Model training method, battery status determination method, device, medium and product

Abstract

The present disclosure relates to the field of battery technology, and in particular to a model training method, a battery status determination method, a device, a medium and a product, wherein the model training method comprises: obtaining sample data marked with a label; training a first model based on the sample data to obtain a second model for predicting the health of the battery, wherein the sample data is indicator data determined based on the discharge data of the battery during a discharge cycle. Since the sample data of the present solution can be determined directly based on the discharge data, the generation of intermediate variables and the complexity of the model training operation can be reduced, thereby improving the model training efficiency and reducing the hardware requirements for the system. In addition, since the sample data of the present solution is indicator data that characterizes the health status of the battery, when the model is trained based on the sample data, the model can more directly capture the key factors that affect the battery performance, thereby improving the accuracy of the model prediction results.

Description

Technical Field

[0001] The present disclosure relates to the field of battery technology, and in particular, to a model training method, a battery status determination method, a device, a medium and a product. Background Art

[0002] With the rapid development of the new energy storage industry, electrochemical energy storage power stations using electrochemical batteries as energy storage media have emerged. In order to ensure the safe operation of electrochemical energy storage power stations, it is necessary to evaluate the battery's state of health (SOH).

[0003] In the related art, the battery health status can be evaluated based on an intelligent model. However, before the battery health status is evaluated based on the intelligent model, the intelligent model needs to be trained first. Summary of the invention

[0004] The purpose of the present disclosure is to provide a model training method, a battery state determination method, a device, a medium and a product to improve the generalization ability of the model and the accuracy of the prediction results.

[0005] To achieve the above objectives, the present disclosure provides a model training method in a first aspect, the model training method comprising:

[0006] Get sample data with labels;

[0007] Training the first model based on the sample data to obtain a second model for predicting battery health;

[0008] The sample data is obtained in the following way:

[0009] Obtain battery health and discharge data during the discharge cycle;

[0010] Determining, according to the discharge data, indicator data for characterizing a battery health state of the battery;

[0011] The battery health is used as a label of the indicator data to obtain the sample data.

[0012] Optionally, the indicator data includes discharge capacity, discharge time, voltage average, discharge platform ratio and / or battery internal resistance, wherein the discharge platform ratio is used to characterize the proportion of the curve interval in the battery discharge curve where the voltage change rate is less than a preset voltage change rate threshold in the discharge curve, and the discharge curve is determined based on the discharge data.

[0013] Optionally, the discharge data includes a sampling interval and a sampling voltage, and determining the discharge platform proportion according to the discharge data includes:

[0014] Determining a voltage change rate according to the sampling interval and the sampling voltage;

[0015] Processing the voltage change rate based on a round-down algorithm to obtain a target voltage change rate;

[0016] Based on the target voltage change rate, a discharge platform proportion is determined.

[0017] Optionally, the target voltage change rate includes a plurality of values, and determining the discharge platform proportion based on the target voltage change rate includes:

[0018] Among the plurality of target voltage change rates, determining the target voltage change rate to be the number of preset voltage change rates;

[0019] The ratio of the target voltage change rate to the number of preset voltage change rates is determined as the discharge platform proportion.

[0020] Optionally, the training of the first model based on the sample data to obtain a second model for predicting the battery health includes:

[0021] Normalizing the sample data to obtain target sample data;

[0022] The first model is trained based on the target sample data to obtain a second model for predicting the battery health.

[0023] Optionally, the sample data includes a plurality of data, and the normalizing the sample data to obtain target sample data includes:

[0024] For each sample data, perform the following steps:

[0025] Determine a first difference between the sample data and the first sample data, and a second difference between the second sample data and the first sample data, wherein the first sample data is used to represent the smallest sample data among the plurality of sample data, and the second sample data is used to represent the largest sample data among the plurality of sample data;

[0026] The ratio of the first difference value to the second difference value is determined as target sample data.

[0027] A second aspect of the present disclosure provides a battery status determination method, the battery status determination method comprising:

[0028] Obtaining indicator data for characterizing the battery health status of the battery;

[0029] Inputting the indicator data into a battery health determination model to obtain the battery health;

[0030] The battery state is determined according to the battery health, wherein the battery health determination model is a second model trained by the model training method described in any one of the first aspects.

[0031] Optionally, before inputting the indicator data into the battery health determination model, the method further includes:

[0032] storing the sample data obtained in each discharge cycle into a target database for storing the sample data;

[0033] Training the battery health determination model based on the sample data in the target database to obtain a battery health determination model with updated parameters;

[0034] The step of inputting the indicator data into a battery health determination model to obtain the battery health includes:

[0035] The indicator data is input into the battery health determination model after the parameter update to obtain the battery health.

[0036] A third aspect of the present disclosure provides an electronic device, including:

[0037] a memory having a computer program stored thereon;

[0038] A processor is used to execute the computer program in the memory to implement the steps of the method in any one of the first aspect or the second aspect.

[0039] A fourth aspect of the present disclosure provides a non-temporary computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any one of the first aspect or the second aspect.

