model based on the encoding vector can accurately reflect the response of the structural node to the current impact.

[0014] Optionally, the control parameter of the protection mechanism of the battery pack is output by a pre-trained parameter prediction model according to the impact feature and the response feature, including: obtain a driving state parameter of the vehicle; output, by the parameter prediction model, the control parameter according to the driving state parameter, the impact feature and the heat map.

[0015] In the embodiments of the present application, since the impact feature can reflect the impact suffered by the battery pack, the heat map can reflect the response of each structural node in the battery pack to the impact, and the driving state parameter can reflect the response of the vehicle to the impact, the parameter prediction model can output the control parameter highly matched with the current impact based on the driving state parameter, the impact feature and the heat map, so as to accurately control each protection mechanism and improve the anti-impact performance of the battery pack under the current impact.

[0016] Optionally, the control parameter of the protection mechanism of the battery pack is output by a pre-trained parameter prediction model according to the impact feature and the response feature, including: determine a target time length according to an impact intensity in the impact feature, the impact intensity being negatively correlated with the target time length; output, by the parameter prediction model, a plurality of continuous control parameters of the protection mechanism in a target time period after the current time, a time length of the target time period being the target time length, according to the impact feature and the response feature.

[0017] In the embodiments of the present application, the target time length is determined according to the impact intensity, so that the determination frequency of the control parameter can be reduced when the impact intensity is small, so as to reduce the power consumption.

[0018] In a second aspect, a battery pack protection device is provided, and the device includes: determining a shock feature of a shock suffered by a battery pack in a vehicle, and determining a response feature of each structural node in the battery pack under the action of the shock; and outputting, according to the shock feature and the response feature, a control parameter of a protection mechanism of the battery pack by a pre-trained parameter prediction model; controlling the protection mechanism according to the control parameter to protect the battery.

[0019] In a third aspect, a vehicle is provided, and the vehicle comprises: a memory configured to store executable program code; a processor configured to invoke and run the executable program code from the memory, so that the vehicle executes the method in any possible implementation manner of the first aspect.

[0020] In a fourth aspect, a program product is provided, and the program product comprises: executable program code, when the executable program code is run on a vehicle, the vehicle executes the method in any possible implementation manner of the first aspect.

[0021] In a fifth aspect, a readable storage medium is provided, and the readable storage medium stores executable program code, when the executable program code is run on a vehicle, the vehicle executes the method in any possible implementation manner of the first aspect. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a step flow chart of a battery pack protection method provided by an embodiment of the present application; Figure 2 is a principle schematic diagram of a shock identification model provided by an embodiment of the present application; Figure 3 is a principle schematic diagram of a control parameter acquisition provided by an embodiment of the present application; Figure 4 is a structural schematic diagram of a battery pack protection device provided by an embodiment of the present application; Figure 5 is a structural schematic diagram of a vehicle provided by an embodiment of the present application. DETAILED DESCRIPTION

[0023] The technical solutions in the present application will be described clearly and exhaustively below with reference to the drawings. In the description of the embodiments of the present application, unless otherwise specified, " / " represents the meaning of or, for example, A / B can represent A or B: "and / or" in the text only describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone. In addition, in the description of the embodiments of the present application, "multiple" means two or more than two.

[0024] Hereinafter, the terms "first" and "second" are used only for descriptive purposes and cannot be understood as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more features.

[0025] As a part of the vehicle, a key problem that the battery pack needs to face during the driving of the vehicle is various impacts on the battery pack caused by the movement of the vehicle. Therefore, the safety of the battery pack is related to the impact resistance of the battery pack, and the higher the impact resistance, the higher the safety of the battery pack.

[0026] In order to improve the impact resistance of the battery pack, the usual way is to improve the hardware structure of the battery pack to improve the impact resistance of the battery pack, and thus improve the safety of the battery pack. Since the hardware structure of the battery pack is fixed, the improvement of the hardware structure can only improve the impact resistance of the battery pack under a small number of impact conditions. However, since the driving state of the vehicle is complex and changeable, the impact conditions of the battery pack are also complex and changeable. When the impact resistance of the battery pack is improved by improving the hardware structure, the impact resistance of the battery pack under part of the impact conditions can be improved, and the battery pack is difficult to face the complex and changeable impact conditions, so the safety of the battery pack is still low.

[0027] In order to solve the above technical problems, the embodiments of the present application provide a battery pack protection method, which detects the impact received by the battery pack during the driving of the vehicle, and detects the response of each structural node in the battery pack to the impact, to obtain the impact characteristics of the impact received by the battery pack and the response characteristics of each structural node. Then, the control parameters of the protection mechanism are determined based on the impact characteristics and the response characteristics, and the protection mechanism is controlled according to the control parameters to protect the battery.

[0028] The impact feature can represent the impact type, impact intensity, and damage level that the battery pack currently suffers, and the response feature can represent the change trend of each structural node in the battery pack when the battery pack suffers the impact. When the control parameter is determined according to the impact feature and the response feature and the protection mechanism is controlled, the dynamic adjustment of the control parameter can be realized, the control parameter of the protection mechanism can be matched with the impact suffered by the battery pack, and the control parameter of the protection mechanism can be matched with the response of each structural node to the impact, so that the anti-impact performance of the battery pack under the current impact working condition can be improved, and the safety of the battery pack can be improved.

[0029] Referring to Figure 1 , Figure 1 is a step flowchart of a battery pack protection method provided by the embodiment of the application. The execution subject of the method can be a controller in a vehicle, for example, a controller in the vehicle for controlling the battery pack, and the method can include the following steps: Step 101, determining an impact feature of an impact suffered by a battery pack in a vehicle, and determining a response feature of each structural node in the battery pack under the impact.

[0030] The impact feature refers to the impact type, impact intensity, and damage level that the battery pack suffers, and the impact-related features. The structural node refers to a mechanical connection point and a force-receiving part in the battery pack, and a key position that will deform when the battery pack suffers the impact, for example, a support point of a tray, a bolt connection point, and a position of a limiting piece, but is not limited thereto. The response feature refers to the change of each structural node under the impact, which can be understood as the response of the structural node to the impact, and can include the activity, risk level, and stress of the structural node under the impact.

