CN115219910A - Analysis method and device for battery residue prediction error - Google Patents
Analysis method and device for battery residue prediction error Download PDFInfo
- Publication number
- CN115219910A CN115219910A CN202210832225.8A CN202210832225A CN115219910A CN 115219910 A CN115219910 A CN 115219910A CN 202210832225 A CN202210832225 A CN 202210832225A CN 115219910 A CN115219910 A CN 115219910A
- Authority
- CN
- China
- Prior art keywords
- margin
- battery
- tree model
- preselected
- integrated tree
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/367—Software therefor, e.g. for battery testing using modelling or look-up tables
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/382—Arrangements for monitoring battery or accumulator variables, e.g. SoC
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Tests Of Electric Status Of Batteries (AREA)
Abstract
The application discloses a method and a device for analyzing a battery residual prediction error. Wherein the integration tree model is used for outputting the battery residual prediction error estimation value. The original data set used for training the integration tree model comprises a plurality of preselected margin factor characteristics and label values, wherein the label values are battery margin prediction errors between the battery real margin and the battery prediction margin output by the battery margin prediction model. And after the integrated tree model is trained, calculating the contribution values corresponding to the margin factor characteristics by using a Treeshap algorithm based on the integrated tree model. The contribution value corresponding to the residue factor characteristic is a quantized value of the influence of the residue factor characteristic on the battery residue prediction error, and can be used for representing the influence of the residue factor characteristic on the battery residue prediction error. In this way, the influence of each residual factor characteristic on the battery residual prediction error can be analyzed in an interpretable manner.
Description
Technical Field
The application relates to the technical field of vehicles, in particular to a method and a device for analyzing a battery residue prediction error.
Background
With the continuous development of automobile technology, new energy automobiles are being widely used. The lithium battery is used as a power source of a new energy automobile, and the problems of short service life, low output power and the like of the battery pack can be caused by overuse of the lithium battery.
The remaining amount of the battery may be represented by a remaining usage time of the battery, or may also be represented by a remaining driving mileage of the vehicle. Generally, a battery residual quantity prediction model is used to predict the battery residual quantity. However, due to the influence of some factors, the prediction of the battery residual quantity may be inaccurate, and a battery residual quantity prediction error exists.
At present, a plurality of factors influencing a battery residual prediction model exist, and how to determine the influence of different factors on a battery residual prediction error is an urgent problem to be solved.
Disclosure of Invention
In order to solve the technical problem, the application provides an analysis method and device for a battery residual prediction error, which can effectively explain the influence of different residual factor characteristics on the battery residual prediction error.
In order to achieve the above purpose, the technical solutions provided in the embodiments of the present application are as follows:
the embodiment of the application provides a method for analyzing a battery residual prediction error, which comprises the following steps:
training the integrated tree model based on the original data set to obtain a trained integrated tree model; the original data set comprises a plurality of preselected margin factor characteristics and a tag value, wherein the tag value is a battery margin prediction error between a battery real margin and a battery prediction margin output by a battery margin prediction model; the integrated tree model is used for outputting a battery margin prediction error estimation value;
calculating the contribution values corresponding to the margin factor characteristics by adopting a Treeshap algorithm based on an integrated tree model, and analyzing the battery margin prediction error based on the contribution values corresponding to the margin factor characteristics; each margin factor characteristic is selected from the plurality of preselected margin factor characteristics.
Optionally, the calculating, based on the integrated tree model, the contribution values corresponding to the margin factor features by using a treshop algorithm includes:
obtaining a plurality of characteristic alliances corresponding to the target allowance factor characteristics; the target margin factor characteristic is each of the margin factor characteristics;
acquiring parameters of the trained integrated tree model;
and calculating the contribution value corresponding to the target margin factor characteristic by adopting a Treeshap algorithm based on the parameters of the integrated tree model and the plurality of characteristic alliances corresponding to the target margin factor characteristic.
Optionally, the hyper-parameters of the trained integrated tree model are target hyper-parameters; the training of the integrated tree model based on the original data set to obtain the trained integrated tree model comprises the following steps:
carrying out feature selection on the original data set to obtain the original data set after feature selection;
dividing the original data set after feature selection into a training data set and a verification data set;
obtaining all preselected hyper-parameters of the integrated tree model;
training the integrated tree models corresponding to the preselected hyper-parameters respectively based on the training data set, and acquiring the integrated tree models corresponding to the trained preselected hyper-parameters respectively;
based on the verification data set, obtaining verification errors of the integrated tree models respectively corresponding to the preselected hyper-parameters;
and searching a target hyper-parameter of the integrated tree model by adopting a tree structure Parrson estimation TPE algorithm based on the verification error of the integrated tree model corresponding to each preselected hyper-parameter, and acquiring the integrated tree model corresponding to the target hyper-parameter.
Optionally, the performing feature selection on the original data set to obtain the original data set after feature selection includes:
selecting the characteristics of the preselected residual factor characteristics, and screening the preselected residual factor characteristics to obtain a plurality of residual factor characteristics;
the plurality of margin factor features and the label values constitute an original data set after feature selection.
Optionally, the performing feature selection on the plurality of preselected residual factor features, and screening the plurality of preselected residual factor features to obtain a plurality of residual factor features includes:
inputting the preselected margin factor characteristics into the integrated tree model to obtain a battery margin prediction error estimation value output by the integrated tree model;
and calculating information gains corresponding to the preselected residual factor characteristics respectively based on the preselected residual factor characteristics and the battery residual prediction error estimation value, determining the preselected residual factor characteristics of which the information gains meet a first preset threshold value as residual factor characteristics, or calculating a correlation coefficient between each preselected residual factor characteristic and the battery residual prediction error estimation value, and determining the preselected residual factor characteristics of which the correlation coefficient meets a second preset threshold value as residual factor characteristics.
Optionally, the training data set includes a training feature set composed of a plurality of margin factor features and a training label value; the training of the integrated tree model corresponding to each preselected hyper-parameter based on the training data set comprises:
inputting the training feature set into the integrated tree models respectively corresponding to the preselected hyper-parameters to obtain battery margin prediction error estimation values respectively output by the integrated tree models;
acquiring training loss values corresponding to the integrated tree models respectively based on the battery residue prediction error estimated values and the training label values;
and training the corresponding integrated tree model based on the loss value for training, and repeatedly executing the steps of inputting the characteristic set for training into the integrated tree model respectively corresponding to each preselected hyper-parameter to obtain the battery residual prediction error estimated value respectively output by each integrated tree model and the subsequent steps until a preset condition is reached.
