Battery static voltage prediction method, electronic device, and storage medium

Abstract

The application provides a battery static voltage prediction method, an electronic device and a storage medium, including: dividing vehicle historical operation data to obtain a plurality of target segments; each target segment includes a static segment and a charging segment, and the static segment and the charging segment are distinguished and connected by a charging flag; a plurality of feature parameters are extracted from each target segment to construct a training sample set; the plurality of feature parameters include static voltage extracted from the static segment, dynamic voltage extracted from the charging segment, a plurality of instantaneous working condition features and a plurality of cumulative working condition features; a preset model is trained based on the training sample set to obtain a static voltage prediction model; and the static voltage prediction model is used to calculate a plurality of instantaneous working condition data and a plurality of cumulative working condition data of a target vehicle under a charging condition to output a predicted battery static voltage value. The application can estimate the static voltage online based on the charging condition to improve the static voltage calculation coverage.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of batteries, in particular to a battery static voltage prediction method, an electronic device and a storage medium. BACKGROUND

[0002] A battery is a core component of a new energy vehicle, and its performance will decrease over time during use. In order to ensure the working reliability of the battery, the SOC (State of Charge) and SOH (State of Health) of the battery need to be monitored in real time and accurately.

[0003] The SOC prediction method of the battery is extremely dependent on the static voltage (OCV), and the SOH prediction method is indirectly dependent on the static voltage. In the related art, the static voltage is usually obtained by an offline detection method, that is, the static voltage is obtained in a state where the battery is at rest.

[0004] However, in a real vehicle working condition scenario, the battery is in various dynamic working conditions (such as driving, charging, standby power consumption) most of the time, and the state where the battery is at rest is not much, and the opportunity to directly measure and obtain the static voltage is less, resulting in low static voltage calculation coverage, and high-frequency estimation of the static voltage cannot be achieved. SUMMARY

[0005] In view of the above problems, the present application provides a battery static voltage prediction method, an electronic device and a storage medium, which can solve the problem of low static voltage calculation coverage in the related art.

[0006] In a first aspect, the present application provides a battery static voltage prediction method, comprising: dividing vehicle historical running data to obtain a plurality of target segments; wherein each target segment includes a resting segment and a charging segment, and the resting segment and the charging segment are distinguished and connected by a charging flag; extracting a plurality of feature parameters from each target segment to construct a training sample set; wherein the plurality of feature parameters include a static voltage extracted from the resting segment and a dynamic voltage, a plurality of instantaneous working condition features and a plurality of cumulative working condition features extracted from the charging segment; training a preset model based on the training sample set to obtain a static voltage prediction model; and calculating a plurality of instantaneous working condition data and a plurality of cumulative working condition data of a target vehicle in a charging working condition by the static voltage prediction model to output a predicted battery static voltage value.

[0007] In the above scheme, by dividing the vehicle historical operation data to obtain a plurality of target segments, and making each target segment include one static segment and one charging segment distinguished and connected by a charging flag bit, the charging segment can be directly associated with the static segment before it, so that the charging segment and the static segment have a highly accurate and reliable corresponding relationship, thereby improving the data quality of the target segment. Further, a plurality of feature parameters are extracted from each target segment to construct a training sample set, and the plurality of feature parameters include static voltage extracted from the static segment, dynamic voltage, a plurality of instantaneous working condition characteristics and a plurality of cumulative working condition characteristics extracted from the charging segment, so that the sample training set includes a plurality of feature parameters that are most core and most relevant to estimating the static voltage, thereby improving the training performance of the sample training set. Further, the preset model is trained based on the training sample set to obtain a static voltage prediction model, which can optimize the preset model based on a plurality of high-quality training samples, thereby obtaining a static voltage prediction model with high precision and strong generalization ability. Further, the plurality of instantaneous working condition data and the plurality of cumulative working condition data of the target vehicle under the charging condition are calculated by the static voltage prediction model to output a predicted battery static voltage value, which can break through the static condition restriction, so that the static voltage prediction model estimates the static voltage online under the charging condition, thereby improving the acquisition frequency of the static voltage, and then improving the static voltage calculation coverage, to realize high-precision and online applicable static voltage estimation.

[0008] In some embodiments, the step of extracting a plurality of feature parameters from each target segment to construct a training sample set includes: extracting the static voltage corresponding to the end of the static segment as the static voltage, and extracting the dynamic voltage corresponding to the starting time of the charging segment as the dynamic voltage; extracting the state of charge, temperature and current corresponding to the starting time of the charging segment to obtain a plurality of instantaneous working condition characteristics; and extracting the cumulative mileage, cumulative effective cycle number and operation life corresponding to the starting time of the charging segment to obtain a plurality of cumulative working condition characteristics; taking the static voltage, dynamic voltage, a plurality of instantaneous working condition characteristics and a plurality of cumulative working condition characteristics corresponding to each target segment as a group of training data; and constructing a training sample set based on a plurality of groups of training data corresponding to a plurality of target segments.

[0009] In the above scheme, by extracting the static voltage corresponding to the end of the static segment as the static voltage and extracting the dynamic voltage corresponding to the starting time of the charging segment as the dynamic voltage, not only can the static voltage and the dynamic voltage be obtained through two continuous data frames with extremely short time intervals, so as to minimize the difference between the dynamic voltage and the static voltage, but also the static voltage obtained is the voltage closest to the true value of the static voltage. Moreover, since the time points of obtaining the static voltage and the dynamic voltage are very close, it can also be considered that the multiple cumulative working condition characteristics corresponding to the static voltage and the dynamic voltage are the same. Further, by extracting the state of charge, temperature and current corresponding to the starting time of the charging segment, and extracting the cumulative mileage, cumulative effective cycle number and operation age corresponding to the starting time of the charging segment, multiple-dimensional characteristics reflecting the instantaneous state and reflecting the aging degree can be obtained. Further, by taking the static voltage, dynamic voltage, multiple instantaneous working condition characteristics and multiple cumulative working condition characteristics corresponding to each target segment as a group of training data, and constructing a training sample set based on multiple groups of training data corresponding to multiple target segments, the instantaneous state characteristics, aging characteristics and other information after the change of the battery state can be associated with the static voltage, thereby constructing a sample training set with high quality and high correlation. Further, since the battery aging will affect the shape of the static voltage-state of charge curve, including multiple cumulative working condition characteristics in the sample training set can also train the model to capture the offset of the aging degree to the static voltage, so that the finally obtained static voltage prediction model can track the whole life cycle of the battery.

[0010] In some embodiments, the step of constructing a training sample set based on multiple groups of training data corresponding to multiple target segments comprises: performing down-sampling processing on the multiple groups of training data based on the current rate interval and the temperature interval to obtain multiple groups of training samples; performing normalization processing on the multiple groups of training samples, and constructing a training sample set based on the multiple groups of training samples after normalization processing.

[0011] In the above scheme, by performing down-sampling processing on the multiple groups of training data based on the current rate interval and the temperature interval to obtain multiple groups of training samples, the sample number of training data obtained based on different rate intervals and different temperature intervals can be balanced, so that the model can better learn and adapt to the voltage variation law under different working conditions. Further, by performing normalization processing on the multiple groups of training samples, and constructing a training sample set based on the multiple groups of training samples after normalization processing, the preset model can be prevented from producing unfair punishment on different dimensional characteristics during subsequent training, thereby improving the reliability of the training sample set.