[0040] A fifth aspect of the present disclosure provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method described in any one of the first aspect or the second aspect.

[0041] Through the above technical solution, the battery health and discharge data of the battery during the discharge cycle can be obtained, and the indicator data used to characterize the battery health status can be determined based on the discharge data, and the first model of the model can be trained based on the indicator data and the battery health status to obtain a second model for predicting the battery health status. Since the sample data of this solution can be directly determined based on the discharge data, the generation of intermediate variables and the complexity of the model training operation can be reduced, the model training efficiency can be improved, and the hardware requirements for the system can be reduced. In addition, since the sample data of this solution is the indicator data that characterizes the battery health status, when the model is trained based on the sample data, the model can more directly capture the key factors that affect the battery performance, thereby improving the accuracy of the model prediction results.

[0042] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present disclosure but do not constitute a limitation of the present disclosure. In the accompanying drawings:

[0044] Figure 1 It is a flow chart of the steps of a lithium-ion power battery health assessment method based on data fusion in the related art;

[0045] Figure 2 is a flowchart of a model training method according to an exemplary embodiment of the present disclosure;

[0046] Figure 3 is a schematic diagram showing a partial correlation between indicator data and battery health according to an exemplary embodiment of the present disclosure;

[0047] Figure 4 is a schematic diagram of a model structure of a first model according to an exemplary embodiment of the present disclosure;

[0048] Figure 5 is a schematic diagram showing a principle of a random deactivation layer according to an exemplary embodiment of the present disclosure;

[0049] Figure 6 is a schematic diagram showing a principle of a learning rate decay mechanism according to an exemplary embodiment of the present disclosure;

[0050] Figure 7 is a schematic diagram showing another principle of a learning rate decay mechanism according to an exemplary embodiment of the present disclosure;

[0051] Figure 8 is a flowchart of a model training according to an exemplary embodiment of the present disclosure;

[0052] Fig. 9 is a schematic diagram showing a comparison effect between a health prediction value of a first model and a label according to an exemplary embodiment of the present disclosure;

[0053] Fig.10 is a flow chart of a method for determining a battery status according to an exemplary embodiment of the present disclosure;

[0054] Fig.11 is a block diagram of a battery status determination system according to an exemplary embodiment of the present disclosure;

[0055] Fig.12 is a flow chart showing another method for determining a battery state according to an exemplary embodiment of the present disclosure;

[0056] Fig.13 is a block diagram of a model training device according to an exemplary embodiment of the present disclosure;

[0057] Fig.14 is a block diagram of a battery status determination device according to an exemplary embodiment of the present disclosure;

[0058] Fig.15 It is a block diagram of an electronic device according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0059] The specific implementation of the present disclosure is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the present disclosure, and is not used to limit the present disclosure.

[0060] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0061] As mentioned in the background technology, in the related technology, the battery health status can be evaluated based on the intelligent model. However, before the battery health status is evaluated based on the intelligent model, the intelligent model needs to be trained first.

[0062] For example, the invention patent with publication number CN116953529A discloses a lithium-ion power battery health assessment method based on data fusion, such as Figure 1As shown, the method includes: performing a cycle aging experiment on the lithium-ion power battery to construct an incremental capacity curve of the lithium-ion power battery; extracting SOH characteristic parameters from the incremental capacity curve of the lithium-ion power battery, and performing correlation analysis between the SOH characteristic parameters of the lithium-ion power battery and the SOH of the lithium-ion power battery by a grey correlation analysis method to obtain the optimal SOH characteristic parameters of the lithium-ion power battery; constructing a SOH estimation model for the lithium-ion power battery; based on a data fusion method, inputting the optimal SOH characteristic parameters of the lithium-ion power battery into the model for training, and outputting the SOH value of the lithium-ion power battery.

[0063] Since the above scheme requires the construction of incremental capacity curves and the gray correlation analysis, the operational complexity of model training is increased and the efficiency of model training is reduced. At the same time, the above operations will generate a large number of intermediate variables, resulting in an increase in the required memory and increasing the hardware requirements for the system.

[0064] In view of this, the embodiments of the present disclosure provide a model training method, a battery status determination method, a device, a medium and a product to overcome the above-mentioned technical problems.

[0065] The embodiments of the present disclosure are further explained below with reference to the accompanying drawings.

[0066] Figure 2 is a flowchart of a model training method according to an exemplary embodiment of the present disclosure, referring to Figure 2 , the model training method may include the following steps:

[0067] S201: Obtain sample data marked with labels.

[0068] S202: Training the first model based on the sample data to obtain a second model for predicting battery health.

[0069] The sample data can be obtained in the following ways:

[0070] The battery health and discharge data of the battery in the discharge cycle are obtained; based on the discharge data, indicator data for characterizing the battery health state of the battery is determined; the battery health is used as a label for the indicator data to obtain sample data.