The embodiment of the present application further provides an analysis apparatus for a battery remaining prediction error, where the apparatus includes:
the acquisition unit is used for training the integrated tree model based on the original data set and acquiring the trained integrated tree model; the original data set comprises a plurality of preselected margin factor characteristics and a tag value, wherein the tag value is a battery margin prediction error between a battery real margin and a battery prediction margin output by a battery margin prediction model; the integrated tree model is used for outputting a battery margin prediction error estimation value;
the calculation unit is used for calculating the contribution values corresponding to the margin factor characteristics by adopting a Treeshap algorithm based on the integrated tree model, and analyzing the battery margin prediction error based on the contribution values corresponding to the margin factor characteristics; each margin factor characteristic is selected from the plurality of preselected margin factor characteristics.
Optionally, the computing unit includes:
the first obtaining subunit is used for obtaining a plurality of characteristic alliances based on the characteristics of all the margin factors; the characteristic alliance comprises a preset number of residual factor characteristics, and the characteristic alliances are different from one another; the preset number is an integer not greater than the characteristic number of the margin factor;
the second acquisition subunit is used for acquiring the parameters of the trained integrated tree model;
and the first calculating subunit is used for calculating the contribution values corresponding to the margin factor features by adopting a Treeshap algorithm based on the parameters of the integrated tree model and the feature alliances.
An embodiment of the present application further provides an electronic device, including:
one or more processors;
a storage device having one or more programs stored thereon,
when executed by the one or more processors, cause the one or more processors to implement the method of analyzing a battery remaining prediction error as described in any of the above.
An embodiment of the present application further provides a computer readable medium, on which a computer program is stored, where the program is executed by a processor to implement any one of the above methods for analyzing a battery remaining prediction error.
According to the technical scheme, the method has the following beneficial effects:
the embodiment of the application provides an analysis method and device for a battery residual prediction error, which train an integrated tree model based on an original data set to obtain the trained integrated tree model. Wherein the integration tree model is used for outputting the battery residual prediction error estimation value. The original data set used for training the integration tree model comprises a plurality of preselected margin factor characteristics and label values, wherein the label values are battery margin prediction errors between the battery real margin and the battery prediction margin output by the battery margin prediction model. And after the integrated tree model is trained, based on the integrated tree model, calculating contribution values corresponding to the residual factor characteristics by using a Treeshap algorithm. The contribution value corresponding to the residue factor characteristic is a quantized value of the influence of the residue factor characteristic on the battery residue prediction error, and can be used for representing the influence of the residue factor characteristic on the battery residue prediction error. In this way, the influence of each residual factor characteristic on the battery residual prediction error can be analyzed in an interpretable manner.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following descriptions are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative efforts.
Fig. 1 is a schematic diagram of an exemplary application scenario provided in an embodiment of the present application;
fig. 2 is a flowchart of an analysis method for a battery remaining prediction error according to an embodiment of the present disclosure;
fig. 3 is a schematic structural diagram of an apparatus for analyzing a battery remaining prediction error according to an embodiment of the present disclosure.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present application more comprehensible, embodiments accompanying figures and detailed description thereof are described in further detail below.
For the convenience of understanding and explaining the technical solutions provided in the embodiments of the present application, the background art related to the embodiments of the present application will be described first.
With the continuous development of automobile technology, new energy automobiles are being widely used. The lithium battery is used as a power source of the new energy automobile, and the problems of short service life, low output power, sudden safety accidents and the like of the battery pack can be caused by excessive use of the lithium battery.
The remaining amount of the battery may be represented by a remaining usage time of the battery, or may also be represented by a remaining driving mileage of the vehicle. Generally, a battery residual quantity prediction model is used to predict the battery residual quantity. However, due to the influence of some factors, the prediction of the battery residual quantity may be inaccurate, and a battery residual quantity prediction error exists.
At present, there are many factors affecting the battery residual quantity prediction model, and after the factors affecting the battery residual quantity prediction model are determined, the battery residual quantity prediction model can be optimized in a targeted manner. Therefore, how to determine the influence of different factors on the battery residual prediction error is an urgent problem to be solved.
Based on this, the embodiment of the application provides an analysis method and device for a battery remaining prediction error. Wherein the integration tree model is used for outputting the battery residual prediction error estimation value. The original data set used for training the integration tree model comprises a plurality of preselected margin factor characteristics and label values, wherein the label values are battery margin prediction errors between the battery real margin and the battery prediction margin output by the battery margin prediction model. And after the integrated tree model is trained, calculating the contribution values corresponding to the margin factor characteristics by using a Treeshap algorithm based on the integrated tree model. The contribution value corresponding to the margin factor characteristic is a quantized value of the influence of the margin factor characteristic on the battery margin prediction error, and can represent the influence of the margin factor characteristic on the battery margin prediction error. In this way, the influence of each margin factor characteristic on the battery margin prediction error is analyzed in an interpretable manner.
In order to facilitate understanding of the analysis method for the battery remaining prediction error provided in the embodiment of the present application, the following description is made with reference to a scenario example shown in fig. 1. Referring to fig. 1, the drawing is a schematic diagram of a framework of an exemplary application scenario provided in an embodiment of the present application.
In practical application, the battery residual capacity prediction model is used for predicting a battery on a vehicle to obtain a battery prediction residual capacity. However, the predicted remaining amount of the battery obtained by the predicted model of the remaining amount of the battery may be inaccurate, so that an error may exist between the predicted remaining amount of the battery and the true remaining amount of the battery. It is recognized that some margin factor characteristics may have an effect on the error, such as battery current, battery voltage, etc. Thus, a determination of the possible margin factor characteristics that affect the error is needed.
In the embodiment of the application, the integrated tree model is trained based on the original data set to obtain the trained integrated tree model. The integrated tree model is used for outputting a battery margin prediction error estimation value. The battery margin prediction error estimate is an estimate of the battery margin error.
Wherein the raw data set used to train the ensemble tree model includes a plurality of preselected residue factor features and label values. The label value is a battery residual prediction error between the battery true residual and a battery predicted residual output by the battery residual prediction model. As an alternative example, the plurality of preselected margin factor characteristics includes a plurality of battery current, battery voltage, battery charge level, and vehicle operating conditions.
After the integrated tree model is trained, the contribution values corresponding to the margin factor features are calculated by using a Treeshap algorithm based on the integrated tree model.
Those skilled in the art will appreciate that the block diagram shown in fig. 1 is only one example in which embodiments of the present application may be implemented. The scope of applicability of the embodiments of the present application is not limited in any way by this framework.