[0031] Optionally, the step of determining the impact feature of the impact suffered by the battery pack in the vehicle can include: obtaining an impact pulse signal and a motion state parameter of the battery pack under the impact, and a strain parameter of each structural node under the impact; outputting the impact feature through a pre-trained impact recognition model based on the impact pulse signal, the motion state parameter, and the strain parameter.

[0032] The impact pulse signal can represent the mechanical response characteristics of the battery pack when the battery pack suffers the impact, and can reflect the impact force, action time, and dynamic characteristics of the impact suffered by the battery pack. The motion state parameter can include the acceleration and angular velocity of the battery pack when the battery pack suffers the impact, and the strain parameter (also referred to as strain data) represents the deformation degree of the structural node when the battery pack suffers the impact.

[0033] In this embodiment, different types of multiple sensors can be respectively arranged at different positions of the battery pack, and impact pulse signals, motion state parameters and strain parameters can be respectively detected by the different types of multiple sensors. For ease of description, the data collected by the sensors, such as impact pulse signals, motion state parameters and strain parameters, will be collectively referred to as sensor data.

[0034] For example, one type of sensor is a piezoelectric impact sensor. A piezoelectric impact sensor can be arranged at the bottom of the battery pack. The piezoelectric impact sensor can detect the impact received by the battery pack to output an impact pulse signal.

[0035] Another type of sensor is an inertial measurement unit (IMU). Three-axis IMUs can be arranged at the front end, middle and bottom of the battery pack, respectively. The IMUs can detect the acceleration and angular velocity of the battery pack in three reference directions when the battery pack is impacted. The three reference directions are the forward direction of the vehicle (i.e., the X direction), the right / left direction of the vehicle (i.e., the Y direction), and the vertical direction (i.e., the Z direction). The IMUs can detect the acceleration of the battery pack in the X, Y and Z directions, and the angular velocity of the battery pack in the X, Y and Z directions.

[0036] Another type of sensor is a strain gauge. A strain gauge can be arranged at each structural node of the battery pack. For each structural node, the strain gauge arranged at the structural node can detect the deformation degree of the structural node when the battery pack is impacted. The data output by the strain gauge is strain data, which can represent the deformation degree of the structural node when the battery pack is impacted.

[0037] For example, a pre-trained impact recognition model (also referred to as an impact encoding model) can be deployed in the controller. The impact recognition model can identify the impact received by the battery pack based on the sensor data collected by the above-described sensors, obtain and output an encoding vector, which can represent the impact type, impact intensity and possible damage level of the impact received by the battery pack.

[0038] For example, the impact recognition model can be constructed based on an Efficient Convolution Operators (ECO) network, which includes multiple convolution channels arranged in parallel and an ECO network as a core network. Each convolution channel is used to process one sensor data, which can avoid feature aliasing between different sensor data and ensure the independence of each sensor data.

[0039] In constructing the impact recognition model, a normalization operation and an activation function are arranged after each convolution layer in the convolution channel, so that the impact recognition model can quickly converge in the training process, and the impact recognition model can have high non-linear modeling capability. At the same time, a dropout mechanism can be introduced in the middle layer of the impact recognition model to prevent overfitting. In addition, the overall structure of the impact recognition model can be controlled within five convolution layers, and the parameter scale is maintained within one million, so that the impact recognition model can meet the real-time requirement and has low power consumption.

[0040] Referring to Figure 2 , Figure 2 is a principle schematic diagram of an impact recognition model provided by an embodiment of the present application. The impact recognition model includes M convolution channels and an ECO network as a core network, and the M convolution channels correspond one-to-one to the above-mentioned multiple sensor data. As shown in Figure 2 , the IMU in the above example can output the acceleration of the battery pack in the X direction, the Y direction and the Z direction, i.e., the X direction acceleration, the Y direction acceleration and the Z direction acceleration as shown in Figure 2 ; and can also output the angular velocity of the battery pack in the X direction, the Y direction and the Z direction, i.e., the X direction angular velocity, the Y direction angular velocity and the Z direction angular velocity as shown in Figure 2 . The piezoelectric impact sensor can output an impact pulse signal. The battery pack includes N structural nodes, and the strain gauge on each structural node can output a strain data, i.e., strain data 1-strain data N as shown in Figure 2 .

[0041] Each convolution channel corresponds to one sensor data in the above example, is used to receive the corresponding sensor data, and performs convolution processing on the received sensor data to obtain a convolution result. The multiple convolution results output by the multiple convolution channels are input into the ECO network, and the ECO network outputs an encoding vector representing the impact type, the impact strength and the possible damage level based on the multiple convolution results.

[0042] During the operation of the vehicle, the controller can obtain the sensor data detected by each sensor at the current time, input each sensor data into a corresponding convolution channel, and the convolution channel performs convolution on the input sensor data to obtain a convolution result. The multiple convolution results output by the multiple convolution channels are spliced and input into the ECO network, and the ECO network outputs an encoding vector based on the multiple convolution results.

[0043] It should be noted that before the sensor data of the plurality of sensors is input into the impact identification model, each sensor data can be pre-processed, the pre-processing including identification and filtering of outliers, for the impact pulse signal, a first-order difference and a Hampel filter can be used to remove pulse noise and distorted wave bands, and each sensor data can be normalized to unify the input scale, so that the sensor data is adapted to the input specification of the impact identification model.

[0044] In the embodiments of the application, the impact pulse signal can represent the mechanical response characteristics of the battery pack when subjected to impact, the motion state parameter can represent the state change of the battery pack when subjected to impact, and the strain parameter can represent the response of the battery pack to impact. Therefore, the impact pulse signal, the motion state parameter and the strain parameter can reflect the impact working condition of the battery pack, and therefore a more accurate impact feature can be obtained based on the impact pulse signal, the motion state parameter and the strain parameter.

[0045] Optionally, the impact pulse signal and the motion state parameter of the battery pack under impact, and the strain parameter of each structural node under impact are obtained, comprising: obtaining a plurality of continuous accelerations and a plurality of continuous angular velocities of the battery pack in a historical time period before the current time to obtain the motion state parameter; obtaining the impact pulse signal of the battery pack in the historical time period; obtaining a plurality of continuous strain parameters of each structural node in the historical time period.