[0012] In some embodiments, the step of calculating the plurality of instantaneous working condition data and the plurality of cumulative working condition data of the target vehicle in the charging working condition by the static voltage prediction model to output the predicted battery static voltage value comprises: obtaining the dynamic voltage, the battery state of charge, the temperature, and the current corresponding to the starting moment of the charging working condition of the target vehicle to obtain the plurality of instantaneous working condition data; and obtaining the cumulative mileage, the cumulative effective cycle number, and the running age corresponding to the starting moment of the charging working condition of the target vehicle to obtain the plurality of cumulative working condition data; and calculating the plurality of instantaneous working condition data and the plurality of cumulative working condition data by the static voltage prediction model to output the predicted battery static voltage value.

[0013] In the above scheme, by immediately obtaining the plurality of instantaneous working condition data and the plurality of cumulative working condition data at the moment when the target vehicle starts charging each time, and outputting the predicted battery static voltage value by the static voltage prediction model, the static voltage acquisition can be changed from a low-frequency occasional event to a high-frequency regular event, thereby providing more data support for real-time and accurate monitoring of the battery state. Further, by making the plurality of instantaneous working condition data the dynamic voltage, the battery state of charge, the temperature, and the current, and making the plurality of cumulative working condition data the cumulative mileage, the cumulative effective cycle number, and the running age, the feature types, the timing points (the starting moment of the charging segment), and the physical meanings collected in online application are completely consistent with those when the training samples are constructed, thereby improving the accuracy and reliability of the static voltage prediction, reducing the safety hazards of the battery, and improving the safety and reliability of the target vehicle.

[0014] In some embodiments, the preset model is an elastic net model; and the step of training the preset model based on the training sample set to obtain the static voltage prediction model comprises: taking the dynamic voltage corresponding to each target segment, the plurality of instantaneous working condition features, and the plurality of cumulative working condition features as a group of input features, taking the static voltage corresponding to each group of input features as an output label; and training the elastic net model based on the plurality of groups of input features and the plurality of output labels corresponding to the plurality of target segments to obtain the static voltage prediction model.

[0015] In the above scheme, taking the dynamic voltage corresponding to each target segment, the plurality of instantaneous working condition features, and the plurality of cumulative working condition features as a group of input features, taking the static voltage corresponding to each group of input features as an output label, and training the elastic net model based on the plurality of groups of input features and the plurality of output labels, can filter out the most core features for static voltage prediction by the mechanism of the elastic net model to simplify the model, and effectively handle the possible multicollinearity between the features to improve the stability of the model, so that the obtained static voltage prediction model is a linear model with simple structure, stable performance, and strong generalization ability, which is convenient for deployment in a vehicle-mounted environment to provide efficient and reliable prediction data.

[0016] In some embodiments, the step of training the elastic net model based on the plurality of sets of input features corresponding to the plurality of target segments and the plurality of output labels to obtain the static voltage prediction model comprises: constructing a target function corresponding to the elastic net model; wherein the target function comprises a prediction error loss term and a regularization term; the prediction error loss term comprises a static voltage predicted value and a static voltage true value, the static voltage predicted value is calculated by an input feature matrix, a feature weight coefficient vector corresponding to the input feature matrix, and a model intercept, and the regularization term is calculated by the feature weight coefficient vector; inputting the plurality of sets of input features corresponding to the plurality of target segments and the plurality of output labels into the elastic net model, taking each set of input features as an input feature matrix, and taking the input label corresponding to each set of input features as the static voltage true value, to calculate the minimum value of the target function; generating a linear prediction formula based on the feature weight coefficient vector and the model intercept corresponding to the minimum value; and verifying the linear prediction formula by using a ten-fold cross-validation method, and determining the linear prediction formula as the static voltage prediction model after verification.

[0017] In the above scheme, by making the target function corresponding to the elastic net model comprise a prediction error loss term and a regularization term, and making the prediction error loss term comprise a static voltage predicted value and a static voltage true value, and making the static voltage predicted value be calculated by an input feature matrix, a feature weight coefficient vector corresponding to the input feature matrix, and a model intercept, and the regularization term be calculated by the feature weight coefficient vector, the difference between the static voltage predicted value and the static voltage true value can be directly measured by the prediction error loss term, and excessive weight coefficients can be penalized by the regularization term to avoid model overfitting. Further, by inputting the plurality of sets of input features corresponding to the plurality of target segments and the plurality of output labels into the elastic net model, taking each set of input features as an input feature matrix, and taking the input label corresponding to each set of input features as the static voltage true value, to calculate the minimum value of the target function, the global optimal solution can be found based on the solved minimum value, and the obtained feature weight coefficient vector is stable and unique, so that each weight coefficient in the feature weight coefficient vector can accurately quantify the contribution degree and influence direction of each corresponding feature in the matrix feature. Further, generating a linear prediction formula based on the feature weight coefficient vector and the model intercept corresponding to the minimum value can make the optimal solution be presented through a clear and reliable linear formula, thereby reducing the complexity of the model and improving the accuracy of the model. Further, verifying the linear prediction formula by using a ten-fold cross-validation method, and determining the linear prediction formula as the static voltage prediction model after verification can further improve the reliability of the static voltage prediction model.

[0018] In some embodiments, the step of adopting the ten-fold cross-validation method to perform model verification on the linear prediction formula, and determining the linear prediction formula as the static voltage prediction model after verification, comprises: dividing the training sample set into ten non-overlapping subsets at random; taking each subset as the test set in turn, and taking the remaining nine subsets as the training set, to perform ten times of training and testing on the linear prediction formula, to obtain ten average absolute errors and ten determination coefficients; obtaining a first average value based on the ten average absolute errors, and obtaining a second average value based on the ten determination coefficients; in response to the first average value being greater than or equal to a first preset threshold value, and the second average value being greater than or equal to a second preset threshold value, determining the linear prediction formula as the static voltage prediction model; in response to the first average value being less than the first preset threshold value, and / or the second average value being less than the second preset threshold value, retraining the elastic net model, and performing model verification on the linear prediction formula obtained after retraining.

[0019] In the above scheme, by adopting the ten-fold cross-validation method to perform model verification on the linear prediction formula, all data in the training sample set can be used for training and verification, and through ten cycles, each training sample has a time as the test set, so as to maximize the performance of the linear prediction formula on unknown data. Further, by calculating the first average value and the second average value corresponding to the ten results, the evaluation variance can be effectively reduced, so that the performance indicators obtained are more stable and reliable. Further, when the first average value and the second average value both satisfy the threshold condition, the linear prediction formula is determined as the static voltage prediction model, which can ensure that only high-quality models with performance up to standard can be finally confirmed and deployed, thereby improving the reliability and safety of the static voltage prediction model, so as to more accurately predict the static voltage. When the first average value and / or the second average value do not satisfy the corresponding threshold condition, the elastic net model is retrained, and the linear prediction formula obtained after retraining is verified, which can realize iterative improvement and automatic optimization of the model based on the closed loop of “verification-feedback-retraining”, thereby improving the robustness and reliability of model training.

[0020] In some embodiments, the step of dividing the vehicle historical operation data to obtain a plurality of target segments comprises: scanning the vehicle historical operation data, identifying a time point at which a charging flag bit jumps from a first state to a second state, and taking the time point as a charging starting point; based on the charging starting time point, scanning the vehicle historical operation data backward until the charging flag bit jumps from the second state back to the first state, to extract a data segment in the corresponding interval as a charging segment; the charging starting point is the starting time point of the charging segment; based on the charging starting time point, tracing the vehicle historical operation data forward to extract a data segment satisfying the static condition as a static segment; based on each charging starting point, associating the corresponding static segment and the charging segment to obtain a plurality of target segments.

