CN112036083B - Similar product residual life prediction method and system - Google Patents
Similar product residual life prediction method and system Download PDFInfo
- Publication number
- CN112036083B CN112036083B CN202010883470.2A CN202010883470A CN112036083B CN 112036083 B CN112036083 B CN 112036083B CN 202010883470 A CN202010883470 A CN 202010883470A CN 112036083 B CN112036083 B CN 112036083B
- Authority
- CN
- China
- Prior art keywords
- data
- test
- training
- sample data
- life
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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]
- G01R31/392—Determining battery ageing or deterioration, e.g. state of health
-
- 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]
- G01R31/396—Acquisition or processing of data for testing or for monitoring individual cells or groups of cells within a battery
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/04—Ageing analysis or optimisation against ageing
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Theoretical Computer Science (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Secondary Cells (AREA)
Abstract
The invention discloses a method and a system for predicting the service life of a similar product, which relate to the field of similar product prediction, and are characterized in that target sample data and training test data are obtained by processing short-term test sample data of a battery to be predicted and full-life test capacity databases of other formula batteries; then intercepting data with the same length as target sample data from the training test data to perform migratable sample mining, and acquiring a neural network model for similar product test data migration and residual life prediction by training and learning the migration relationship from the migratable sample data to the target sample data; the residual life of the cross-formula battery is predicted in the neural network model by applying the data of the residual length in the training test data, the prediction precision of the method can reach more than 99 percent, the cost for analyzing the formula performance of the battery is saved, the data sharing is realized, and the method has good economy and practicability.
Description
Technical Field
The invention relates to the technical field of similar product prediction, in particular to a method, a system and a device for predicting the residual service life of similar products.
Background
With the development of energy storage technology and energy industry, lithium ion batteries are widely used as main energy storage devices of military electronic products, avionic devices, electric vehicles and various portable electronic devices (such as notebook computers, digital cameras, tablet computers, mobile phones and the like) due to the advantages of light weight, low discharge rate, long life and the like. Cycle life is an important design property for lithium battery products. In the design and development process of the lithium battery, in order to accurately obtain the service life conditions of batteries with different formulas in a design matrix and provide feedback for formula selection and design optimization, a cycle life test needs to be carried out on the batteries with various formulas in the design matrix. This test is continued until the battery capacity retention reaches a prescribed threshold, i.e., the end-of-life point of the lithium battery. However, the number of lithium battery formulas in the design matrix is large, so that the existing cycle life testing time and capital cost are too high, and particularly for power lithium batteries with the life cycle of more than years, the design and development efficiency is too low, and enterprises are difficult to bear.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a battery life prediction method more suitable for different formulas in the battery design and development process by utilizing short-term actual measurement data and a transfer learning prediction method.
In order to achieve the technical object of the present invention, in one aspect, the present invention provides a method for predicting a lifetime of a similar product, including:
carrying out short-term cycle life test on similar products with different formulas to be predicted to obtain short-term test sample data;
processing the obtained short-term test sample data and data in a full-life test capacity database of other formula batteries to obtain target sample data and training test data;
intercepting data with the same length as target sample data from the training test data to carry out migratable sample digging, and taking the data with higher migration degree as the migratable sample data;
learning a migration relation from migratable sample data to target sample data through forward network training to obtain a neural network model for similar product test data migration and residual life prediction;
and applying the data of the residual length in the training test data to predict the residual life of the cross-formula battery in the neural network model, and obtaining the residual life prediction results of similar products with different formulas to be predicted.
The short-term cycle life test refers to stopping at a preset test stopping threshold value in the early stage of the test when cycle life tests are carried out on similar products.
Wherein the test stop threshold is greater than a life test stop threshold.
Wherein, the stop threshold value refers to a certain percentage value of the battery capacity decaying to the initial capacity.
Wherein the test stop threshold is greater than 85%.
Preferably, the test stop threshold is not less than 87%.
Further preferably, the test stop threshold is not less than 90%.
In particular, a similar product is a lithium battery.
In particular, the processing of the obtained short-term test sample data and the data in the database of the full-life test capacity of the other formula batteries comprises:
rejecting unstable initial data and data which do not show a degeneration trend in short-term test sample data and a full-life test capacity database;
and performing smoothing treatment on the removed short-term test sample data and the data in the full-life test capacity database to obtain target sample data and prediction training test data.
In particular, the smoothing process employs a local weighted regression method.