[0071] It should be understood that the battery in this embodiment may be a single cell or a battery pack composed of multiple single cells connected in series and / or in parallel. When the battery in this embodiment is a battery pack, the discharge data of the battery in the discharge cycle may refer to the discharge data of all single cells constituting the battery pack. Accordingly, the index data is the index data characterizing the health status of the battery pack, and the battery health is the battery health of the battery pack.

[0072] The discharge data and index data may be set according to actual conditions, and the embodiments of the present disclosure do not impose any limitation on this.

[0073] For example, when the discharge cycle is used to characterize the discharge from the rated voltage of the battery (fully charged state) to the preset cut-off voltage, the discharge data may include the sampling interval, the discharge start time, the discharge end time, the battery voltage, the discharge current, the battery voltage at the end of the discharge (i.e., the end of the discharge), the battery voltage at the next moment after the start of the discharge (i.e., the start of the discharge), and the instantaneous current value at the end of the discharge, etc.

[0074] For example, when the discharge cycle is used to characterize a complete charge and discharge cycle (i.e., the battery is discharged from a fully charged state to a depleted state, and then recharged to a fully charged state), the discharge data may include the sampling interval, the discharge start time, the discharge end time, the battery voltage, the discharge current, the battery voltage at the end of the discharge (i.e., the discharge end time), the battery voltage at the next moment after the start of the discharge (i.e., the discharge start time), and the current value at the end of the discharge, the charging current, the charge start time, the charge end time, the battery voltage at the start of the charge (i.e., the charge start time), the battery voltage at the previous moment after the start of the charge, and the current value at the instant of charging, etc.

[0075] For example, the indicator data may include discharge capacity, discharge duration, voltage average value, battery internal resistance, discharge rate and / or discharge energy, etc.

[0076] In a possible way, in order to screen out the indicator data that can better reflect the battery health status from multiple indicator data and improve the prediction accuracy of the model, the inventors conducted a series of experiments to test the correlation between different indicator data and battery health, among which the results with higher correlation can be as follows: Figure 3 As shown. Figure 3 It can be seen that the correlation between discharge time and battery health is 97.43%, the correlation between discharge platform proportion and battery health is 86.98%, the correlation between discharge average voltage and battery health is 66.99%, the correlation between battery internal resistance and battery health is 72.99%, and the correlation between discharge capacity and battery health is 100%. Since each correlation is greater than 50%, in a possible way, the indicator data may include discharge capacity, discharge time, voltage average, discharge platform proportion and / or battery internal resistance. The discharge platform proportion is used to characterize the proportion of the curve interval in the battery discharge curve where the voltage change rate is less than the preset voltage change rate threshold in the discharge curve. The discharge curve is determined based on the discharge data.

[0077] Accordingly, determining the indicator data for characterizing the health status of the battery according to the discharge data may include: determining the discharge capacity, discharge duration, voltage average, discharge platform ratio and / or battery internal resistance according to the discharge data.

[0078] For example, when determining the discharge capacity based on the discharge data, if the discharge current is constant, the discharge duration can be determined from the discharge start time to the discharge end time, and then the product of the discharge current and the discharge duration can be calculated to obtain the discharge capacity. If the discharge current is not constant, the discharge current can be integrated from the discharge start time to the discharge end time to obtain the discharge capacity. That is, when the discharge current is not constant, the discharge capacity can be determined by the following formula:

[0079]

[0080] in, represents the discharge capacitance, Indicates the discharge start time, Indicates the end time of discharge. Indicates the discharge current.

[0081] For example, when determining the voltage average value according to the discharge data, the discharge duration can be determined according to the discharge start time to the discharge end time, and then the battery voltage sampled at each sampling time is summed, and then the ratio of the battery voltage to the discharge duration is calculated to obtain the voltage average value, that is, the voltage average value can be determined by the following formula:

[0082]

[0083] in, Indicates the average voltage. Indicates The battery voltage at the sampling moment, Indicates the number of sampling times, Indicates the discharge time.

[0084] Alternatively, the sampling times may be counted first, and then the battery voltages sampled at each sampling moment may be summed, and then the ratio of the sum of the battery voltages to the sampling times may be calculated to obtain the voltage average value, that is, the voltage average value may be determined by the following formula:

[0085]

[0086] in, Indicates the average voltage. Indicates The battery voltage at the sampling moment, Indicates the number of sampling times.

[0087] For example, when determining the battery internal resistance according to the discharge data, the charging internal resistance and the discharging internal resistance may be determined based on the discharge data first, and then the battery internal resistance may be determined based on the charging internal resistance and the discharging internal resistance.

[0088] Specifically, the difference between the battery voltage at the start of charging and the battery voltage at the last moment of the start of charging can be determined first, and then the charging internal resistance can be determined based on the difference and the instantaneous current value at the charging moment. That is, the charging internal resistance can be determined by the following formula:

[0089]

[0090] in, Indicates the charging internal resistance, Indicates the battery voltage at the beginning of charging. Indicates the battery voltage at the last moment of charging start. Indicates the instantaneous current value at the moment of charging.