[0021] In the above scheme, by scanning the vehicle historical operation data, the time point at which the charging flag bit jumps from the first state to the second state is identified, and the time point is taken as the charging starting point, so as to extract the static segment and the charging segment respectively based on the charging starting point, each charging condition can be accurately associated with the static condition before it, and at the moment of the charging starting point, the aging state of the battery is almost completely the same as the aging state at the end of the static segment, which can minimize variable interference. And since the selection of the static segment needs to meet the static condition, the quality of the static segment can be improved to ensure that the voltage at the end of the static segment is the voltage that is fully relaxed and closest to the static voltage true value, thereby improving the data quality of the training sample.

[0022] In some embodiments, before the step of dividing the vehicle historical operation data to extract the static segment and the charging segment, the method further includes: obtaining vehicle historical raw data; performing a preprocessing operation on the vehicle historical raw data to obtain the vehicle historical operation data; wherein the preprocessing operation includes at least one of outlier rejection, null value filling, state of charge jump cleaning, time jump cleaning, and data format standardization.

[0023] In the above scheme, by performing at least one preprocessing operation on the vehicle raw data, the abnormal values and jump parameters in the vehicle raw data can be removed, and the data missing problem can be solved, so as to improve the accuracy and consistency of the data.

[0024] In a second aspect, the present application provides an electronic device, comprising: a memory for storing program data, the program data being executed to implement the steps of the battery static voltage prediction method as described above; and a processor for executing the program instructions stored in the memory to implement the steps of the battery static voltage prediction method as described above.

[0025] In a third aspect, the present application provides a computer readable storage medium, the computer readable storage medium storing a computer program, the computer program being executed by a processor to implement the steps of the battery static voltage prediction method as described above.

[0026] The beneficial effects of the present application are: Different from the related art, the present application provides a battery static voltage prediction method, an electronic device and a storage medium. By dividing the historical operation data of the vehicle, a plurality of target segments are obtained, and each target segment includes one static segment and one charging segment distinguished and connected by a charging flag bit. The charging segment can be directly associated with the static segment before it, so that the corresponding relationship between the charging segment and the static segment is highly accurate and reliable, thereby improving the data quality of the target segment. Further, a plurality of feature parameters are extracted from each target segment to construct a training sample set, and the plurality of feature parameters include static voltage extracted from the static segment, dynamic voltage extracted from the charging segment, a plurality of instantaneous working condition characteristics and a plurality of cumulative working condition characteristics. The sample training set can include a plurality of feature parameters that are most core and most relevant to estimating static voltage, thereby improving the training performance of the sample training set. Further, the preset model is trained based on the training sample set to obtain a static voltage prediction model. The preset model can be optimized based on a plurality of training samples of high quality, thereby obtaining a static voltage prediction model with high precision and strong generalization ability. Further, the plurality of instantaneous working condition data and the plurality of cumulative working condition data of the target vehicle under the charging condition are calculated by the static voltage prediction model to output a predicted battery static voltage value. The static voltage prediction model can break through the static condition restriction and estimate the static voltage online under the charging condition, thereby improving the acquisition frequency of the static voltage, and then improving the static voltage calculation coverage, so as to realize high-precision and online static voltage estimation. BRIEF DESCRIPTION OF DRAWINGS

[0027] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the specification, serve to explain the technical solutions of the present application.

[0028] Figure 1 is a flowchart of the first embodiment of the battery static voltage prediction method of the present application;

[0029] Figure 2 is a flowchart of the second embodiment of the battery static voltage prediction method of the present application;

[0030] Figure 3 is a flowchart of the third embodiment of the battery static voltage prediction method of the present application;

[0031] Figure 4 is a flowchart of the fourth embodiment of the battery static voltage prediction method of the present application;

[0032] Figure 5 is a flowchart of the fifth embodiment of the battery static voltage prediction method of the present application;

[0033] Figure 6 is a flowchart of a sixth embodiment of the battery static voltage prediction method of the present application;

[0034] Figure 7 is a workflow diagram of an application scenario of the battery static voltage prediction method of the present application;

[0035] Figure 8 is a schematic diagram of the verification result of the static voltage prediction model in Embodiment 1 of the present application;

[0036] Figure 9 is a structural schematic diagram of an embodiment of the electronic device of the present application;

[0037] Figure 10 is a structural schematic diagram of an embodiment of the computer-readable storage medium of the present application.

[0038] Reference signs:

[0039] 90 - electronic device, 91 - memory, 92 - processor;

[0040] 100 - computer-readable storage medium, 101 - computer program. DETAILED DESCRIPTION

[0041] Hereinafter, embodiments of the battery cell, the battery, and the electric device of the present application are specifically disclosed with appropriate reference to the accompanying drawings. However, there will be cases where unnecessary detailed description is omitted. For example, there will be cases where detailed description of matters that are well known, repeated description of substantially the same structure, are omitted. This is to avoid the following description from becoming unnecessarily lengthy and to facilitate understanding by those skilled in the art. Furthermore, the accompanying drawings and the following description are provided so that those skilled in the art can fully understand the present application, and are not intended to limit the subject matter recited in the claims.

[0042] The ranges disclosed herein are intended to be "open" ranges, i.e., the end values are not included in the range. For example, if a range is listed as 60-120 and 80-110, it is intended that 60-110 and 80-120 are also contemplated. In addition, if a minimum range value of 1 and 2 is listed, and if a maximum range value of 3, 4, and 5 is listed, then the following ranges are all contemplated: 1-3, 1-4, 1-5, 2-3, 2-4, and 2-5. In this application, unless otherwise indicated, a numerical range "a-b" indicates a shorthand way of describing each and every number that is contained in the range, wherein a and b are both real numbers. For example, the numerical range "0-5" indicates that all real numbers between 0 and 5 are contemplated, and "0-5" is merely a shorthand way of describing such numerical range. Also, when a parameter is stated to be an integer ≥ 2, it is equivalent to disclose that the parameter is, for example, integer 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, etc.

[0043] All embodiments and optional embodiments of the present application can be combined with each other to form new technical solutions, unless otherwise specified.

[0044] All technical features and optional technical features of the present application can be combined with each other to form new technical solutions, unless otherwise specified.

[0045] All steps of the present application can be performed in sequence or randomly, preferably in sequence, unless otherwise specified. For example, a method comprises steps (a) and (b) means that the method can comprise steps (a) and (b) in sequence, or steps (b) and (a) in sequence. For example, it is mentioned that the method can further comprise step (c), which means that step (c) can be added to the method in any order, for example, the method can comprise steps (a), (b) and (c), or steps (a), (c) and (b), or steps (c), (a) and (b), etc.

[0046] Unless otherwise specified, "including" and "comprising" mentioned in the present application means open-ended, and can also be closed-ended. For example, "including" and "comprising" can mean that other components not listed can also be included or contained, or only the listed components can be included or contained.

[0047] If not specifically stated, the term "or" is inclusive in this application, by way of example, the phrase "A or B" means "A, B, or both A and B." More specifically, any of the following satisfy the condition "A or B": A is true (or present) and B is false (or not present); A is false (or not present) and B is true (or present); or both A and B are true (or present).

[0048] To ensure the reliability of the battery, the SOC (State of Charge) and the SOH (State of Health) of the battery need to be monitored in real time and accurately. The SOC prediction method of the battery is extremely dependent on the static voltage (OCV), and the SOH prediction method is indirectly dependent on the static voltage. In the related art, the static voltage is usually obtained by offline detection, that is, in the state of static battery.

[0049] However, in the real vehicle working condition scene, the battery is in various dynamic working conditions (such as driving, charging, standby power consumption) most of the time, and the state of static battery is not much, and the opportunity of directly measuring the static voltage is less, resulting in low static voltage calculation coverage, and the high-frequency estimation of the static voltage cannot be realized.

