Device control method and electronic device

Abstract

The application discloses a device control method and an electronic device, and relates to the field of energy storage. The device control method comprises the following steps: acquiring a first parameter of an energy storage unit, wherein the first parameter is an operation parameter capable of reflecting a current working condition of the energy storage unit; determining a target parameter corresponding to the first parameter based on a target mapping relationship; the target mapping relationship represents a mapping relationship between different operation parameters and target parameters; the target mapping relationship satisfies a change rule based on actual operation parameters under different working conditions; and the energy storage unit is configured based on the target parameter.

Description

Technical Field

[0001] This application relates to the field of energy storage, and more particularly to a device control method and electronic device. Background Technology

[0002] With the acceleration of the global energy transition and the booming development of 3C products (such as laptops, mobile phones, tablets, etc.), electric vehicles, smart grids, large-scale energy storage, aerospace and other fields, the intelligence and sophistication of the management system of high-performance chemical power sources such as lithium-ion batteries, as core power sources / energy storage units, has become the key to industrial competition.

[0003] During equipment operation, the energy storage unit provides electrical energy to the equipment. Therefore, by controlling the core parameters of the energy storage unit, the operation of the energy storage unit can be refined to improve the control effect of the equipment.

[0004] The current control method for the core parameters during the operation of energy storage units is to establish a continuous representation of the mathematical function relationship between the core parameter and related parameters, so as to determine the value of the core parameter under the corresponding operating conditions through the mathematical function relationship.

[0005] However, although this method achieves continuous prediction, it is prone to overfitting or Runge's phenomenon in sparse data regions or at the boundary of operating conditions. This causes the prediction curve to exhibit non-physical oscillations, which deviate significantly from the true characteristics of the energy storage unit, resulting in poor robustness and high risks in engineering applications. Summary of the Invention

[0006] The first aspect of this application provides a device control method, comprising:

[0007] Obtain the first parameter of the energy storage unit, wherein the first parameter is an operating parameter that can reflect the current operating condition of the energy storage unit;

[0008] Based on the target mapping relationship, the target parameter corresponding to the first parameter is determined; the target mapping relationship represents the mapping relationship between different operating parameters and target parameters; the target mapping relationship satisfies the variation law of actual operating parameters under different working conditions.

[0009] Configure the energy storage unit based on the target parameters.

[0010] In one possible implementation, determining the target parameter corresponding to the first parameter based on the target mapping relationship includes:

[0011] The first intelligent agent determines the target parameter corresponding to the first parameter based on the target mapping relationship. The first intelligent agent is trained based on training data, which includes the actual operating parameters of the energy storage unit under various operating conditions.

[0012] One possible implementation also includes:

[0013] Obtain a second parameter of the energy storage unit; the second parameter characterizes the type of the energy storage unit.

[0014] Based on the second parameter of the energy storage unit, a target mapping relationship corresponding to the second parameter is determined.

[0015] One possible implementation also includes:

[0016] Determine the operating mode of the equipment;

[0017] Based on the device's operating mode, a target mapping relationship is determined, with different operating modes corresponding to different discharge rates of the energy storage unit.

[0018] In one possible implementation, the process of determining the target mapping relationship includes:

[0019] Obtain a training dataset, which contains training data corresponding to each working condition. The training data for each working condition includes training values ​​of the working condition parameters and target parameters for that working condition.

[0020] Input the working condition parameters of each training working condition in the training dataset into the model to be trained to obtain the first predicted value of the target parameter of the corresponding training working condition.

[0021] Based on the training value and the first predicted value corresponding to each training condition, the model to be trained is adjusted to obtain the target model;

[0022] Using the target model, a target mapping relationship is constructed based on the predicted values ​​corresponding to the target parameters generated for each training condition and the corresponding condition parameters.

[0023] In one possible implementation, based on each training condition containing at least two condition parameters, the step of adjusting the model to be trained according to the training value and the first predicted value corresponding to each training condition to obtain the target model includes:

[0024] One of the at least two operating condition parameters is selected as the first operating condition parameter in turn;

[0025] The non-first working condition parameters among the at least two working condition parameters are sequentially taken as preset values ​​corresponding to each training working condition. The first working condition parameter is used as a variable, and the values ​​of the first working condition parameter are sorted according to preset change rules to determine the change curve of the first working condition parameter and the target parameter under each training working condition. The target parameter in the change curve adopts the first predicted value.

[0026] Based on the change curve, the training values ​​corresponding to the target parameters and the first condition parameters under each training condition are determined sequentially.

[0027] Based on the training values ​​and the change curve, adjust the first predicted value corresponding to the first condition parameter under the corresponding training condition;

[0028] Based on the adjusted first prediction value, the model to be trained is adjusted to obtain the adjusted model to be trained.

[0029] Based on the adjusted model to be trained, the process returns to input the working condition parameters of each training condition in the training dataset into the model to be trained, and obtains the first predicted value of the target parameter for the corresponding training condition, until the termination condition is met, and the target model is obtained.

[0030] In one possible implementation, adjusting the first predicted value corresponding to the first condition parameter under the corresponding training condition based on the training value and the change curve includes:

[0031] Based on the change curves of the first working condition parameters and the target parameters and the training values, analyze whether the first predicted value conforms to the change trend of the change curves;

[0032] If the first predicted value matches the trend of the change curve, retain the first predicted value;

[0033] If the first predicted value does not conform to the trend of the change curve, a target value is determined based on the training value and the first predicted value, and the first predicted value corresponding to the first condition parameter under the corresponding training condition is adjusted based on the target value.

[0034] One possible implementation also includes:

[0035] Obtain a validation dataset, which contains validation data for multiple validation scenarios. The validation data for each validation scenario contains validation values ​​for the scenario parameters and target parameters. The validation scenarios correspond to the training scenarios.

[0036] Input the working condition parameters of each verification working condition in the verification dataset into the adjusted model to be trained to obtain the second predicted value of the target parameter of the corresponding verification working condition and the confidence level of the second predicted value.

[0037] Based on the confidence level corresponding to the second predicted value of each verification condition, at least two target verification conditions are determined. The confidence level corresponding to the second predicted value of the target verification condition is less than the confidence level corresponding to the second predicted value of the non-target verification condition. The target verification condition is any one of the verification conditions.

[0038] Based on the target verification conditions, multiple target training conditions are determined. The interval between the condition parameters of any two adjacent training conditions is a first value, and the interval between the condition parameters of any two adjacent verification conditions is a second value. The first value is less than the second value.

[0039] Based on each target training condition, the energy storage unit is controlled to operate, and the training values ​​of the target parameters under the target training condition are obtained;

[0040] The training dataset is updated based on the at least two target training conditions and their corresponding training values.

[0041] In one possible implementation, a target mapping relationship is constructed based on the predicted values ​​of the target parameters generated by the target model for each training condition and the corresponding condition parameters, including:

[0042] If the equipment resources do not meet the preset resource requirements, the target model generates a target surface based on the predicted values ​​of the target parameters generated for each training condition and the corresponding condition parameters. The preset resource requirements are the resources needed to deploy the target model, and the target surface is generated based on the predicted values ​​of the target parameters and the condition parameters for each training condition. A target table is generated based on the target surface, and the target table contains parameters for several condition points. Each condition point contains condition parameters and a second target parameter. The target table is used as a target mapping relationship.

[0043] If the equipment's resources meet the preset resource requirements, the control target model generates a target surface based on the predicted values ​​of the target parameters generated for each training condition and the corresponding condition parameters, and uses the target surface as the target mapping relationship.

[0044] A second aspect of this application provides an electronic device, comprising:

[0045] Energy storage unit;

[0046] The processor is configured to acquire a first parameter of the energy storage unit, wherein the first parameter is an operating parameter that reflects the current operating condition of the energy storage unit; determine a target parameter corresponding to the first parameter based on a target mapping relationship; wherein the target mapping relationship represents the mapping relationship between different operating parameters and target parameters; wherein the target mapping relationship satisfies the variation law of actual operating parameters under different operating conditions; and configure the energy storage unit based on the target parameter.

[0047] A third aspect of this application provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the device control method described in the first aspect or any implementation thereof.

[0048] A fourth aspect of this application provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the device control method described in the first aspect or any implementation thereof. Attached Figure Description

[0049] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0050] Figure 1 This is a schematic flowchart of a device control method provided in an embodiment of this application;

[0051] Figure 2 This is a flowchart illustrating the process of determining the target mapping relationship provided in an embodiment of this application;

[0052] Figure 3 This is a schematic diagram of a multi-dimensional curved surface provided in an embodiment of this application;

[0053] Figure 4 This is a flowchart illustrating how, based on a first predicted value and a change curve, the training value corresponding to the first condition parameter under the corresponding training condition is adjusted.