In order to facilitate understanding of the present application, a method for analyzing a battery remaining prediction error provided in an embodiment of the present application is described below with reference to the accompanying drawings.
Referring to fig. 2, which is a flowchart of a method for analyzing a battery remaining prediction error according to an embodiment of the present disclosure, as shown in fig. 2, the method may include S201-S202:
s201: training the integrated tree model based on the original data set to obtain a trained integrated tree model; the original data set comprises a plurality of preselected margin factor characteristics and a tag value, wherein the tag value is a battery margin prediction error between the real battery margin and a battery prediction margin output by a battery margin prediction model; the integrated tree model is used for outputting a battery margin prediction error estimation value.
Generally, a battery residual quantity prediction model is used to predict the battery residual quantity. However, the predicted battery remaining amount obtained by the battery remaining amount prediction model may be inaccurate, so that an error exists between the predicted battery remaining amount and the true battery remaining amount, which is a predicted battery remaining amount error. Thus, it is necessary to determine a margin factor characteristic that may affect the battery margin prediction error.
The battery residual prediction error is a real error value caused by a battery residual prediction model. For example, when the remaining amount of the battery is expressed by the remaining use time of the battery, the error value may be 0.5h; when the remaining amount of the battery is expressed using the remaining driving range of the vehicle, the error value may be 15km.
In the embodiment of the application, an integrated tree model is trained firstly, and the integrated tree model is used for inputting the margin factor characteristics and outputting the estimated value of the battery margin prediction error. The battery margin prediction error estimate is an estimate of the battery margin prediction error. When the battery residual prediction error is 0.5h, the estimated value of the battery residual prediction error output by the integrated tree model may be a value close to 0.5 h.
It is understood that, in the embodiment of the present application, a specific algorithm structure of the integrated tree model is not limited, and may be selected according to actual situations.
The raw data set used to train the ensemble tree model includes a plurality of preselected residue factor features and label values. In one or more embodiments, the plurality of preselected margin factor characteristics includes a plurality of battery currents, battery voltages, battery capacities, and vehicle operating conditions. The label value is a battery residual prediction error between the battery real residual and the battery prediction residual output by the battery residual prediction model. It is understood that the specific content and the number of the preselected margin factor features are not limited in the embodiments of the present application, and can be determined according to actual situations.
It is understood that a plurality of preselected margin factor characteristics are included in the raw data set. Some of these preselected margin factor characteristics may not or may not have a small impact on the margin prediction error. Therefore, to reduce data throughput, some margin factor features may be screened from a plurality of pre-selected margin factor features prior to training the integrated tree model. The influence degree of the screened margin factor characteristics on the margin prediction error meets a certain condition. Then, only the influence of the screened residual factor characteristics on the residual prediction error is analyzed subsequently.
In a possible implementation manner, an embodiment of the present application provides a specific implementation manner for training an ensemble tree model based on an original data set to obtain the trained ensemble tree model, and see the following descriptions A1 to A6.
S202: based on the integrated tree model, calculating the contribution values corresponding to the margin factor characteristics by adopting a Treeshap algorithm, and analyzing the battery margin prediction error based on the contribution values corresponding to the margin factor characteristics; each margin factor characteristic is selected from a plurality of preselected margin factor characteristics.
In this step, each of the margin factor features is selected from a plurality of preselected margin factor features, and the degree of influence of the margin factor features on the margin prediction error satisfies a certain condition.
After the integrated tree model after training is obtained, the contribution values corresponding to the margin factor features can be calculated by using a Treeshap algorithm on the basis of the integrated tree model. The Treeshap algorithm is one of the sharp algorithms, is used for estimating the sharp value, has low computational complexity and high computational efficiency, and can efficiently estimate the contribution values corresponding to the margin factor features by using the integrated tree model.
The contribution value corresponding to the margin factor characteristic is a quantized value of the influence degree of the margin factor characteristic on the margin prediction error. The influence degree of the margin factor characteristic on the margin prediction error can be intuitively known through the margin factor characteristic.
Therefore, the battery residual capacity prediction error can be analyzed based on the contribution value corresponding to each residual capacity factor characteristic. It is understood that the margin factor characteristic having a large contribution value has a large influence on the battery margin prediction error. The obtained margin factor characteristic with a large contribution value is an analysis result which needs to be obtained in the embodiment of the application, and then the battery margin prediction model can be adjusted according to the analysis result so as to improve the prediction accuracy of the battery margin prediction model.
In a possible implementation manner, an embodiment of the present application provides a specific implementation manner for calculating the contribution values corresponding to the margin factor features respectively by using a Treeshap algorithm based on an integrated tree model, which is specifically referred to as B1 to B3 below.
Based on the contents of S201-S202, the embodiment of the present application provides an analysis method for a battery remaining prediction error, which trains an integrated tree model based on an original data set to obtain the trained integrated tree model. Wherein the integration tree model is used for outputting the battery residual prediction error estimation value. The original data set used for training the integration tree model comprises a plurality of preselected margin factor characteristics and label values, wherein the label values are battery margin prediction errors between the battery real margin and the battery predicted margin output by the battery margin prediction model. And after the integrated tree model is trained, calculating the contribution values corresponding to the residual factor characteristics by using a Treeshap algorithm based on the integrated tree model. The contribution value corresponding to the margin factor characteristic is a quantized value of the influence of the margin factor characteristic on the battery margin prediction error, and can represent the influence of the margin factor characteristic on the battery margin prediction error. In this way, different margin factor characteristics are subjected to attribution interpretation, and the influence of each margin factor characteristic on the battery margin prediction error is analyzed in an interpretable manner. The determined margin factor characteristics are greatly helpful for improving the design and the system performance.
In a possible implementation manner, an embodiment of the present application provides a specific implementation manner for training an ensemble tree model based on an original data set in S201 to obtain a trained ensemble tree model, including:
a1: and performing feature selection on the original data set to obtain the original data set after feature selection.
The raw data set includes a plurality of preselected margin factor characteristics and a label value. In order to improve the data processing efficiency and the training efficiency of the integrated tree model, the residual factor features that have an influence degree on the battery residual prediction error that meets a certain condition are screened from the preselected residual factor features, and then the screened residual factor features are analyzed in the subsequent S202.
In a possible implementation manner, an embodiment of the present application provides a specific implementation manner of A1, including a11-a12:
a11: and selecting the characteristics of the plurality of preselected residual factor characteristics, and screening the plurality of preselected residual factor characteristics to obtain the plurality of residual factor characteristics.
A plurality of margin factor characteristics obtained by screening meet certain conditions, which can be seen in detail as A102.