[0046] The starting time of the historical time period is a historical time before the current time, the ending time of the historical time period is the current time, and the interval between the historical time and the current time is a first preset time length.

[0047] In one embodiment, in the process of determining the impact feature based on the above-mentioned sensor data by the impact identification model, a plurality of continuous sensor data of each sensor in a historical time period before the current time can be collected, and the impact identification model can output an encoding vector representing the impact type, the impact intensity and the damage level based on the plurality of continuous sensor data collected by each sensor.

[0048] Exemplarily, the first preset time length can be 500 ms, and each sensor can be sampled at a preset sampling frequency during the driving of the vehicle to obtain continuous multiple sensor data. In the process of determining the impact feature, the controller can obtain the continuous multiple sensor data collected by each sensor in the last 500 ms (i.e., in the historical time period). For example, for the X-direction acceleration collected by the IMU, the controller can obtain, from the IMU, the continuous multiple X-direction accelerations collected by the IMU in the last 500 ms. In addition, the controller can obtain an impact pulse signal collected by the piezoelectric impact sensor in the last 500 ms, and the time length of the impact pulse signal is 500 ms. In addition, the controller can obtain the continuous multiple strain data collected by each strain gauge in the last 500 ms.

[0049] Then, the controller inputs the continuous multiple sensor data collected by each sensor in the last 500 ms into a corresponding convolution channel of the impact recognition model. Each convolution channel convolves the input continuous multiple sensor data to obtain a convolution result. The multiple convolution results output by the multiple convolution channels are spliced and input into the ECO network, and the ECO network outputs an encoding vector representing the impact type, impact strength and damage level based on the multiple convolution results.

[0050] In addition, for the continuous multiple sensor data collected by each sensor, the multiple sensor data can be sequentially input into the impact recognition model according to the collection time of the multiple sensor data. For example, for the X-direction acceleration, after obtaining the continuous multiple X-direction accelerations, the multiple X-direction accelerations are arranged in the order of collection and are sequentially input into the corresponding convolution channel.

[0051] Optionally, after obtaining the continuous multiple sensor data collected by each sensor in the historical time period, the continuous multiple sensor data collected by each sensor can be filtered to eliminate outliers in the continuous multiple sensor data.

[0052] In the embodiment of the application, the pre-trained impact recognition model outputs the impact feature based on the continuous multiple sensor data of each sensor in the historical time period before the current time. Since the multiple sensor data of each sensor in the historical time period can accurately reflect the dynamic changes of the impact suffered by the battery pack, the impact feature can be obtained based on the continuous multiple sensor data in the historical time period.

[0053] In practical applications, the impact type can be divided based on the waveform characteristics of the impact pulse signal collected by the piezoelectric impact sensor, and can be divided into bottom impact, side impact, frontal impact, bottom compression, and rollover, but is not limited thereto. The impact intensity can be divided according to the peak value of the acceleration of the battery pack when subjected to an impact, the duration and energy density of the impact pulse signal, and can be divided into no impact, slight impact, moderate impact, and severe impact, but is not limited thereto. The damage level can be divided according to the damage degree of the battery pack, and can be divided into 1st, 2nd, 3rd, and 4th levels, and the higher the level, the higher the damage degree.

[0054] Optionally, determining the response characteristics of each structural node in the battery pack under the action of the impact includes: Obtaining the topological structure of each structural node; Based on the topological structure and the impact characteristics, outputting a heat map by the trained response identification model, the heat map including the response characteristics of each structural node.

[0055] The topological structure includes the mechanical connection relationship and mutual influence degree of each structural node in the battery pack. Optionally, each structural node in the battery pack can be represented by graph data with weights, each point (Vertex / Node) in the graph data representing a structural node, the edge (Edge) between two points representing the connection relationship between the two structural nodes, and the weight of the edge representing the distance between the two structural nodes.

[0056] In an embodiment, a pre-trained response identification model can be deployed in the controller, and the response identification model can output the response characteristics of each structural node based on the graph data and the impact characteristics. Exemplarily, the response identification model can be a pre-trained Spiking Neural Network (SNN) model. The neurons in the SNN model can be pulse neurons constructed based on the “leaky integration-firing” mechanism, simulating the excitation and reset behavior of biological neurons after receiving input electrical signals. Each structural node in the battery pack is mapped to a pulse neuron, receives input stimuli encoded in vectors, and accumulates voltage according to the node attributes and edge weights. When a node reaches the firing threshold, it triggers a pulse emission, indicating a high structural response probability at that location under that impact scenario. The entire SNN model propagates electrical signals in the time dimension, achieving time sequence mapping from the encoded vector to the structural response.

[0057] In addition, the SNN model can be modeled in a time step expansion manner, and can include five to ten time steps to simulate the dynamic response in the impact propagation process. Each neuron unit internally integrates a leakage rate, a voltage threshold and a firing reset mechanism, and the system models complex structural dynamic characteristics such as local delay, signal propagation order and response sensitivity through the time dimension.

[0058] Referring to Figure 3 , Figure 3 is a schematic diagram of a control parameter acquisition principle provided by an embodiment of the present application. As Figure 3 indicated, the graph data can be compressed and stored in the controller in a sparse storage manner to improve the loading efficiency and meet the demand for low-power operation of the controller. After the controller obtains an encoding vector output by the impact recognition model based on the sensor data of each sensor, the controller can acquire the pre-stored graph data, and then encode the graph data and the encoding vector respectively, encode each element in the encoding vector into a pulse sequence, and encode the attribute value of each point and the weight of each edge in the graph data into a pulse sequence. Each pulse sequence obtained by encoding is input into a corresponding neuron in the input layer of the response recognition model (i.e., the SNN model) to input the graph data and the encoding vector into the response recognition model. Correspondingly, after the graph data and the encoding vector are input into the response recognition model, the response recognition model can output a heat map including the response characteristics of each structural node based on the graph data and the encoding vector.