In particular, the data extracted in the training test data with the same length as the target sample data is obtained by extracting the predicted training test data according to the number of charge and discharge cycles of the target sample data.
Particularly, the migratable sample mining is to evaluate the similarity between the training test data and the target sample data in a multidimensional manner, perform migratable degree calculation on the evaluated similarity result, and take the training test data with higher migratable degree as the migratable sample data.
In particular, the training test data with higher migratability is the training test data with the top 20% of the ascending ranking of migratability.
In particular, the similarity between the multidimensional evaluation training test data and the target sample data comprises:
evaluating the dimensionality of the Chebyshev distance between each piece of training test data and the target sample data;
evaluating the dimension of the mean value of the difference between each piece of training test data and the target sample data;
and evaluating the dimension of the change rate of the tail point of the difference between each training test data and the target sample data.
To achieve the technical object of the present invention, in another aspect, the present invention provides a system for predicting life of similar products, including:
data unit, data processing unit, data mining unit, model training unit and life prediction unit, wherein:
the data unit is used for acquiring and storing predicted cycle life data of similar products with different formulas;
the data processing unit is used for processing the short-term test sample data to obtain target sample data and processing data in the full-life test capacity database of other formula batteries to obtain training test data;
the data mining unit is used for intercepting data with the same length as target sample data from the training test data to perform migratable sample mining, and taking the data with higher migration degree in the mined migratable samples as the migratable sample data;
the model training unit is used for learning the migration relationship from the transferable sample data to the target sample data through forward network training to obtain a neural network model for similar product test data migration and residual life prediction;
and the service life prediction unit is used for predicting the residual service life of the cross-formula battery in the neural network model by using the data of the residual length in the training test data to obtain the residual service life prediction results of similar products of different formulas to be predicted.
To achieve the technical object of the present invention, the present invention also provides a use of the above method or system for battery formulation optimization.
Has the advantages that:
1. the method can predict the residual life of the battery to be tested only by utilizing the existing full-life data of the known battery and the short-term cycle life test data of the battery with different formulas to be tested, has simple method and low cost, and only needs to carry out short-term cycle life test on the battery.
2. The method and the system provided by the invention have high prediction precision which can reach more than 99 percent at most, save the cost for analyzing the performance of the battery formula and accelerate the progress of the battery research and development stage.
3. The method and the system save a large amount of test cost, realize effective data sharing, bring a new choice for the battery cycle life test, and have good economy and practicability.
Drawings
FIG. 1 is a flow chart of the method of the present invention provided in example 1 of the present invention;
FIG. 2 is a flow chart of the system of the present invention provided in embodiment 2 of the present invention;
FIG. 3 is a graph of degradation before and after 25DC smoothing provided in an example of an application of the present invention;
FIG. 4 is a graph showing degradation before and after a 45DC smoothing process provided in an applied embodiment of the present invention;
FIG. 5 is a graph of degradation before and after 60DC smoothing provided in an example of an application of the present invention;
FIG. 6 is a graph showing a mobility change curve provided in an example of application of the present invention;
FIG. 7 is a graph comparing the degradation trend of a 25 DC-like sample and a target sample provided in an example of the application of the present invention;
FIG. 8 is a graph comparing the degradation trend of a 45DC similar sample and a target sample provided in an applied example of the present invention;
FIG. 9 is a graph comparing the degradation trend of a 60 DC-like sample and a target sample provided in an example of the application of the present invention;
FIG. 10 is a schematic diagram of a BPNN structure used in an embodiment of the present invention.
Detailed Description
The method and system of the present invention will now be described in greater detail in connection with the accompanying schematic drawings, in which preferred embodiments of the invention are shown, it being understood that those skilled in the art may modify the invention herein described while still achieving the advantageous results of the invention. Accordingly, the following description should be construed as broadly as possible to those skilled in the art and not as limiting the invention.
The invention is described in more detail in the following paragraphs by way of example with reference to the accompanying drawings. Advantages and features of the present invention will become apparent from the following description and from the claims. It is to be noted that the drawings are in a very simplified form and are not to precise scale, which is merely for the purpose of facilitating and distinctly claiming the embodiments of the present invention.
The experimental procedures used in the following examples are all conventional procedures unless otherwise specified. The structures, materials and the like used in the following examples are commercially available unless otherwise specified.
Example 1
As shown in fig. 1, the method for predicting the remaining life of a cross-formulation similar product (such as a lithium ion battery) provided by the invention comprises the following steps:
step S101, performing short-term cycle life test on similar products to be predicted with different formulas to obtain short-term test sample data.
The short-term cycle life test means that when cycle life test is carried out on similar products, the test is stopped at a preset test stop threshold value in the early stage of the test.
Wherein the test stop threshold is greater than a life test stop threshold.
Wherein, the test stop threshold value and the life test stop threshold value both refer to a certain percentage value of the battery capacity decaying to the initial capacity.
Wherein the test stop threshold is greater than 85%.
Preferably, the test stop threshold is not less than 87%.
Further preferably, the test stop threshold is not less than 90%.
Specifically, the battery refers to a battery soft package in a battery product design stage, and can also be a finished battery with different formulas.
Specifically, the different formulations are such that the cell under test has the same cathode and separator materials as the cell that has completed the life test, but the cell anode material and electrolyte solution are different.
It should be noted that the battery cycle life test referred to in the present invention can be performed by using a test standard commonly used for similar products, such as the GB/T18287 standard, and can also be performed by using a test method commonly used in the art, and in order to reduce errors, the data referred to in the present invention are all from data measured under the same test condition.
Step S102, the obtained short-term test sample data and data in the full-life test capacity database of other formula batteries are processed to obtain target sample data and training test data.
Further, the database of the total life test capacity of the other formula batteries is the sum of the historical total life test data of the existing different formula batteries, namely, each battery test in the existing battery library is operated until the battery capacity retention rate is degraded to the national standard to the life threshold TS Failture Sum of capacity retention ratio data of (1).
Specifically, the processing of the obtained short-term test sample data and the data in the database of the full-life test capacity of the other formula batteries includes:
removing unstable initial data and data which do not show a degeneration trend in short-term test sample data and life test capacity data;
and smoothing the removed target sample data and the life-cycle test capacity data to obtain target sample data with random noise removed and prediction training test data.
Furthermore, the smoothing process is to perform local weighted regression on the rejected target sample data and the full-life test capacity data, remove the influence of random noise and keep the degradation trend of the capacity retention rate.
Due to factors such as fluctuation of test conditions, measurement errors, cross tests and the like, test data of similar products have certain fluctuation, but the similar products do not influence the degradation trend of the battery under a large scale. If the noise is preserved, the prediction model is at risk of overfitting. In order to improve the prediction robustness, the invention utilizes the local weighted regression algorithm to carry out smoothing processing on the data, thereby reserving the capacity retention rate degradation trend and removing the influence of random noise.
The local weighted regression algorithm used by the invention is a regression algorithm for improving the realization effect of the common regression algorithm. The principle is that polynomial weighted fitting is carried out on local observation data, and a least square method is used for estimation, so that points needing fitting are obtained finally. The method comprises the following specific steps:
for each data q i Determining a window range (i.e., data length) within which all q's are within a window k K =1,2, \ 8230, n, the weight ω can be obtained by the weight function k (q i ) Using a weight ω k (q i ) Weighted least squares multiplication pair q i D-order polynomial fitting is carried out to obtain a fitting value p i . Using omega k (q i ) To obtain p i It is called local weighted regression.
Wherein alpha is 0 (q i )，α 1 (q i )，…，α m (q i ) Is relative to q i Unknown parameter, ∈ i I =1,2, \ 8230, n is an independent and equally distributed random error term. The β value is set according to the purpose and requirement of the experiment, and the larger the value, the higher the smoothness of the curve, and the smaller the value, the lower the smoothness of the curve.
Step S103, intercepting data with the same length as target sample data from the training test data to perform migratable sample mining, and taking the data with higher migration degree as the migratable sample data.
Specifically, the data with the same length as the target sample data is extracted from the training test data by predicting the number of charge/discharge cycles of the target sample data.
Specifically, the migratable sample mining is to evaluate the similarity between the training test data and the target sample data in a multidimensional manner, perform migratable degree calculation on the evaluated similarity result, and use the training test data with higher migratable degree as the migratable sample data.
Further, the multi-dimensionally evaluating the similarity between the training test data and the target sample data comprises:
evaluating the dimensionality of the Chebyshev distance between each piece of training test data and the target sample data;
evaluating the dimension of the mean value of the difference between each training test data and the target sample data;
and evaluating the dimension of the change rate of the difference end point between each training test data and the target sample data.
Specifically, the dimensionality evaluation of the Chebyshev distance between each piece of training test data and the target sample data is to calculate the Chebyshev distance between each piece of intercepted data and the target sample data, rank the obtained Chebyshev distance values in an ascending order, mark a sequence number and use the sequence number as a first criterion;
that is to say that the temperature of the molten steel,
wherein dis chebychev A value representing the chebyshev distance,
it is shown that each of the target sample data,
each training test data is represented.
Specifically, the dimensional evaluation of the difference mean value between each piece of training test data and the target sample data is to calculate the difference mean value between each piece of intercepted data and the target sample data, rank the obtained difference mean values in an ascending order, and mark a sequence number as a second criterion;
namely dis capacity ＝mean(|D s -D t |)；
Wherein the content of the first and second substances,
wherein dis capacity A value representing the chebyshev distance,
each of the target sample data is represented by,
represents each training test data, D s Representing sets of training test data, D t Represents a set of target sample data, and T represents transposition.
Specifically, the dimension evaluation of the change rate of the difference end point between each piece of training test data and the target sample data is to calculate the change rate of the difference end point between each piece of intercepted data and the target sample data, rank the obtained change rate values of the difference end points in an ascending order, and mark a sequence number as a third criterion.
Namely dis gradient ＝diff(D s -D t ) r ；
Wherein the content of the first and second substances,
wherein dis gradient A value representing the chebyshev distance,
each of the target sample data is represented by,
represents each training test data, D s Representing sets of training test data, D t Represents a set of target sample data, and T represents transposition.
Further, the mobility calculation of the evaluated similarity result is to perform weight calculation on the dimensional evaluation result of the chebyshev distance, the dimensional evaluation result of the difference mean value and the dimensional evaluation result of the change rate of the difference end point to obtain the mobility value of each training test data and the target sample data.
Specifically, the step of performing weight calculation on the dimensional evaluation result of the chebyshev distance, the dimensional evaluation result of the mean difference value, and the dimensional evaluation result of the change rate of the end point of the difference value is to calculate the obtained first criterion, second criterion, and third criterion according to a set weight value, and the result is the mobility value of each training test data and each target sample data.
Namely rank s ＝rank·W f T
Wherein, rank s Representing the mobility value, rank is the position number obtained by each criterion, rank = (rank) in the example of the invention chebychev ,rank capacity ,rank gradient )，W f The weight of the sequence obtained by each criterion in the comprehensive judgment of the similarity, and T represents transposition.
Further, the training test data with higher migratability is the training test data with the top 20% of the ascending ranking of the migratability value.
Step S104, learning a migration relation from transferable sample data to target sample data through forward network training, and obtaining a neural network model of similar product test data migration and residual life prediction.
Further, the forward network training takes similar sample data as input and takes target sample data as output to perform forward network training.
The Neural Network type adopted by the invention is a BPNN (Back Propagation Neural Network) type Neural Network. BPNN is a neural network of forward structure that maps a set of input vectors to a set of output vectors. BPNN except for the input layer, each layer of neurons carries an activation function. Meanwhile, each layer of the BPNN is a fully connected layer and is trained by adopting a back propagation algorithm.
The batteries with different formulas have certain similarity, the battery with each formula can be regarded as a task domain, the prediction of the battery capacity of a certain task domain (namely target sample data) is to be realized, and when the target domain sample is insufficient to support the prediction task, the prediction task of the battery capacity of the target domain can be realized by using the knowledge of the similar sample (source domain) through the migration based on the sample.
In the invention, the neural network is adopted to realize corresponding migration, and the neural network is trained by using sample data with the length r of the target domain and the source domain, so that the migration weight of the neural network is determined.
In the prediction task, the target battery is at cycle number c j Capacity retention ratio of
And the capacity retention rate of the similar samples corresponding to the cycle number is obtained by mapping and changing through a neural network. The type of the Neural Network adopted by the invention is a BPNN (Back Propagation Neural Network, BPNN) type Neural Network. BPNN is a neural network of forward structure that maps a set of input vectors to a set of output vectors. BPNN except for the input layer, each layer of neurons carries an activation function. Meanwhile, each layer of the BPNN is a fully connected layer and is trained by adopting a reverse propagation algorithm.
The training sample for BPNN training is [ d ] ck,1 ,d ck,2 ,…,d ck,i ,d ck,t ]K =1,2 \8230, and the battery capacity retention rate corresponding to each cycle number of the r, i samples and the battery capacity retention rate corresponding to each cycle number of the target battery form a training sample.
Supposing that the neural network has a T-layer structure, wherein the 1 st layer is an input layer, the T-th layer is an output layer, the 2 nd to the T-1 th layers are hidden layers, and the connection weight matrix between the T-th and the T +1 th layers is
The offset between the t-th and t + 1-th layers is
Wherein s is t ,s t+1 Respectively representing the number of the t-th layer node and the t + 1-th layer node.
The BP algorithm updates each weight in the network by using a gradient descent algorithm to obtain a minimum value of the target function. And (3) using a batch updating algorithm, setting the size of the batch as p, and adopting a mean square error and a calculation formula, wherein the total error of the batch at one time is as follows:
in the formula
For computational convenience. We use the mean square error as the objective function, then the objective function is:
according to the gradient descent formula, the updating equation of the connection weight and the bias between each time of batch to each layer is as follows:
where α ∈ (0, + ∞) is the learning rate, i denotes the ith node of the t-th layer, and j denotes the jth node of the t + 1-th layer. The weight gradient of each layer is equal to the input of the previous layer to which the weight of this layer is connected multiplied by the inverted output of the connected next layer.
After the BPNN training is finished, the mature network can be trained to predict the service life of the battery.
Step S105, residual life prediction of the cross-formula battery is carried out on the neural network model by using the data of the residual length in the training test data, and a residual life prediction result of similar products of different formulas to be predicted is obtained.
Specifically, the data with the remaining length in the training test data is the data remaining after being intercepted in the prediction training test data according to the charge-discharge cycle number of the target sample data.
Example 2 similar product prediction System
As shown in fig. 2, the system for predicting the remaining life of a cross-formulation similar product provided by the present invention comprises: data unit 1, data processing unit 2, data mining unit 3, model training unit 4, and life prediction unit 5.
Further, the data unit 1 is used for collecting and storing short-term cycle life data of similar products of different formulas to be predicted; the data processing unit 2 is used for processing the short-term test sample data to obtain target sample data, and processing data in the life test capacity database of other similar products with different formulas from the similar products to be predicted to obtain training test data; the data mining unit 3 is used for intercepting data with the same length as target sample data from the training test data to perform migratable sample mining so as to mine the migratable sample data serving as a similar product to be predicted from the training test data; the model training unit 4 is used for learning the migration relationship from the transferable sample data to the target sample data through forward network training to obtain a neural network model for similar product test data migration and residual life prediction; and the service life prediction unit 5 is used for applying the data of the residual length in the training test data to predict the residual service life of the cross-formula battery in the neural network model, and obtaining the residual service life prediction results of similar products of different formulas to be predicted.
Application example remaining life prediction across recipe similarity products
1. Data collection
In order to verify the life migration prediction capability of the battery with different formulas, the battery life test method screens ten groups of battery sample data with different formulas from similar products with the same battery platform and different formulas, which are subjected to the battery life test, and performs the residual life prediction of the battery with different formulasMeasuring, wherein each group of sample data comprises test data at three different temperatures of 25 ℃,45 ℃ and 60 ℃, and one group of samples is subjected to length r interception, wherein the length is a stop threshold TS (transport stream) when the battery capacity retention rate is degraded to 90% after the test is operated Failture The capacity retention rate data of the battery pack is used as a sample to be predicted, and other nine groups of samples are used as a battery library, namely a life test capacity database. And predicting a group of batteries with the intercepted length. The battery to be tested in the sample to be predicted at the temperature of 25 ℃ is named as: 25DC; the cell to be tested at a temperature of 45 ℃ is named: 45DC; the battery to be tested at a temperature of 60 ℃ is named: at the time of the start of the 60DC,
the above data are all measured under the same experimental condition, for example, a standard constant current/constant voltage mode is adopted, the constant current rate is 1C before the voltage reaches 4.2V, then the voltage value is kept to be 4.2V until the charging current is reduced to be below 0.1A, the cyclic discharge process is carried out by the constant discharge current 2A, the discharge cut-off voltage is 2.78V, and the life cycle test data is that the test is operated until the battery capacity retention rate is degraded to 80% and reaches the life threshold TS Failture When the capacity retention ratio data of all formulations of (1), i.e., the battery capacity retention ratio, falls to 80% of the initial capacity, the battery is considered to reach the end of life. The method can also be carried out according to the test conditions of the national similar product test standard, such as the GB/T18287 standard.
2. Data processing
2.1 data culling
Removing the sample to be tested and the initial small section of unstable part which does not show the degeneration trend in the sample library to obtain target sample data
Corresponding charge-discharge cycle C t ＝ (c 1 ,c 2 ,…,c r ) And training test data
Where i represents the ith sample.