[0091] After that, the difference between the battery voltage at the next moment after the discharge start and the battery voltage at the discharge end can be determined first, and then the discharge internal resistance can be determined based on the difference and the instantaneous current value at the discharge end. That is, the discharge internal resistance can be determined by the following formula:

[0092]

[0093] in, Indicates the discharge internal resistance, Indicates the battery voltage at the next moment when discharge begins. represents the battery voltage at the end of discharge, Indicates the instantaneous current value at the end of discharge.

[0094] After obtaining the discharge internal resistance and charge internal resistance, the battery resistance can be obtained by the following formula:

[0095]

[0096] in, Indicates the battery resistance.

[0097] For example, when determining the discharge platform proportion based on the discharge data, the battery voltage curve can be first constructed based on the battery voltage and the discharge time, and then the target interval with smaller voltage change in the battery voltage curve is identified, and the target discharge time corresponding to the target interval is determined. Finally, the discharge platform proportion is obtained by calculating the ratio of the target discharge time to the discharge time.

[0098] It should be understood that when determining the discharge platform ratio based on the above method, the target interval is generally determined by the operator based on experience or by observing the battery voltage curve, which leads to the problem of inaccurate discharge platform ratio. To overcome the above problem, this embodiment can calculate the voltage change rate between two adjacent sampled voltages based on the sampling interval and the sampled voltage, and then determine the discharge platform ratio based on the voltage change rate, thereby improving the accuracy of the discharge platform ratio.

[0099] That is, in a possible manner, the discharge data may include a sampling interval and a sampling voltage. Accordingly, determining the discharge platform proportion according to the discharge data may include:

[0100] The voltage change rate is determined according to the sampling interval and the sampling voltage. The voltage change rate is processed based on a rounding-down algorithm to obtain a target voltage change rate. The discharge platform proportion is determined based on the target voltage change rate.

[0101] The sampling interval can be set according to actual conditions, and the embodiments of the present disclosure do not impose any restrictions on this. For example, the sampling interval can be set to 1 minute, so that the battery voltage can be collected every 1 minute. After collecting all the voltage data, the voltage change rate between two adjacent battery voltages is obtained by substituting the voltage data of two adjacent batteries into the following formula:

[0102]

[0103] in, It represents the voltage change rate, represents the sampling interval, Indicates the collected The battery voltage, Indicates the collected A battery voltage.

[0104] It should be understood that during the battery discharge process, there may be some random fluctuations caused by measurement errors or environmental factors. In order to reduce the impact of these random fluctuations on the voltage change rate, this embodiment rounds down the voltage change rate, that is, truncates the decimal part of the value to reduce the impact of these random fluctuations on the voltage change rate, thereby obtaining the target voltage change rate. After that, the discharge platform proportion can be obtained based on the target voltage change rate.

[0105] In a possible manner, the target voltage change rate includes multiple ones. Accordingly, based on the target voltage change rate, determining the discharge platform proportion may include:

[0106] Among multiple target voltage change rates, determine the number of target voltage change rates that are preset voltage change rates; and determine the ratio of the number of target voltage change rates that are preset voltage change rates to the number of target voltage change rates as the discharge platform proportion.

[0107] The preset voltage change rate can be set according to actual conditions, and the embodiment of the present disclosure does not impose any limitation on this. In a possible manner, the preset voltage change rate is set to 0.

[0108] It should be understood that since the discharge capacity and discharge time can reflect the battery's ability to store energy, the voltage average and discharge platform ratio can reflect the battery's ability to maintain voltage stability during discharge, and the battery's internal resistance can reflect the resistance to the transmission of electrons and ions inside the battery, by using the discharge capacity, discharge time, voltage average, discharge platform ratio, and battery internal resistance as indicator data, the battery's health status can be reflected from different angles, and then when the model is trained based on the above indicator data, the model can more accurately identify and predict the battery's health status and performance changes, thereby improving the accuracy of the prediction results.

[0109] It should be understood that the battery health is generally used to characterize the percentage of capacity that a battery can maintain relative to its initial capacity (i.e., factory capacity), and is generally determined by calculating the ratio between the current capacity and the initial capacity. Therefore, this embodiment can obtain the battery health of the battery during the discharge cycle by obtaining the target battery capacity of the battery during the discharge cycle and calculating the ratio between the target battery capacity and the initial capacity.

[0110] The target battery capacity of the battery during the discharge cycle may be obtained through testing or prediction, and the embodiments of the present disclosure do not impose any limitation on this.

[0111] For example, the battery may be charged to its rated voltage first, and then an electronic load may be used to discharge the battery at a constant current until the battery voltage drops to a preset cut-off voltage, and finally the target battery capacity may be calculated based on the discharge time and the discharge current.

[0112] For example, the neural network model can be trained so that the neural network model can learn the relationship between battery usage time and battery capacity, or can learn the relationship between the number of charge and discharge times and battery capacity, so that the battery capacity can be estimated based on the battery usage time or the number of charge and discharge times.

[0113] After obtaining the indicator data and the battery health in the above manner, the battery health can be used as a label for the indicator data to train the first model and obtain a second model for predicting the battery health.