[0059] Exemplarily, the heat map can be a three-dimensional heat map, and each region in the heat map represents one or more structural nodes in the battery pack. Each pixel in the heat map has three channels, and the channel values of the three channels respectively represent the activity, risk level and force of the structural node when the structural node is subjected to an impact. The region with a high excitation frequency in the heat map (i.e., the region with high activity, risk level and force) corresponds to a stress concentration area or an initial deformation position in the battery pack.

[0060] It can be understood that for each structural node, the higher the activity of the structural node when the structural node is subjected to an impact, the greater the reflection of the structural node to the impact, the higher the risk level and force of the structural node, the greater the deformation occurring at the structural node, and the higher the probability of failure of the battery pack at the structural node.

[0061] According to the embodiments of the present application, the response recognition model outputs the response characteristics of each structural node based on the encoding vector, and the impact recognition model does not need to process the original sensor data, which can reduce the data processing amount of the response recognition model. In addition, the encoding vector can accurately represent the impact characteristics of the impact received by the battery pack, so the response characteristics output by the response recognition model based on the encoding vector can accurately reflect the response of the structural node to the current impact.

[0062] In actual applications, the response recognition model can also output a heat map based on the graph data and / or the sensor data collected by the plurality of sensors. The manner in which the response recognition model outputs the heat map based on the sensor data is similar to the manner in which the impact recognition model outputs the encoded vector based on the sensor data, and thus will not be described herein.

[0063] At step 102, the control parameter of the protection mechanism of the battery pack is output by the pre-trained parameter prediction model based on the impact feature and the response feature.

[0064] At step 103, the protection mechanism is controlled according to the control parameter to protect the battery.

[0065] In this embodiment, after obtaining the impact feature map and the response feature, the controller can determine the control parameter of each protection mechanism of the battery pack based on the impact feature and the response feature. For example, the protection mechanism of the battery pack can include an energy absorption mechanism (such as an energy absorption shell), a support mechanism (such as an active restraint mechanism), and a damper and other active protection mechanisms, and the controller can adjust the deformation amount of the energy absorption mechanism, the displacement amount of the support mechanism, the damping coefficient of the damper, and other control parameters.

[0066] When the deformation amount of the energy absorption mechanism changes, the protection effect of the energy absorption mechanism on the battery pack changes, and the impact resistance performance of the battery pack changes. Similarly, when the displacement amount of the support mechanism changes, the protection effect of the support mechanism on the battery pack changes, and the impact resistance performance of the battery pack changes; when the damping coefficient of the damper changes, the protection effect of the damper on the battery pack changes, and the impact resistance performance of the battery pack changes. It should be understood that the protection mechanism of the battery pack can include but is not limited to the above examples.

[0067] Optionally, outputting the control parameter of the protection mechanism of the battery pack by the pre-trained parameter prediction model based on the impact feature and the response feature can include: Obtaining a driving state parameter of the vehicle; Outputting the control parameter by the parameter prediction model based on the driving state parameter, the impact feature, and the heat map.

[0068] The driving state parameter can include a driving speed of the vehicle, a slope of a road currently traveled by the vehicle, a pose of the vehicle, and other parameters related to the driving state of the vehicle.

[0069] For example, Figure 3As shown, a pre-trained parameter prediction model can be deployed in the controller, which can be a policy network (Actor) in a Deep Deterministic Policy Gradient (DDPG) model. The policy network can be composed of three layers of feedforward neural networks, with an activation function for nonlinear modeling, and the number of hidden layer neurons can be between 128 and 256. The input layer of the policy network is used to uniformly normalize and encode the input multi-source data, and the output layer generates control parameters through a policy mapping function.

[0070] Optionally, to prevent oscillation of the control parameters, an action smoothing processing layer can be added at the end of the policy network. The action smoothing processor can smooth the control parameters output by the policy network to improve the stability and execution accuracy of the control mechanism.

[0071] For example, during vehicle driving, the controller can obtain continuous driving speeds collected by the speed sensor in the vehicle within a historical time period, continuous poses collected by the pose sensor within the historical time period, and continuous slopes collected by the slope sensor within the historical time period, while obtaining continuous sensor data collected by each of the above sensors within the historical time period, to obtain driving state parameters composed of the multiple driving speeds, the multiple poses, and the multiple slopes.

[0072] After the controller inputs the sensor data within the historical time period into the impact recognition model to obtain the encoding vector output by the impact recognition model, and inputs the encoding vector and the graph data into the response recognition model to obtain the heat map output by the response recognition model, the controller can simultaneously input the heat map and the encoding vector, and the driving state parameters within the historical time period into the parameter prediction model (i.e., the policy network) to obtain the control parameters of each protection mechanism output by the parameter prediction model.

[0073] For example, the control parameters output by the parameter prediction model can include a target deformation amount of the energy absorption mechanism, a target displacement amount of the support mechanism, and a target damping coefficient of the damper. After obtaining the target deformation amount, the target displacement amount, and the target damping coefficient, the controller can adjust the deformation amount of the energy absorption mechanism to the target deformation amount, adjust the displacement amount of the support mechanism to the target displacement amount, and adjust the damping coefficient of the damper to the target damping coefficient, to achieve dynamic control of each protection mechanism, so that the control parameters of the protection mechanism match the current impact of the battery pack, to improve the anti-impact performance of the battery pack.

[0074] Optionally, in the process of outputting the control parameter, the parameter prediction model can output a plurality of continuous control parameters of each protection mechanism in a target time period after the current time. The starting time of the target time period is the current time, the ending time of the target time period is a time after the current time, and the interval between the starting time and the ending time of the target time period is a second preset time length.

[0075] For example, for the energy absorption mechanism, when the parameter prediction model outputs the target deformation amount, it can output a plurality of continuous target deformation amounts in the target time period, which are arranged in time sequence. In the process of controlling the energy absorption mechanism, the controller can control the deformation amount of the energy absorption mechanism to reach each target deformation amount in the order of the arrangement of the plurality of target deformation amounts.

[0076] Among the plurality of continuous control parameters, the interval between two adjacent control parameters is a preset interval time length. When the protection mechanism is controlled according to the plurality of continuous control parameters, the controller adjusts the parameter of the protection mechanism once every preset interval time length. For example, after obtaining a plurality of continuous target deformation amounts of the energy absorption mechanism, a target deformation amount is selected from the plurality of target deformation amounts every preset interval time length in the order of the arrangement of the plurality of target deformation amounts, and the deformation amount of the energy absorption mechanism is adjusted to the selected target deformation amount.