In one or more embodiments, A11 can include the following steps A101-A102:
a101: and inputting the characteristics of the preselected margin factors into the integrated tree model to obtain a battery margin prediction error estimation value output by the integrated tree model.
It is understood that the ensemble tree model in this step may be the ensemble tree model to be trained. And inputting the characteristics of the preselected margin factors into the integrated tree model to obtain a battery margin prediction error estimation value output by the integrated tree model.
A102: based on a plurality of preselected residual factor characteristics and the battery residual prediction error estimation value, calculating information gains corresponding to the preselected residual factor characteristics respectively, determining the preselected residual factor characteristics of which the information gains meet a first preset threshold value as the residual factor characteristics, or calculating a correlation coefficient between each preselected residual factor characteristic and the battery residual prediction error estimation value, and determining the preselected residual factor characteristics of which the correlation coefficient meets a second preset threshold value as the residual factor characteristics.
As an alternative example, the preselected margin factor characteristic is filtered according to the information gain. The information gain can represent the amount of information that a preselected margin factor characteristic can bring to the system, and the larger the information gain, the more information is brought, which means that the preselected margin factor characteristic is more important.
The information gain is the difference between the information entropy and the conditional entropy. In specific implementation, the information entropy and the conditional entropy corresponding to each preselected margin factor characteristic can be calculated based on a plurality of preselected margin factor characteristics and the estimated battery margin prediction error value. And calculating the information gain corresponding to each preselected margin factor characteristic according to the information entropy and the conditional entropy corresponding to each preselected margin factor characteristic. And then, comparing the information gain corresponding to each pre-selected margin factor characteristic with a first preset threshold value respectively, and determining the pre-selected margin factor characteristic of which the information gain meets the first preset threshold value as the margin factor characteristic.
It is understood that the first preset threshold may be set according to practical situations, and is not limited herein.
As another alternative example, the preselected margin factor characteristic is filtered according to the correlation coefficient. In specific implementation, a correlation coefficient between each preselected margin factor characteristic and the battery margin prediction error estimation value is calculated, and the preselected margin factor characteristic of which the correlation coefficient meets a second preset threshold value is determined as the margin factor characteristic.
It will be appreciated that the larger the correlation coefficient, the more important the preselected margin factor characteristic. The second preset threshold may be set according to actual conditions, and is not limited herein.
A12: the plurality of margin factor features and the label values constitute an original data set after feature selection.
After the plurality of margin factor characteristics are obtained through screening, the plurality of margin factor characteristics and the label value form an original data set after characteristic selection.
Thus, based on A11-A12, the original data set after feature selection can be obtained. The ensemble tree model is trained based on the original data set after feature selection, and training efficiency can be improved.
A2: and dividing the original data set after feature selection into a training data set and a verification data set.
And optimizing the hyper-parameters of the integration tree model while training the integration tree model. Based on the method, the original data set after feature selection is divided into a training data set and a verification data set.
A3: and acquiring each preselected hyper-parameter of the integrated tree model.
Each preselected hyper-parameter is determined, or, alternatively, may be understood as determining a range of values for the hyper-parameter. The hyper-parameter can be determined according to a specific structure of the integrated tree model, for example, one hyper-parameter of the integrated tree model is the depth of the tree of the integrated tree model.
A4: and training the integrated tree models respectively corresponding to the preselected hyper-parameters based on the training data set, and acquiring the integrated tree models respectively corresponding to the trained preselected hyper-parameters.
The different pre-selected hyper-parameters correspond to respective integration tree models. And based on the training data set, independently training the integrated tree models respectively corresponding to the preselected hyper-parameters to obtain the integrated tree models respectively corresponding to the preselected hyper-parameters after training.
The training data set comprises a training feature set and a training label value, wherein the training feature set is composed of a plurality of margin factor features.
In a possible implementation manner, an embodiment of the present application provides a specific implementation manner of A4, including:
a41: and inputting the training feature set into the integrated tree models respectively corresponding to the preselected hyper-parameters to obtain the battery margin prediction error estimation values respectively output by the integrated tree models.
And the integrated tree model is used for inputting the training feature set and outputting the battery margin prediction error estimation value. And inputting the training feature set into the integrated tree models respectively corresponding to the preselected hyper-parameters, so as to obtain the battery margin prediction error estimation value output by each integrated tree model.
A42: and acquiring loss values for training respectively corresponding to the integrated tree models based on the battery residue prediction error estimated values and the label values for training.
The training label value is a battery remaining prediction error, and can be understood as an expected value of a battery remaining prediction error estimated value. Taking an integrated tree model corresponding to one preselected hyper-parameter as an example, after obtaining a battery residual prediction error estimation value output by the integrated tree model, based on the battery residual prediction error estimation value and a training label value, a training loss value corresponding to the integrated tree model can be obtained. Similarly, the training loss values corresponding to the respective ensemble tree models can be obtained.
A43: and training the corresponding integrated tree model based on the loss value for training, repeatedly inputting the feature set for training into the integrated tree model respectively corresponding to each preselected hyper-parameter to obtain the battery residual prediction error estimation value respectively output by each integrated tree model and the subsequent steps until the preset condition is reached.
And training the integrated tree model based on the training loss value corresponding to the integrated tree model, and judging whether the preset condition is met. And when a preset condition is reached, stopping the training process of the integrated tree model. And when the preset condition is not met, inputting the training feature set into the integrated tree models respectively corresponding to the preselected hyper-parameters to obtain the battery residual prediction error estimated values respectively output by the integrated tree models and the subsequent steps repeatedly until the preset condition is met.
As an alternative example, the preset condition is that the loss value for training satisfies a preset threshold. As another alternative example, the preset condition is a preset number of training times reached. It can be understood that the embodiment of the present application does not limit the specific content of the preset condition, and can be determined according to the actual situation.
A5: and acquiring the verification errors of the integrated tree models respectively corresponding to the preselected hyper-parameters based on the verification data set.
And after the integrated tree models corresponding to the preselected hyper-parameters are trained, verifying the integrated tree models based on the verification data set so as to test the training effect of the integrated tree models.
It is understood that the verification data set includes a verification-use feature set composed of a plurality of margin factor features and a verification-use tag value. And the training feature set and the verification feature set form all feature sets in the original data set after feature selection. The training label values and the verification label values constitute all the label values in the original data set after feature selection.
In specific implementation, the feature set for verification is input into the integrated tree model corresponding to each preselected hyper-parameter, and the battery margin prediction error estimation value output by each integrated tree model is obtained. And acquiring the verification errors of the integrated tree models respectively corresponding to the preselected hyper-parameters based on the battery residual prediction error estimated values and the tag values for verification.