[0114] The first model in this embodiment may be an existing model in the related art, or may be a model that is improved from an existing model in the related art, and the embodiment of the present disclosure does not impose any restrictions on this. For example, the first model may be a deep neural network (DNN) in the related art, such as Figure 4As shown in Figure 1, it can include 1 input layer, 4 hidden layers and 1 output layer. At the same time, in order to improve the generalization ability and training effect of the model, a random dropout layer and a learning rate decay mechanism can be introduced as much as possible. The schematic diagram of the random dropout layer can be shown as follows: Figure 5 As shown in , by making the nodes of the hidden layer fail with a certain probability at each iteration, the model can avoid overfitting. The principle diagram of the learning rate decay mechanism is as follows Figure 6 and Figure 7 As shown in the figure, when training the model parameters, the learning rate will decay with the training rounds. If the current error value of the model is at the bottom arrow, if the learning rate is too large, the back propagation step size will be too large, thereby increasing the model error value. Therefore, setting the learning rate decay mechanism can help obtain model parameters with better training effects.

[0115] After the model structure and training mechanism of the first model are determined, the first model can be trained based on the sample data.

[0116] In a possible manner, training the first model based on sample data to obtain a second model for predicting battery health may include:

[0117] The sample data is normalized to obtain target sample data; the first model is trained based on the target sample data to obtain a second model for predicting battery health.

[0118] It should be understood that different indicator data may have different dimensions and value ranges. For example, the discharge capacity may be measured in milliampere hours (mAh), while the battery internal resistance may be measured in ohms (Ω). Therefore, this embodiment normalizes the sample data to eliminate the influence of different dimensions, so that data with different characteristics can be compared and analyzed under the same metric.

[0119] In a possible manner, the sample data may include multiple data. Accordingly, normalizing the sample data to obtain target sample data may include:

[0120] For each sample data, perform the following steps:

[0121] Determine a first difference between sample data and first sample data, and a second difference between second sample data and first sample data, wherein the first sample data is used to characterize the smallest sample data among multiple sample data, and the second sample data is used to characterize the largest sample data among multiple sample data; determine the ratio of the first difference to the second difference as the target sample data.

[0122] For example, the internal resistance of the battery may include internal resistance r1, internal resistance r2, internal resistance r3, internal resistance r4 and internal resistance r5, wherein internal resistance r2 is the largest and internal resistance r5 is the smallest. Therefore, when normalizing the internal resistance r1, the first difference between the internal resistance r1 and the internal resistance r5 and the second difference between the internal resistance r2 and the internal resistance r5 may be determined first, and then the ratio of the first difference to the second difference may be calculated to obtain the normalized internal resistance r1. In other words, the sample data may be normalized by the following formula:

[0123]

[0124] in, Indicates After normalization, the sample data No. Sample data before normalization, represents the first sample data, Represents the second sample data.

[0125] After the sample data is normalized, in order to improve the training effect of the first model, in a possible implementation, the target sample data can be divided into a training set, a test set, and a validation set according to a preset ratio, for example, a ratio of 6:2:2. The training set is used for training the first model. After the training is completed, the best hyperparameters of the first model are determined based on the validation set. After the best hyperparameters are determined, the generalization accuracy of the first model is verified based on the test set to see if it meets the requirements. When the generalization accuracy of the first model meets the requirements, the training of the first model is considered to be completed.

[0126] For example, Figure 8As shown, the discharge capacity, discharge time, voltage average, discharge platform ratio and battery internal resistance in the training set can be first input into the first model to obtain a first battery health prediction value, and then based on the first battery health prediction value and the label, the first loss function value is calculated, and the parameters of the first model are adjusted based on the first loss function value until a preset number of iterations or a preset accuracy is reached. After that, the discharge capacity, discharge time, voltage average, discharge platform ratio and battery internal resistance in the validation set can be input into the first model trained by the training set to obtain a second battery health prediction value, and then based on the second battery health prediction value and the label, the second loss function value is calculated. If the second loss function value is greater than the first loss function value, the hyperparameters of the first model, such as the learning rate, training rounds, hidden layer input and output dimensions and / or the number of hidden layers, are adjusted, and the above process is re-executed; if the second loss function value is less than or equal to the first loss function value, the discharge capacity, discharge time, voltage average, discharge platform ratio and battery internal resistance in the training set are input into the first model trained by the training set to obtain a third battery health prediction value, and then the generalization accuracy of the model is determined based on the third battery health prediction value and label calculation accuracy and / or recall rate and other evaluation indicators. When the generalization accuracy meets the preset generalization accuracy threshold, the first model training is determined to be completed. By way of example, the first model training can be determined to be completed when the error between the health prediction value of the first model and the label is less than 1%, such as Fig. 9 shown.

[0127] The loss function value may be calculated based on the mean squared error, hinge loss, or cross entropy loss function, and the disclosed embodiment does not impose any restrictions on this. In a possible implementation, the loss function value may be determined based on the mean absolute error function, that is, by substituting the battery health prediction value and the label into the following formula to obtain the loss function value:

[0128]

[0129] in, l represents the loss function value, Indicates the number of predicted values ​​of battery health. Indicates tags, Indicates The battery health prediction value.