[0077] It can be understood that during the driving of the vehicle, at each time (i.e., the current time), the controller obtains sensor data in a historical time period before the current time, determines an encoding vector including an impact feature based on the sensor data, and determines a heat map including a response feature based on the encoding vector. Then, the controller predicts a plurality of control parameters of each protection mechanism in a target time period after the current time based on the heat map and the encoding vector, and controls the protection mechanism according to the plurality of control parameters in the target time period, so that the control parameter of the protection mechanism matches the impact suffered by the battery pack in the target time period.

[0078] After the protection mechanism is controlled according to the plurality of control parameters in the target time, the sensor data in the historical time period before the current time is obtained again, and the above steps are repeated, and the above steps are executed in a loop. Dynamic adjustment of the control parameter can be achieved.

[0079] In the embodiments of the present application, since the impact feature can reflect the impact suffered by the battery pack, the heat map can reflect the response of each structural node in the battery pack to the impact, and the driving state parameter can reflect the response of the vehicle to the impact, the parameter prediction model can output a control parameter that matches the current impact height based on the driving state parameter, the impact feature and the heat map, and then accurately control each protection mechanism, thereby improving the anti-impact performance of the battery pack under the current impact.

[0080] In another implementation, after obtaining the heat map output by the response recognition model, the controller can also only input the impact feature and the heat map to the parameter prediction model, so that the parameter prediction model outputs the control parameter of the energy absorption mechanism based on the impact feature and the heat map.

[0081] Optionally, the control parameter of the protection mechanism of the battery pack is output by the pre-trained parameter prediction model according to the impact feature and the response feature, including: The target duration is determined according to the impact intensity in the impact feature, and the impact intensity is negatively correlated with the target duration. The control parameter of the protection mechanism in the target time period after the current time is output by the parameter prediction model according to the impact feature and the response feature, and the duration of the target time period is the target duration.

[0082] For example, different target durations can be set for each impact intensity in the above example. As in the above example, the impact intensity is included in the impact feature, and after obtaining the encoding vector output by the impact recognition model, the controller can extract the impact intensity from the encoding vector, and determine the corresponding target duration according to the impact intensity. The impact intensity is negatively correlated with the target duration, and the smaller the impact intensity, the longer the target duration.

[0083] Then, in the process of outputting the control parameter of the protection mechanism by the parameter prediction model, the number of control parameters output by the parameter prediction model can be controlled for each protection mechanism. The longer the target duration, the more control parameters. In combination with the above example, the interval duration between two adjacent control parameters is the preset interval duration, so the target duration is positively correlated with the number of control parameters, and the longer the target duration, the more the number of control parameters.

[0084] It can be understood that when the impact intensity is small, the impact on the battery pack is small, and the damage caused is small. The smaller the impact intensity, the longer the corresponding target, and the number of control parameters output by the parameter prediction model can be reduced when the impact intensity is small to reduce power consumption.

[0085] In the embodiments of the present application, the target duration is determined according to the impact intensity, and the number of control parameters is reduced when the impact intensity is small to reduce power consumption.

[0086] Optionally, the control parameter of the protection mechanism of the battery pack is output by the pre-trained parameter prediction model according to the impact feature and the response feature, including: The impact intensity of the battery pack under the impact is predicted based on the impact feature. In the case where the impact intensity is not lower than the preset intensity threshold, the control parameter is output by the parameter prediction model according to the impact feature and the response feature.

[0087] In one implementation, after obtaining the output encoding vector of the impact identification model, the controller can extract the impact intensity from the encoding vector. After extracting the impact intensity, the controller compares the extracted impact intensity with a preset intensity threshold (the preset intensity threshold is, for example, a minor impact). When the extracted impact intensity is a minor impact, a moderate impact, or a severe impact (i.e., not lower than the preset intensity threshold), the controller inputs the impact characteristics and response characteristics into the parameter prediction model to obtain the control parameters output by the parameter prediction model, and controls the protection mechanism according to the control parameters.

[0088] Conversely, when the extracted impact intensity is no impact (i.e., below the preset intensity threshold), the controller does not determine the control parameters and does not control the protection mechanism. Afterward, the controller acquires sensor data again, determines the new impact characteristics based on the sensor data, and executes subsequent actions.

[0089] In the implementation of this application, when the impact intensity is not lower than the preset intensity threshold, the control parameters are determined based on the impact characteristics and response characteristics to control the protection mechanism. When the impact intensity is lower than the preset intensity threshold, the protection mechanism is not controlled, which can reduce unnecessary control of the protection mechanism and extend its service life.

[0090] In this embodiment, during vehicle operation, the impact characteristics of the impact on the battery pack are determined, as well as the response characteristics of each structural node in the battery pack under impact. Based on the impact and response characteristics, a pre-trained parameter prediction model outputs control parameters for the battery pack's protection mechanism. The protection mechanism is then controlled according to these control parameters. The impact characteristics characterize the type and intensity of the impact currently experienced by the battery pack, while the response characteristics characterize the state of each structural node in the battery pack when it is impacted. When control parameters are determined based on the impact and response characteristics and the protection mechanism is controlled, dynamic adjustment of the control parameters can be achieved. This ensures that the control parameters of the protection mechanism match the impact experienced by the battery pack and the response of each structural node to the impact, thereby improving the battery pack's impact resistance under the current impact conditions and ultimately enhancing its safety.

[0091] in, Figure 3 The impact recognition model, response recognition model, and parameter prediction model shown can be pre-trained and deployed in the vehicle's controller, such as an in-vehicle artificial neural network (AI) acceleration platform or an edge controller.

[0092] It should be noted that the core network of the impact recognition model can be an ECO network or other network with similar functions. When the core network of the impact recognition model is an ECO network, the impact recognition model not only has lower delay and higher robustness, but also has sparse modeling capability and edge computing efficiency, which adapts to the deployment requirements of automotive-grade chips.