A6: and searching target hyper-parameters of the integrated tree model by adopting a tree structure Parerson estimation TPE algorithm based on the verification errors of the integrated tree models respectively corresponding to the preselected hyper-parameters, and acquiring the integrated tree model corresponding to the target hyper-parameters.
The Tree-structured Parzen Estimator (TPE) algorithm is an algorithm for optimizing hyper-parameters by using a gaussian mixture model. After obtaining the verification error of the integrated tree model corresponding to each preselected hyper-parameter, setting an error threshold. And determining the preselected hyperparameter with the verification error smaller than the error threshold as a first preselected hyperparameter, wherein the first preselected hyperparameter is a better preselected hyperparameter, the rest preselected hyperparameters are second preselected hyperparameters, and the second preselected hyperparameters are the rest hyperparameters.
The TPE algorithm maintains two gaussian mixture models, i.e., a first gaussian mixture model l (x) and a second gaussian mixture model g (x). In specific implementation, the first gaussian mixture model l (x) is fitted based on the first preselected hyper-parameter and the verification error of the integrated tree model corresponding to the first preselected hyper-parameter. And fitting a second Gaussian mixture model g (x) based on the second preselected hyperparameter and the verification error of the integrated tree model corresponding to the second preselected hyperparameter. Wherein x is a preselected hyperparameter. l (x) and g (x) are two Gaussian mixture models maintained by the TPE algorithm.
Further, the objective function is fitted based on the verification errors of the integrated tree models corresponding to the first Gaussian mixture model l (x), the second Gaussian mixture model g (x), each preselected hyper-parameter and each preselected hyper-parameter. It can be seen that the maximization of the objective function can be achieved by minimizing g (x)/l (x) in the objective function. Thus, by minimizing g (x)/l (x), a new hyper-parameter x is obtained.
And then, training the integrated tree model corresponding to the new hyper-parameter x based on the training data set, and acquiring the verification error of the integrated tree model corresponding to the new hyper-parameter x through the verification data set. And re-fitting the first Gaussian mixture model l (x) and the second Gaussian mixture model g (x) according to the obtained new hyper-parameter x, the verification error of the integrated tree model corresponding to the new hyper-parameter x, the previous preselected hyper-parameter and the verification error of the integrated tree model corresponding to the previous preselected hyper-parameter, further re-fitting the objective function and re-minimizing g (x)/l (x). Therefore, g (x)/l (x) is continuously minimized until the preset super parameter adjusting times are reached, and the target super parameter is obtained.
After the target hyper-parameter is obtained, the integrated tree model corresponding to the trained target hyper-parameter can also be obtained. It can be understood that the hyper-parameters of the integrated tree model trained in S201 are target hyper-parameters.
Based on the contents of the above-mentioned A1-A6, in the process of training the ensemble tree model, the hyper-parameters in the ensemble tree model can be optimized based on the TPE algorithm, so that the target hyper-parameters of the ensemble tree model and the trained ensemble tree model can be obtained.
In a possible implementation manner, an embodiment of the present application provides a specific implementation manner that, in S202, based on an integrated tree model, a treshop algorithm is adopted to calculate contribution values corresponding to each margin factor feature, including:
b1: obtaining a plurality of characteristic alliances corresponding to the target allowance factor characteristics; the target margin factor characteristic is each of the respective margin factor characteristics.
After the plurality of margin factor features are screened from the plurality of preselected margin factor features, the contribution value corresponding to each margin factor feature can be calculated by adopting a Treeshap algorithm based on the integrated tree model. In particular, for convenience of description, each of the margin factor features is described as a target margin factor feature. The target margin factor characteristic is used as an example for explanation.
Wherein the plurality of margin factor features may constitute the full set N. And the plurality of feature unions corresponding to the target margin factor features are a plurality of subsets of the feature set consisting of the remaining margin factor features except the target margin factor features in the full set N. The number of subsets is the number of feature unions corresponding to the target margin factor features.
For example, when the plurality of margin factor characteristics are specifically a battery voltage and a battery current, the full set N is { battery voltage, battery current }, and the target margin factor characteristic is either the battery voltage or the battery current. When the target margin factor characteristic is a battery voltage, a plurality of characteristics corresponding to the battery voltage are united as { battery current } and an empty set. When the target margin factor characteristic is a battery current, the plurality of characteristics corresponding to the battery current are united as { battery voltage } and an empty set.
B2: and acquiring parameters of the trained integrated tree model.
After the training of the integrated tree model is completed, the parameters of the trained integrated tree model can be obtained. The parameters of the integrated tree model are specifically parameters of a decision tree in the integrated tree model. In one or more embodiments, the parameters of the integrated tree model include primarily { v, a, b, t, r, d }.
Wherein, the vector v is a label value at each node in the integrated tree model, and the dimension of v is q. For an internal node, the tag value is designated as "internal", and if not, the tag value is designated as the output value of the node. The internal nodes are non-leaf nodes, and the non-internal nodes are leaf nodes. Vectors a and b represent left and right node index values for each internal node in the integrated tree model. The vector t contains the threshold value for each internal node in the integrated tree model. d is an index vector used to represent the features split in the interior node. The vector r represents the coverage (i.e., the number of data samples in the sub-tree) of each node in the integrated tree model.
B3: and calculating the contribution value corresponding to the target margin factor characteristic by adopting a Treeshap algorithm based on the parameters of the integrated tree model and a plurality of characteristic alliances corresponding to the target margin factor characteristic.
Treeshap algorithm, a specific algorithm of the sharp algorithm, is an attribution analysis algorithm which can be used for explaining model output, and is derived from the sharp value in the idea of the joint game theory to quantify the contribution value of each player. In the embodiment of the application, the method can be used for quantifying the contribution values corresponding to the margin factor characteristics respectively.
The calculation of the original sharley value is described below.
Specifically, the contribution value is an average of marginal contributions of the margin factor feature under different feature alliances of the margin factor feature. The calculation formula of the contribution value corresponding to the ith margin factor characteristic is specifically as follows:
where N is the full set, S is a subset of N, f x (S) is the performance under the subset. N \ i is a set of features except the ith feature. N is the number of features in set N, N! Is a factorial of | N |. S is the number of features in the set S, S |! Is a factorial of | S |.
Is the contribution of the ith margin factor feature.
For example, the full set N is { cell voltage, cell current }, and i is 1 or 2. Let the margin factor at i =1 be characterized as the battery voltage, and the margin factor at i =2 be characterized as the battery current. Then, when i =1, { i } is { battery voltage }, and the subset S is { battery current } or an empty set, i.e., the plurality of characteristics corresponding to the battery voltage are associated with { battery current } or an empty set. It can be known that, when the subset S is { battery current }, (S utou { i }) is { battery voltage, battery current }; when the subset S is an empty set, (S utoxy { i }) is { battery voltage }. The feature federation may also be denoted by S.
When i =2, { i } is { battery current }, and the subset S is { battery voltage } or an empty set, that is, a plurality of characteristics corresponding to the battery current are associated together as { battery voltage } or an empty set. I.e., the plurality of characteristics corresponding to the battery current are associated as { battery voltage } or empty set. It can be known that, when the subset S is { battery voltage }, (S { [ i }) is { battery voltage, battery current }; when the subset S is an empty set, (S utou { i }) is { battery current }.
The verification shows that the complexity of the Shapley formula in calculation is in an exponential relation with the feature quantity, the time complexity is high, and the NP-hard problem exists. The NP-hard problem refers to the problem that all non-deterministic polynomial NP problems can encounter within polynomial time complexity.
Based on this, in this embodiment, under the condition of following the above-mentioned sharley formula principle, the contribution values corresponding to the respective margin factor features are calculated by using the treshop algorithm with higher calculation efficiency. The Treeshap algorithm is used for calculating based on model characteristics and samples at the same time, so that the calculation efficiency can be improved while the result is ensured, and the time complexity is greatly shortened.
There are many Treeshap algorithms, and the basic Treeshap algorithm is used as an example for brief explanation. In the basic Treeshap algorithm, f is used x (S)＝E[f(x)|x s ]. Wherein x is s Are elements in the subset S. That is, for the ith margin factor feature, a plurality of different feature unions corresponding to the margin factor feature are obtained, and the conditional expectation E [ f (x) | x ] under each feature union corresponding to the ith margin factor feature is estimated s ]And substituting the estimation result into the Shapley formula to calculate the contribution value corresponding to the ith margin factor characteristic.
Wherein E [ f (x) | x ] is estimated based on a decision tree in the ensemble tree model and the parameters { v, a, b, t, r, d } of the ensemble tree model s ]. In specific implementation, after the characteristic alliance S corresponding to the ith margin factor characteristic is determined, regression is conducted on each node in a decision tree in the whole integrated number modelAnd if the current node is a leaf node, the value of the current node is the label value of the leaf node. If the current node is not a leaf node, judging whether an index vector d corresponding to the current node is in S, if so, taking the value of the current node as a left node index value or a right node index value of the node; if not in S, calculating a first coverage rate r1 of the left node index value of the current node and the node corresponding to the left node index value, calculating a second coverage rate r2 of the pole corresponding to the right node index value of the current node and the right node index value, calculating a sum of the first coverage rate and the second coverage rate, determining the quotient of the sum and the coverage rate of the current node as the value of the current node, and finally estimating E [ f (x) | x through the value of each node s ]。
It can be understood that the basic treshap algorithm is only one basic Treeshap algorithm, and on this basis, the length of each possible subset may be tracked in the regression process, instead of counting all subsets of the feature set composed of the remaining margin factor features except the target margin factor feature in the full set N, so as to reduce the time complexity, which is not described herein again.
A Treeshap algorithm based on an integrated tree model is used as a specific algorithm of a Shap algorithm, and the three properties of the SHAP value are guaranteed, and meanwhile, the structure of the integrated tree is used for efficiently estimating the SHAP value. In the embodiment of the present application, the SHAP value is the contribution value. Wherein, three properties of the SHAP value are: local accuracy, i.e. the sum of the contribution values of the features equals the model output; consistency, i.e., a feature that is important to a model does not degrade important properties by changing a model; the deficiency, i.e. the feature with a SHAP value of zero, has no effect on the model (a SHAP value of zero, i.e. a contribution value of zero).
In the embodiment of the application, the contribution values corresponding to the margin factor characteristics can be efficiently calculated by adopting a Treeshap algorithm. The influence of each margin factor characteristic on the battery margin prediction error is quantized through the contribution value, and the method is made to be interpretable.
Based on the contents of B1 to B3, it can be seen that the contribution values corresponding to the respective margin factor features can be calculated by the treshop algorithm. Thus, the influence of each margin factor characteristic on the battery margin prediction error is interpretable.
Based on the method for analyzing the battery residual quantity prediction error provided by the above method embodiment, the embodiment of the present application further provides an analysis device for the battery residual quantity prediction error, and the analysis device for the battery residual quantity prediction error will be described with reference to the accompanying drawings.
Referring to fig. 3, the figure is a schematic structural diagram of an analysis apparatus for a battery remaining prediction error according to an embodiment of the present disclosure. As shown in fig. 3, the apparatus for analyzing a battery residual prediction error includes:
an obtaining unit 301, configured to train an ensemble tree model based on an original data set, and obtain the trained ensemble tree model; the original data set comprises a plurality of preselected margin factor characteristics and a tag value, wherein the tag value is a battery margin prediction error between a battery real margin and a battery prediction margin output by a battery margin prediction model; the integrated tree model is used for outputting a battery margin prediction error estimation value;
a calculating unit 302, configured to calculate, based on the integrated tree model, contribution values corresponding to the margin factor features by using a treshop algorithm, and analyze the battery margin prediction error based on the contribution values corresponding to the margin factor features; each margin factor characteristic is selected from the plurality of preselected margin factor characteristics.
In a possible implementation manner, the computing unit 302 includes:
the first obtaining subunit is used for obtaining a plurality of characteristic alliances corresponding to the target allowance factor characteristics; the target margin factor characteristic is each of the margin factor characteristics;
the second acquisition subunit is used for acquiring the parameters of the trained integrated tree model;
and the first calculating subunit is configured to calculate, based on the parameters of the integrated tree model and the plurality of feature unions corresponding to the target residue factor feature, a contribution value corresponding to the target residue factor feature by using a treshop algorithm.
In a possible implementation manner, the obtaining unit 301 includes:
the first selection subunit is used for carrying out feature selection on the original data set and acquiring the original data set after the feature selection;
the dividing subunit is used for dividing the original data set after the feature selection into a training data set and a verification data set;
the third acquisition subunit is used for acquiring all preselected hyper-parameters of the integrated tree model;
the training subunit is used for training the integrated tree models respectively corresponding to the preselected hyper-parameters based on the training data set, and acquiring the integrated tree models respectively corresponding to the trained preselected hyper-parameters;
a fourth obtaining subunit, configured to obtain, based on the verification data set, verification errors of the integrated tree models respectively corresponding to the preselected hyper-parameters;
and the searching subunit is used for searching the target hyper-parameters of the integrated tree model by adopting a tree structure Parson estimation TPE algorithm based on the verification errors of the integrated tree models respectively corresponding to the preselected hyper-parameters, and acquiring the integrated tree models corresponding to the target hyper-parameters.