[0054] Figure 5 This is a flowchart illustrating the process of generating the training dataset provided in the embodiments of this application;

[0055] Figure 6 This is a schematic diagram of the device control method provided in this application in an application scenario;

[0056] Figure 7 This is a schematic diagram of the device control method provided in this application in another application scenario;

[0057] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0058] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0059] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0060] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0061] Reference Figure 1 , Figure 1 This is a flowchart illustrating a device control method provided in an embodiment of this application, as shown below. Figure 1 As shown in the embodiment of this application, a device control method may include steps 101 to 103, which are described in detail below.

[0062] 101. Obtain the first parameter of the energy storage unit. The first parameter is the operating parameter that can reflect the current operating condition of the energy storage unit.

[0063] The energy storage unit can be an energy storage structure such as a battery in the device. There can be one or more energy storage structures in the electronic device. The scheme in this application can be implemented for each energy storage structure. This application does not limit the specific structure of the energy storage battery.

[0064] The first parameter may include various operating parameters of the energy storage unit when it is in operation, providing electrical energy to the device during the discharge process; the first parameter may also include various operating parameters of the energy storage unit during the charging process.

[0065] The first parameter can include various operating parameters of the energy storage unit under its current operating conditions.

[0066] As an example, the first parameter may include operating parameters such as battery temperature, discharge current, discharge voltage / discharge efficiency, etc.; the first parameter may also include operating parameters such as battery temperature, charging current, charging voltage / charging efficiency, etc.

[0067] 102. Based on the target mapping relationship, determine the target parameter corresponding to the first parameter; the target mapping relationship represents the mapping relationship between different operating parameters and target parameters; the target mapping relationship satisfies the variation law of the actual operating parameters under different working conditions;

[0068] A target mapping relationship is predetermined, which can characterize the actual operating parameter variation pattern of the energy storage unit under various operating conditions. Moreover, the energy storage unit conforms to this variation pattern during operation and meets the operating requirements of the energy storage unit.

[0069] As an example, the first parameter includes the operating current and the battery temperature, which characterize the operating condition of the energy storage unit. Then, the target parameter is the operating voltage. Accordingly, when the battery temperature, operating current and operating voltage of the energy storage unit meet the target mapping relationship, the operating state of the energy storage unit is better than when the target mapping relationship is not met.

[0070] In one possible implementation, the target parameter could be the battery's discharge efficiency or the battery's terminal voltage during constant current discharge. This target parameter is one of the core parameters of the battery's instantaneous operating state, and it can indicate the smoothness of the electrochemical reaction inside the battery and the energy loss.

[0071] The target parameter corresponding to the first parameter can be determined by querying the target mapping relationship.

[0072] In one possible implementation, if the first parameter includes two parameters, then the target mapping relationship may include physical constraints characterizing the changes of the first parameter and the target parameter in actual operation, as well as the possible value ranges of each parameter.

[0073] In one possible implementation, the first parameter includes the current battery temperature, the current operating current, and the current operating voltage. The target parameter is the target operating voltage. The current battery temperature and the current operating current characterize the current operating conditions of the energy storage device. By querying the target mapping relationship, the target operating voltage that best matches the current operating conditions can be obtained.

[0074] If the target operating voltage is different from or not close to the current operating voltage, it indicates that the current battery operating voltage is not compatible with the current operating conditions. In this case, the battery energy loss is large or the internal chemical reaction is not smooth. Therefore, it is necessary to reconfigure the operating parameters of the energy storage unit and execute the subsequent step 103.

[0075] If the target operating voltage is the same as or close to the current operating voltage, it indicates that the current battery operating voltage is compatible with the current operating conditions and no reconfiguration is required.

[0076] 103. Configure energy storage units based on target parameters.

[0077] After determining the target parameter, if the value of the target parameter is different from the corresponding parameter in the first parameter, it indicates that the parameter generated by the current operation of the energy storage unit does not meet the change pattern. Accordingly, the energy storage unit is configured using the target parameter.

[0078] In one possible implementation, the settings of the corresponding parameters in the energy storage unit are adjusted based on the target parameters.

[0079] As an example, the target parameter is the target operating voltage, which is different from the current operating voltage of the battery, and the battery's operating voltage is configured to be the target operating voltage.

[0080] In this embodiment, a target mapping relationship is pre-set. This target mapping relationship represents the mapping relationship between different operating parameters and target parameters, which conforms to the actual physical characteristics of the energy storage unit during operation. Moreover, this target mapping relationship satisfies the changing patterns of the energy storage unit's actual operating parameters under different operating conditions. During the operation of the energy storage unit, a first parameter of the energy storage unit is acquired, which reflects the current operating condition of the energy storage unit. Based on the target mapping relationship, target parameters corresponding to the first parameter are determined. Based on the target parameters, the energy storage unit is configured. The determined target parameters are parameters that satisfy the target mapping relationship with each parameter in the first parameter, and are actual operating parameters that conform to the changing patterns of the first parameter under the corresponding operating conditions. Therefore, configuring the energy storage unit according to these target parameters can better match the actual situation of the energy storage unit under the corresponding operating conditions, resulting in more accurate control.

[0081] In one possible implementation, the target parameters corresponding to the first parameter are determined based on the target mapping relationship, including:

[0082] The first intelligent agent determines the target parameters corresponding to the first parameter based on the target mapping relationship. The first intelligent agent is trained based on training data, which includes the actual operating parameters of the energy storage unit under various working conditions.

[0083] The first intelligent agent is used to carry the target mapping relationship. The first intelligent agent is trained in advance based on the actual operating parameters of the energy storage unit under various working conditions as training data.

[0084] The first intelligent agent adopts a Gaussian process regression model. The kernel function (covariance function) of this model can be multi-level composite kernel functions to accurately capture the characteristics of different scales and types of parameters during the operation of the energy storage unit.

[0085] The composite kernel function is composed of multiple kernel functions combined linearly or multiplicatively. This composite kernel function includes a constant kernel, a main effect kernel, a trend kernel, and a noise kernel, and the main effect kernel has automatic correlation determination enabled.

[0086] The main effect kernel is either an RBF (Radial Basis Function) kernel or a Matérn 5 / 2 kernel. RBF is infinitely differentiable and suitable for modeling very smooth variations; while Matérn 5 / 2 kernel is twice differentiable and can better handle phenomena with certain roughness or abrupt changes (such as the performance inflection point of a battery at low temperatures), which is generally more consistent with the actual characteristics of the battery.

[0087] The ARD (Automatic Relevance Determination) extension introduces ARD functionality to the main effect kernel, forming an RBF (length_scale=[l_I, l_T], ARD=True). This allows the first agent to automatically learn independent length scale parameters (l_I, l_T) for the two input dimensions of current (I) and temperature (T). The size of the length scale intuitively reflects the importance of the variable's influence on the output (the larger the length scale, the smoother the function changes along that dimension, and the smaller the relative influence).

[0088] The trend kernel, which can optionally be superimposed with a linear kernel (Linear Kernel), can be in the form of DotProduct(sigma_0) to explicitly capture the global linear trend of voltage / efficiency with current or temperature.

[0089] Among them, a noise kernel is superimposed with a white noise kernel, which can be in the form of White(noise_level) to explicitly model measurement noise and random fluctuations in the data.

[0090] As an example, the composite function kernel can be represented by the following formula:

[0091] kernel=ConstantKernel()*(RBF(length_scale=[1.0,1.0],ARD=True)+Linear())+ WhiteKernel() (1)

[0092] Among them, ConstantKernel() represents the constant kernel; RBF(length_scale=[1.0, 1.0], ARD=True) represents the main effect RBF kernel with ARD functionality, Linear() represents the trend kernel, and WhiteKernel() represents the noise kernel.

[0093] It should be noted that the above formula is for illustrative purposes only and does not limit the specific implementation of the kernel function in the first intelligent agent.

[0094] Since the kernel function includes the ARD kernel function, the length scale parameters obtained through ARD kernel function analysis can clearly and quantitatively reveal the relative significance of the influence of current and temperature on discharge efficiency. This not only increases the credibility of the model in the first intelligent agent, but also provides a theoretical reference for the optimization of battery materials, structural design and thermal management.

[0095] Because the kernel function in the first intelligent agent adopts a multi-level composite kernel function, it can accurately capture the characteristics of different scales and types of parameters during the operation of the energy storage unit. Correspondingly, the first intelligent agent can carry the target mapping relationship corresponding to the energy storage unit, so as to determine the target parameter corresponding to the first parameter through the first intelligent agent.