[0093] Similarly, the response recognition model can be an SNN model or other model with similar functions. When the response recognition model is an SNN model, the response recognition model has stronger modeling capability in processing burst, nonlinear, and sparse time sequence signals, can simulate the dynamic response characteristics of the biological nervous system to external stimuli, is suitable for modeling the dynamic response behavior of the battery pack under varying stress conditions, can improve the accuracy of the response features of the final output, and thus improve the accuracy of the control parameters to improve the impact resistance of the battery pack.

[0094] The parameter prediction model can be constructed based on the policy network in DDPG or other network with similar functions. When constructed by the policy network in DDPG, the parameter prediction model has the ability to process high-dimensional continuous action space and can adapt to different intensity and direction of impact situations, providing differentiated and real-time control parameters for the battery pack.

[0095] For ease of understanding, the training process of the impact recognition model, the response recognition model, and the parameter prediction model shown in FIG. 1 is introduced as follows. Figure 3

[0096] For the impact recognition model, in the sample construction process, first, an experimental vehicle is selected, various sensors in the above examples are installed in the experimental vehicle, all sensors are time-synchronized by a unified time source, and the time error between the sensors is ensured to be less than 1 ms. The experimental vehicle is tested by leap landing, deep pit bottom impact, high-speed impact platform experiment, etc. to simulate various impact conditions that the battery pack may suffer.

[0097] During the test, the impact pulse signal, the motion state parameter of the battery pack, and the strain parameter of each structure node are continuously collected by the above sensors. For each test, the sensor data collected by each sensor is standardized and preprocessed. First, the outliers are identified and filtered, and the first-order difference and Hanning filter are used to remove pulse noise and distorted wave segments. Then, a fixed time length sliding window is used to cut the collected continuous sensor data, and the impact process is cut from the continuous sensor data. Each segment includes continuous multiple sensor data when the battery pack is impacted. At the same time, all sensor data are normalized to unify the input scale, so that the sensor data adapt to the input specification of the impact recognition model. ​

[0098] The sensor data of each segment obtained by cutting during the covering impact process constitutes a training sample, and the training sample includes various sensor data of the battery pack in the experimental vehicle under the impact working condition. The sensors in the training sample are processed sensor data, and the input specification of the impact identification model is adapted.

[0099] For each training sample, a sample label can be added to the training sample, and the sample label can include impact type, impact strength, and damage level. Among them, for the training sample generated for each test, the scanning result of the battery pack can be obtained by X-ray scanning, and the inspection result of the battery pack can be obtained by manual inspection. According to the scanning result and the inspection result, the impact type, the impact strength and the damage level suffered by the battery pack are determined, and the corresponding sample label is added to the training sample.

[0100] In the above manner, a sample set composed of a large number of training samples can be obtained, and each training sample in the sample set is provided with a corresponding sample label. The training samples in the sample set are divided into a training set, a validation set and a test set.

[0101] After the training samples are constructed, the impact identification model can be trained by the training samples. In the training process, the parameters of the impact identification model can be iteratively updated by using an adaptive optimization algorithm, and a momentum-based optimizer can be selected to speed up the convergence speed. The initial learning rate can be set to one thousandth, and the learning rate can be gradually reduced by using a cosine annealing strategy to prevent oscillatory convergence. At the same time, a learning rate warm restart mechanism is introduced, and the learning rate is reset every fixed number of rounds to effectively jump out of the local optimal solution. In addition, to prevent gradient explosion or disappearance, a gradient clipping threshold is set to limit the network update amplitude, ensuring the stability of the training process.

[0102] In addition, the hyperparameter tuning stage can use a combination of grid search and Bayesian optimization to systematically explore the key hyperparameters of the impact identification model. This includes convolution kernel size, network depth, Dropout ratio, batch size, and activation function type. During the parameter tuning process, the classification accuracy on the validation set is used as the main evaluation indicator, and the inference time and model size are also considered. The performance of each set of parameter configurations is quantitatively compared, and the model structure that balances accuracy and efficiency is finally selected as the deployment baseline.

[0103] After training is completed, the test set is used to verify the performance of the impact identification model, and the impact identification accuracy, class balance, misjudgment rate and inference delay of the impact identification model are evaluated to ensure that the impact identification accuracy of the impact identification model is higher than 95%, the misjudgment rate is lower than 3%, and the inference delay is within 10 milliseconds.

[0104] After the training of the impact recognition model is completed, the impact recognition model can be compressed, for example, weight quantization, structure pruning and tensor compression can be performed. In the weight quantization process, the floating point weight is compressed into a fixed point format to adapt to the hardware in the vehicle; in the structure pruning, the redundant channels and connections are removed through sparsity analysis, and the inference delay is reduced while keeping the function of the model unchanged. Through the tensor compression technology, the storage space occupied by the impact recognition model is reduced, and the loading speed is improved. The model size of the impact recognition model is controlled within five megabytes, and the average inference time is controlled within ten milliseconds, meeting the real-time requirements of the vehicle controller. After the compression of the impact recognition model is completed, the impact recognition model is deployed in the controller.

[0105] For the response recognition model (i.e. SNN model), in the sample construction process, the impact recognition model can output an encoding vector based on each training sample. For each training sample, a sample label is set for the training sample, which can be obtained through physical experiments or simulation. By spatial mapping of strain, stress and deformation of each structure point in the battery pack, a label corresponding to the training is constructed. Among them, for the SNN model, all sample labels are mapped to discrete levels of pulse firing frequency, thereby converting into a pulse frequency supervision signal to train the SNN model.

[0106] In the training process of the response recognition model, a multi-label binary classification objective function with pulse coding can be used as a loss function. Each structure node in the battery pack corresponds to an output pulse sequence, and its frequency corresponds to the response of the structure node to the impact. The average pulse firing frequency in the time window is used as the supervision signal to construct the pulse error function, and the network parameters of the response recognition model are adjusted through the back propagation algorithm.

[0107] In the training process, to solve the influence of the non-continuity of the pulse signal on the gradient calculation, the surrogate gradient method can be used to approximate the derivative, so that the training process is differentiable and stable. The training target of the response recognition model is to maximize the firing accuracy of the high-risk area while suppressing the false trigger rate of the non-target nodes.