[0130] In summary, through the above technical solution, the battery health and discharge data of the battery during the discharge cycle can be obtained, and the indicator data used to characterize the battery health status can be determined based on the discharge data, and the first model of the model can be trained based on the indicator data and the battery health status to obtain the second model for predicting the battery health status. Since the sample data of this solution can be directly determined based on the discharge data, the generation of intermediate variables and the complexity of the model training operation can be reduced, thereby improving the model training efficiency and reducing the hardware requirements for the system. In addition, since the sample data of this solution is the indicator data that characterizes the battery health status, when the model is trained based on the sample data, the model can more directly capture the key factors that affect the battery performance, thereby improving the accuracy of the prediction results.

[0131] Based on the same concept, the present disclosure also provides a method for determining a battery status. Fig.10 As shown, the battery status determination method may include:

[0132] S1001: Obtain indicator data for characterizing a battery health status of a battery.

[0133] S1002: Input the indicator data into a battery health determination model to obtain the battery health, wherein the battery health determination model is a second model trained by the above-mentioned model training method.

[0134] S1003: Determine the battery status according to the battery health.

[0135] Among them, the battery state is determined according to the battery health, and the corresponding relationship between the battery health and the battery state can be pre-built. After the battery health is predicted, the battery state can be obtained based on the battery health and the corresponding relationship.

[0136] Among them, the corresponding relationship between the battery health and the battery status can be determined according to actual conditions, and the embodiments of the present disclosure do not impose any restrictions on this.

[0137] For example, when the battery health is 100%-85%, the battery status can be set to a healthy state; when the battery health is 85%-60%, the battery status can be set to a medium healthy state; when the battery health is 60%-40%, the battery status can be set to an aging state; when the battery health is less than 40%, the battery status can be set to a replacement required state.

[0138] In summary, through the above technical scheme, the indicator data used to characterize the battery health state can be input into the battery health determination model to obtain the battery health, and then the battery state can be determined according to the battery health. In this way, the battery state can be quickly determined. In addition, since the sample data can be directly determined based on the discharge data when training the battery health determination model based on the above method, the generation of intermediate variables and the complexity of the model training operation can be reduced, thereby improving the model training efficiency and reducing the hardware requirements for the system. In addition, since the sample data of this scheme is the indicator data that characterizes the battery health state, when the model is trained based on the sample data, the model can more directly capture the key factors that affect the battery performance, thereby improving the accuracy of the model prediction results.

[0139] In a possible manner, before the indicator data is input into the battery health determination model, the following may also be included:

[0140] The sample data obtained in each discharge cycle is stored in a target database for storing the sample data; the battery health determination model is trained based on the sample data in the target database to obtain the battery health determination model with updated parameters;

[0141] Accordingly, the indicator data is input into the battery health determination model to obtain the battery health, which may include:

[0142] The indicator data is input into the battery health determination model after the parameters are updated to obtain the battery health.

[0143] The method of obtaining the sample data in each discharge cycle can refer to the above-mentioned related description, which will not be repeated here.

[0144] It should be understood that since battery aging is affected by many factors, in order to improve the prediction accuracy of battery health, this embodiment continuously expands the training data set by storing the battery health and corresponding indicator data of each discharge cycle, so that when the battery status evaluation is required, all stored battery health and indicator data can be used as sample data, and the battery health determination model can be retrained, so that the parameters of the battery health determination model can be updated according to the actual attenuation of the battery, and then a more accurate health estimation result can be obtained based on the battery health determination model with updated parameters.

[0145] In a possible implementation, a battery status determination system may also be pre-built, such as Fig.11As shown, the battery status determination system 1100 may include a data acquisition unit 1101, a feature extraction unit 1102, a deep learning unit 1103 and a prediction execution unit 1104. When the battery undergoes a complete charge and discharge cycle, the data acquisition unit 1101 collects discharge data such as the maximum voltage value, the minimum voltage value, the discharge current and the discharge time of the battery. Thereafter, the feature extraction unit 1102 processes the collected discharge data to obtain indicator data, and accumulates the saved indicator data to obtain a characteristic value data set or an indicator data set. When it is necessary to evaluate the battery status, the deep learning unit 1103 trains the first model based on the characteristic value data set or the indicator data set to obtain a second model. Thereafter, the prediction execution unit 1104 estimates the battery health based on the acquired indicator data and the second model to obtain the battery health, and determines the battery status based on the battery health, such as Fig.12 shown.

[0146] In a possible manner, in order to further reduce the occupancy of system memory, after the feature extraction unit 1102 calculates the indicator data based on the discharge data such as the maximum voltage value, the minimum voltage value, the discharge current and the discharge time collected by the data acquisition unit 1101, it can delete the discharge data collected by the data acquisition unit 1101, thereby further reducing the hardware requirements for the battery status determination system 1100.