[0050] In order to solve the problem of low static voltage calculation coverage in the related art, by dividing the historical running data of the vehicle, a plurality of target segments are obtained, and each target segment includes a static segment and a charging segment distinguished and connected by a charging flag. The charging segment can be directly associated with the static segment before it, so that the charging segment and the static segment have a highly accurate and reliable corresponding relationship, thereby improving the data quality of the target segment. Further, a plurality of feature parameters are extracted from each target segment to construct a training sample set, and the plurality of feature parameters include a static voltage extracted from the static segment and a dynamic voltage, a plurality of instantaneous working condition characteristics and a plurality of cumulative working condition characteristics extracted from the charging segment. The sample training set can include a plurality of core and most relevant feature parameters for estimating the static voltage, thereby improving the training performance of the sample training set. Further, the preset model is trained based on the training sample set to obtain a static voltage prediction model, which can optimize the preset model based on a plurality of high-quality training samples, thereby obtaining a static voltage prediction model with high precision and strong generalization ability. Further, the plurality of instantaneous working condition data and the plurality of cumulative working condition data of the target vehicle in the charging working condition are calculated by the static voltage prediction model to output a predicted battery static voltage value, which can break through the static condition restriction and enable the static voltage prediction model to estimate the static voltage online in the charging working condition, thereby improving the acquisition frequency of the static voltage, and in turn improving the static voltage calculation coverage, to realize high-precision and online applicable static voltage estimation.

[0051] The battery static voltage prediction method disclosed by the embodiments of the present application can be used for health state prediction of different types of lithium batteries in the new energy vehicle industry. The different types of lithium batteries can be, but are not limited to, lithium iron phosphate systems or ternary lithium systems.

[0052] To this end, the first embodiment of the present application provides a battery static voltage prediction method, as shown in the following formula (1), comprising: Figure 1

[0053] S101: dividing vehicle historical operation data to obtain a plurality of target segments; wherein each target segment comprises a static segment and a charging segment, and the static segment and the charging segment are distinguished and connected by a charging flag.

[0054] In the embodiment, the static segment is a segment that meets the static condition.

[0055] Specifically, the static condition is that the charging flag is continuously 0, the vehicle speed is continuously 0, and the absolute value of the current is less than a preset threshold.

[0056] Specifically, the charging flag being 0 indicates that the vehicle is in a non-charging working condition (including driving, static, and engine off), and the charging flag being 1 indicates that the vehicle is in a charging working condition.

[0057] Specifically, the preset threshold is 10 mA. The static current of a new energy vehicle is usually less than 10 mA after the vehicle is locked and hibernated. By setting the preset threshold to 10 mA, it can be better determined whether the vehicle is in a static working condition.

[0058] In the embodiment, the time point when the charging flag changes from 0 to 1 is the starting time of the charging segment, and the charging flag in the charging segment is continuously 1.

[0059] S102: extracting a plurality of feature parameters from each target segment to construct a training sample set; wherein the plurality of feature parameters comprise a static voltage extracted from the static segment and a dynamic voltage, a plurality of instantaneous working condition features and a plurality of cumulative working condition features extracted from the charging segment.

[0060] Specifically, the static voltage refers to the voltage of the battery in a static state, which is only affected by chemical reactions, and usually reflects the SOC and SOH of the battery.

[0061] Specifically, the dynamic voltage refers to the voltage of the battery when current flows. When current flows through the battery, the resistance and reactance inside the battery will produce voltage drop, resulting in a dynamic voltage lower than the static voltage.

[0062] S103: training a preset model based on the training sample set to obtain a static voltage prediction model.

[0063] ​S104: calculating the multiple instantaneous working condition data and the multiple cumulative working condition data of the target vehicle under the charging working condition by the static voltage prediction model, to output a predicted battery static voltage value.

[0064] In the technical solution of the embodiment, the vehicle historical running data is divided to obtain multiple target segments, and each target segment includes a static segment and a charging segment distinguished and connected by a charging flag. The charging segment and the static segment before it are directly associated, so that the target segments have a highly accurate and reliable corresponding relationship, thereby improving the data quality of the target segments. Further, multiple feature parameters are extracted from each target segment to construct a training sample set, and the multiple feature parameters include static voltage extracted from the static segment, dynamic voltage extracted from the charging segment, multiple instantaneous working condition features, and multiple cumulative working condition features. The sample training set includes the most core and most relevant multiple feature parameters for estimating the static voltage, thereby improving the training performance of the sample training set. Further, the preset model is trained based on the training sample set to obtain a static voltage prediction model. The preset model is optimized based on the high-quality multiple training samples, thereby obtaining a static voltage prediction model with high precision and strong generalization ability. Further, the multiple instantaneous working condition data and the multiple cumulative working condition data of the target vehicle under the charging working condition are calculated by the static voltage prediction model to output a predicted battery static voltage value. The static voltage prediction model can break through the static condition restriction and estimate the static voltage online under the charging working condition, thereby improving the acquisition frequency of the static voltage and the static voltage calculation coverage, to realize high-precision and online static voltage estimation.

[0065] The second embodiment provides a battery static voltage prediction method, as shown in Figure 2 The method comprises the following steps:

[0066] S201: dividing the vehicle historical running data to obtain multiple target segments; wherein each target segment includes a static segment and a charging segment, and the static segment and the charging segment are distinguished and connected by a charging flag.

[0067] S202: extracting the static voltage corresponding to the end of the static segment as the static voltage, and extracting the dynamic voltage corresponding to the starting time of the charging segment as the dynamic voltage.

[0068] In the embodiment, the end of the static segment corresponds to the last frame of data of the static segment, and the starting time of the charging segment corresponds to the first frame of data of the charging segment.

[0069] Specifically, the data frame corresponding to the end of the static segment is continuous with the data frame corresponding to the start time of the charging segment, and the time interval is extremely short, so the dynamic voltage corresponding to the start time of the charging segment is the dynamic voltage with the smallest difference between the static voltage corresponding to the end of the static segment.

[0070] S203: Extract the state of charge, temperature, and current corresponding to the start time of the charging segment to obtain multiple instantaneous working condition characteristics; and extract the cumulative mileage, cumulative effective cycle number, and operation age corresponding to the start time of the charging segment to obtain multiple cumulative working condition characteristics.

[0071] In this embodiment, the dynamic voltage also belongs to the instantaneous working condition characteristics, and in combination with the remaining multiple instantaneous working condition characteristics, can be used to represent the instantaneous state of the battery after the battery is converted from static to dynamic.

[0072] In this embodiment, the multiple cumulative working condition characteristics are used to represent the aging degree of the battery.

[0073] Specifically, according to the electrochemical impedance spectrum of the battery, the following formula (1) is known:

[0074] (1)

[0075] wherein, is the dynamic voltage, is the static voltage, is the current, is the impedance, is the ohmic impedance, is the concentration impedance, is the electrochemical impedance.

[0076] Specifically, the ohmic impedance is mainly affected by multiple cumulative working condition characteristics such as the cumulative mileage (Odometer), the cumulative effective cycle number (Cycle), and the operation age (Year), and gradually increases with the aging of the battery. The concentration impedance is mainly affected by the state of charge (SOC) and the current When the SOC is in the low-end interval or the high-end interval, the concentration impedance is relatively high; when the SOC is in the middle interval, the concentration impedance is relatively low; and the concentration impedance increases with the increase of the current rate. The electrochemical impedance is mainly affected by the temperature (Temperature, Temp) and the SOC, and increases with the increase of the temperature.

[0077] ​Based on the correlation between the impedance and the plurality of instantaneous working condition characteristics and the plurality of cumulative working condition characteristics, formula (1) can be represented as formula (2) as follows:

[0078] (2)

[0079] wherein, is a dynamic voltage, is a static voltage, is a current, is a relationship equation between the plurality of cumulative working conditions and the ohmic impedance, is a relationship equation between the state of charge, the current and the concentration impedance, is a relationship equation between the temperature, the state of charge and the electrochemical impedance.

[0080] In the embodiment, since the time points of obtaining the static voltage and the dynamic voltage are extremely close, it can be considered that the plurality of cumulative working condition characteristics corresponding to the static voltage and the dynamic voltage are the same, that is, the ohmic impedance corresponding to the static voltage and the dynamic voltage is the same. S204: Taking the static voltage, the dynamic voltage, the plurality of instantaneous working condition characteristics and the plurality of cumulative working condition characteristics corresponding to each target segment as a group of training data.

[0081] S205: Constructing a training sample set based on the plurality of groups of training data corresponding to the plurality of target segments.

[0082] S206: Training the preset model based on the training sample set to obtain a static voltage prediction model.

[0083] S207: Calculating the plurality of instantaneous working condition data and the plurality of cumulative working condition data of the target vehicle in the charging working condition by the static voltage prediction model to output a predicted battery static voltage value.

[0084] In the technical solution of the embodiments of the present application, the static voltage is extracted from the static voltage corresponding to the end of the static segment, and the dynamic voltage is extracted from the dynamic voltage corresponding to the starting time of the charging segment. The static voltage and the dynamic voltage are obtained through two continuous data frames with a very short time interval, so that the difference between the dynamic voltage and the static voltage is minimized, and the static voltage obtained is the voltage closest to the true value of the static voltage. Moreover, since the time points of obtaining the static voltage and the dynamic voltage are very close, it can also be considered that the multiple cumulative working condition characteristics corresponding to the static voltage and the dynamic voltage are the same. Further, by extracting the state of charge, temperature and current corresponding to the starting time of the charging segment, and extracting the cumulative mileage, cumulative effective cycle number and operation life corresponding to the starting time of the charging segment, multiple-dimensional features reflecting the instantaneous state and reflecting the aging degree can be obtained. Further, by taking the static voltage, dynamic voltage, multiple instantaneous working condition characteristics and multiple cumulative working condition characteristics corresponding to each target segment as a set of training data, and constructing a training sample set based on multiple sets of training data corresponding to multiple target segments, the instantaneous state features, aging features and other information after the change of the battery state can be associated with the static voltage, thereby constructing a high-quality and high-correlation sample training set. Further, since the battery aging will affect the shape of the static voltage-state of charge curve, the multiple cumulative working condition characteristics are included in the sample training set, which can also train the model to capture the offset of the aging degree to the static voltage, so that the finally obtained static voltage prediction model can track the whole life cycle of the battery.

[0085] Specifically, the step of constructing a training sample set based on multiple sets of training data corresponding to multiple target segments in step S205 specifically includes the following embodiments:

[0086] The multiple sets of training data are down-sampled based on the current rate interval and the temperature interval to obtain multiple sets of training samples.

[0087] The multiple sets of training samples are normalized, and a training sample set is constructed based on the normalized multiple sets of training samples.

[0088] In some specific embodiments, the number of samples of small-rate training data obtained based on small-rate current working conditions and / or conventional temperature working conditions is reduced, so that the number of samples of large-rate training data obtained based on large-rate current working conditions and / or extreme temperature working conditions increases in the training sample set.

[0089] Specifically, in the real vehicle operation data, the data points under the conditions of small current ratio, normal temperature, etc. account for the majority, while the data points under the conditions of large current ratio, extreme temperature, etc. are very rare. If the original data is directly used for training, the model will tend to optimize the prediction of the majority class data (small current ratio data, normal temperature data), and for the minority class data (large current ratio data, extreme temperature data), the model may directly ignore or fit poorly, and cannot learn and adapt to the voltage variation law under different conditions in a balanced manner.

[0090] In the technical solution of the embodiment of the application, the plurality of groups of training data are down-sampled based on the current ratio interval and the temperature interval to obtain a plurality of groups of training samples, so that the number of samples of the training data obtained based on different current ratio intervals and different temperature intervals is balanced, thereby enabling the model to better learn and adapt to the voltage variation law under different conditions. Further, by normalizing the plurality of groups of training samples and constructing a training sample set based on the plurality of groups of normalized training samples, the preset model can be prevented from unfairly punishing different dimensions of features during subsequent training, thereby improving the reliability of the training sample set.

[0091] The third embodiment provides a battery static voltage prediction method, as shown in the formula (I), comprising: Figure 3

[0092] S301: The vehicle historical operation data is divided to obtain a plurality of target segments; wherein each target segment includes a static segment and a charging segment, and the static segment and the charging segment are distinguished and connected by a charging flag.

[0093] S302: A plurality of feature parameters are extracted from each target segment to construct a training sample set; wherein the plurality of feature parameters include static voltage extracted from the static segment and dynamic voltage, a plurality of instantaneous working condition features and a plurality of cumulative working condition features extracted from the charging segment.

[0094] S303: The preset model is trained based on the training sample set to obtain a static voltage prediction model.

[0095] S304: The dynamic voltage, the battery state of charge, the temperature and the current corresponding to the starting time of the target vehicle under the charging condition are obtained to obtain a plurality of instantaneous working condition data.

[0096] S305: The cumulative mileage, the cumulative effective cycle number and the running age corresponding to the starting time of the target vehicle under the charging condition are obtained to obtain a plurality of cumulative working condition data.

[0097] S306: The plurality of instantaneous working condition data and the plurality of cumulative working condition data are calculated by the static voltage prediction model to output a predicted battery static voltage value. ​

[0098] In the technical solution of the embodiment of the application, the plurality of instantaneous working condition data and the plurality of cumulative working condition data are obtained immediately at the moment when the target vehicle starts charging each time, and the predicted static voltage value of the battery is output through the static voltage prediction model, so that the static voltage acquisition is changed from a low-frequency occasional event to a high-frequency regular event, thereby providing more data support for real-time and accurate monitoring of the battery state. Further, by making the plurality of instantaneous working condition data be dynamic voltage, battery state of charge, temperature and current, and making the plurality of cumulative working condition data be cumulative mileage, cumulative effective cycle number and operation life, the feature type, time sequence point (charging segment starting time) and physical meaning collected in online application are ensured to be completely consistent with those in the training sample construction, thereby improving the accuracy and reliability of the static voltage prediction, reducing the safety hazard of the battery, and then improving the safety and reliability of the target vehicle.

[0099] The fourth embodiment provides a battery static voltage prediction method, as shown in the following formula (1): Figure 4 The method comprises the following steps:

[0100] S401: Obtain vehicle historical original data.

[0101] In the embodiment, the vehicle historical original data is historical original data generated in the actual operation process of the vehicle.

[0102] Specifically, the vehicle historical original data can be obtained based on cloud big data.

[0103] S402: Perform a preprocessing operation on the vehicle historical original data to obtain vehicle historical operation data; wherein the preprocessing operation comprises at least one of the following operations: outlier elimination, null value filling, state of charge jump cleaning, time jump cleaning and data format standardization.

[0104] In the technical solution of the embodiment of the application, at least one preprocessing operation is performed on the vehicle original data, so as to eliminate the abnormal values and jump parameters in the vehicle original data, and solve the data missing problem, thereby improving the accuracy and consistency of the data.

[0105] S403: Scan the vehicle historical operation data to identify the time when the charging flag bit jumps from the first state to the second state, and take the time as the charging starting point.

[0106] In the embodiment, the first state of the charging flag bit indicates that the charging flag bit is 0, and the second state of the charging flag bit indicates that the charging flag bit is 1.

[0107] S404: Scan the vehicle historical operation data backward based on the charging starting time until the charging flag bit jumps back to the first state from the second state, and extract the data segment in the corresponding interval as a charging segment; the charging starting point is the starting time of the charging segment.

[0108] In a specific embodiment, the data is scanned to find the time point t0 at which the charging flag bit changes from 0 to 1, the time point t0 is taken as the starting time of the charging segment, and the scanning is continued from the time point t0 to find the time point t1 at which the charging flag bit changes from 1 to 0. The data frames with timestamps in the interval [t0, t1] are extracted to form the charging segment.

[0109] S405: Based on the charging starting time, the historical running data of the vehicle is traced back to extract a data segment that meets the static condition as a static segment.

[0110] In the embodiment, the static condition is that the charging flag bit is continuously 0, the vehicle speed is continuously 0, and the absolute value of the current is less than a preset threshold.

[0111] In some specific embodiments, the data is scanned to find the time point t0 at which the charging flag bit changes from 0 to 1, and the data is traced back from the time point t0 until the time point t2 at which the above static condition is no longer met. The data frames with timestamps in the interval [t2, t0) are extracted to form the static segment.

[0112] S406: Based on each charging starting point, the corresponding static segment and the charging segment are associated to obtain a plurality of target segments.

[0113] S407: A plurality of feature parameters are extracted from each target segment to construct a training sample set; wherein the plurality of feature parameters include static voltage extracted from the static segment, dynamic voltage, a plurality of instantaneous working condition features and a plurality of cumulative working condition features extracted from the charging segment.

[0114] S408: Based on the training sample set, a preset model is trained to obtain a static voltage prediction model.

[0115] S409: The static voltage prediction model is used to calculate a plurality of instantaneous working condition data and a plurality of cumulative working condition data of the target vehicle under the charging condition to output a predicted battery static voltage value.

[0116] In the technical solution of the embodiment of the application, the time point at which the charging flag bit jumps from the first state to the second state is identified by scanning the historical running data of the vehicle, and the time point is taken as the charging start point, so that the static segment and the charging segment are extracted respectively based on the charging start point, each charging condition can be accurately associated with the static condition before it, and at the moment of the charging start point, the aging state of the battery is almost completely the same as the aging state at the end of the static segment, so that variable interference can be reduced to the greatest extent. Moreover, the selection of the static segment needs to meet the static condition, so that the quality of the static segment can be improved to ensure that the voltage at the end of the static segment is the voltage that is fully relaxed and closest to the true value of the static voltage, thereby improving the data quality of the training sample.

[0117] The fifth embodiment provides a battery static voltage prediction method, as shown in the formula (1), comprising: Figure 5

[0118] S501: dividing the historical running data of the vehicle to obtain a plurality of target segments; wherein each target segment includes a static segment and a charging segment, and the static segment and the charging segment are distinguished and connected by a charging flag bit.

[0119] S502: extracting a plurality of feature parameters from each target segment to construct a training sample set; wherein the plurality of feature parameters include static voltage extracted from the static segment and dynamic voltage, a plurality of instantaneous working condition features and a plurality of cumulative working condition features extracted from the charging segment.

[0120] S503: taking the dynamic voltage, the plurality of instantaneous working condition features and the plurality of cumulative working condition features corresponding to each target segment as a group of input features, and taking the static voltage corresponding to each group of input features as an output label.

[0121] S504: training the elastic net model based on the plurality of groups of input features and the plurality of output labels corresponding to the plurality of target segments to obtain a static voltage prediction model.

[0122] S505: calculating the plurality of instantaneous working condition data and the plurality of cumulative working condition data of the target vehicle under the charging condition by the static voltage prediction model to output a predicted battery static voltage value.

[0123] ​In the technical solution of this application embodiment, the dynamic voltage, multiple instantaneous operating condition features, and multiple cumulative operating condition features corresponding to each target segment are taken as a set of input features, and the static voltage corresponding to each set of input features is taken as the output label. The elastic network model is trained by multiple sets of input features and multiple input labels. The mechanism of the elastic network model can be used to select the most core features for static voltage prediction to simplify the model, and effectively handle the possible multicollinearity between features to improve the stability of the model. Thus, the obtained static voltage prediction model is a linear model with simple structure, stable performance, and strong generalization ability, which is easy to deploy in the vehicle environment to provide efficient and reliable prediction data.

[0124] The sixth embodiment provides a method for predicting battery static voltage, such as... Figure 6 As shown, it includes:

[0125] S601: Divide the vehicle's historical operating data to obtain multiple target segments; each target segment includes a stationary segment and a charging segment, which are distinguished and connected by a charging flag bit.

[0126] S602: Extract multiple feature parameters from each target segment to construct a training sample set; wherein, the multiple feature parameters include static voltage extracted from the stationary segment and dynamic voltage extracted from the charging segment, multiple instantaneous operating condition features and multiple cumulative operating condition features.

[0127] S603: Take the dynamic voltage, multiple instantaneous operating condition features and multiple cumulative operating condition features corresponding to each target segment as a set of input features, and take the static voltage corresponding to each set of input features as the output label.

[0128] S604: Construct the objective function corresponding to the elastic network model; wherein, the objective function includes a prediction error loss term and a regularization term; the prediction error loss term includes the static voltage prediction value and the static voltage true value, the static voltage prediction value is calculated through the input feature matrix, the feature weight coefficient vector corresponding to the input feature matrix and the model intercept, and the regularization term is calculated through the feature weight coefficient vector.

[0129] In this embodiment, the regularization terms include L1 (LASSO Regression) regularization and L2 (Ridge Regression) regularization, and the regularization terms only constrain the feature weight coefficient vector.

[0130] Specifically, the L1 regularization term is used to generate a sparse model, directly compressing the coefficients of some unimportant features to 0, thereby achieving automatic feature selection. The L2 regularization term is used to handle the problem of feature collinearity, making the model more stable. The elastic net model combines the feature selection capability of L1 with the stability of L2.

[0131] Specifically, the regularization term also includes two hyperparameters: the regularization strength coefficient and the mixed parameter of L1 and L2 regularization.

[0132] Specifically, the feature weight coefficient vector includes multiple feature weight coefficients, each of which corresponds to a different input feature in the input feature matrix.

[0133] S605: Input multiple sets of input features corresponding to multiple target segments and multiple output labels into the elastic network model, use each set of input features as the input feature matrix, and use the input label corresponding to each set of input features as the true value of static voltage to calculate the minimum value of the objective function.

[0134] S606: Generate a linear prediction formula based on the feature weight coefficient vector corresponding to the minimum value and the model intercept.

[0135] In this embodiment, the linear prediction formula is characterized as follows:

[0136] (3)

[0137] in, This is the predicted static voltage value. This represents the dynamic voltage at the start of the charging segment. This represents the current at the start of the charging segment. The temperature corresponding to the start of the charging segment. This represents the cumulative mileage corresponding to the start time of the charging segment. This represents the cumulative number of valid cycles corresponding to the start time of the charging segment. The operating life corresponding to the start time of the charging segment. In a charged state, For model intercept, for The feature weight coefficients, for The feature weight coefficients, for The feature weight coefficients, for The feature weight coefficients, for The feature weight coefficients, for The feature weight coefficients, is a feature weight coefficient.

[0138] Specifically, the dynamic voltage and the plurality of instantaneous working condition features have a greater impact on the calculation, so the values of , , are larger, and the value of is the largest, is in the middle, is the smallest. , , , are smaller, and are all smaller than .

[0139] In some specific embodiments, the value range of is 0.06-0.07, the value range of

[0140] S607: The linear prediction formula is verified by ten-fold cross-validation method, and after verification, the linear prediction formula is determined as the static voltage prediction model.

[0141] S608: The plurality of instantaneous working condition data and the plurality of cumulative working condition data of the target vehicle under the charging working condition are calculated by the static voltage prediction model to output the predicted battery static voltage value.

[0142] ​In the technical scheme of the embodiment of the present application, the target function corresponding to the elastic net model includes a prediction error loss term and a regularization term, the prediction error loss term includes a static voltage predicted value and a static voltage true value, and the static voltage predicted value is calculated by using an input feature matrix, a feature weight coefficient vector corresponding to the input feature matrix, and a model intercept, and the regularization term is calculated by using the feature weight coefficient vector, so that the gap between the static voltage predicted value and the static voltage true value can be directly measured by the prediction error loss term, and an excessively large weight coefficient can be punished by the regularization term to avoid model overfitting. Further, a plurality of groups of input features corresponding to a plurality of target segments and a plurality of output labels are input into the elastic net model, each group of input features is used as an input feature matrix, and an input label corresponding to each group of input features is used as a static voltage true value, and the minimum value of the target function is calculated, so that a global optimal solution can be found based on the minimum value, and the feature weight coefficient vector obtained is stable and unique, so that each weight coefficient in the feature weight coefficient vector can accurately quantify the contribution degree and influence direction of each corresponding feature in the matrix feature. Further, a linear prediction formula is generated based on the feature weight coefficient vector and the model intercept corresponding to the minimum value, so that the optimal solution can be presented by a clear and reliable linear formula, thereby reducing the complexity of the model and improving the accuracy of the model. Further, the ten-fold cross-validation method is used to verify the linear prediction formula, and the linear prediction formula is determined as the static voltage prediction model after verification, which can further improve the reliability of the static voltage prediction model.

[0143] Specifically, the step of verifying the linear prediction formula by using the ten-fold cross-validation method in step S607 and determining the linear prediction formula as the static voltage prediction model after verification specifically includes the following implementation manners.

[0144] The training sample set is randomly divided into ten non-overlapping subsets.

[0145] Each subset is sequentially used as a test set, and the remaining nine subsets are combined as a training set, and the linear prediction formula is trained and tested ten times to obtain ten mean absolute errors and ten determination coefficients.

[0146] A first average value is obtained based on the ten mean absolute errors, and a second average value is obtained based on the ten determination coefficients.

[0147] In response to the first average value being greater than or equal to a first preset threshold value and the second average value being greater than or equal to a second preset threshold value, the linear prediction formula is determined as the static voltage prediction model.

[0148] In response to the first average value being less than the first preset threshold value and / or the second average value being less than the second preset threshold value, the elastic net model is retrained, and the linear prediction formula obtained by retraining is verified.

[0149] In the technical solution of the embodiment of the application, the ten-fold cross-validation method is used to verify the linear prediction formula, so that all data in the training sample set can be used for training and verification. Through ten cycles, each training sample has one as a test set to maximize the evaluation of the performance of the linear prediction formula on unknown data. Further, by calculating the first average value and the second average value corresponding to the ten results, the evaluation variance can be effectively reduced, so that the performance index obtained is more stable and reliable. Further, when the first average value and the second average value both satisfy the threshold condition, the linear prediction formula is determined as a static voltage prediction model, which can ensure that only high-quality models with performance up to standard can be finally confirmed and deployed, thereby improving the reliability and safety of the static voltage prediction model, so as to more accurately predict the static voltage. When the first average value and / or the second average value do not satisfy the corresponding threshold condition, the elastic net model is retrained, and the linear prediction formula obtained by retraining is verified, which can realize iterative improvement and automatic optimization of the model based on the closed loop of "verification-feedback-retraining", thereby improving the robustness and reliability of model training.

[0150] In some specific embodiments, retraining the elastic net model includes further optimizing the model or adjusting the hyperparameters.

[0151] Specifically, the first preset threshold value corresponding to the mean absolute error (MAE) is set based on a reference table of OCV and MAE in a specific project.

[0152] In some specific embodiments, the value range of the first preset threshold value is 20-50 mV.

[0153] Specifically, the value range of the coefficient of determination (R 2 The value range of the coefficient of determination (R

[0154] In some specific embodiments, the value range of the second preset threshold value is 0.7-0.9.

[0155] In the technical solution of the embodiment of the application, by setting the value range of the first preset threshold value and the second preset threshold value, the linear prediction formula can be better verified.

[0156] Please refer to Figure 7 , Figure 7is a work flow diagram of an application scenario of the battery static voltage prediction method of the present application. In a specific application scenario, first, vehicle historical raw data is acquired. Then, a pretreatment operation is performed on the vehicle historical raw data to obtain vehicle historical running data. Next, the vehicle historical running data is divided to obtain a plurality of target segments. Each target segment includes a static segment and a charging segment, and the static segment and the charging segment are distinguished and connected by a charging flag. Then, a plurality of feature parameters are extracted from each target segment to construct a training sample set. The plurality of feature parameters include static voltage extracted from the static segment and dynamic voltage, a plurality of instantaneous working condition features and a plurality of cumulative working condition features extracted from the charging segment. The plurality of training samples in the training sample set are obtained by down-sampling and normalization processing. The preset model is trained based on the training sample set to obtain a static voltage prediction model. Finally, the static voltage prediction model is used to calculate a plurality of instantaneous working condition data and a plurality of cumulative working condition data of a target vehicle under a charging condition to output a predicted battery static voltage value.

[0157] In order to facilitate the understanding of the embodiments of the present application, the present application provides the following non-limiting examples for further detailed description of the present application.

[0158] Embodiment 1

[0159] The static voltage prediction model obtained by the above model training method is acquired. Massive vehicle historical raw data is acquired from the cloud, and 100,000 target segments including charging segments and static segments are obtained based on the massive vehicle historical raw data. The static voltage prediction model is verified by the 100,000 target segments.

[0160] The verification result is shown in Figure 8 , and the final calculated determination coefficient is 0.96 and the average absolute error is 8mV.

[0161] The above verification result shows that the fitting degree of the static voltage prediction model provided by the present application is very high, so the embodiment of the present application can realize high-precision and online application of static voltage estimation.

[0162] The present application provides an electronic device.

[0163] Please refer to Figure 9 , Figure 9 is a structural schematic diagram of an embodiment of the electronic device of the present application. As shown in Figure 9 , in the embodiment, the electronic device 90 includes a memory 91 and a processor 92.

[0164] In this embodiment, the memory 91 is configured to store program data, which when executed implements the steps of the method for predicting the static voltage of a battery as described above; and the processor 92 is configured to execute the program instructions stored in the memory 91 to implement the steps of the method for predicting the static voltage of a battery as described above.

[0165] In particular, the processor 92 is configured to control itself and the memory 91 to implement the steps of the method for testing as described above. The processor 92 can also be referred to as a CPU (Central Processing Unit). The processor 92 can be an integrated circuit chip capable of carrying out a series of operations. The processor 92 can also be a general purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component. The general purpose processor can be a microprocessor or the processor can also be any conventional processor or the like. In addition, the processor 92 can be implemented by a plurality of integrated circuit chips working together.