[0108] The Adam-based pulse time unfolding optimizer is used in the optimization algorithm, and the periodic learning rate scheduler is used for dynamic adjustment. The initial learning rate is set to one thousandth, and linear decay is performed every fixed number of rounds to prevent late training shock. In each training iteration, the firing frequency of the key nodes, the total activation time and the spatial distribution stability are recorded as training state monitoring references.

[0109] During the hyperparameter adjustment process, the following dimensions are focused on: first, the firing threshold of neurons, which determines the sensitivity of the node, generally between 0.5 and 1; second, the synaptic delay range, which is configured according to the geometric distance between nodes, usually 0-3 ms; third, the selection of pulse coding mode, which uses pulse frequency coding to obtain better response resolution; fourth, the time step length and time window setting, for example, using 5-10 time steps per training round, and the time window does not exceed 1 second, to balance physical interpretability and computational load.

[0110] After completing the training of the response recognition model, the response recognition model is compressed, and the graph structure storage of the heat map is optimized, and then the trained response recognition model is deployed in the controller.

[0111] For the parameter prediction model, training samples can be constructed based on the encoding vectors output by the impact recognition model during training, the heat map output by the response recognition model during training, and the driving state parameters corresponding to each encoding vector. As described above, when constructing the training samples of the impact recognition model, the driving state parameters of the experimental vehicle can be obtained at the same time, and the driving state parameters, the encoding vectors corresponding to the training samples of the impact recognition model, and the heat maps output based on the encoding vectors are combined into a training sample. Set a sample label for each training sample, and the sample label is the control parameter of the impact of the adaptive battery pack measured by experiment.

[0112] During the training process, the Actor-Critic framework is used, the policy network is responsible for generating control parameters, and the evaluation network (Critic) evaluates the value of the control parameters according to the current state and the control parameters. The parameters of the policy network are updated by alternating optimization, and the policy gradient method is used to improve the output of the value function, and the evaluation policy network uses the time difference method to fit the deviation between the actual reward and the estimated value. The training process is carried out in an offline training pool with an experience replay mechanism, which improves sample diversity by shuffling the time sequence and avoids converging to a local optimum.

[0113] In terms of hyperparameter adjustment, key adjustment items include learning rate, network layer number, neuron number, discount factor, and target network soft update rate. The learning rates of the policy network and the evaluation network are set to 0.001 and 0.005, respectively, and the discount factor is controlled at about 0.95 to balance short-term and long-term benefits; the experience replay pool capacity is set to 100,000, and the batch size is set to 128. In order to improve the stability of the model, a soft target network update mechanism is introduced, which updates the target network with a very small proportion each time to reduce policy fluctuations.

[0114] The training results were validated in a simulation environment to observe the rationality of the control parameters of the policy network under different impact scenarios, and to evaluate the maximum structural stress, response delay, false trigger rate, and energy consumption performance. Under automotive-grade requirements, the parameter output delay of the policy network was controlled within ten milliseconds, the false trigger rate was controlled below five percent, and the protection accuracy was maintained above ninety percent. The trained policy network can be exported to an edge deployment format, fine-tuned and optimized in the onboard controller, and then deployed.

[0115] It should be understood that the above are merely illustrative examples, and the specific training process can be set according to specific needs. This embodiment does not impose any restrictions on this.

[0116] The above text combined Figures 1 to 3 The battery pack protection method provided in the embodiments of this application is described in detail below; the following will be combined with Figure 4 and Figure 5 The apparatus embodiments of this application are described in detail below. It should be understood that the apparatus in the embodiments of this application can perform the various methods described in the foregoing embodiments of this application, that is, the specific working processes of the various products described below can be referred to the corresponding processes in the foregoing method embodiments.

[0117] See Figure 4 , Figure 4 This is a schematic diagram of the structure of a battery pack protection device provided in an embodiment of this application. Figure 4 As shown, the battery pack protection device 400 may include: The determination module 401 is used to determine the impact characteristics of the impact on the battery pack in the vehicle, and to determine the response characteristics of each structural node in the battery pack under the impact; and, based on the impact characteristics and the response characteristics, to output the control parameters of the protection mechanism of the battery pack through a pre-trained parameter prediction model. The control module 402 is used to control the protection mechanism according to the control parameters in order to protect the battery.

[0118] Optionally, the determining module 401 is specifically used to acquire the impact pulse signal and motion state parameters of the battery pack under impact, as well as the strain parameters of each structural node under impact; based on the impact pulse signal, the motion state parameters and the strain parameters, the impact features are output through a pre-trained impact recognition model.

[0119] Optionally, the determining module 401 is specifically used to predict the impact intensity of the battery pack under impact based on the impact characteristics; and when the impact intensity is not lower than a preset intensity threshold, output the control parameters through the parameter prediction model according to the impact characteristics and the response characteristics.

[0120] Optionally, the determination module 401 is specifically configured to obtain a plurality of continuous accelerations and a plurality of continuous angular velocities of the battery pack in a historical time period before the current time, to obtain the motion state parameter; obtain the impact pulse signal of the battery pack in the historical time period; and obtain a plurality of continuous strain parameters of each structural node in the historical time period.

[0121] Optionally, the determination module 401 is specifically configured to obtain a topological structure of each structural node; and output a heat map including the response feature of each structural node based on the topological structure and the impact feature through a trained response identification model.

[0122] Optionally, the determination module 401 is specifically configured to obtain a driving state parameter of the vehicle; and output the control parameter based on the driving state parameter, the impact feature and the heat map through the parameter prediction model.

[0123] Optionally, the determination module 401 is specifically configured to determine a target time length according to an impact intensity in the impact feature, the impact intensity being negatively correlated with the target time length; and output a plurality of continuous control parameters of the protection mechanism in a target time period after the current time based on the impact feature and the response feature through the parameter prediction model, a time length of the target time period being the target time length.

[0124] Referring to Figure 5 , Figure 5 is a structural diagram of a vehicle provided by an embodiment of the present application. As shown in the figure, the vehicle 500, for example, a server, includes a memory 501 and a processor 502, wherein the memory 501 stores executable program code 5011, and the processor 502 is configured to invoke and execute the executable program code 5011 to execute a battery pack protection method. Figure 5

[0125] In addition, an embodiment of the present application also protects a battery pack protection device, which can include a memory and a processor, wherein the memory stores executable program code, and the processor is configured to invoke and execute the executable program code to execute a battery pack protection method provided by an embodiment of the present application.