[0096] The training data used in training the first intelligent agent can be the operating parameters corresponding to a specific type of energy storage unit. The target mapping relationship carried by the first intelligent agent is then determined only for the target parameters of that specific type of energy storage unit. Alternatively, it can be the operating parameters corresponding to multiple types of energy storage units. The target mapping relationship carried by the intelligent agent can then determine the target parameters for all types of energy storage units. Technicians can select the appropriate training data to train the first intelligent agent based on the actual situation.

[0097] In one possible implementation, the user of the device can input a request to query the operating parameters of the energy storage unit through a first intelligent agent. In response to the request, the intelligent agent requests a first parameter from the device's system. After obtaining the first parameter, it determines the target parameter corresponding to the first parameter based on the target mapping relationship and displays it to the user.

[0098] In this embodiment, a first intelligent agent is set up, which is pre-trained based on training data. This training data consists of parameters from the actual operation of the energy storage unit under various operating conditions. The first intelligent agent is trained based on these actual operating parameters. This first intelligent agent carries the target mapping relationship. Based on this relationship, the first intelligent agent determines the target parameter corresponding to the first parameter. Due to the agent's strong data processing capabilities, the accuracy of determining the corresponding target parameter under the current operating condition using the agent is high.

[0099] One possible implementation also includes:

[0100] Obtain the second parameter of the energy storage unit; the second parameter characterizes the type of the energy storage unit; based on the second parameter of the energy storage unit, determine the target mapping relationship corresponding to the second parameter.

[0101] Different mapping relationships can be set for different types of energy storage units.

[0102] The type of the energy storage unit is characterized by a second parameter, and correspondingly, the second parameter of the energy storage unit is also obtained.

[0103] The type of energy storage unit can be based on its chemical system. For energy storage units with the same chemical system, the same mapping relationship is used. The chemical system of a battery is determined by its electrode materials and electrolyte system; different models of batteries with the same chemical system can use the same mapping relationship.

[0104] The second parameter can be a parameter characterizing the chemical system of the energy storage unit, or it can be a parameter related to the identity of the energy storage unit, such as its model.

[0105] When the second parameter is the model of the energy storage unit, the correspondence between the model and the mapping relationship can be saved in advance. After obtaining the second parameter of the energy storage unit, the target mapping relationship corresponding to the second parameter can be determined by querying the correspondence.

[0106] When the second parameter adopts the chemical system of the energy storage unit, the correspondence between the chemical system and the mapping relationship can be predetermined. After obtaining the second parameter of the energy storage unit, the target mapping relationship corresponding to the second parameter can be determined by querying the correspondence.

[0107] In this embodiment, the second parameter of the energy storage unit characterizes its type. Different mapping relationships are set for different types of energy storage units. These mapping relationships can more accurately reflect the changing patterns of the corresponding type of energy storage unit during actual operation. After obtaining the second parameter of the energy storage unit, the corresponding target mapping relationship is determined based on the second parameter. This allows for the precise determination of the target parameters corresponding to the energy storage unit, which is more closely aligned with the actual situation of the energy storage unit under corresponding operating conditions, resulting in more accurate control.

[0108] One possible implementation also includes:

[0109] Determine the operating mode of the equipment; based on the operating mode of the equipment, determine the target mapping relationship, and different operating modes correspond to different discharge rates of the energy storage unit.

[0110] Correspondingly, different mapping relationships are set for different operating modes. These mapping relationships can more accurately reflect the changing patterns of energy storage units during actual operation under the corresponding operating modes.

[0111] The control focus of the energy storage unit varies depending on the operating mode of the equipment.

[0112] For example, this focus can include performance and lifespan. The discharge rate of the energy storage unit is positively correlated with the performance of the energy storage unit, but negatively correlated with the lifespan of the energy storage unit.

[0113] The operating mode can include a performance mode and a balanced mode. The performance mode focuses on higher performance, while the balanced mode focuses on balancing performance and lifespan.

[0114] In one possible implementation, the operating mode of the device can be automatically determined based on the device's operating status, which is determined by at least one piece of information such as the remaining battery power and the current operating mode.

[0115] As an example, during the charging process, when the remaining power of the energy storage unit is low and it is in a high power consumption mode, a high charging speed performance mode is adopted; when the remaining power of the energy storage unit is high and it is in a low power consumption mode, a balanced mode is adopted.

[0116] In one possible implementation, the device's operating mode can be selected based on user input. The user inputs a selection operation through the device's input device, and in response to this selection operation, controls the device to switch to the corresponding operating mode. For example, the user controls the device to switch from a balanced mode to a performance mode in order to improve the device's performance.

[0117] In this embodiment, different mapping relationships are pre-set for different operating modes of the device. These mapping relationships, set for different operating modes, can more accurately reflect the changing patterns of the energy storage unit during actual operation under the corresponding operating mode. After determining the operating mode of the device, the corresponding target mapping relationship is determined based on the operating mode. This allows for the precise determination of the target parameters of the energy storage unit under the corresponding operating mode, which can better match the actual situation of the energy storage unit under the corresponding operating conditions, resulting in more accurate control.

[0118] Figure 2 This is a flowchart illustrating the process of determining the target mapping relationship provided in the embodiments of this application, which may include steps 201 to 204. These steps are described in detail below.

[0119] 201. Obtain the training dataset. The training dataset contains training data corresponding to each working condition. The training data for each working condition contains the working condition parameters and the training values ​​of the target parameters.

[0120] These operating parameters construct the operating conditions of the energy storage module. The energy storage module operates under each operating condition to obtain the training values ​​of the target parameters.

[0121] Determine the operating range of each operating parameter of the energy storage module participating in the training, set an interval value for the operating range of each operating parameter, select several values ​​in the operating range according to the interval value, and combine the values ​​of each operating parameter to obtain the corresponding operating condition.

[0122] As an example, the operating parameters for a battery setting include temperature and current, with voltage as the target parameter. Accordingly, the current range is determined (which can include the range from near-open circuit to the maximum permissible discharge current), and the temperature range is determined (from the minimum operating temperature to the maximum operating temperature). For example, the current value interval could be 0.2C (C-rate, charge / discharge rate), and the temperature value interval could be 10℃ (degrees Celsius). Training operating points can be implemented using a two-dimensional (current, temperature) operating point model. For each (I, T) operating point, a test is performed to obtain its corresponding voltage, where I represents current and T represents temperature.

[0123] Considering that overfitting or Runge's phenomenon is very likely to occur in sparse data regions or at the boundary of operating conditions (such as high current and low temperature), the predicted curve will exhibit non-physical oscillations, which will deviate significantly from the true characteristics of the battery and have extremely poor robustness.

[0124] Therefore, after obtaining the initial data for each training operating point, it is necessary to perform outlier handling and normalization. Outlier handling can be achieved using statistical methods (such as the 3σ criterion) or based on physical ranges (such as voltage not exceeding open-circuit voltage) to identify and remove obviously abnormal data. Data normalization aims to eliminate the impact of differences in the dimensions and numerical ranges of current and temperature on model training by preprocessing the input features. The preferred normalization process is Z-score standardization, which transforms the original data into a standard distribution with a mean of 0 and a variance of 1 by subtracting the dataset mean and dividing by the standard deviation. This eliminates differences in the dimensions and numerical magnitudes between different features, thereby improving the training stability and prediction accuracy of the model.

[0125] In one possible implementation, training data can be determined separately for different types of energy storage units to obtain corresponding training datasets. This allows for training of target models for different types of energy storage units, thereby obtaining corresponding mapping relationships. The type of energy storage unit can include NMC (Nickel Manganese Cobalt) ternary lithium, LFP (Lithium Iron Phosphate), etc. This application does not limit the type of energy storage unit.

[0126] In the process of generating training values ​​for target parameters based on various training conditions, a test platform is first constructed. This test platform includes an automated test platform consisting of a high and low temperature control chamber, a battery charge and discharge tester, a data acquisition unit, and upper computer control software. The temperature control chamber must have a temperature control accuracy better than ±0.5℃, and the current / voltage measurement accuracy must be better than ±0.1%FS (Full Scale).

[0127] During testing using the test platform, each training condition is treated as a test point. For each (I, T) test point, the battery is allowed to stand at the target temperature to reach thermal equilibrium, and then discharged at a constant current I to the cutoff voltage. The voltage-time curves throughout the discharge process are recorded, and the average value of the steady-state discharge voltage plateau period is taken as the voltage value V at that condition point. To calculate the discharge efficiency η, the input / output energy must be measured simultaneously. This results in a structured dataset, which records the current, temperature, voltage, and discharge efficiency at each test point.