[0147] Based on the same concept, the present disclosure also provides a model training device, such as Fig.13 As shown, the model training device 1300 may include:

[0148] The first acquisition module 1301 is used to acquire sample data marked with labels;

[0149] A first training module 1302 is used to train the first model based on sample data to obtain a second model for predicting battery health;

[0150] The first acquisition module 1301 may include:

[0151] The acquisition submodule is used to obtain the battery health and discharge data of the battery during the discharge cycle;

[0152] A first determination submodule, used to determine indicator data for characterizing a battery health state of the battery according to the discharge data;

[0153] The second determination submodule is used to use the battery health as a label of the indicator data to obtain sample data.

[0154] Through the above-mentioned model training device 1300, the battery health and discharge data of the battery during the discharge cycle can be obtained, and the indicator data used to characterize the battery health status can be determined based on the discharge data, and the first model of the model can be trained based on the indicator data and the battery health status to obtain a second model for predicting the battery health status. Since the sample data of the present solution can be directly determined based on the discharge data, the generation of intermediate variables and the complexity of the model training operation can be reduced, thereby improving the model training efficiency and reducing the hardware requirements for the system. In addition, since the sample data of the present solution is the indicator data characterizing the battery health status, when the model is trained based on the sample data, the model can more directly capture the key factors affecting the battery performance, thereby improving the accuracy of the model prediction results.

[0155] In a possible manner, the indicator data includes discharge capacity, discharge time, average voltage, discharge platform ratio and / or battery internal resistance, wherein the discharge platform ratio is used to characterize the proportion of the curve interval in the battery discharge curve where the voltage change rate is less than a preset voltage change rate threshold, and the discharge curve is determined based on the discharge data.

[0156] In a possible manner, the discharge data may include a sampling interval and a sampling voltage. Accordingly, the first determination submodule may include:

[0157] A first determining unit, used to determine a voltage change rate according to a sampling interval and a sampling voltage;

[0158] A processing unit, used for processing the voltage change rate based on a round-down algorithm to obtain a target voltage change rate;

[0159] The second determining unit is used to determine the discharge platform proportion based on the target voltage change rate.

[0160] In a possible manner, the target voltage change rate includes multiple ones, and accordingly, the second determining unit may include:

[0161] A first determining subunit is used to determine the number of target voltage change rates that are preset voltage change rates among a plurality of target voltage change rates;

[0162] The second determining subunit is used to determine the ratio of the number of preset voltage change rates and the number of target voltage change rates as the discharge platform proportion.

[0163] In a possible manner, the first training module 1302 may include:

[0164] A processing submodule is used to normalize the sample data to obtain target sample data;

[0165] The training submodule is used to train the first model based on the target sample data to obtain a second model for predicting the battery health.

[0166] In a possible manner, the processing submodule may include:

[0167] a second determining unit, configured to determine a first difference between the sample data and the first sample data, and a second difference between the second sample data and the first sample data, wherein the first sample data is used to represent the smallest sample data among the plurality of sample data, and the second sample data is used to represent the largest sample data among the plurality of sample data;

[0168] The third determining unit is used to determine the ratio of the first difference value to the second difference value as the target sample data.

[0169] Based on the same concept, the embodiment of the present disclosure also provides a battery status determination device, such as Fig.14 As shown, the battery status determining device 1400 may include:

[0170] The second acquisition module 1401 is used to acquire indicator data for representing the battery health status of the battery;

[0171] A processing module 1402 is used to input the indicator data into a battery health determination model to obtain the battery health, wherein the battery health determination model is a second model trained by the above-mentioned model training device;

[0172] The determination module 1403 is used to determine the battery status according to the battery health.

[0173] Through the above-mentioned battery status determination device 1400, the indicator data used to characterize the battery health status can be input into the battery health determination model to obtain the battery health, and then the battery status can be determined according to the battery health. In this way, the battery status can be quickly determined. In addition, since the sample data can be directly determined based on the discharge data when training the battery health determination model based on the above-mentioned method, the generation of intermediate variables and the complexity of the model training operation can be reduced, thereby improving the model training efficiency and reducing the hardware requirements for the system. In addition, since the sample data of this scheme is the indicator data characterizing the battery health status, when the model is trained based on the sample data, the model can more directly capture the key factors affecting the battery performance, thereby improving the accuracy of the model prediction results.

[0174] In a possible manner, the battery status determining device 1400 may further include:

[0175] A storage module, used for storing the sample data obtained in each discharge cycle into a target database for storing the sample data;

[0176] A second training module is used to train the battery health determination model based on the sample data in the target database to obtain the battery health determination model with updated parameters;

[0177] Accordingly, the processing module 1402 is used to input the indicator data into the battery health determination model after the parameters are updated to obtain the battery health.

[0178] Based on the same concept, an embodiment of the present disclosure also provides a non-temporary computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned model training method or battery status determination method are implemented.

[0179] Based on the same concept, an embodiment of the present disclosure also provides a computer program product, including a computer program, which implements the steps of the above-mentioned model training method or battery status determination method when executed by a processor.