In a possible implementation manner, the first selecting subunit includes:
the second selection subunit is used for performing feature selection on the plurality of preselected residual factor features and screening the plurality of preselected residual factor features to obtain a plurality of residual factor features;
and the composition subunit is used for composing the original data set after feature selection by the plurality of margin factor features and the label value.
In one possible implementation manner, the second selecting subunit includes:
the first input subunit is used for inputting the preselected margin factor characteristics into the integrated tree model and acquiring a battery margin prediction error estimation value output by the integrated tree model;
and the second calculating subunit is configured to calculate, based on the plurality of preselected residual factor features and the battery residual prediction error estimate, information gains corresponding to the preselected residual factor features, respectively, determine, as a residual factor feature, a preselected residual factor feature in which the information gain satisfies a first preset threshold, or calculate a correlation coefficient between each preselected residual factor feature and the battery residual prediction error estimate, and determine, as a residual factor feature, a preselected residual factor feature in which the correlation coefficient satisfies a second preset threshold.
In one possible implementation manner, the training data set includes a training feature set composed of a plurality of margin factor features and a training label value; the training subunit includes:
the second input subunit is used for inputting the training feature set into the integrated tree models respectively corresponding to the preselected hyper-parameters so as to obtain battery residual prediction error estimated values respectively output by the integrated tree models;
a fifth obtaining subunit, configured to obtain, based on each battery residual prediction error estimation value and the training label value, a training loss value corresponding to each ensemble tree model;
and the execution subunit is used for training the corresponding integrated tree model based on the loss value for training, and repeatedly executing the integrated tree model corresponding to the input of the feature set for training into each preselected hyper-parameter so as to obtain the battery residue prediction error estimation value output by each integrated tree model and subsequent steps until a preset condition is reached.
In addition, an embodiment of the present application further provides an electronic device, including:
one or more processors;
a storage device having one or more programs stored thereon,
when executed by the one or more processors, cause the one or more processors to implement a method of analyzing battery residue prediction error as described in any one of the above.
In addition, an embodiment of the present application further provides a computer readable medium, on which a computer program is stored, where the program is executed by a processor to implement any one of the above methods for analyzing a battery remaining prediction error.
The embodiment of the application provides an analysis device for a battery residual prediction error, which trains an integration tree model based on an original data set to obtain the integrated tree model after training. Wherein the integration tree model is used for outputting the battery residual prediction error estimation value. The original data set used for training the integration tree model comprises a plurality of preselected margin factor characteristics and label values, wherein the label values are battery margin prediction errors between the battery real margin and the battery prediction margin output by the battery margin prediction model. And after the integrated tree model is trained, calculating the contribution values corresponding to the margin factor characteristics by using a Treeshap algorithm based on the integrated tree model. The contribution value corresponding to the residue factor characteristic is a quantized value of the influence of the residue factor characteristic on the battery residue prediction error, and can be used for representing the influence of the residue factor characteristic on the battery residue prediction error. In this way, the influence of each residual factor characteristic on the battery residual prediction error can be analyzed in an interpretable manner.
From the above description of the embodiments, it is clear to those skilled in the art that all or part of the steps in the method of the above embodiments may be implemented by software plus a necessary general hardware platform. Based on such understanding, the technical solution of the present application may be essentially or partially implemented in the form of a software product, which may be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the method according to the embodiments or some parts of the embodiments of the present application.
It should be noted that, in the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments may be referred to each other. The method disclosed by the embodiment corresponds to the system disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the system part for description.
It should also be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of additional like elements in a process, method, article, or apparatus that comprises the element.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. A method for analyzing a prediction error of a remaining battery level, the method comprising:
training the integrated tree model based on the original data set to obtain a trained integrated tree model; the original data set comprises a plurality of preselected margin factor characteristics and a tag value, wherein the tag value is a battery margin prediction error between a battery real margin and a battery prediction margin output by a battery margin prediction model; the integrated tree model is used for outputting a battery margin prediction error estimation value;
calculating the contribution values corresponding to the margin factor characteristics by adopting a Treeshap algorithm based on an integrated tree model, and analyzing the battery margin prediction error based on the contribution values corresponding to the margin factor characteristics; each margin factor characteristic is selected from the plurality of preselected margin factor characteristics.
2. The method according to claim 1, wherein the calculating, based on the ensemble tree model, the contribution values corresponding to the margin factor features respectively by using a Treeshap algorithm includes:
obtaining a plurality of characteristic alliances corresponding to the target allowance factor characteristics; the target margin factor characteristic is each of the margin factor characteristics;
acquiring parameters of the trained integrated tree model;
and calculating the contribution value corresponding to the target margin factor characteristic by adopting a Treeshap algorithm based on the parameters of the integrated tree model and the plurality of characteristic alliances corresponding to the target margin factor characteristic.
3. The method of claim 1, wherein the hyper-parameters of the trained ensemble tree model are target hyper-parameters; the training of the integrated tree model based on the original data set to obtain the trained integrated tree model comprises the following steps:
carrying out feature selection on the original data set to obtain the original data set after feature selection;
dividing the original data set after the feature selection into a training data set and a verification data set;
obtaining all preselected hyper-parameters of the integrated tree model;
training the integrated tree models corresponding to the preselected hyper-parameters respectively based on the training data set, and acquiring the integrated tree models corresponding to the preselected hyper-parameters respectively after training;
based on the verification data set, obtaining verification errors of the integrated tree models respectively corresponding to the preselected hyper-parameters;
and searching a target hyper-parameter of the integrated tree model by adopting a tree structure Parrson estimation TPE algorithm based on the verification error of the integrated tree model corresponding to each preselected hyper-parameter, and acquiring the integrated tree model corresponding to the target hyper-parameter.
4. The method of claim 3, wherein the performing feature selection on the original data set to obtain the feature-selected original data set comprises:
selecting the characteristics of the preselected residual factor characteristics, and screening the preselected residual factor characteristics to obtain a plurality of residual factor characteristics;
and combining the plurality of margin factor characteristics and the label value into an original data set after characteristic selection.