[0128] During the generation of training data, the operating parameters for each training condition are standardized, and a Gaussian process prior is defined so that the mean function value can reach 0 after the operating parameters of each training condition point are standardized. The purpose of using the back-calculation function of the data established by the model is to calibrate the accuracy of the data obtained by the model.

[0129] 202. Input the working condition parameters of each training working condition in the training dataset into the model to be trained to obtain the first predicted value of the target parameter for the corresponding training working condition.

[0130] The model to be trained is an incompletely trained model. This embodiment describes the process of training the model to obtain the target model.

[0131] The model to be trained and the target model are Gaussian regression process models with the first parameter of the energy storage unit as input and the target parameter as output. The kernel function used is a composite kernel function that includes automatic correlation determination function. The kernel function used is as described in the previous embodiment, and will not be repeated in this embodiment.

[0132] After obtaining the training dataset, the working condition parameters of each training working condition point are sequentially input into the model to be trained. After processing, the model outputs the first predicted value of the target parameter corresponding to each training working condition, with one first predicted value corresponding to one training working condition.

[0133] The first predicted value output by the model to be trained and the operating parameters can form a multidimensional prediction surface (GPR prediction surface), which is used to adjust the model to be trained.

[0134] 203. Based on the training values ​​and the first predicted value corresponding to each training condition, adjust the model to be trained to obtain the target model;

[0135] The first predicted value output by the model to be trained has a certain gap with the training value of the corresponding training condition parameters. The data relationship between the condition parameters and the target parameters can be determined by the multi-dimensional prediction surface output by the model to be trained. Then, the model to be trained can be adjusted by using this data relationship.

[0136] As an example, the operating parameters of this training condition include temperature and current, and the target parameter is voltage. The voltage is composed of a first predicted value, the current value used at each training condition point, and the temperature, forming a three-dimensional surface. Then, the voltage corresponding to each training condition is adjusted by using the relationship between voltage and current values ​​and the relationship between voltage and temperature in this three-dimensional surface, and finally, the relationships that conform to the constraints of physical laws are obtained.

[0137] In one possible implementation, based on each training condition containing at least two condition parameters, step 203 includes:

[0138] 2031. Select one of at least two operating condition parameters as the first operating condition parameter in sequence;

[0139] When the training conditions include two or more operating condition parameters, one is selected as the first operating condition parameter in turn, and the subsequent steps are executed until all of the multiple operating condition parameters have been selected as the first operating condition parameter, thus completing this iteration of training.

[0140] For example, if the operating parameters include temperature and current, then firstly, temperature can be selected as the first operating parameter and current as a non-first operating parameter to execute steps 2032-2034; then, current can be selected as the first operating parameter and temperature as a non-first operating parameter to execute steps 2032-2034. Of course, the order in which the two are selected can also be interchanged, and this application does not impose any restrictions.

[0141] 2032. Take the non-first working condition parameter from at least two working condition parameters and take the corresponding preset value under each training working condition in turn. Take the first working condition parameter as a variable and sort the value of the first working condition parameter according to the preset change rule to determine the change curve of the first working condition parameter and the target parameter under each training working condition. The target parameter in the change curve adopts the first predicted value.

[0142] The target model uses the Gaussian Process Regression (GPR) model. Since the GPR model is a purely data-driven model, its predictions may violate known physical constraints in areas where the training data coverage is insufficient. Therefore, this embodiment employs an iterative post-processing algorithm to physically correct the initial GPR prediction surface generated by the model to be trained. This correction process includes steps 2302 to 2034.

[0143] As a Bayesian nonparametric model, GPR's powerful generalization ability enables it to achieve high-precision fitting results even in small sample scenarios with tens to hundreds of data points, greatly reducing battery testing costs and cycles. Therefore, in this embodiment, the number of training conditions can be limited, and a massive number of training conditions are not required to train the target model.

[0144] Each training condition is treated as a point in a multi-dimensional space. The dimensions of this point include the number of condition parameters and the determination of the target parameter. For example, if the condition parameters include temperature and current, and the target parameter is voltage, then the point corresponding to this training condition is in a three-dimensional space, and the dimensions of this training condition include three dimensions: temperature, current, and voltage.

[0145] After determining the first working condition parameter, the remaining working condition parameters are used as non-first working condition parameters. Each non-first working condition parameter is assigned a corresponding preset value according to each training working condition. The preset value is fixed, and the first working condition parameter is used as a variable. The values ​​of the first working condition parameter are sorted according to the preset change rules to obtain a series of points corresponding to the training working conditions. The points are then connected to obtain the change curves of the first working condition parameter and the target parameter under each training working condition.

[0146] The preset change rule can be in ascending order. For example, if the first operating condition parameter is temperature, the temperature is sorted from low to high; if the first operating condition parameter is current, the current is sorted from low to high.

[0147] As an example, the first operating condition parameter is current, the non-first operating condition parameter is temperature, and the target parameter is voltage. Accordingly, after obtaining the corresponding points of each training operating condition in three-dimensional space, the temperature dimension is fixed, and values ​​are taken for each temperature to obtain the change curves of current and voltage.

[0148] For example, if the temperature range includes seven values: -10℃, 0℃, 10℃, 20℃, 30℃, 40℃, and 50℃, then for -10℃, we obtain the current and voltage variation curves; for 0℃, we obtain the current and voltage variation curves; and so on, resulting in seven variation curves. These variation curves can be called temperature slice curves. This curve characterizes the relationship between voltage and current at that temperature and can be represented by V(T_j, I), where V represents voltage, T_j represents the corresponding temperature (where j can be the j-th temperature value), and I represents current.

[0149] As an example, the first operating condition parameter is temperature, the non-first operating condition parameter is current, and the target parameter is voltage. Correspondingly, after obtaining the points corresponding to each training operating condition in three-dimensional space, the current dimension is fixed, and for each current value, the temperature and voltage variation curves are obtained. These variation curves can be called current slice curves. This curve characterizes the relationship between voltage and temperature under that current, and can be represented by V(I_j, T), where V represents voltage, I_j represents the corresponding current I_j (where j represents the j-th current value), and T represents temperature.

[0150] 2033. Based on the change curve, determine the training values ​​corresponding to the target parameters and the first condition parameters under each training condition in turn;

[0151] For each change curve, the training value corresponding to each first operating condition parameter is determined sequentially. This training value is the value of the target parameter during the actual operation of the energy storage unit under the corresponding operating conditions of each training condition. This training value can characterize the actual operation of the energy storage unit. Therefore, the relationship between this training value and the first operating condition parameter can conform to the constraints of physical laws.

[0152] Since the accuracy of the model to be trained is not high enough, its first predicted value may not conform to the constraints of physical laws. Therefore, the training value is used for adjustment to improve the predicted value output by the model to be trained and the first working condition parameter under the corresponding training conditions so that they can better conform to the constraints of physical laws.

[0153] For example, this physical law constraint could be that the voltage should not decrease as the current increases, or that the voltage should decrease as the temperature increases. The setting of this physical law constraint can be determined based on the operating parameters and target parameters, and this application does not limit the specific content of the physical law constraint.

[0154] For example, by traversing each current point from left to right (in the direction of increasing current) according to the curve, the voltage value corresponding to each current point can be obtained; by traversing each temperature point from left to right (in the direction of increasing temperature) according to the curve, the voltage value corresponding to each temperature point can be obtained.

[0155] 2034. Based on the training values ​​and the change curve, adjust the first predicted value corresponding to the first condition parameter under the corresponding training condition;

[0156] The change curve represents the corresponding change of the first predicted value as the first operating condition parameter changes. By combining this change curve with the various training operating conditions, the training value corresponding to each first operating condition parameter can be determined sequentially. Each training value characterizes the changing trend of the target parameter constrained by physical laws under the corresponding training operating condition. The change curve represents the changing trend predicted by the model. Accordingly, using the training value and the change curve, the first predicted value corresponding to the first operating condition parameter is adjusted so that the change curve composed of the adjusted first predicted value can conform to the physical law constraint and match the original prediction. The specific adjustment process will be described in detail in subsequent embodiments and will not be elaborated here.

[0157] After each iteration of the physical law constraint, a slight two-dimensional Gaussian filter is applied to the corrected surface to eliminate the slight discontinuities that may be caused by local correction, making the surface smoother and more natural overall.

[0158] 2035. Based on the adjusted first prediction value, adjust the model to be trained to obtain the adjusted model to be trained;

[0159] Based on the adjusted model to be trained, return to step 202 until the termination condition is met, and obtain the target model.

[0160] Through the aforementioned steps, the first predicted values ​​of each target parameter corresponding to the first condition parameter under each training condition are adjusted to obtain the adjusted change curves. All the adjusted change curves form a new multidimensional surface.

[0161] By adjusting each first predicted value, the parameters in the model to be trained are adjusted to achieve one iteration of training of the model.

[0162] After multiple rounds of iterative training, training stops when the termination condition is met, and the target model is obtained.

[0163] The trained target model can output first predicted values ​​for each input training dataset. Moreover, the first predicted values ​​of the working parameters of each training condition and the target parameters can form a multi-dimensional surface that conforms to the constraints of physical laws.

[0164] One approach is to determine whether the training of the model meets the termination condition by verifying the results after each round of training iterations.

[0165] By setting up a validation dataset, the prediction performance of the adjusted model can be verified to determine whether the termination condition is met.

[0166] The validation dataset can be obtained simultaneously with the training dataset. When obtaining the dataset, it is randomly divided into a training dataset and a validation dataset according to a set ratio to ensure that the validation set can cover the entire operating range to evaluate the generalization ability of the model.

[0167] The model processes the input validation data to obtain a second predicted value for the corresponding working condition. A validation value corresponding to this validation data is set, and this validation value serves as the label for the second predicted value, thus determining the validation index. By using preset evaluation conditions, it is determined whether the validation index meets these conditions. If it does, the termination condition is met; otherwise, the termination condition is not met.

[0168] Pre-set evaluation conditions can include both quantitative indicators and qualitative evaluation.

[0169] The quantitative metrics include accuracy, physical consistency, and uncertainty calibration. Accuracy is calculated on an independent test set using MAE (Mean Absolute Error), RMSE (Root Mean Squared Error), and the coefficient of determination (R²). MAE and RMSE measure the model's error, while R² characterizes the model's ability to explain data fluctuations. Physical consistency measures the proportion of data points on the validation dataset that violate the monotonicity constraint of the aforementioned physical laws; this value should approach 0%. Uncertainty calibration checks whether the predicted 95% confidence interval truly covers approximately 95% of the test data points to assess the accuracy of the uncertainty measure.

[0170] Qualitative evaluation involves comparing the predicted curve with the measured data points using two-dimensional slice plots (fixed I or T) to observe the goodness of fit; and observing the rationality of the overall shape using three-dimensional surface plots.

[0171] If both qualitative and quantitative indicators are met, and the preset evaluation conditions are satisfied, training can end. Otherwise, if the preset evaluation conditions are not met, the next round of iterative training will continue.

[0172] 204. Using the target model, construct a target mapping relationship based on the predicted values ​​corresponding to the target parameters generated for each training condition and the corresponding condition parameters.

[0173] Once the target model is trained, it can construct a target mapping relationship based on the predicted values ​​corresponding to the training conditions and the parameters of multiple working points constructed from the working condition parameters. Furthermore, it can visualize the target mapping relationship in the form of a three-dimensional surface.

[0174] The target mapping relationship can be a multi-dimensional surface obtained through multiple iterations and adjustments. This multi-dimensional surface conforms to the physical laws constraining the actual operation of the energy storage unit and also meets the predictions of the model.

[0175] This target model can be set in the device as the processing model adopted by the first intelligent agent. Through this target model, the corresponding target parameters of the energy storage unit under various operating conditions can be determined to be suitable for the current operating parameters, so as to provide a better utilization effect of the energy storage unit.

[0176] In one possible implementation, a two-dimensional plane is first constructed by combining an operating parameter and a target parameter. After normalizing this two-dimensional plane, a grid of points is generated on it. The target parameter of each grid point can be predicted using the target model, and the mean and variance of the target parameter can be calculated. The predicted result of the target parameter is then denormalized to restore it to its original dimensions. A scientific plotting library is used to create an image representing the target mapping relationship. This image can include surface plots, contour maps / heat maps, and slice maps with confidence intervals. The surface plot is used to visually display the overall topography as the target parameter changes with the operating parameter; the contour map / heat map is used to clearly show the performance distribution of different regions from a top-down perspective; and the slice map is used to display the predicted curve and its uncertainty range under specific operating parameters (such as temperature or current), and is overlaid with measured data.

[0177] Figure 3 This is a schematic diagram of a multi-dimensional surface provided in an embodiment of this application. The surface shown is a three-dimensional surface, with dimensions including temperature, current, and voltage. The red dots in the diagram represent the first predicted values ​​output by the model to be trained. This three-dimensional surface is the final three-dimensional surface obtained. The three-dimensional surface adopts the color temperature correspondence shown on the right side of the diagram. The three-dimensional surface is smooth, continuous, and strictly conforms to the monotonicity of physical laws.

[0178] Subsequently, this target mapping relationship can be applied to predict the values ​​of target parameters under actual working conditions, so as to obtain values ​​that conform to the physical laws of the energy storage unit.

[0179] In one possible implementation, to improve model stability, particularly to suppress prediction variance under extreme conditions, a Bagging (Bootstrap Aggregating) ensemble learning strategy can be employed. This strategy involves randomly sampling several subsets of the original training dataset with replacement (Bootstrap samples), and independently training a GPR model (base model) for each subset. The final prediction result is the mean of the predictions from all base GPR models. The variance of this model can be estimated by combining the variances of the base models and the differences between models, thereby reducing the risk of overfitting during training and improving generalization ability.

[0180] In one possible implementation, if the target parameters are multiple highly correlated parameters, such as discharge voltage, surface temperature rise, and internal resistance, a multi-output Gaussian process (Multi-output GP) can be used. This process can model the correlation between different outputs, and when one output is measured accurately, this correlation can be used to improve the prediction accuracy of other outputs.

[0181] During model training, it is necessary to determine the hyperparameters of the kernel function. These hyperparameters can be determined by maximizing the log marginal likelihood. This involves adjusting all hyperparameters of the kernel function (such as constant factor, length scale, noise level, etc.) through optimization algorithms to maximize the probability of observed training data appearing under this set of hyperparameters. A high probability of appearance indicates a high success rate of positive learning for the model.

[0182] Since the likelihood function may have multiple local optima, a multi-starting-point random initialization optimization strategy is adopted. For example, 10 to 20 different initial hyperparameter values ​​are randomly selected for independent optimization, and finally the set with the largest likelihood value is selected as the final model parameters, thereby increasing the probability of finding the global optimum.

[0183] In this embodiment, a training dataset is obtained, which contains training data corresponding to each operating condition. The training data for each operating condition includes the operating condition parameters and training values ​​of the target parameters. The operating condition parameters for each operating condition in the training dataset are input into the model to be trained to obtain the first predicted value of the target parameter for the corresponding operating condition. Based on the training value and the first predicted value for each operating condition, the model to be trained is adjusted to obtain the target model. Using the target model, a target mapping relationship is constructed based on the predicted values ​​corresponding to the target parameters generated for each operating condition and the corresponding operating condition parameters. The model to be trained is trained using the training dataset. Furthermore, during the training process, the predicted values ​​are adjusted according to physical laws to ensure that the predicted values ​​output by the model conform to the physical laws of energy storage unit operation, thereby improving the accuracy of model training.

[0184] Figure 4 This is a flowchart illustrating how, based on a first predicted value and a change curve, the training values ​​corresponding to the parameters of the first training condition are adjusted under the corresponding training condition. The flowchart may include steps 401 to 403, which are described in detail below. These steps include:

[0185] 401. Based on the change curves of the first working condition parameters and the target parameters and the training values, analyze whether the first predicted value conforms to the change trend of the change curves;

[0186] This variation curve represents how the target parameter follows the change of the first operating condition parameter when the non-first operating condition parameter is fixed.

[0187] It can be done by comparing the training value corresponding to the training condition with the first predicted value to determine whether the first predicted value conforms to the trend of change.

[0188] For example, this variation curve is a temperature slice curve, representing the voltage change as the current increases for the same temperature. For each training condition in this slice curve, a corresponding training value is determined. This training value is compared with a first predicted value to determine whether the first predicted value is less than the training value. The purpose of this judgment is to determine whether the prediction for this training condition violates the physical law that "voltage should not decrease as current increases." If it is less than the training value, then this physical law is violated, and the first predicted value does not conform to the trend of the variation curve. If the first predicted value is not less than the training value, then the prediction for this training condition conforms to the physical law, and the first predicted value conforms to the trend of the variation curve.

[0189] 402. If the first predicted value matches the trend of the change curve, retain the first predicted value;

[0190] If the first predicted value corresponding to the training condition conforms to the changing trend, it can be determined that the model to be trained predicts the target parameter under the training condition relatively accurately, and the first predicted value is retained.

[0191] In one possible implementation, the retained first prediction value can be used as the training value for the corresponding training case in the next iteration of training, so as to achieve iterative training for each training case.

[0192] If the training result of the model to be trained meets the termination condition after this training, the first predicted value is also used to generate the target mapping relationship.

[0193] 403. If the first predicted value does not conform to the trend of the change curve, determine the target value based on the training value and the first predicted value, and adjust the first predicted value corresponding to the first condition parameter under the corresponding training condition based on the target value.

[0194] If the first predicted value corresponding to the training condition conforms to the changing trend, it can be determined that the model to be trained is not accurately predicting the target parameter under the training condition, and the predicted value under the training condition needs to be adjusted.

[0195] The training value and the first predicted value of this training scenario are weighted and calculated to achieve a smooth correction of the training scenario, which conforms to the changing trend of the change curve.

[0196] Weights can be set to sum the training value and the first predicted value to obtain the target value, which is then used as the first predicted value for the training scenario and retained.

[0197] The retained first predicted value can be used as the training value for the corresponding training condition in the next iteration of training, so as to achieve iterative training for each training condition.

[0198] The value of this weight can be set according to the actual situation to control the correction intensity. This weighted calculation aims to achieve a balance between respecting the original prediction and enforcing physical constraints. Generally, the weight of the training value is greater than the weight of the first predicted value; for example, the weight of the training value is usually between 0.5 and 0.8.

[0199] In a space composed of multiple dimensions, the points formed by each training condition and target parameter can be used as a grid in this space. For each first condition parameter, after completing a full grid traversal, the traversal process is repeated 2 to 3 times until the number of violation points (condition points that do not conform to the constraints of physical laws) in the direction of the first condition parameter on the entire surface is less than a preset value. At this point, the training process for the first condition parameter is determined to be over, the next first condition parameter is determined, and the above process is repeated until the termination condition is met.

[0200] Because of the use of an iterative physical constraint post-processing algorithm, non-physical prediction results can be prevented, ensuring that the model's output under any operating condition is consistent with the basic electrochemical principles of the battery.

[0201] In this embodiment, based on the change curves of the first operating condition parameter and the target parameter, and the training values, it is analyzed whether the first predicted value of the target parameter conforms to the change trend of the change curve. If it does, the first predicted value is retained; if it does not, a target value is determined based on the training values ​​and the first predicted value, and the first predicted value corresponding to the first operating condition parameter under the corresponding training condition is adjusted according to the target value. If the first predicted value satisfies the change trend, it indicates that the prediction of the model to be trained is accurate; otherwise, it is inaccurate. Combining the training values, it is determined whether the first predicted value of the target parameter under the training condition satisfies the change trend of the change curve. If it does not, a target value is determined based on the first predicted value and the training values, and the first predicted value corresponding to the first operating condition parameter under the corresponding training condition is adjusted according to the target value. This ensures that the predicted value of the training condition conforms to both the original prediction in the model to be trained and the physical laws of energy storage unit operation, thereby improving the accuracy of model training.

[0202] In one possible implementation, to reduce the amount of training data and improve training efficiency, a training process combining active learning can be adopted. When determining the training dataset, a sparse training condition matrix can be obtained by initially using larger interval values. Iterative training is then performed using this training condition matrix. After each iteration, the predictive performance of the model is validated using a validation dataset to determine the uncertainty of the model's prediction across all conditions. Several training conditions with high uncertainty levels are identified as requiring further training, and the corresponding training data for these conditions is increased. The process of increasing the training dataset can be illustrated in the following example.

[0203] Figure 5 This is a flowchart illustrating the process of generating the training dataset provided in this application embodiment, which may include steps 501 to 506. These steps are described in detail below.

[0204] 501. Obtain the validation dataset. The validation dataset contains validation data for multiple validation scenarios. The validation data for each validation scenario contains validation values ​​for the scenario parameters and the target parameters. The validation scenarios correspond to the training scenarios.

[0205] The fact that the verification scenario corresponds to the training scenario means that the process of obtaining the verification dataset can be the same as the process of obtaining the training dataset in the aforementioned embodiments, and will not be described in detail here.

[0206] In one possible implementation, a dataset can be obtained and divided according to a set ratio, with one part used as a training dataset and the other part as a validation dataset.

[0207] 502. Input the working condition parameters of each verification working condition in the verification dataset into the adjusted model to be trained to obtain the second predicted value of the target parameter of the corresponding verification working condition and the confidence level of the second predicted value.

[0208] The working condition parameters of each verification working condition in the verification dataset are sequentially input into the adjusted training model in order to verify the predictive ability of the adjusted training model.

[0209] The model to be trained is trained to obtain the target model, which can also provide the confidence level of its predicted values. The target model provides an uncertainty measure (confidence interval) for each predicted operating condition point, which provides a crucial information dimension for the BMS (battery management system) to achieve risk adaptive control and formulate conservative or aggressive strategies, thereby improving the robustness and intelligence level of the system.

[0210] In one possible implementation, the model to be trained outputs a second predicted value for the target parameter for each validation condition, and also outputs the confidence level of the second prediction.

[0211] In one possible implementation, in the initial stage, the predictive ability of the model to be trained can be verified according to the second predicted value and the verification value in the verification dataset. Based on the two, a verification index is determined, and the predictive ability of the model to be trained is quantitatively analyzed using the verification index to determine whether the quantitative index is met. Alternatively, two-dimensional slices and multi-dimensional surfaces can be determined in combination with each verification condition to perform a qualitative evaluation.

[0212] In one possible implementation, the model to be trained can calculate the variance of the second predicted value of the target parameter for each verification condition within the entire verification condition space, and use the variance as the confidence level of the second predicted value for the corresponding verification condition.

[0213] Please refer to the corresponding explanation of step 203 above for the quantitative and qualitative assessments; they will not be detailed here.

[0214] 503. Based on the confidence level corresponding to the second predicted value of each verification condition, determine at least two target verification conditions. The confidence level corresponding to the second predicted value of the target verification condition is less than the confidence level corresponding to the second predicted value of the non-target verification condition. The target verification condition is any one of the verification conditions.

[0215] Based on the confidence level corresponding to the predicted value, one or more target verification conditions are determined. If the confidence level of the target verification condition is low, more training is required for the test mine.

[0216] In one possible implementation, a confidence threshold can be set, and it can be determined in turn whether the confidence of each verification condition is less than the confidence threshold. If the confidence of the second predicted value corresponding to any verification condition is less than the confidence threshold, the verification condition is taken as the target verification condition. The accuracy of the model to be trained is low in the target verification condition, and it needs to be trained for the target verification condition.

[0217] 504. Based on the target verification conditions, determine multiple target training conditions. The interval between the condition parameters of any two adjacent training conditions is the first value, and the interval between the condition parameters of any two adjacent verification conditions is the second value. The first value is less than the second value.

[0218] To increase the training data for this target verification condition, more condition parameters can be determined based on this target verification condition.

[0219] In one possible implementation, to improve the precision of training, a smaller interval is selected relative to the interval between the values ​​of the previous training dataset and validation dataset, and more values ​​for the operating condition parameters are determined.

[0220] As an example, when the training and validation datasets were determined last time, the current value interval was 0.2C. This time, the value interval will be 0.1C. Based on the current value of the target validation condition, several new current values ​​will be determined around this current value. For example, if the current value of the target validation condition is 1C, then the current values ​​determined this time include 0.81C, 0.83C, ... 1.17C, 1.19C and 1.21C (excluding the current values ​​in the existing training dataset for each training condition).

[0221] As an example, if the temperature values ​​in the training and validation datasets were set at 10°C intervals in the previous test, then the intervals will be set at 5°C this time. Based on the temperature values ​​of the target validation condition, multiple new temperature values ​​will be determined around this temperature value. For example, if the temperature value of the target validation condition is 45°C, then the current values ​​determined this time will include 25°C, 35°C, ..., 55°C, 60°C (excluding the temperature values ​​of the existing training conditions in the training dataset).

[0222] The first value is less than the second value. The relationship between the two is not limited to twice as in the example above. It can be set according to the actual situation. For example, the current value interval can be 0.1C, 0.2C, 0.3C, etc.

[0223] 505. Based on the training conditions of each target, control the operation of the energy storage unit to obtain the training values ​​of the target parameters under the target training conditions;

[0224] After determining the target training condition, the energy storage unit is controlled to operate under the target training condition to obtain the training value of the target parameter. This training value is the actual value of the target parameter when the energy storage unit operates under the target training condition.

[0225] For each target training scenario, the corresponding target parameter training values ​​are obtained.

[0226] 506. Update the training dataset based on at least two target training scenarios and their corresponding training values.

[0227] The at least two target training scenarios are higher-precision training scenarios. Accordingly, after determining their training values, the training scenario and its corresponding training values ​​are used as new training data and added to the training dataset, which enriches the training dataset and can provide higher-precision training samples for subsequent training of the model to be trained, thereby realizing the optimization of the training process through active learning.

[0228] By iteratively updating the training dataset until the model accuracy of the model to be trained meets the requirements or the experimental budget is exhausted, the training process ends, enabling intelligent allocation of experimental resources and improving the efficiency of resource utilization.

[0229] In one possible implementation, after determining the target verification condition, more energy storage units can be selected, and more energy storage units can be controlled to operate under the target verification condition to obtain the corresponding training values. The training values ​​and the corresponding target verification condition can be used as the new training condition and training values. Each energy storage unit can be used to operate under the target verification condition to obtain the first predicted value of the target parameters. The first predicted value and the training values ​​can be used to update the model to be trained.

[0230] In this embodiment, the predictive ability of the adjusted training model is verified using a validation dataset. Based on the confidence level of the second predicted value output by the training model, several validation conditions with low confidence are identified as target validation conditions. Multiple values ​​are selected around the parameter values ​​corresponding to the target validation conditions at smaller intervals. These selected values ​​form multiple target training conditions. The energy storage unit is controlled to operate according to each target training condition to obtain the training values ​​of the target parameters under those conditions. The training dataset is updated using at least two target training conditions and their corresponding training values. This process gradually refines the determination of training samples during training, and the dataset is updated iteratively to achieve intelligent allocation of experimental resources and improve resource utilization efficiency.

[0231] After the training of the model is completed, it needs to be packaged and deployed to the device.

[0232] In one possible implementation, a target mapping relationship is constructed based on the predicted values ​​of the target parameters generated by the target model for each training condition and the corresponding condition parameters, including:

[0233] If the equipment resources do not meet the preset resource requirements, the target model generates a target surface based on the predicted values ​​of the target parameters generated for each training condition and the corresponding operating condition parameters. The target surface is generated based on the predicted values ​​of the target parameters and the operating condition parameters for each training condition. The preset resource requirements are the resources needed to deploy the target model. A target table is generated based on the target surface. The target table contains parameters for several operating condition points. Each operating condition point contains the operating condition parameter and the second target parameter. The target table is used as a target mapping relationship.

[0234] If the equipment's resources meet the preset resource requirements, the control target model generates target surfaces based on the predicted values ​​of target parameters for each training condition and the corresponding condition parameters, and uses the target surfaces as target mapping relationships.

[0235] First, determine the preset resource requirements for deploying the target model. If the device's resources can meet these requirements, the target model can be deployed on the device, and the model within the target model can be encapsulated as an independent, lightweight prediction function library. This prediction function library can then be deployed on a device with sufficient resources. If the device's resources cannot meet the preset resource requirements, the target model can be lightweighted before deployment. This lightweighting process can involve dividing the target surface into high-precision multi-dimensional tables according to high-precision partitioning rules.

[0236] For resource-constrained embedded BMS, continuous surfaces can be discretized into high-precision lookup tables. However, these tables are generated based on the GPR model and their accuracy far exceeds that of traditional experimental lookup table methods.

[0237] Because of the target model, the generated target surface is a continuous multidimensional analytical surface covering the entire specified range, which can make instantaneous and smooth predictions of the performance at any point within the range, thus solving the discreteness and interpolation error problems of the lookup table method.

[0238] In one possible implementation, an API interface can be developed to interact with system hardware in the device. This API interface can use standard C / C++ interface functions.

[0239] In one possible implementation, a BMS integration approach can be adopted, compiling and integrating the function library into the BMS application layer software. The BMS acquires the battery's current and temperature signals in real time, and by calling this interface function, the corresponding voltage prediction value and its confidence interval can be instantly retrieved.

[0240] This BMS integration can be used for scenarios such as: SOC (State of Charge) calibration, SOP (State of Power) estimation, thermal management triggering, or SOH (State of Health) inference.

[0241] Among them, SOC calibration is used to compare the predicted voltage with the measured voltage and perform SOC correction; SOP estimation is used to calculate the maximum allowable charge and discharge current in combination with the limiting voltage; thermal management triggering is used to activate the cooling or heating system in advance when the prediction efficiency is too low or the uncertainty is too high; and SOH inference is used to monitor the changes in model prediction bias over a long period of time and can serve as an indicator of SOH decay.

[0242] It should be noted that, through the above process, a high-precision, physically consistent battery voltage prediction model that provides uncertainty information can be constructed using hundreds of data points (training conditions). This model performs excellently on the test dataset, and the post-processed surface conforms to physical laws.

[0243] The final encapsulated prediction function can be directly called by the BMS, achieving a perfect transition from laboratory data to applications in 3C products such as laptops.

[0244] In this embodiment, the method for constructing the target mapping relationship can be determined by considering whether the device's resources meet the preset resource requirements for setting the target model. If the device's resources meet the preset resource requirements, the target model generates target surfaces based on the predicted values ​​of the target parameters for each training condition and the corresponding condition parameters. These target surfaces serve as the target mapping relationship, and the target model can then be directly configured onto the device. If the device's resources do not meet the preset resource requirements, the target model first generates target surfaces based on the predicted values ​​of the target parameters for each training condition and the corresponding condition parameters. Then, a target table is generated based on the target surfaces, and this target table serves as the target mapping relationship, reducing the resource requirements for setting this target mapping relationship.

[0245] The first intelligent agent involved in the device control method of this application can be trained through the aforementioned model training process and deployed into the device in the form of an intelligent agent.

[0246] The model training process includes steps such as training data acquisition, model construction and hyperparameter optimization, training process, surface generation, model encapsulation and BMS integration and deployment. The training process includes post-processing of physical consistency constraints and comprehensive evaluation and verification of model performance. Each step has been described in the preceding embodiments; the examples listed here are to illustrate the execution order of each step in the application scenario. Of course, the training data acquisition and model construction and hyperparameter optimization processes can be executed simultaneously or in any order; this application does not impose any restrictions.

[0247] Because the model training and deployment technology chain is complete and easy to engineer, the complete technical solution from laboratory to product can be derived from the above content, including model lightweighting, packaging and integration guidelines, enabling advanced machine learning models to be successfully implemented on actual BMS embedded platforms, and promoting the intelligent upgrade of battery management technology.

[0248] Figure 6 This is a schematic diagram of the device control method provided in this application in an application scenario. In this application scenario, the device includes an intelligent agent 601 and a battery module 602. The battery module has a battery management unit, an MCU (Micro Controller Unit), and battery management firmware FW nested in sequence. The first intelligent agent is represented by an intelligent agent.

[0249] The agent sends a command to the host to request battery parameters (such as current temperature, current output current, and possibly current output voltage); the host sends a command to the battery module to request battery parameters; the battery module feeds back the battery parameters to the host, which then feeds them back to the agent; the agent determines the optimal battery parameters based on these parameters (the optimal output voltage corresponding to the current temperature and current output current); the agent feeds back these optimal parameters to the host, which outputs the optimal battery parameters and the acquired battery parameters, and can generate a query message asking whether to change the battery parameters; after receiving feedback from the user confirming the change, the agent adjusts the battery's output voltage using the output voltage from the optimal parameters.

[0250] Figure 7 This is a schematic diagram of the device control method provided in this application in another application scenario. This scenario includes a local device 701 and a cloud device 702. The local device includes an intelligent agent 7011 and a battery module 7012. The battery module contains a battery management unit, an MCU, and battery management firmware (FW) nested in a stacked manner. The first intelligent agent in this application is located on the cloud device and is represented as intelligent agent 2. This intelligent agent 2 serves as part of the cloud database within the cloud device. The local device includes intelligent agent 1 for interaction.

[0251] The intelligent agent 1 sends a command to the host to request battery parameters (such as current temperature, current output current, and possibly current output voltage); the host sends a command to the battery module to request battery parameters; the battery module feeds back the battery parameters to the host, which then feeds them back to intelligent agent 1; intelligent agent 1 uploads the battery parameters to the cloud; intelligent agent 2 on the cloud determines the optimal battery parameters (the optimal output voltage corresponding to the current temperature and current output current) based on these parameters; intelligent agent 2 feeds back these optimal parameters to intelligent agent 1 on the local device, which then sends them to the host; the host outputs the optimal battery parameters and the acquired battery parameters, and can generate a query message asking whether to change the battery parameters; after receiving feedback from the user confirming the change, the host adjusts the battery's output voltage using the output voltage from the optimal parameters.