[0180] Based on the same concept, the embodiment of the present disclosure further provides an electronic device, including:

[0181] a memory having a computer program stored thereon;

[0182] A processor is used to execute a computer program in a memory to implement the steps of the above-mentioned model training method or battery status determination method.

[0183] Fig.15 1 is a block diagram of an electronic device 1500 according to an exemplary embodiment. For example, the electronic device 1500 may be provided as a server. Fig.15 The electronic device 1500 includes a processor 1522, which may be one or more, and a memory 1532 for storing a computer program executable by the processor 1522. The computer program stored in the memory 1532 may include one or more modules each corresponding to a set of instructions. In addition, the processor 1522 may be configured to execute the computer program to perform the above-mentioned model training method or battery state determination method.

[0184] In addition, the electronic device 1500 may further include a power supply component 1526 and a communication component 1550, wherein the power supply component 1526 may be configured to perform power management of the electronic device 1500, and the communication component 1550 may be configured to implement communication of the electronic device 1500, for example, wired or wireless communication. In addition, the electronic device 1500 may further include an input / output (I / O) interface 1558. The electronic device 1500 may operate based on an operating system stored in the memory 1532.

[0185] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by the processor, the steps of the above-mentioned model training method or battery state determination method are implemented. For example, the non-temporary computer-readable storage medium can be the above-mentioned memory 1532 including program instructions, and the above-mentioned program instructions can be executed by the processor 1522 of the electronic device 1500 to complete the above-mentioned model training method or battery state determination method.

[0186] The preferred embodiments of the present disclosure are described in detail above in conjunction with the accompanying drawings; however, the present disclosure is not limited to the specific details in the above embodiments. Within the technical concept of the present disclosure, a variety of simple modifications can be made to the technical solution of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0187] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations.

[0188] In addition, various embodiments of the present disclosure may be arbitrarily combined, and as long as they do not violate the concept of the present disclosure, they should also be regarded as the contents disclosed by the present disclosure.

Claims

1. A model training method, characterized in that: The model training method comprises: Get sample data with labels; Training the first model based on the sample data to obtain a second model for predicting battery health; The sample data is obtained in the following way: Obtain battery health and discharge data during the discharge cycle; Determine, according to the discharge data, indicator data for characterizing the battery health status of the battery, the indicator data including discharge capacity, discharge duration, voltage average value, discharge platform ratio and / or battery internal resistance, wherein the discharge platform ratio is used to characterize the proportion of a curve interval in a battery discharge curve whose voltage change rate is less than a preset voltage change rate threshold in the discharge curve, and the discharge curve is determined based on the discharge data; The battery health is used as a label of the indicator data to obtain the sample data.

2. The model training method according to claim 1, characterized in that: The discharge data includes a sampling interval and a sampling voltage, and determining the discharge platform proportion according to the discharge data includes: Determining a voltage change rate according to the sampling interval and the sampling voltage; Processing the voltage change rate based on a round-down algorithm to obtain a target voltage change rate; Based on the target voltage change rate, a discharge platform proportion is determined.

3. The model training method according to claim 2, characterized in that: The target voltage change rate includes multiple ones, and determining the discharge platform proportion based on the target voltage change rate includes: Among the plurality of target voltage change rates, determining the target voltage change rate to be the number of preset voltage change rates; The ratio of the target voltage change rate to the number of preset voltage change rates and the target voltage change rate is determined as the discharge platform proportion.

4. The model training method according to any one of claims 1 to 3, characterized in that: The training of the first model based on the sample data to obtain a second model for predicting the battery health includes: Normalizing the sample data to obtain target sample data; The first model is trained based on the target sample data to obtain a second model for predicting battery health.

5. The model training method according to claim 4, characterized in that: The sample data includes a plurality of data, and the normalization process is performed on the sample data to obtain target sample data, including: For each sample data, perform the following steps: Determine a first difference between the sample data and the first sample data, and a second difference between the second sample data and the first sample data, wherein the first sample data is used to represent the smallest sample data among the plurality of sample data, and the second sample data is used to represent the largest sample data among the plurality of sample data; The ratio of the first difference value to the second difference value is determined as target sample data.

6. A method for determining a battery status, characterized in that: The battery status determination method comprises: Obtaining indicator data for characterizing the battery health status of the battery; Inputting the indicator data into a battery health determination model to obtain battery health, wherein the battery health determination model is a second model trained by the model training method according to any one of claims 1 to 5; A battery state is determined according to the battery health.

7. The method for determining the battery status according to claim 6, characterized in that: Before inputting the indicator data into the battery health determination model, the method further includes: storing the sample data obtained in each discharge cycle into a target database for storing the sample data; Training the battery health determination model based on the sample data in the target database to obtain a battery health determination model with updated parameters; The step of inputting the indicator data into a battery health determination model to obtain the battery health includes: The indicator data is input into the battery health determination model after the parameter update to obtain the battery health.

8. An electronic device, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images