[0166] In the technical scheme of the embodiment of the application, the vehicle historical operation data is divided by the processor 92 to obtain a plurality of target segments, and each target segment includes one static segment and one charging segment distinguished and connected by a charging flag bit, so that the charging segment can be directly associated with the static segment before it, so that the charging segment and the static segment have a highly accurate and reliable corresponding relationship, thereby improving the data quality of the target segment. Further, a plurality of feature parameters are extracted from each target segment to construct a training sample set, and the plurality of feature parameters include static voltage extracted from the static segment and dynamic voltage, a plurality of instantaneous working condition features and a plurality of cumulative working condition features extracted from the charging segment, so that the sample training set includes a plurality of feature parameters that are most core and most relevant for estimating the static voltage, thereby improving the training performance of the sample training set. Further, the preset model is trained based on the training sample set to obtain a static voltage prediction model, which can optimize the preset model based on a plurality of high-quality training samples, thereby obtaining a static voltage prediction model with high precision and strong generalization ability. Further, the plurality of instantaneous working condition data and the plurality of cumulative working condition data of the target vehicle under the charging condition are calculated by the static voltage prediction model to output a predicted battery static voltage value, which can break through the static condition restriction, so that the static voltage prediction model can estimate the static voltage online under the charging condition, thereby improving the acquisition frequency of the static voltage, and then improving the static voltage calculation coverage, to realize high-precision and online static voltage estimation.

[0167] The application provides a computer readable storage medium.

[0168] Please refer to Figure 10 , Figure 10 is a structural schematic diagram of an embodiment of the computer readable storage medium of the application.

[0169] The computer readable storage medium 100 includes a computer program 101 stored on the computer readable storage medium 100, and the computer program 101 is executed by the processor to realize the steps in the battery static voltage prediction method as described above.

[0170] The above description is only an embodiment of the application, and does not limit the patent scope of the application, and any equivalent structure or equivalent process transformation using the content of the specification and drawings, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the application.​

Claims

1. A method for predicting the static voltage of a battery, characterized in that, include: The vehicle's historical operating data is divided into multiple target segments; each target segment includes a static segment and a charging segment, and the static segment and the charging segment are distinguished and connected by a charging flag bit; Multiple feature parameters are extracted from each of the target segments to construct a training sample set; wherein, the multiple feature parameters include static voltage extracted from the stationary segment and dynamic voltage extracted from the charging segment, multiple instantaneous operating condition features and multiple cumulative operating condition features; The step of extracting multiple feature parameters from each of the target segments to construct a training sample set includes: The static voltage corresponding to the end of the static segment is extracted as the static voltage, and the dynamic voltage corresponding to the start time of the charging segment is extracted as the dynamic voltage. Extract the state of charge, temperature, and current corresponding to the start time of the charging segment to obtain multiple instantaneous operating condition features; and extract the cumulative mileage, cumulative effective cycle count, and operating years corresponding to the start time of the charging segment to obtain multiple cumulative operating condition features. The static voltage, dynamic voltage, multiple instantaneous operating condition features, and multiple cumulative operating condition features corresponding to each target segment are used as a set of training data. The training sample set is constructed based on multiple sets of training data corresponding to multiple target segments; The preset model is trained based on the training sample set to obtain a static voltage prediction model; The static voltage prediction model calculates multiple instantaneous and cumulative operating condition data of the target vehicle under charging conditions to output the predicted battery static voltage value.

2. The battery static voltage prediction method according to claim 1, characterized in that, The step of constructing the training sample set based on multiple sets of training data corresponding to multiple target segments includes: Based on the current rate range and temperature range, multiple sets of training data are downsampled to obtain multiple sets of training samples. The training samples are normalized, and the training sample set is constructed based on the normalized training samples.

3. The battery static voltage prediction method according to claim 1, characterized in that, The step of calculating multiple instantaneous and cumulative operating condition data of the target vehicle under charging conditions using the static voltage prediction model to output a predicted battery static voltage value includes: The dynamic voltage, battery state of charge, temperature, and current of the target vehicle at the start of the charging condition are acquired to obtain multiple instantaneous operating condition data; and, The cumulative mileage, cumulative effective cycle count, and operating years of the target vehicle at the start time of the charging condition are obtained to obtain multiple cumulative operating condition data. The static voltage prediction model calculates multiple instantaneous operating condition data and multiple cumulative operating condition data to output the predicted battery static voltage value.

4. The battery static voltage prediction method according to claim 1, characterized in that, The preset model is an elastic network model; The step of training a preset model based on the training sample set to obtain a static voltage prediction model includes: The dynamic voltage, multiple instantaneous operating condition features, and multiple cumulative operating condition features corresponding to each target segment are taken as a set of input features, and the static voltage corresponding to each set of input features is taken as the output label. The elastic network model is trained based on multiple sets of input features corresponding to multiple target segments and multiple output labels to obtain the static voltage prediction model.

5. The battery static voltage prediction method according to claim 4, characterized in that, The step of training the elastic network model based on multiple sets of input features corresponding to multiple target segments and multiple output labels to obtain the static voltage prediction model includes: Construct the objective function corresponding to the elastic network model; wherein, the objective function includes a prediction error loss term and a regularization term; the prediction error loss term includes the static voltage prediction value and the static voltage true value, the static voltage prediction value is calculated by the input feature matrix, the feature weight coefficient vector corresponding to the input feature matrix and the model intercept, and the regularization term is calculated by the feature weight coefficient vector; The elastic network model is input with multiple sets of input features corresponding to multiple target segments and multiple sets of output labels. Each set of input features is used as the input feature matrix, and the output label corresponding to each set of input features is used as the true value of the static voltage. The minimum value of the objective function is calculated. A linear prediction formula is generated based on the feature weight coefficient vector corresponding to the minimum value and the model intercept. The linear prediction formula was validated using the ten-fold cross-validation method, and after successful validation, the linear prediction formula was determined as the static voltage prediction model.

6. The battery static voltage prediction method according to claim 5, characterized in that, The step of using the ten-fold cross-validation method to verify the linear prediction formula and determining the linear prediction formula as the static voltage prediction model after successful verification includes: The training sample set is randomly divided into ten non-overlapping subsets; Each subset is used as a test set in turn, and the remaining nine subsets are combined as a training set. The linear prediction formula is trained and tested ten times to obtain ten mean absolute errors and ten coefficients of determination. A first average value is obtained based on the ten mean absolute errors, and a second average value is obtained based on the ten coefficients of determination; In response to the first average value being greater than or equal to a first preset threshold and the second average value being greater than or equal to a second preset threshold, the linear prediction formula is determined as the static voltage prediction model; In response to the first average value being less than the first preset threshold, and / or the second average value being less than the second preset threshold, the elastic network model is retrained, and the linear prediction formula obtained from the retraining is validated.

7. The battery static voltage prediction method according to claim 1, characterized in that, The step of dividing historical vehicle operation data into multiple target segments includes: Scan the vehicle's historical operating data to identify the moment when the charging flag changes from the first state to the second state, and use that moment as the charging start point; Based on the charging start point, the vehicle's historical operating data is scanned backward until the charging flag changes from the second state back to the first state, and the data segment within the corresponding interval is extracted as the charging segment; the charging start point is the start time of the charging segment. Based on the charging start point, the vehicle's historical operating data is traced backward to extract data segments that meet the static condition as the static segment; Based on each charging start point, the corresponding static segment is associated with the charging segment to obtain multiple target segments.

8. The battery static voltage prediction method according to claim 1, characterized in that, Before the step of dividing the vehicle's historical operating data to extract static and charging segments, the following steps are included: Obtain raw historical data of the vehicle; Preprocessing operations are performed on the raw historical vehicle data to obtain the historical vehicle operation data; wherein, the preprocessing operations include at least one of the following operations: outlier removal, null value filling, state of charge transition cleaning, time transition cleaning, and data format standardization.

9. An electronic device, characterized in that, include: A memory for storing program data, which, when executed, implements the steps in the battery static voltage prediction method as described in any one of claims 1 to 8; A processor for executing program instructions stored in the memory to implement the steps in the battery static voltage prediction method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the battery static voltage prediction method as described in any one of claims 1 to 8.

Images