[0126] The embodiment can divide the device into functional modules according to the above method examples, for example, corresponding to each functional module, or two or more functions can be integrated into one output module, and the integrated module can be realized in the form of hardware. It should be noted that the division of modules in the embodiment is illustrative, and is only a logical function division, and another division mode can be used in actual implementation.

[0127] ​In the case of adopting the respective functional modules corresponding to the respective functions, the apparatus can further include a determining module, a replacing module, a controlling module, and the like. It should be noted that all the related content of the respective steps involved in the above method embodiments can be referred to the function description of the corresponding functional modules, and will not be repeated here.

[0128] It should be understood that the apparatus provided by the embodiment is used to execute the above battery pack protection method, and thus can achieve the same effects as the above implementation method.

[0129] In the case of adopting the integrated unit, the apparatus can include a determining module and a controlling module. When the apparatus is applied to a vehicle, the output module can be used to control and manage the actions of the vehicle. The storage module can be used to support the vehicle to execute the related program codes and the like.

[0130] The output module can be a processor or a vehicle body setting module, which can realize or execute various exemplary logical blocks, modules and circuits shown in combination with the disclosure of the present application. The processor can also be a combination of computing functions, such as including one or more microprocessor combinations, a combination of digital signal processing (digital signal processing, DSP) and microprocessor, and the like, and the storage module can be a memory.

[0131] The embodiment also provides a readable storage medium, which stores executable program codes, and when the executable program codes are run on a vehicle, the vehicle executes the related method steps to realize the battery pack protection method provided by the above embodiment.

[0132] The embodiment also provides a program product, which, when run on a vehicle, causes the vehicle to execute the related steps to realize the battery pack protection method provided by the above embodiment.

[0133] The apparatus, readable storage medium, program product or chip provided by the embodiment are used to execute the corresponding method provided above, and thus the beneficial effects that can be achieved can refer to the beneficial effects of the corresponding method provided above, which will not be repeated here.

[0134] Through the description of the above embodiments, those skilled in the art can understand that, for the convenience and brevity of description, only the above division of the functional modules is taken as an example, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the apparatus is divided into different functional modules to complete all or part of the functions described above.

[0135] In the embodiments of the present disclosure, it should be understood that the disclosed apparatus and method can be implemented in other ways. For example, the apparatus embodiments described above are merely schematic, and the division of the modules or units is merely a logical function division. In actual implementation, another division manner can be adopted, for example, a plurality of units or components can be combined or integrated into another apparatus, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, apparatuses or units, and can be electrical, mechanical or in other forms.

[0136] The above merely describes specific embodiments of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present disclosure, which should be covered by the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Claims

1. A battery pack protection method, characterized by, The method comprises: determining an impact feature of an impact on a battery pack in a vehicle, and determining a response feature of each structural node in the battery pack under the impact; outputting, according to the impact feature and the response feature, a control parameter of a protection mechanism of the battery pack by a pre-trained parameter prediction model; controlling the protection mechanism according to the control parameter to protect the battery.

2. The method of claim 1, wherein, The determination of the impact feature of the impact on the battery pack in the vehicle comprises: obtaining an impact pulse signal and a motion state parameter of the battery pack under the impact, and a strain parameter of each structural node under the impact; outputting the impact feature by a pre-trained impact recognition model based on the impact pulse signal, the motion state parameter and the strain parameter.

3. The method of claim 2, wherein, The outputting of the control parameter of the protection mechanism of the battery pack by the pre-trained parameter prediction model according to the impact feature and the response feature comprises: predicting an impact intensity of the battery pack under the impact based on the impact feature; in the case that the impact intensity is not lower than a preset intensity threshold, outputting the control parameter by the parameter prediction model according to the impact feature and the response feature.

4. The method of claim 2, wherein, The obtaining of the impact pulse signal and the motion state parameter of the battery pack under the impact, and the strain parameter of each structural node under the impact comprises: obtaining a plurality of continuous accelerations and a plurality of continuous angular velocities of the battery pack in a historical time period before the current time to obtain the motion state parameter; obtaining the impact pulse signal of the battery pack in the historical time period; obtaining a plurality of continuous strain parameters of each structural node in the historical time period.

5. The method of claim 2, wherein, The determination of the response feature of each structural node in the battery pack under the impact comprises: obtaining a topological structure of the structural nodes; outputting a heat map by a trained response recognition model based on the topological structure and the impact feature, wherein the heat map comprises the response feature of each structural node.

6. The method of claim 5, wherein, The outputting of the control parameter of the protection mechanism of the battery pack by the pre-trained parameter prediction model according to the impact feature and the response feature comprises: obtaining a driving state parameter of the vehicle; outputting the control parameter by the parameter prediction model according to the driving state parameter, the impact feature and the heat map.

7. The method of any one of claims 1-6, wherein, The outputting of the control parameter of the protection mechanism of the battery pack by the pre-trained parameter prediction model according to the impact feature and the response feature comprises: determining a target time length according to an impact intensity in the impact feature, wherein the impact intensity and the target time length are negatively correlated; outputting a plurality of continuous control parameters of the protection mechanism in a target time period after the current time by the parameter prediction model according to the impact feature and the response feature, wherein a time length of the target time period is the target time length.

8. A battery pack protection apparatus, characterized by, The device comprises: determining a shock feature of a shock suffered by a battery pack in a vehicle, and determining a response feature of each structural node in the battery pack under the action of the shock; and outputting, according to the shock feature and the response feature, a control parameter of a protection mechanism of the battery pack by a pre-trained parameter prediction model; controlling the protection mechanism according to the control parameter to protect the battery.

9. A vehicle characterized by comprising: The vehicle comprises: a memory for storing executable program code; a processor for calling and running the executable program code from the memory, so that the vehicle executes the method according to any one of claims 1 to 7.

10. A readable storage medium, characterized by, The readable storage medium stores executable program code, which, when executed on a vehicle, causes the vehicle to execute the method according to any one of claims 1 to 7.

Images