2.2 smoothing
According to the method, the data is smoothed by using a local weighted regression algorithm, so that the overall trend of a degradation curve can be kept, and the results of smoothing the target sample data 25DC, 45DC and 60DC are shown in FIGS. 2-4.
Fig. 3 is a degradation graph before and after smoothing of a sample 25DC, fig. 4 is a degradation graph before and after smoothing of a sample 45DC, and fig. 5 is a degradation graph before and after smoothing of a sample 60DC, in which the abscissa represents the number of cycles and the ordinate represents capacity data, and it can be seen from fig. 3 to 5 that the overall trend of the battery degradation curve after smoothing is more significant.
And smoothing the training test data by the same method to obtain the training test data with more obvious overall trend.
3. Migratable sample mining
Data with the same length as that of target sample data 25DC, 45DC and 60DC is intercepted in the training test data, transferable sample mining is carried out on each capacity data of the 25DC, 45DC and 60DC in the training test data, and the corresponding length is 1 st to 1503 th charging and discharging cycles by taking the battery data of the 25DC as an example. And intercepting the capacity degradation data between 1503 cycles from 25 ℃ temperature data in the training test data to 29 degradation data of different formula batteries (total 29) meeting the length requirement, smoothly denoising to obtain 29 1503 vectors with the same dimension as the target battery, respectively calculating each criterion between the 29 vectors and target sample data, converting into similarity and sorting, and obtaining the similarity ranking as shown in table 1.
The migration results shown in table 1 and the migratability variation curve shown in fig. 6 were obtained.
Table 1 migration results
As can be seen from table 1 and fig. 6, the similarity of the similar curves is distributed in a staircase shape, and in order to ensure that the selected similar curves make a positive contribution to prediction as much as possible, the similar curves with the stable similarity change, which are ranked 20% of the first, are selected for prediction. Meanwhile, in order to avoid abnormal similar curves caused by accidental errors, the cycle number of the capacity retention rate of 0.9 is adopted for further elimination, and the cycle number of the capacity retention rate of 0.9 of the selected similar curves is ensured not to deviate from the 1 st 15% or more, namely, the batteries 1,2, 4 and 5 are used as similar samples. The life-cycle degradation curves for the similar and target samples are shown as 8. It can be seen that the degradation trends of the four similar samples in the reference segment are substantially consistent with that of the target sample, and the four similar samples can be used for predicting missing data of the target sample. The results of the screening at 45 ℃ and 60 ℃ (i.e., 45DC, 60 DC) are shown in FIGS. 7-9.
4. Neural network learning
Taking a 25 ℃ battery (namely 25 DC) as an example, 1503 capacity retention rate data of 1 st to 1503 th cycles of four similar samples of the batteries 1,2, 4 and 5 are intercepted as training input data 25DC data and taken as training output data, and the BPNN neural network is trained.
The BPNN configuration selected for use in one embodiment of the present invention is illustrated in fig. 10.
The network training loss function selects Mean Square Error (MSE), the adam optimizer has higher convergence speed, and can enable the objective function to converge to be close to the optimal point in the shortest time, so that the adam optimizer is firstly adopted to optimize 30 epochs to enable the objective function to quickly reach to be close to the optimal point. Although the convergence speed of the sgd optimizer is low, the sgd optimizer is extremely stable, so that the sgd optimizer is suitable for being used after the objective function reaches the vicinity of an optimal point, and after the adam optimizer is used, the sgd optimizer is selected to retrain 50 epochs, so that the loss value of the current BPNN network is converged to an ideal state.
The trained BPNN neural network has the capability of transferring similar sample knowledge to the target sample, and the trained network transfer weight can be used for predicting the service life of the target sample.
5. Remaining life prediction and result analysis
And (5) migrating similar sample data according to the neural network obtained in the step (4), predicting data with the capacity retention rate degraded to be below 0.9, and obtaining degradation curves shown in the figures 7-9 by the batteries at 25 ℃ (namely 25 DC), 45 ℃ (namely 45 DC) and 60 ℃ (namely 60 DC).
In FIGS. 7-9, the gray curves on the right side of the black vertical dashed line are all predicted results.
It can be seen from fig. 7 that the predicted results at 25 c (i.e. 25 DC) substantially match the true curve. And searching the intersection points of the prediction curve, the real curve and the failure threshold value to obtain the predicted service life of 2370 cycles, the real service life of 2365 cycles and the prediction accuracy of 99.79%.
As can be seen from fig. 8, the predicted results for 3 of the samples at 45 ℃ (i.e., 45 DC) substantially fit the true curve. And searching the intersection points of the prediction curve, the real curve and the failure threshold value to obtain the predicted service life of 718 cycles, the real service life of 757 cycles and the prediction accuracy of 94.85%.
As can be seen from fig. 9, the predicted results for 5 of the samples at 60 ℃ (i.e., 60 DC) substantially fit the true curve. And searching the intersection points of the prediction curve, the real curve and the failure threshold value to obtain the predicted service life of 791 cycles, the real service life of 778 cycles and the prediction accuracy of 98.33%.
In conclusion, the prediction algorithm provided by the invention has good performance, meanwhile, from the data driving perspective, the cost for analyzing the battery formula performance is saved, and by analyzing the data similarity of batteries with different formulas, a large amount of test cost can be saved, and effective data sharing is realized. The method brings a new choice for the battery cycle life test, and has good economical efficiency and practicability.
Although the present invention has been described in detail hereinabove, the present invention is not limited thereto, and various modifications can be made by those skilled in the art in light of the principle of the present invention. Thus, modifications made in accordance with the principles of the present invention should be understood to fall within the scope of the present invention.
Claims (8)
1. A method for predicting life of similar products, comprising:
performing short-term cycle life test on similar products to be predicted to obtain short-term test sample data;
processing the short-term test sample data to obtain target sample data, and processing data in the database of the full-life test capacity of other similar products with different formulas from the similar product to be predicted to obtain a plurality of training test data;
intercepting data with the same length as target sample data from each training test data to carry out migratable sample mining so as to mine a plurality of migratable sample data serving as similar products to be predicted from a plurality of training test data;
training a neural network of a forward structure by taking the plurality of migratable sample data as training input data and the target sample data as training output data to obtain a neural network model of similar product test data migration and residual life prediction;
and predicting the residual life of the cross-formulation battery in a neural network model by using the data of the residual length in the training test data to obtain the residual life prediction result of the similar products of different formulations to be predicted, wherein the data of the residual length in the training test data is the data which is remained after being intercepted in the predicted training test data according to the charge-discharge cycle number of target sample data.
2. The method of claim 1, wherein the short cycle life test is a test that stops at a predetermined test stop threshold early in the test when cycle life tests are performed on similar products;
wherein the test stop threshold is less than a life test stop threshold.
3. The method of claim 1, wherein said processing the obtained short term test sample data and the data in the database of the full life test capacity of the other battery formulas comprises:
rejecting unstable initial data and data which does not show a degeneration trend in the short-term test sample data and the full-life test capacity database;
and smoothing the removed short-term test sample data and the data in the full-life test capacity database to obtain target sample data and prediction training test data.
4. The method of claim 3, wherein the smoothing process uses a locally weighted regression method.
5. The method of claim 1, wherein the migratable sample mining is multi-dimensionally evaluating similarity between training test data and target sample data, performing migratable degree calculation on the evaluated similarity result, and using training test data with higher migratable degree as migratable sample data.
6. The method of claim 5, wherein the multidimensional evaluating the similarity of training test data to target sample data comprises:
evaluating the dimensionality of the Chebyshev distance between each piece of training test data and the target sample data;
evaluating the dimension of the mean value of the difference between each piece of training test data and the target sample data;
and evaluating the dimension of the change rate of the tail point of the difference between each training test data and the target sample data.
7. The method of any of claims 1-6, wherein the similar product is a lithium battery.
8. A similar product life prediction system, comprising: data unit, data processing unit, data mining unit, model training unit and life prediction unit, wherein:
the data unit is used for acquiring and storing short-term cycle life data of similar products with different formulas to be predicted;
the data processing unit is used for processing the short-term test sample data to obtain target sample data, and processing data in the life test capacity database of other similar products with different formulas from the similar products to be predicted to obtain a plurality of training test data;
the data mining unit is used for intercepting data with the same length as the target sample data from each training test data to perform migratable sample mining so as to mine a plurality of migratable sample data serving as similar products to be predicted from the plurality of training test data;
the model training unit is used for training a neural network by taking the plurality of migratable sample data as training input data and the target sample data as training output data to obtain a neural network model for similar product test data migration and residual life prediction;
and the service life prediction unit is used for predicting the residual service life of the cross-formulation battery in the neural network model by using the data of the residual length in the training test data to obtain the residual service life prediction result of similar products of different formulations to be predicted, wherein the data of the residual length in the training test data is the data which is remained after the data is intercepted in the prediction training test data according to the charge-discharge cycle number of the target sample data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010883470.2A CN112036083B (en) | 2020-08-28 | 2020-08-28 | Similar product residual life prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010883470.2A CN112036083B (en) | 2020-08-28 | 2020-08-28 | Similar product residual life prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112036083A CN112036083A (en) | 2020-12-04 |
| CN112036083B true CN112036083B (en) | 2023-01-13 |
Family
ID=73586748
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010883470.2A Active CN112036083B (en) | 2020-08-28 | 2020-08-28 | Similar product residual life prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112036083B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112734002B (en) * | 2020-12-28 | 2023-01-13 | 北京航空航天大学 | Service life prediction method based on data layer and model layer joint transfer learning |
| CN112834945A (en) * | 2020-12-31 | 2021-05-25 | 东软睿驰汽车技术（沈阳）有限公司 | Evaluation model establishing method, battery health state evaluation method and related product |
| CN113297787B (en) * | 2021-05-12 | 2021-12-17 | 北京航空航天大学 | Method for predicting remaining life of aircraft engine based on transfer learning |
| CN115470717A (en) * | 2022-10-31 | 2022-12-13 | 四川工程职业技术学院 | Method, device, equipment and storage medium for predicting remaining life of robot |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103399276B (en) * | 2013-07-25 | 2016-01-20 | 哈尔滨工业大学 | A kind of capacity of lithium ion battery is estimated and cycles left life-span prediction method |
| CN107797067B (en) * | 2016-09-05 | 2019-09-27 | 北京航空航天大学 | Lithium ion battery life migration prediction method based on deep learning |
| CN110188920A (en) * | 2019-04-26 | 2019-08-30 | 华中科技大学 | A kind of lithium battery method for predicting residual useful life |
-
2020
- 2020-08-28 CN CN202010883470.2A patent/CN112036083B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112036083A (en) | 2020-12-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112036083B (en) | Similar product residual life prediction method and system | |
| CN109061504B (en) | Method and system for predicting remaining service life of lithium ion batteries of same type and different types | |
| Tong et al. | Early prediction of remaining useful life for Lithium-ion batteries based on a hybrid machine learning method | |
| Guo et al. | Online estimation of SOH for lithium-ion battery based on SSA-Elman neural network | |
| CN110224192B (en) | Method for predicting service life of power battery by gradient utilization | |
| Xu et al. | A novel data-driven method for predicting the circulating capacity of lithium-ion battery under random variable current | |
| CN112036084B (en) | Similar product life migration screening method and system | |
| Xu et al. | An improved CNN-LSTM model-based state-of-health estimation approach for lithium-ion batteries | |
| Li et al. | Data-driven battery state of health estimation based on interval capacity for real-world electric vehicles | |
| CN111680848A (en) | Battery life prediction method based on prediction model fusion and storage medium | |
| WO2022213789A1 (en) | Lithium battery soc estimation method and apparatus, and computer-readable storage medium | |
| CN108009585B (en) | Lead-acid battery health state prediction method based on local information fusion | |
| CN112051506B (en) | Similar product transferable sample screening method, system and application | |
| Zhang et al. | State of health estimation method for lithium-ion batteries using incremental capacity and long short-term memory network | |
| Chang et al. | Micro-fault diagnosis of electric vehicle batteries based on the evolution of battery consistency relative position | |
| CN115201686A (en) | Lithium ion battery health state assessment method under incomplete charging and discharging data | |
| CN115994441A (en) | Big data cloud platform online battery life prediction method based on mechanism information | |
| CN114200333A (en) | Lead-acid storage battery capacity prediction method | |
| CN116008844A (en) | Vehicle lithium battery health state online prediction method adapting to quick charge strategy | |
| Long et al. | Multi-parameter optimization method for remaining useful life prediction of lithium-ion batteries | |
| CN110232432B (en) | Lithium battery pack SOC prediction method based on artificial life model | |
| CN113361692A (en) | Lithium battery residual life combined prediction method | |
| CN113376541A (en) | Lithium ion battery health state prediction method based on CRJ network | |
| Satyan et al. | A Survey on data-driven methods for state of charge estimation of battery | |
| EP4270033A1 (en) | Method and apparatus for estimating state of health of battery |
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 |