[0252] The above describes a device control method provided by an embodiment of this application. The following will describe an electronic device that performs the above device control method.

[0253] Please see Figure 8 , Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. For example... Figure 8 As shown, the electronic device 800 includes:

[0254] Energy storage unit 801;

[0255] The processor 802 is used to acquire the first parameter of the energy storage unit, which is an operating parameter that reflects the current operating condition of the energy storage unit; determine the target parameter corresponding to the first parameter based on the target mapping relationship; the target mapping relationship represents the mapping relationship between different operating parameters and target parameters; the target mapping relationship satisfies the change law of the actual operating parameters under different operating conditions; and configure the energy storage unit based on the target parameter.

[0256] In one possible implementation, the processor can be a functional structure with data processing capabilities in an electronic device, such as an NPU (Neural-network Processing Unit), a battery management unit, etc. This application does not limit the specific functional structure for implementing the processor.

[0257] For details on the specific steps by which the processor implements its functions and the detailed explanations of each step, please refer to the explanations in the foregoing method embodiments; they will not be repeated here.

[0258] The processor can carry the first intelligent agent and can also set the target model. The training process of the target model can be executed in other electronic devices. In this embodiment, the electronic device 800 only uses the target model.

[0259] In this embodiment, the electronic device includes an energy storage unit and a processor. The processor pre-sets a target mapping relationship, which represents the mapping relationship between different operating parameters and target parameters. This target mapping relationship conforms to the actual physical characteristics of the energy storage unit during operation and satisfies the changing patterns of the energy storage unit's actual operating parameters under different operating conditions. During the operation of the energy storage unit, a first parameter of the energy storage unit is acquired, which reflects the current operating condition of the energy storage unit. Based on the target mapping relationship, target parameters corresponding to the first parameter are determined. Based on the target parameters, the energy storage unit is configured. The determined target parameters are parameters that satisfy the target mapping relationship with each parameter in the first parameter and are actual operating parameters that conform to the changing patterns of the first parameter under the corresponding operating conditions. Therefore, configuring the energy storage unit according to the target parameters can better match the actual situation of the energy storage unit under the corresponding operating conditions, resulting in more accurate control.

[0260] This application also provides a computer program product including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the device control methods provided in this application.

[0261] This application also provides a computer-readable storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the device control methods provided in this application.

[0262] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0263] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0264] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the accompanying drawings of the device embodiments provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0265] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0266] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0267] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A device control method, comprising: Obtain the first parameter of the energy storage unit, wherein the first parameter is an operating parameter that can reflect the current operating condition of the energy storage unit; Based on the target mapping relationship, determine the target parameter corresponding to the first parameter; The target mapping relationship represents the mapping relationship between different operating parameters and target parameters; The target mapping relationship satisfies the variation law of actual operating parameters under different working conditions; Configure the energy storage unit based on the target parameters.

2. The device control method according to claim 1, wherein determining the target parameter corresponding to the first parameter based on the target mapping relationship includes: The first intelligent agent determines the target parameter corresponding to the first parameter based on the target mapping relationship. The first intelligent agent is trained based on training data, which includes the actual operating parameters of the energy storage unit under various operating conditions.

3. The equipment control method according to claim 1 further includes: Obtain the second parameter of the energy storage unit; The second parameter characterizes the type of the energy storage unit; Based on the second parameter of the energy storage unit, a target mapping relationship corresponding to the second parameter is determined.

4. The equipment control method according to claim 1 further includes: Determine the operating mode of the equipment; Based on the device's operating mode, a target mapping relationship is determined, with different operating modes corresponding to different discharge rates of the energy storage unit.

5. The device control method according to claim 1, wherein the process of determining the target mapping relationship includes: Obtain a training dataset, which contains training data corresponding to each working condition. The training data for each working condition includes training values ​​of the working condition parameters and target parameters for that working condition. Input the working condition parameters of each training working condition in the training dataset into the model to be trained to obtain the first predicted value of the target parameter of the corresponding training working condition. Based on the training value and the first predicted value corresponding to each training condition, the model to be trained is adjusted to obtain the target model; Using the target model, a target mapping relationship is constructed based on the predicted values ​​corresponding to the target parameters generated for each training condition and the corresponding condition parameters.

6. The equipment control method according to claim 5, wherein each training condition includes at least two condition parameters, and the step of adjusting the model to be trained based on the training value and the first predicted value corresponding to each training condition to obtain the target model includes: One of the at least two operating condition parameters is selected as the first operating condition parameter in turn; The non-first working condition parameters among the at least two working condition parameters are sequentially taken as preset values ​​corresponding to each training working condition. The first working condition parameter is used as a variable, and the values ​​of the first working condition parameter are sorted according to preset change rules to determine the change curve of the first working condition parameter and the target parameter under each training working condition. The target parameter in the change curve adopts the first predicted value. Based on the change curve, the training values ​​corresponding to the target parameters and the first condition parameters under each training condition are determined sequentially. Based on the training values ​​and the change curve, adjust the first predicted value corresponding to the first condition parameter under the corresponding training condition; Based on the adjusted first prediction value, the model to be trained is adjusted to obtain the adjusted model to be trained. Based on the adjusted model to be trained, the process returns to input the working condition parameters of each training condition in the training dataset into the model to be trained, and obtains the first predicted value of the target parameter for the corresponding training condition, until the termination condition is met, and the target model is obtained.

7. The equipment control method according to claim 6, wherein adjusting the first predicted value corresponding to the first condition parameter under the corresponding training condition based on the training value and the change curve includes: Based on the change curves of the first working condition parameters and the target parameters and the training values, analyze whether the first predicted value conforms to the change trend of the change curves; If the first predicted value matches the trend of the change curve, retain the first predicted value; If the first predicted value does not conform to the trend of the change curve, a target value is determined based on the training value and the first predicted value, and the first predicted value corresponding to the first condition parameter under the corresponding training condition is adjusted based on the target value.

8. The device control method according to claim 6 further includes: Obtain a validation dataset, which contains validation data for multiple validation scenarios. The validation data for each validation scenario contains validation values ​​for the scenario parameters and target parameters. The validation scenarios correspond to the training scenarios. Input the working condition parameters of each verification working condition in the verification dataset into the adjusted model to be trained to obtain the second predicted value of the target parameter of the corresponding verification working condition and the confidence level of the second predicted value. Based on the confidence level corresponding to the second predicted value of each verification condition, at least two target verification conditions are determined. The confidence level corresponding to the second predicted value of the target verification condition is less than the confidence level corresponding to the second predicted value of the non-target verification condition. The target verification condition is any one of the verification conditions. Based on the target verification conditions, multiple target training conditions are determined. The interval between the condition parameters of any two adjacent training conditions is a first value, and the interval between the condition parameters of any two adjacent verification conditions is a second value. The first value is less than the second value. Based on each target training condition, the energy storage unit is controlled to operate, and the training values ​​of the target parameters under the target training condition are obtained; The training dataset is updated based on the at least two target training conditions and their corresponding training values.

9. The equipment control method according to claim 5, comprising constructing a target mapping relationship based on the predicted values ​​of target parameters generated by the target model for each training condition and the corresponding condition parameters, including: If the equipment resources do not meet the preset resource requirements, the target model generates a target surface based on the predicted values ​​of the target parameters generated for each training condition and the corresponding condition parameters. The preset resource requirements are the resources needed to deploy the target model, and the target surface is generated based on the predicted values ​​of the target parameters and the condition parameters for each training condition. A target table is generated based on the target surface, and the target table contains parameters for several condition points. Each condition point contains condition parameters and a second target parameter. The target table is used as a target mapping relationship. If the equipment's resources meet the preset resource requirements, the control target model generates a target surface based on the predicted values ​​of the target parameters generated for each training condition and the corresponding condition parameters, and uses the target surface as the target mapping relationship.

10. An electronic device, comprising: Energy storage unit; The processor is used to acquire the first parameter of the energy storage unit, wherein the first parameter is an operating parameter that can reflect the current operating condition of the energy storage unit; Based on the target mapping relationship, determine the target parameter corresponding to the first parameter; The target mapping relationship represents the mapping relationship between different operating parameters and target parameters; The target mapping relationship satisfies the variation law of actual operating parameters under different working conditions; the energy storage unit is configured based on the target parameters.

Images