5. The method of claim 4, wherein the selecting the features from the plurality of preselected margin factor features and the screening the plurality of preselected margin factor features to obtain the plurality of margin factor features comprises:
inputting the preselected margin factor characteristics into the integrated tree model to obtain a battery margin prediction error estimation value output by the integrated tree model;
and calculating information gains corresponding to the preselected residual factor characteristics respectively based on the preselected residual factor characteristics and the battery residual prediction error estimation value, determining the preselected residual factor characteristics of which the information gains meet a first preset threshold value as residual factor characteristics, or calculating a correlation coefficient between each preselected residual factor characteristic and the battery residual prediction error estimation value, and determining the preselected residual factor characteristics of which the correlation coefficient meets a second preset threshold value as residual factor characteristics.
6. The method according to claim 3, wherein the training data set comprises a training feature set composed of a plurality of margin factor features and a training label value; the training of the integrated tree model corresponding to each preselected hyper-parameter based on the training data set comprises:
inputting the training feature set into the integrated tree models respectively corresponding to the preselected hyper-parameters to obtain battery margin prediction error estimation values respectively output by the integrated tree models;
acquiring training loss values corresponding to the integrated tree models respectively based on the battery residue prediction error estimated values and the training label values;
and training the corresponding integrated tree model based on the loss value for training, and repeatedly executing the steps of inputting the characteristic set for training into the integrated tree model respectively corresponding to each preselected hyper-parameter to obtain the battery residual prediction error estimated value respectively output by each integrated tree model and the subsequent steps until a preset condition is reached.
7. An apparatus for analyzing a prediction error of a remaining battery level, the apparatus comprising:
the acquisition unit is used for training the integrated tree model based on the original data set and acquiring the trained integrated tree model; the original data set comprises a plurality of preselected margin factor characteristics and a tag value, wherein the tag value is a battery margin prediction error between a battery real margin and a battery prediction margin output by a battery margin prediction model; the integrated tree model is used for outputting a battery margin prediction error estimation value;
the calculation unit is used for calculating the contribution values corresponding to the margin factor characteristics by adopting a Treeshap algorithm based on the integrated tree model, and analyzing the battery margin prediction error based on the contribution values corresponding to the margin factor characteristics; each margin factor characteristic is selected from the plurality of preselected margin factor characteristics.
8. The apparatus of claim 7, wherein the computing unit comprises:
the first obtaining subunit is used for obtaining a plurality of characteristic alliances corresponding to the target allowance factor characteristics; the target margin factor characteristic is each of the margin factor characteristics;
the second acquisition subunit is used for acquiring the parameters of the trained integrated tree model;
and the first calculating subunit is configured to calculate, based on the parameters of the integrated tree model and the multiple feature unions corresponding to the target residue factor feature, a contribution value corresponding to the target residue factor feature by using a Treeshap algorithm.
9. An electronic device, comprising:
one or more processors;
a storage device having one or more programs stored thereon,
when executed by the one or more processors, cause the one or more processors to implement the method of analyzing battery remaining prediction error as recited in any of claims 1-6.
10. A computer-readable medium, on which a computer program is stored, wherein the program, when executed by a processor, implements the method of analyzing a battery residual prediction error according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210832225.8A CN115219910A (en) | 2022-07-15 | 2022-07-15 | Analysis method and device for battery residue prediction error |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210832225.8A CN115219910A (en) | 2022-07-15 | 2022-07-15 | Analysis method and device for battery residue prediction error |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115219910A true CN115219910A (en) | 2022-10-21 |
Family
ID=83611347
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210832225.8A Pending CN115219910A (en) | 2022-07-15 | 2022-07-15 | Analysis method and device for battery residue prediction error |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115219910A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115544900A (en) * | 2022-11-28 | 2022-12-30 | 深圳联友科技有限公司 | Method for analyzing influence factors of endurance mileage of electric vehicle based on Shap algorithm |
-
2022
- 2022-07-15 CN CN202210832225.8A patent/CN115219910A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115544900A (en) * | 2022-11-28 | 2022-12-30 | 深圳联友科技有限公司 | Method for analyzing influence factors of endurance mileage of electric vehicle based on Shap algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Meng et al. | Long short-term memory network with Bayesian optimization for health prognostics of lithium-ion batteries based on partial incremental capacity analysis | |
| CN110488202B (en) | Vehicle battery state of charge estimation method based on deep neural network | |
| Grover et al. | Best arm identification in multi-armed bandits with delayed feedback | |
| CN110658460B (en) | Battery life prediction method and device for battery pack | |
| Qin et al. | State of health prediction for lithium-ion battery using a gradient boosting-based data-driven method | |
| Huang et al. | A comprehensively optimized lithium-ion battery state-of-health estimator based on Local Coulomb Counting Curve | |
| Qi et al. | A new adaptation method based on adaptability under k-nearest neighbors for case adaptation in case-based design | |
| Myshkov et al. | Posterior distribution analysis for bayesian inference in neural networks | |
| CN115236519A (en) | Lithium battery health state prediction method and device based on hidden Markov model | |
| Zhang et al. | State of health estimation method for lithium-ion batteries using incremental capacity and long short-term memory network | |
| CN113721159A (en) | Data-driven battery health estimation method oriented to random local charging | |
| CN113128671A (en) | Service demand dynamic prediction method and system based on multi-mode machine learning | |
| Tian et al. | State-of-charge estimation for lithium-ion batteries based on attentional sequence-to-sequence architecture | |
| Gao et al. | HFCM-LSTM: A novel hybrid framework for state-of-health estimation of lithium-ion battery | |
| Liu et al. | Wasserstein generative learning of conditional distribution | |
| CN115219910A (en) | Analysis method and device for battery residue prediction error | |
| CN112801231B (en) | Decision model training method and device for business object classification | |
| CN113376540B (en) | LSTM battery health state estimation method based on evolutionary attention mechanism | |
| Hatherall et al. | Remaining discharge energy estimation for lithium-ion batteries using pattern recognition and power prediction | |
| Li et al. | State-of-health estimation method for fast-charging lithium-ion batteries based on stacking ensemble sparse Gaussian process regression | |
| Anandhakumar et al. | Extreme learning machine model with honey badger algorithm based state-of-charge estimation of lithium-ion battery | |
| CN117310504A (en) | Battery health state evaluation method, system, equipment and medium | |
| CN116796821A (en) | Efficient neural network architecture searching method and device for 3D target detection algorithm | |
| CN116008815A (en) | Method, device, equipment, storage medium and vehicle for detecting short circuit in battery cell | |
| Abbas et al. | A three-way classification with game-theoretic N-soft sets for handling missing ratings in context-aware recommender systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |