CN108287317B - Battery power prediction model generation method and system and power prediction method and system - Google Patents
Battery power prediction model generation method and system and power prediction method and system Download PDFInfo
- Publication number
- CN108287317B CN108287317B CN201810068746.4A CN201810068746A CN108287317B CN 108287317 B CN108287317 B CN 108287317B CN 201810068746 A CN201810068746 A CN 201810068746A CN 108287317 B CN108287317 B CN 108287317B
- Authority
- CN
- China
- Prior art keywords
- power
- battery
- data
- prediction
- module
- 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.)
- Active
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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Charge And Discharge Circuits For Batteries Or The Like (AREA)
Abstract
The invention discloses a method and a system for generating a battery power prediction model and a power prediction methodThe method and the system, the generation method comprises the following steps: s1Acquiring power characteristic data of each sample battery in a training sample set, wherein the power characteristic data comprises discharge power data of each charge and discharge of the sample battery; s2Preprocessing the power characteristic data to generate a corresponding power data matrix; s3And performing model training on a convolutional neural network algorithm by using the power data matrix of the sample battery in the training sample set to generate a prediction model of the battery power. The invention uses the convolution neural network algorithm to carry out multi-layer dimensionality reduction and characteristic extraction on the input power data matrix, and can obtain a prediction model of the battery power. The prediction model can be used for predicting the power of the power battery to be predicted, particularly the retired power battery, and the full life cycle prediction and real-time monitoring of the power battery to be predicted can be realized according to the power prediction result.
Description
Technical Field
The invention relates to the field of power batteries, in particular to a battery power prediction model generation method and system and a power prediction method and system.
Background
With the rapid development and popularization of new energy vehicles, the demand of power batteries used by the new energy vehicles is increasing day by day. Limited by the technical level of the current power battery, when the loss of the power battery reaches a certain degree, the power supply characteristic of the power battery cannot reach the power supply standard of the electric automobile, and the power battery must be eliminated to form a retired power battery. The retired power battery is often applied to some fields with relatively low requirements on battery characteristics due to the capability of storing electricity and charging and discharging, such as energy storage power stations, charging piles and the like, so as to realize secondary utilization of the power battery, and the secondary utilization of the retired battery is called battery cascade utilization.
The retired automobile power battery has certain loss in battery characteristics compared with a new battery due to secondary utilization, and the reliability and safety of the retired automobile power battery are different from those of the new battery. Therefore, when the retired battery is actually used, in order to maximize the utilization of the retired battery and ensure the safety and reliability of the battery, it is necessary to monitor and display the internal characteristic data of the battery in real time so that a user or a maintenance worker can know the dynamic information of the battery in real time.
In addition, the lifetime of a retired power battery is often limited, and the aging of a retired battery is highly uncertain. An accidental emergency of the retired battery, such as the situation of battery scrapping or battery deep discharging, will bring inconvenience to the user and increase the maintenance cost for the manufacturer. Therefore, the service life and the safety of the battery are predicted according to the internal characteristic information and the external characteristic information of the battery, and the battery is managed more reliably and safely. The dynamic information guidance of the retired power battery can be provided for the user, and the periodic maintenance reference information can also be provided for the maintenance personnel of the retired power battery manufacturer, so that the dynamic information guidance has great significance for the comfort of the user on the use of the battery and the guidance of the maintenance personnel.
The power batteries have individual differences, and even if the power batteries are brand-new batteries with the same model and the same batch, the consistency of the power batteries also has a high-low problem. In addition, due to the use of various complex working conditions, variables are brought to the life prediction and health management of the battery, and the decay rule of the battery cannot be simply obtained. In the power battery, the retired battery is a battery with a loss of service life and safety, and therefore, displaying and monitoring information such as retired battery power, service life and health data are more necessary. The service life of the battery is often invisible or unpredictable for users, and sudden battery failure or rejection is a problem which is undoubtedly troublesome for users or manufacturers, so that prediction and estimation of the service life and safety of the battery are of great instructive significance for the users or manufacturers.
Disclosure of Invention
The invention aims to overcome the defect that a corresponding effective prediction scheme does not exist for the decay rule of a power battery, particularly a retired power battery in the prior art, and provides a battery power prediction model generation method and system, and a power prediction method and system, which can predict the power characteristic of the power battery and have higher accuracy.
The invention solves the technical problems through the following technical scheme:
the invention provides a generation method of a battery power prediction model, which is characterized by comprising the following steps of:
S1acquiring power characteristic data of each sample battery in a training sample set, wherein the power characteristic data comprises discharge power data of each charge and discharge of the sample battery;
S2preprocessing the power characteristic data to generate a corresponding power data matrix;
S3and performing model training on a convolutional neural network algorithm by using the power data matrix of the sample battery in the training sample set to generate a prediction model of the battery power.
In this scheme, the discharge power data is data including a time sequence variable, and the power data matrix is also a data matrix including a time sequence variable.
In the scheme, a convolutional neural network algorithm is used for carrying out multi-layer dimension reduction and characteristic extraction on the input power data matrix, and a mechanism model, namely a prediction model of the battery power, can be obtained. The convolutional neural network algorithm is adopted to train the prediction model of the battery power, so that individual groups which are most similar to the performance of the individual batteries are searched for the single batteries, the decay condition of the individual is deduced according to the historical data of the similar individuals, and the mutation condition in the historical data is learned to form a mechanism model, so that the decay rule of the power battery is effectively and accurately predicted.
Preferably, step S2The method comprises the following steps:
S21to the workExtracting data from the rate characteristic data to obtain a periodic component and a trend component;
S22and preprocessing the trend component to generate the corresponding power data matrix.
In this embodiment, step S3Generated is a predictive model of the trend component in battery power.
In the scheme, the discharge power data of the battery is divided into a trend component and a period component. The periodic component is a translational component which reflects the inherent characteristics of the variable of the discharge power data. I.e. this component is neither attenuated nor deformed, but is not itself a function of time that can be expressed in a monomial form. Conventionally, the periodic component is represented by a sum or an integral of a set of sinusoids.
In the scheme, in the battery state prediction, the discharge power data is decomposed into a periodic component and a trend component, and only the trend component is used as the model input of the convolutional neural network algorithm, so that the dimension reduction in advance is realized, and the operation efficiency of the convolutional neural network is improved.
Preferably, the discharge power data includes discharge voltage data and discharge current data, and the power characteristic data further includes at least one of a number of times of use, a frequency of use, and a charge/discharge condition.
In the scheme, variables such as the use times, the use frequency, the charge and discharge working conditions and whether abnormal working conditions such as overcharge and overdischarge exist in the power battery can be used as coefficients to participate in an iteration process when a convolutional neural network algorithm is used for model training.
In the scheme, at least two-dimensional data formed by the discharge voltage data and the discharge current data is used for model training of a convolutional neural network algorithm, so that the accuracy of a generated prediction model is higher.
Preferably, step S2The preprocessing comprises data invalidation processing, data normalization processing and data matrixing processing.
In the scheme, the power characteristic data of the sample battery acquired by collection cannot be directly used as input information of a convolutional neural network algorithm, and the power characteristic data needs to be reused after data invalidation processing, data normalization processing and data matrixing input processing. When the input data volume is large, the calculation amount of the algorithm is improved, and the calculation efficiency of the algorithm can be improved by matrixing power characteristic data for convenient calculation.
Preferably, the periodic component is data corresponding to a discharge power curve of the corresponding sample battery in a brand new state.
In the scheme, for a specific battery, the discharge power curve under a brand new state can be determined through empirical values, and the group of data is identified as a periodic component.
Preferably, step S1The method also comprises the following steps:
S0acquiring a total sample set, sampling the total sample set by using a sampling rate r to obtain a sample subset, and setting the sample subset as the training sample set;
step S3The method also comprises the following steps:
S4verifying the prediction model by using the power data matrix of the sample battery which does not participate in the model training in the total sample set, adjusting the parameters of the model training if the output error is greater than the preset error, and executing the step S3。
In the scheme, a prediction model generated by verifying a sample battery which does not participate in model training in a total sample set is used, and when an output error is within a reasonable range, a prediction mechanism is completed; otherwise, repeating the model training step by adjusting the model parameters so as to enable the generated prediction model to be more accurate.
The invention also provides a battery power prediction model generation system which is characterized by comprising a data acquisition module, a preprocessing module and a model generation module;
the data acquisition module is used for acquiring power characteristic data of each sample battery in a training sample set, wherein the power characteristic data comprises discharge power data of each charge and discharge of the sample battery;
the preprocessing module is used for preprocessing the power characteristic data to generate a corresponding power data matrix;
the model generation module is used for performing model training on a convolutional neural network algorithm by using the power data matrix of the sample battery in the training sample set so as to generate a prediction model of the battery power.
Preferably, the preprocessing module comprises a data extraction module and a component preprocessing module;
the data extraction module is used for extracting data from the power characteristic data to obtain a periodic component and a trend component;
the component preprocessing module is used for preprocessing the trend component to generate the corresponding power data matrix.
Preferably, the discharge power data includes discharge voltage data and discharge current data, and the power characteristic data further includes at least one of a number of times of use, a frequency of use, and a charge/discharge condition.
Preferably, the preprocessing in the preprocessing module includes data invalidation processing, data normalization processing and data matrixing processing.
Preferably, the periodic component is data corresponding to a discharge power curve of the corresponding sample battery in a brand new state.
Preferably, the battery power prediction model generation system further comprises a sampling module and a verification module;
the sampling module is used for acquiring a total sample set before the data acquisition module executes, sampling the total sample set by using a sampling rate r to obtain a sample subset, and setting the sample subset as the training sample set;
the verification module is used for verifying the prediction model by using the power data matrix of the sample battery which does not participate in model training in the total sample set after the model generation module executes, and if the output error is larger than a preset error, adjusting the parameter of model training and calling the model generation module.
The invention also provides a battery power prediction method, which is characterized by comprising the following steps:
and T, performing power prediction on the battery to be predicted by using the prediction model generated by the battery power prediction model generation method to obtain a power prediction result of the battery to be predicted.
According to the scheme, the full life cycle prediction and real-time monitoring of the battery to be predicted can be realized according to the power prediction result.
Preferably, step T comprises the steps of:
T1acquiring power characteristic data of the battery to be predicted;
T2preprocessing the power characteristic data to generate a corresponding power data matrix;
T3and inputting the power data matrix into the prediction model to perform power prediction so as to obtain a power prediction result of the battery to be predicted.
The invention also provides a battery power prediction system which is characterized by comprising a prediction module and the battery power prediction model generation system;
the prediction module is used for performing power prediction on the battery to be predicted by using the prediction model so as to obtain a power prediction result of the battery to be predicted.
Preferably, the prediction module comprises a to-be-predicted battery data acquisition module, a to-be-predicted battery preprocessing module and a prediction execution module;
the battery data acquisition module to be predicted is used for acquiring power characteristic data of the battery to be predicted;
the battery pre-processing module to be predicted is used for pre-processing the power characteristic data to generate a corresponding power data matrix;
and the prediction execution module is used for inputting the power data matrix into the prediction model to perform power prediction so as to obtain a power prediction result of the battery to be predicted.
The positive progress effects of the invention are as follows: the battery power prediction model generation method and system, and the power prediction method and system provided by the invention use the convolutional neural network algorithm to perform multi-layer dimension reduction and feature extraction on the input power data matrix, so that a prediction model of the battery power can be obtained. The prediction model can be used for predicting the power of the power battery to be predicted, particularly the retired power battery, and the full life cycle prediction and real-time monitoring of the power battery to be predicted can be realized according to the power prediction result. Furthermore, the discharge power data are decomposed into periodic components and trend components, and only the trend components are used as model input of the convolutional neural network algorithm, so that the dimension reduction in advance is realized, and the operation efficiency of the convolutional neural network is improved.
Drawings
Fig. 1 is a flowchart of a method for generating a battery power prediction model according to embodiment 1 of the present invention.
Fig. 2 is a schematic block diagram of a battery power prediction model generation system according to embodiment 2 of the present invention.
Fig. 3 is a flowchart of a battery power prediction method according to embodiment 3 of the present invention.
Fig. 4 is a block diagram of a battery power prediction system according to embodiment 4 of the present invention.
Fig. 5 is a flow chart of the present invention applied to battery state of health prediction.
FIG. 6 is a flow chart of a battery prediction method when applying the present invention.
Detailed Description
The invention is further illustrated by the following examples, which are not intended to limit the scope of the invention.
Example 1
As shown in fig. 1, the present embodiment provides a method for generating a battery power prediction model, including the following steps:
step S0, acquiring a total sample set, sampling the total sample set by using a sampling rate r to obtain a sample subset, and setting the sample subset as a training sample set;
step S1, obtaining power characteristic data of each sample battery in the training sample set, wherein the power characteristic data comprise the use times, the use frequency, the charge and discharge working conditions and the discharge power data of each charge and discharge of the sample battery, and the discharge power data comprise discharge voltage data and discharge current data;
step S2, performing data extraction on the power characteristic data to obtain a periodic component and a trend component, wherein the periodic component is data corresponding to a discharge power curve of a corresponding sample battery in a brand new state;
step S3, preprocessing the trend component to generate the corresponding power data matrix, wherein the preprocessing comprises data invalidation processing, data normalization processing and data matrixing processing;
step S4, performing model training on a convolutional neural network algorithm by using the power data matrix of the sample battery in the training sample set to generate a prediction model of battery power;
step S5, the power data matrix of the sample battery which does not participate in model training in the total sample set is used for verifying the prediction model, whether the output error is larger than a preset error or not is judged, if yes, the step S6 is executed, and if not, the process is ended;
step S6, adjusting the parameters of the model training, and executing step S4.
In this embodiment, the preprocessing of step S3 may also be performed in exchange with the data extraction of step S2, that is, the data is preprocessed first, and then the periodic component and the trend component are extracted from the preprocessed data.
In this embodiment, the discharge power data, the power data matrix, the discharge voltage data, and the discharge current data are all data including time-series variables. At least two-dimensional data formed by the discharge voltage data and the discharge current data is used for model training of a convolutional neural network algorithm, so that the generated prediction model has higher accuracy than that of a one-dimensional discharge power data directly used.
The gradual attenuation of the power battery along with the use times is a necessary trend, and meanwhile, the discharge power of the battery has periodicity. In this embodiment, the discharge power data of the battery is divided into a trend component and a periodic component. The periodic component is a translational component which reflects the inherent characteristics of the variable of the discharge power data. I.e. this component is neither attenuated nor deformed, but is not itself a function of time that can be expressed in a monomial form. Conventionally, the periodic component is represented by a sum or an integral of a set of sinusoids. In the battery state prediction, only the trend component is used as the model input of the convolutional neural network algorithm, so that the dimension reduction in advance is realized, and the operation efficiency of the convolutional neural network is improved.
In this embodiment, a convolutional neural network algorithm is used to perform multi-layer dimensionality reduction and feature extraction on the input power data matrix, so that a mechanism model, that is, a prediction model of battery power, can be obtained. The training of the prediction model of the battery power is carried out by adopting a convolutional neural network algorithm, so that individual groups which are most similar to the performance of the individual groups are searched for a single battery, the decay condition of the individual is deduced according to the historical data of the similar individual, and the mutation condition in the historical data is learned to form a mechanism model, so that the decay rule of the power battery is effectively and accurately predicted.
Example 2
As shown in fig. 2, the present embodiment provides a battery power prediction model generation system, which includes a sampling module 0, a data acquisition module 1, a preprocessing module 2, a model generation module 3, and a verification module 4.
The sampling module 0 is configured to obtain a total sample set, sample the total sample set by using a sampling rate r to obtain a sample subset, and set the sample subset as a training sample set.
The data acquisition module 1 is configured to acquire power characteristic data of each sample battery in the training sample set, where the power characteristic data includes the number of times of use, the frequency of use, the charging and discharging conditions of the sample battery, and the discharging power data of each charging and discharging, and the discharging power data includes discharging voltage data and discharging current data.
The preprocessing module 2 is configured to preprocess the power characteristic data to generate a corresponding power data matrix.
The model generating module 3 is configured to perform model training on a convolutional neural network algorithm by using the power data matrix of the sample battery in the training sample set, so as to generate a prediction model of battery power.
The verification module 4 is configured to verify the prediction model by using the power data matrix of the sample battery not participating in model training in the total sample set after the model generation module 3 executes the power data matrix, and if an output error is greater than a preset error, adjust a parameter of model training and call the model generation module 3.
In this embodiment, the preprocessing module 2 includes a data extraction module 201 and a component preprocessing module 202. The data extraction module 201 is configured to perform data extraction on the power characteristic data to obtain a periodic component and a trend component, where the periodic component is data corresponding to a discharge power curve of a corresponding sample battery in a brand new state. The component preprocessing module 202 is configured to preprocess the trend component to generate the corresponding power data matrix, where the preprocessing includes data invalidation processing, data normalization processing, and data matrixing processing.
In this embodiment, the discharge power data of the battery is divided into a trend component and a periodic component. In the battery state prediction, only the trend component is used as the model input of the convolutional neural network algorithm, so that the dimension reduction in advance is realized, and the operation efficiency of the convolutional neural network is improved. In this embodiment, a convolutional neural network algorithm is used to perform multi-layer dimensionality reduction and feature extraction on the input power data matrix, so that a mechanism model, that is, a prediction model of battery power, can be obtained. The prediction model can be used for predicting the power of the power battery to be predicted, particularly the retired power battery, and the full life cycle prediction and real-time monitoring of the power battery to be predicted can be realized according to the power prediction result.
Example 3
As shown in fig. 3, the present embodiment provides a method for predicting battery power, including the following steps:
t1, acquiring power characteristic data of the battery to be predicted;
t2, preprocessing the power characteristic data to generate a corresponding power data matrix;
and T3, inputting the power data matrix into the prediction model generated by the battery power prediction model generation method in the embodiment 1 to perform power prediction so as to obtain a power prediction result of the battery to be predicted.
In this embodiment, a decay rule of the battery to be predicted can be obtained according to the result of the power prediction, and the life prediction of the battery is realized according to the result of the power prediction, so as to further perform health management.
The battery power prediction method provided by the embodiment can realize the full life cycle prediction and real-time monitoring of the battery to be predicted.
Example 4
As shown in fig. 4, the present embodiment provides a battery power prediction system, which includes a prediction module 5 and a battery power prediction model generation system 6 in embodiment 2;
the prediction module 5 is configured to perform power prediction on the battery to be predicted by using the prediction model to obtain a result of power prediction of the battery to be predicted. The prediction module 5 comprises a to-be-predicted battery data acquisition module 501, a to-be-predicted battery preprocessing module 502 and a prediction execution module 503;
the to-be-predicted battery data obtaining module 501 is configured to obtain power characteristic data of the to-be-predicted battery;
the battery pre-processing module 502 to be predicted is configured to pre-process the power characteristic data to generate a corresponding power data matrix;
the prediction execution module 503 is configured to input the power data matrix to the prediction model to perform power prediction, so as to obtain a result of power prediction of the battery to be predicted.
In this embodiment, the prediction module 5 can be used to predict the power of the power battery to be predicted, especially the retired power battery, and according to the result of power prediction, the prediction and real-time monitoring of the full life cycle of the battery to be predicted can be achieved.
The following further illustrates the technical solutions and effects of the present invention by means of specific examples.
The method is applied to the prediction of the full life cycle of the power battery during specific implementation, and the specific technical scheme is as follows:
the whole prediction process is divided into two major aspects of a forming mechanism and an application mechanism. The forming mechanism is divided into four steps of data acquisition and processing, model design, model training and model verification, and is detailed in figure 5. The method of power prediction and health management is embodied in model design and model training, and is detailed in fig. 6.
Mechanism of I formation
I.1 data acquisition and processing
I.1.1 measurement data
Common indicators for determining battery performance based on industry experience are: a charging voltage, a charging current, a discharging voltage, a discharging current, an internal resistance of the battery, an SOC (state of charge) and the like. The conventional measurement means is to perform one or more complete charging and discharging processes on the retired battery, and record corresponding experimental numerical charging voltage Ucharge(t), charging Current Icharge(t), discharge voltage Udischarge(t), discharge current Idischarge(t), internal resistance Rinner(t)，SOC。
For the end user, the variable that intuitively affects the sense of use of the battery is the discharge power, which can be measured by the discharge voltage Udischarge(t), discharge current Idischarge(t) the calculation shows that when the discharge power of the battery decays to a critical value or sharply, which indicates that the battery has expired, the battery cannot be used continuously. Therefore, it is the most direct method to calculate the discharge power using the discharge voltage and the discharge current and then input the calculated discharge power as the model input. But at the same time it is not excluded that the remaining measurement data have no relevance to battery life and health.
I.1.2 recording data
Even if the same battery is used, the discharge performance and the residual service life of the battery are related to the previous use times, frequency, charge and discharge working conditions and whether the normal working conditions (over-charge and over-discharge) are expected or not. These variables will participate as coefficients in the iteration.
I.1.3 data processing
The collected data cannot be directly used as input information of the algorithm, and invalid data deleting processing, data normalization processing and data matrix input processing are required.
Data invalidation processing
For input data xiWherein x isi∈{x1,x2…NGet it before
Then to xiUpdating:
data normalization processing
To prevent the gradient from disappearing or diverging in the algorithm, the collected data need to be normalized before being input, xi∈{x1,2…NData after normalization }
Is composed of
Input data matrixing
When the input data amount is large, the calculation amount of the algorithm is increased, and the calculation efficiency is improved by matrixing the input data for convenient calculation.
I.2 model design
The gradual attenuation of the power battery along with the use times is a necessary trend, and meanwhile, the discharge power of the battery has periodicity. Therefore, the discharge power of the battery is divided into a trend component and a period component, i.e. P (t) ═ Pa(t)+Pb(t) wherein PaAs a trend component, PbIs a periodic component.
I.2.1 extraction of periodic components
The periodic component is a translational component that manifests its inherent characteristics with respect to the variable (i.e., the discharge power). I.e. this component is neither attenuated nor deformed, but is not itself a function of time that can be expressed in a monomial form. Conventionally, the periodic component is represented by a sum or an integral of a set of sinusoids. However, in the method, for a specific battery, the discharge power curve P under the brand new state can be determined through empirical values0(t) and identifying the set of data as a periodic component.
I.2.2 extraction of trend component
Pa(t)＝P(t)-P0(t)。
I.3 model training
The convolutional neural network algorithm is a mechanism model obtained by performing multi-layer dimensionality reduction and feature extraction on an n-x-n input matrix. The matrix of n × n may be feature data extracted from a picture, or may be a matrix of pure values (including time-series variables). The convolutional neural network algorithm is widely applied to image recognition, and as a result, the image recognition is also one of supervised learning. The method can also be understood as that individual groups which are most similar to the individual groups are searched for the single battery, the decay condition of the individual is deduced according to the historical data of the similar individuals, and the mutation condition in the historical data is learned to form a mechanism.
I.3.1 model input matrix
The discharge power array of a single cell m in a sample is Pam，e(t)∈{Pam,1(t)，Pam,2(t)，Pam,3(t)，…，Pam,Em(t), wherein the subscript E ═ 1, 2, 3, …, EmExpressed as charge-discharge times, EmThe number of times of the full life charge and discharge of the battery m. And a single charge-discharge cycle Pam，e(t) is a trend component of a set of consecutive data or high frequency sampled data.
For the total number of samples, cell M, the cell power matrix is:
i.3.2 dimension reduction and feature extraction
In an excellent convolutional neural algorithm network, the dimensionality reduction layer and the feature extraction layer are used alternately and repeatedly to achieve the effect of deep learning. And the learning result is normalized in the last layer, so that the consistency of the measurement standard is ensured.
The dimensionality reduction is to reduce data redundancy, increase the speed of computation and ensure reliability. The main means of reducing dimension is low frequency sampling, and the average of the peripheral values is equal.
Feature extraction is the filtering out of unwanted data by filtering tools. The alternate use of dimension reduction and feature extraction for multiple times can ensure the retention of feature data.
I.3.3 model output
The model output is a prediction mechanism of the discharge power matrix.
I.4 model validation
The unused data in the sample S is validated against the model.
II mechanism of application
And (3) carrying out real-time monitoring and full life cycle prediction on a new sample in practical application, comparing real-time data with a past predicted value, and bringing the sample into the total sample to renew a mechanism under the condition of a great difference value.
While specific embodiments of the invention have been described above, it will be appreciated by those skilled in the art that this is by way of example only, and that the scope of the invention is defined by the appended claims. Various changes and modifications to these embodiments may be made by those skilled in the art without departing from the spirit and scope of the invention, and these changes and modifications are within the scope of the invention.
Claims (14)
1. A method for generating a battery power prediction model is characterized by comprising the following steps:
S1acquiring power characteristic data of each sample battery in a training sample set, wherein the power characteristic data comprises discharge power data of each charge and discharge of the sample battery;
S2preprocessing the power characteristic data to generate a corresponding power data matrix;
S3performing model training on a convolutional neural network algorithm by using the power data matrix of the sample battery in the training sample set to generate a prediction model of the battery power;
step S2The method comprises the following steps:
S21performing data extraction on the power characteristic data to obtain a periodic component and a trend component;
S22and preprocessing the trend component to generate the corresponding power data matrix.
2. The method of generating a battery power prediction model according to claim 1, wherein the discharge power data includes discharge voltage data and discharge current data, and the power characteristic data further includes at least one of a number of uses, a frequency of uses, and a charge/discharge condition.
3. The battery power prediction model generation method of claim 1, wherein step S2The preprocessing comprises data invalidation processing, data normalization processing and data matrixing processing.
4. The method according to claim 1, wherein the periodic component is data corresponding to a discharge power curve of the corresponding sample battery in a brand new state.
5. The battery power prediction model generation method of claim 1, wherein step S1The method also comprises the following steps:
S0acquiring a total sample set, sampling the total sample set by using a sampling rate r to obtain a sample subset, and setting the sample subset as the training sample set;
step S3The method also comprises the following steps：
S4Verifying the prediction model by using the power data matrix of the sample battery which does not participate in the model training in the total sample set, adjusting the parameters of the model training if the output error is greater than the preset error, and executing the step S3。
6. A battery power prediction model generation system is characterized by comprising a data acquisition module, a preprocessing module and a model generation module;
the data acquisition module is used for acquiring power characteristic data of each sample battery in a training sample set, wherein the power characteristic data comprises discharge power data of each charge and discharge of the sample battery;
the preprocessing module is used for preprocessing the power characteristic data to generate a corresponding power data matrix;
the model generation module is used for performing model training on a convolutional neural network algorithm by using the power data matrix of the sample battery in the training sample set so as to generate a prediction model of the battery power;
the preprocessing module comprises a data extraction module and a component preprocessing module;
the data extraction module is used for extracting data from the power characteristic data to obtain a periodic component and a trend component;
the component preprocessing module is used for preprocessing the trend component to generate the corresponding power data matrix.
7. The battery power prediction model generation system of claim 6, wherein the discharge power data comprises discharge voltage data and discharge current data, and the power characteristic data further comprises at least one of a number of uses, a frequency of uses, and a charge-discharge condition.
8. The battery power prediction model generation system of claim 6, wherein the pre-processing in the pre-processing module comprises a data invalidation process, a data normalization process, and a data matrixing process.
9. The system according to claim 6, wherein the periodic component is data corresponding to a discharge power curve of the corresponding sample battery under a brand new condition.
10. The battery power prediction model generation system of claim 6, further comprising a sampling module and a validation module;
the sampling module is used for acquiring a total sample set before the data acquisition module executes, sampling the total sample set by using a sampling rate r to obtain a sample subset, and setting the sample subset as the training sample set;
the verification module is used for verifying the prediction model by using the power data matrix of the sample battery which does not participate in model training in the total sample set after the model generation module executes, and if the output error is larger than a preset error, adjusting the parameter of model training and calling the model generation module.
11. A method for predicting battery power, comprising the steps of:
and T, performing power prediction on the battery to be predicted by using the prediction model generated by the battery power prediction model generation method of any one of claims 1 to 5 to obtain a power prediction result of the battery to be predicted.
12. The battery power prediction method of claim 11, wherein step T comprises the steps of:
T1acquiring power characteristic data of the battery to be predicted;
T2preprocessing the power characteristic data to generate a corresponding power data matrix;
T3the work is carried outAnd inputting the rate data matrix into the prediction model to perform power prediction so as to obtain a power prediction result of the battery to be predicted.
13. A battery power prediction system comprising a prediction module and the battery power prediction model generation system of any one of claims 6 to 10;
the prediction module is used for performing power prediction on the battery to be predicted by using the prediction model so as to obtain a power prediction result of the battery to be predicted.
14. The battery power prediction system of claim 13, wherein the prediction module comprises a battery data to be predicted acquisition module, a battery pre-processing module to be predicted, and a prediction execution module;
the battery data acquisition module to be predicted is used for acquiring power characteristic data of the battery to be predicted;
the battery pre-processing module to be predicted is used for pre-processing the power characteristic data to generate a corresponding power data matrix;
and the prediction execution module is used for inputting the power data matrix into the prediction model to perform power prediction so as to obtain a power prediction result of the battery to be predicted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810068746.4A CN108287317B (en) | 2018-01-24 | 2018-01-24 | Battery power prediction model generation method and system and power prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810068746.4A CN108287317B (en) | 2018-01-24 | 2018-01-24 | Battery power prediction model generation method and system and power prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108287317A CN108287317A (en) | 2018-07-17 |
| CN108287317B true CN108287317B (en) | 2020-08-14 |
Family
ID=62835677
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810068746.4A Active CN108287317B (en) | 2018-01-24 | 2018-01-24 | Battery power prediction model generation method and system and power prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108287317B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109738809A (en) * | 2019-01-10 | 2019-05-10 | 上海动力储能电池***工程技术有限公司 | The estimation method of power and energy-storage battery power characteristic |
| CN110069810B (en) * | 2019-03-11 | 2023-04-07 | 北京百度网讯科技有限公司 | Battery failure prediction method, device, equipment and readable storage medium |
| CN110059377B (en) * | 2019-04-02 | 2022-07-05 | 西南交通大学 | Fuel cell life prediction method based on deep convolutional neural network |
| CN112240952A (en) * | 2019-10-24 | 2021-01-19 | 北京新能源汽车技术创新中心有限公司 | Power testing method, system, computer device and storage medium |
| CN111025041A (en) * | 2019-11-07 | 2020-04-17 | 深圳供电局有限公司 | Electric vehicle charging pile monitoring method and system, computer equipment and medium |
| CN113219341B (en) * | 2021-03-23 | 2023-04-07 | 陈九廷 | Model generation and battery degradation estimation device, method, medium, and apparatus |
| CN112798963B (en) * | 2021-04-14 | 2021-07-09 | 杭州宇谷科技有限公司 | Method, apparatus and medium for detecting battery charging characteristic abnormality based on time series |
| CN113805064B (en) * | 2021-09-18 | 2022-09-20 | 北京航空航天大学 | Lithium ion battery pack health state prediction method based on deep learning |
| CN115276177B (en) * | 2022-08-18 | 2023-09-26 | 上海采日能源科技有限公司 | Method and device for controlling charge and discharge power of energy storage battery and battery control system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103698710A (en) * | 2013-12-12 | 2014-04-02 | 中南大学 | Prediction method for life cycle of battery |
| CN103954913A (en) * | 2014-05-05 | 2014-07-30 | 哈尔滨工业大学深圳研究生院 | Predication method of electric vehicle power battery service life |
| CN104537268A (en) * | 2015-01-19 | 2015-04-22 | 重庆长安汽车股份有限公司 | Estimation method and device for maximum discharge power of battery |
| CN107329088A (en) * | 2016-04-29 | 2017-11-07 | 株式会社日立制作所 | The health status diagnostic device and method of battery |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10338153B2 (en) * | 2015-03-03 | 2019-07-02 | Samsung Electronics Co., Ltd. | Method and apparatus for automatically estimating remaining useful life (RUL) of battery in real time |
-
2018
- 2018-01-24 CN CN201810068746.4A patent/CN108287317B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103698710A (en) * | 2013-12-12 | 2014-04-02 | 中南大学 | Prediction method for life cycle of battery |
| CN103954913A (en) * | 2014-05-05 | 2014-07-30 | 哈尔滨工业大学深圳研究生院 | Predication method of electric vehicle power battery service life |
| CN104537268A (en) * | 2015-01-19 | 2015-04-22 | 重庆长安汽车股份有限公司 | Estimation method and device for maximum discharge power of battery |
| CN107329088A (en) * | 2016-04-29 | 2017-11-07 | 株式会社日立制作所 | The health status diagnostic device and method of battery |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108287317A (en) | 2018-07-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108287317B (en) | Battery power prediction model generation method and system and power prediction method and system | |
| Chinomona et al. | Long short-term memory approach to estimate battery remaining useful life using partial data | |
| Chen et al. | Remaining useful life prediction of battery using a novel indicator and framework with fractional grey model and unscented particle filter | |
| Ma et al. | Remaining useful life prediction of lithium battery based on capacity regeneration point detection | |
| CN114372417A (en) | Electric vehicle battery health state and remaining life evaluation method based on charging network | |
| CN115943516B (en) | Method and device for detecting unbalance of vehicle battery, electronic equipment and storage medium | |
| CN111025041A (en) | Electric vehicle charging pile monitoring method and system, computer equipment and medium | |
| CN116794543A (en) | Lithium battery performance prediction method based on GBLS boost multitask learning model | |
| CN114646888A (en) | Assessment method and system for capacity attenuation of power battery | |
| Zhou et al. | A light-weight feature extractor for lithium-ion battery health prognosis | |
| CN117630672A (en) | Method and device for predicting a device battery of a diagnostic device using a converter model | |
| CN116609686B (en) | Battery cell consistency assessment method based on cloud platform big data | |
| CN117169761A (en) | Battery state evaluation method, apparatus, device, storage medium, and program product | |
| US20230305073A1 (en) | Method and apparatus for providing a predicted aging state of a device battery based on a predicted usage pattern | |
| CN116901707A (en) | Power battery pack fault early warning method, system and vehicle | |
| CN116754960A (en) | Cloud SOH estimation and residual life prediction method, device and program for sodium ion battery | |
| CN115808627A (en) | Lithium battery SOH prediction method and device | |
| CN113255215B (en) | Lithium battery health state estimation method based on voltage segments | |
| CN114460484B (en) | Rechargeable battery life prediction method and device based on accumulated wear quantity | |
| CN114879054A (en) | Battery safety monitoring method and device, electronic equipment and storage medium | |
| Channegowda et al. | An attention‐based synthetic battery data augmentation technique to overcome limited dataset challenges | |
| Liu et al. | An end-to-end learning framework for early prediction of battery capacity trajectory | |
| CN112083340A (en) | Storage battery pack performance detection method and device, storage medium and terminal | |
| CN116908720B (en) | Battery pack consistency state diagnosis method, device and storage medium | |
| CN112485676B (en) | Battery energy storage system state estimation early warning method under digital mirror image |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |