CN115078999B - Lithium battery condition prediction method based on self-adaptive hidden layer BP neural network - Google Patents
Lithium battery condition prediction method based on self-adaptive hidden layer BP neural network Download PDFInfo
- Publication number
- CN115078999B CN115078999B CN202210554879.9A CN202210554879A CN115078999B CN 115078999 B CN115078999 B CN 115078999B CN 202210554879 A CN202210554879 A CN 202210554879A CN 115078999 B CN115078999 B CN 115078999B
- Authority
- CN
- China
- Prior art keywords
- lithium battery
- training
- hidden layer
- neural network
- output
- 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
- WHXSMMKQMYFTQS-UHFFFAOYSA-N Lithium Chemical compound [Li] WHXSMMKQMYFTQS-UHFFFAOYSA-N 0.000 title claims abstract description 117
- 229910052744 lithium Inorganic materials 0.000 title claims abstract description 117
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 72
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000012549 training Methods 0.000 claims abstract description 106
- 230000036541 health Effects 0.000 claims description 35
- 238000012360 testing method Methods 0.000 claims description 32
- 238000005070 sampling Methods 0.000 claims description 30
- 208000037004 Myoclonic-astatic epilepsy Diseases 0.000 claims description 21
- 238000000874 microwave-assisted extraction Methods 0.000 claims description 21
- 230000003044 adaptive effect Effects 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000007599 discharging Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 230000008859 change Effects 0.000 abstract description 4
- 238000003062 neural network model Methods 0.000 abstract 1
- 238000004088 simulation Methods 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 230000003862 health status Effects 0.000 description 3
- 241000544061 Cuculus canorus Species 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 101100493820 Caenorhabditis elegans best-1 gene Proteins 0.000 description 1
- 101100004280 Caenorhabditis elegans best-2 gene Proteins 0.000 description 1
- 208000032953 Device battery issue Diseases 0.000 description 1
- HBBGRARXTFLTSG-UHFFFAOYSA-N Lithium ion Chemical compound [Li+] HBBGRARXTFLTSG-UHFFFAOYSA-N 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000005431 greenhouse gas Substances 0.000 description 1
- 229910001416 lithium ion Inorganic materials 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
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]
- G01R31/3644—Constructional arrangements
- G01R31/3648—Constructional arrangements comprising digital calculation means, e.g. for performing an algorithm
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Secondary Cells (AREA)
Abstract
The invention provides a lithium battery condition prediction method based on a self-adaptive hidden layer BP neural network, and belongs to the technical field of lithium batteries. According to the method, the charge time is introduced into the training of the BP neural network, so that the accuracy of SOH estimation of the lithium battery is improved; the hidden layer self-adaptive setting is utilized to establish an accurate neural network model, so that the network can rapidly and accurately realize the SOH prediction of the lithium battery under the condition of data set change, and the method has the advantages of wide application range, high prediction precision, strong tracking property and the like, meanwhile, the dependence of the traditional BP neural network on the initial structure setting of the network is reduced, and the accuracy of the model is effectively improved.
Description
Technical Field
The invention relates to the technical field of lithium batteries, in particular to a lithium battery condition prediction method based on a self-adaptive hidden layer BP neural network, wherein the lithium battery condition is the health condition of the lithium battery.
Background
In recent years, the rapid development of lithium ion battery technology drives the rapid development of electrification of vehicles, and the popularity of electric automobiles is rapidly increasing. Currently, in order to reduce the emission of greenhouse gases and save non-renewable energy sources, many countries are out of the counter with related policies to drive the development of electric vehicles.
The Battery Management System (BMS) is an important system for securing safety and reliability of lithium batteries, of which battery state of charge (SOC) estimation and state of health (SOH) estimation are the most important. In order to ensure healthy operation of the battery, and to prevent sudden battery failure from causing a series of possible disaster that could be avoided, accurate SOH estimation is critical. The SOH change of the battery may be reflected on changes in parameters inside the battery, such as a decrease in capacity and an increase in ohmic internal resistance.
The current SOH estimation method for lithium batteries can be roughly divided into the following 2 types:
1. The model-based method comprises two methods, namely an electrochemical model-based method and an equivalent circuit model-based method. Although the electrochemical model method has high prediction accuracy, the model structure is very complex and difficult to build because of considering the complex aging mechanism of the lithium battery, and the method has large calculation amount and usually needs to be combined with other artificial intelligence algorithms. The method based on the equivalent model is simple in structure and easy to build, but has the cost of large error, low precision, poor robustness and the like.
2. Data driving methods, including machine learning, artificial intelligence algorithms, etc., consider lithium batteries as a black box model, and mine rules of battery performance evolution from a large amount of data. However, the method requires a large amount of data training, the accuracy of the data and the structure of the algorithm determine the accuracy of the health condition estimation, and finding a proper algorithm and correctly processing the original data are key problems of estimating the SOH of the lithium battery by a data driving method.
Chinese patent application publication (CN 112881914A) discloses a method for predicting the health status of a lithium battery. The method mainly comprises the following steps: the structural characteristics of the lithium battery of the electric automobile are represented by a BP neural network improved by a cuckoo algorithm, a battery health condition prediction model is established aiming at the data driving method, the capacity parameter of the battery pack is tracked and predicted, a capacity estimated value of the lithium battery at a certain moment is obtained, and the capacity estimated value is compared with an actual value to predict the health condition of the lithium battery of the electric automobile. Although the method for searching the optimal initial value of the network by the cuckoo can effectively avoid the problem that the model is in local optimum, the hidden layer is set to be a fixed value, and when the number of the input training sets is reduced, the accuracy of the method is affected to a certain extent, and the method has the characteristic of poor robustness.
Disclosure of Invention
Aiming at the problems that the accuracy cannot reach the best caused by the original structural setting under the condition that the number of training sets is reduced or the types of batteries are changed in the neural network structure in the prior art, the invention provides the BP neural network with the self-adaptive hidden layer setting for predicting the health condition of the lithium battery, and the method has the advantages of wide application range, high prediction accuracy, strong tracking property and the like, simultaneously reduces the dependence of the traditional BP neural network on the initial structural setting of the network, and effectively improves the accuracy of the model.
In order to solve the technical problems of the invention, the adopted technical scheme is as follows: a lithium battery condition prediction method based on a self-adaptive hidden layer BP neural network, wherein the lithium battery condition is the lithium battery health condition, comprises the following steps:
Step 1, a lithium battery is charged and discharged for n times, the lithium battery in the n times of charging and discharging cycles is sampled, and sampling data form a lithium battery charging and discharging data set;
The following lithium battery sampling parameters related to the lithium battery health condition are extracted from the charge-discharge data set and recorded as sampling health parameters: sampling end voltage in n charge-discharge cycles, sampling end current in n charge-discharge cycles, sampling lithium battery temperature in n charge-discharge cycles, sampling lithium battery charging time in n charge-discharge cycles and sampling lithium battery capacity in n charge-discharge cycles, so as to obtain five groups of sampling health parameters, and marking any one of the five groups as sampling health parameter groups j, j=1, 2,3,4 and 5; any one of the n charge-discharge cycles is denoted as cycle i, i=1, 2,..n; the sampled health parameter set j is used as a column vector to construct a health parameter data matrix, and the health parameter data matrix is recorded as a health matrix H n×5, and the expression is as follows:
Wherein V i is the sample terminal voltage measured in cycle i; i i is the sample side current measured in cycle I; t i is the sample lithium battery temperature measured in cycle i; t i is the sample lithium battery charge time measured in cycle i; c i is the sample lithium battery charge capacity measured in cycle i.
Step 2, carrying out normalization processing on the 5 groups of column vectors to obtain the following data after normalization processing:
n normalized lithium battery terminal voltages, and recording the lithium battery terminal voltage in the cycle i as terminal voltage N normalized end currents, and the end current in the cycle i is recorded as end current/>N normalized lithium battery temperatures, and recording the lithium battery temperature in cycle i as lithium battery temperature/>N normalized lithium battery charging times, and recording the lithium battery charging time in cycle i as lithium battery charging time/>N normalized lithium battery charge capacities, and recording the lithium battery charge capacity in cycle i as lithium battery charge capacity/>
Voltage of n terminalsN terminal currents/>N lithium battery temperatures/>And n lithium battery charging times/>Composition of input dataset X n×4, n lithium Battery Charge Capacity/>An output dataset Y n×1 is composed.
Step 3, dividing the input data set X n×4 into two groups, namely an input training set X train and an input testing set X test in sequence;
Dividing the output data set Y n×1 into two groups, namely an output training set Y train and an output testing set Y test in sequence;
Wherein, the proportion of the two training sets to the original data set is given preset value gamma, and gamma=50-90%;
Let the output training set Y train include m data in total, and record any one of the m data as an output training value Y f, f=1, 2.
And 4, establishing a three-layer BP neural network A and training.
Step 4.1, the structure of the three-layer BP neural network A comprises an input layer, a hidden layer and an output layer; the input layer comprises 4 input layer units, and the output layer comprises only 1 output layer unit; the input layer is fully connected with the hidden layer, and the hidden layer is fully connected with the output layer;
the number of hidden layer units is determined by an empirical formula and an adaptive adjustment formula, wherein the empirical formula is used for estimating the maximum value of the number of hidden layer units, and the empirical formula is as follows:
wherein h is the number of hidden layer units, h is an integer, r is the number of input layer units, s is the number of output layer units, a is a constant less than or equal to 10, i.e. the number of hidden layer units h is less than or equal to 12;
Step 4.2, setting 12 training hidden layer unit numbers h 'in a mode of adding 1 unit at a time, wherein the 1 st training hidden layer unit number h' is 1; substituting the number h 'of the 12 training hidden layer units into the BP neural network A established in the step 4.1 successively to obtain 12 training hidden layers, and then training until the number h' of the training hidden layer units is 12; the process of one training is as follows:
Inputting the input training set X train into BP neural network A comprising training hidden layer to train to obtain a set of output predicted values Wherein/>To output the predicted values, a set of predicted values/>Consistent with the Y train dimension of the output training set, the output predicted value/>Substituting the training error MAE with an output training value y f into an average absolute value error formula to calculate a training error MAE, wherein the average absolute value error formula has the following expression:
After the 12 times of training are finished, the number h 'of training hidden layer units in the 12 th time of training is recorded as a new hidden layer unit number h new, a training error MAE corresponding to the new hidden layer unit number h new is recorded as a new error MAE new, the minimum value in the first 11 training error MAEs is recorded as a minimum error MAE pre, and the number h' of training hidden layer units corresponding to the minimum error MAE pre is recorded as a minimum error hidden layer unit number h pre;
The optimal hidden layer unit number h best is determined according to an adaptive adjustment formula, wherein the expression of the adaptive adjustment formula is as follows:
Substituting the optimal hidden layer unit number h best as the hidden layer unit number into the BP neural network A obtained in the step 4.1, and finishing the construction of the three-layer BP neural network A.
And 5, adding the input training set X train and the output training set Y train obtained in the step 3 into the BP neural network A established in the step 4, training the weight and the threshold of the BP neural network A, completing the final training of the BP neural network A, and marking as the BP neural network B.
And 6, adding the input test set X test and the output test set Y test obtained in the step 3 into the BP neural network B trained in the step 5 for testing, and realizing the prediction of the health state of the lithium battery.
Preferably, in the step 4, the three-layer BP neural network a is built and trained, the goal of training the BP neural network is 10 -6, the hidden layer activation function is poslin, the output layer activation function is purelin, the training function is trainlm, and the learning rate is 0.0001.
The lithium battery condition prediction method based on the self-adaptive hidden layer BP neural network rapidly and accurately realizes lithium battery SOH prediction under the condition of data set change, and has the beneficial effects that:
1. the type of battery and the number of data sets of the training network are not fixed, and the structure of the network can be changed in real time to adapt to the change of the training set.
2. The charging time is added as an input parameter, so that the accuracy of SOH prediction of the lithium battery is effectively improved.
3. The method reduces the calculation complexity, shortens the calculation time, improves the application range, and has the advantages of high precision and strong tracking property.
Drawings
FIG. 1 is a block diagram of a BP neural network constructed in a lithium battery condition prediction method based on an adaptive hidden layer BP neural network of the invention;
FIG. 2 is a flow chart of a method for predicting the condition of a lithium battery based on an adaptive hidden layer BP neural network;
Fig. 3 is a comparison chart of the predicted result of the condition of the BP neural network lithium battery without the adaptive hidden layer under the training of the same data set in the embodiment of the present invention.
Fig. 4 is a graph comparing the predicted result of the condition of the BP neural network lithium battery with the condition of the BP neural network without the adaptive hidden layer under the training of different battery data sets.
Detailed Description
The technical scheme of the invention is further described in detail below with reference to the accompanying drawings.
The invention relates to a lithium battery condition prediction method based on a self-adaptive hidden layer BP neural network, wherein the lithium battery condition is a lithium battery health condition, the flow is shown in figure 2, and the steps of the prediction method are as follows:
Step 1, a lithium battery is charged and discharged for n times, the lithium battery in the n times of charging and discharging cycles is sampled, and sampling data form a lithium battery charging and discharging data set;
The following lithium battery sampling parameters related to the lithium battery health condition are extracted from the charge-discharge data set and recorded as sampling health parameters: sampling end voltage in n charge-discharge cycles, sampling end current in n charge-discharge cycles, sampling lithium battery temperature in n charge-discharge cycles, sampling lithium battery charging time in n charge-discharge cycles and sampling lithium battery capacity in n charge-discharge cycles, so as to obtain five groups of sampling health parameters, and marking any one of the five groups as sampling health parameter groups j, j=1, 2,3,4 and 5; any one of the n charge-discharge cycles is denoted as cycle i, i=1, 2,..n; the sampled health parameter set j is used as a column vector to construct a health parameter data matrix, and the health parameter data matrix is recorded as a health matrix H n×5, and the expression is as follows:
Wherein V i is the sample terminal voltage measured in cycle i; i i is the sample side current measured in cycle I; t i is the sample lithium battery temperature measured in cycle i; t i is the sample lithium battery charge time measured in cycle i; c i is the sample lithium battery charge capacity measured in cycle i.
Step 2, carrying out normalization processing on the 5 groups of column vectors to obtain the following data after normalization processing:
n normalized lithium battery terminal voltages, and recording the lithium battery terminal voltage in the cycle i as terminal voltage N normalized end currents, and the end current in the cycle i is recorded as end current/>N normalized lithium battery temperatures, and recording the lithium battery temperature in cycle i as lithium battery temperature/>N normalized lithium battery charging times, and recording the lithium battery charging time in cycle i as lithium battery charging time/>N normalized lithium battery charge capacities, and recording the lithium battery charge capacity in cycle i as lithium battery charge capacity/>
Voltage of n terminalsN terminal currents/>N lithium battery temperatures/>And n lithium battery charging times/>Composition of input dataset X n×4, n lithium Battery Charge Capacity/>An output dataset Y n×1 is composed.
Step 3, dividing the input data set X n×4 into two groups, namely an input training set X train and an input testing set X test in sequence;
Dividing the output data set Y n×1 into two groups, namely an output training set Y train and an output testing set Y test in sequence;
Wherein, the proportion of the two training sets to the original data set is given preset value gamma, and gamma=50-90%;
Let the output training set Y train include m data in total, and record any one of the m data as an output training value Y f, f=1, 2.
And 4, establishing a three-layer BP neural network A and training.
Step 4.1, the structure of the three-layer BP neural network A comprises an input layer, a hidden layer and an output layer; the input layer comprises 4 input layer units, and the output layer comprises only 1 output layer unit; the input layer is fully connected with the hidden layer, and the hidden layer is fully connected with the output layer;
the number of hidden layer units is determined by an empirical formula and an adaptive adjustment formula, wherein the empirical formula is used for estimating the maximum value of the number of hidden layer units, and the empirical formula is as follows:
wherein h is the number of hidden layer units, h is an integer, r is the number of input layer units, s is the number of output layer units, a is a constant less than or equal to 10, i.e. the number of hidden layer units h is less than or equal to 12;
Step 4.2, setting 12 training hidden layer unit numbers h 'in a mode of adding 1 unit at a time, wherein the 1 st training hidden layer unit number h' is 1; substituting the number h 'of the 12 training hidden layer units into the BP neural network A established in the step 4.1 successively to obtain 12 training hidden layers, and then training until the number h' of the training hidden layer units is 12; the process of one training is as follows:
Inputting the input training set X train into BP neural network A comprising training hidden layer to train to obtain a set of output predicted values Wherein/>To output the predicted values, a set of predicted values/>Consistent with the Y train dimension of the output training set, the output predicted value/>Substituting the training error MAE with an output training value y f into an average absolute value error formula to calculate a training error MAE, wherein the average absolute value error formula has the following expression:
After the 12 times of training are finished, the number h 'of training hidden layer units in the 12 th time of training is recorded as a new hidden layer unit number h new, a training error MAE corresponding to the new hidden layer unit number h new is recorded as a new error MAE new, the minimum value in the first 11 training error MAEs is recorded as a minimum error MAE pre, and the number h' of training hidden layer units corresponding to the minimum error MAE pre is recorded as a minimum error hidden layer unit number h pre;
The optimal hidden layer unit number h best is determined according to an adaptive adjustment formula, wherein the expression of the adaptive adjustment formula is as follows:
Substituting the optimal hidden layer unit number h best as the hidden layer unit number into the BP neural network A obtained in the step 4.1, and finishing the construction of the three-layer BP neural network A.
And 5, adding the input training set X train and the output training set Y train obtained in the step 3 into the BP neural network A established in the step 4, training the weight and the threshold of the BP neural network A, completing the final training of the BP neural network A, and marking as the BP neural network B.
And 6, adding the input test set X test and the output test set Y test obtained in the step 3 into the BP neural network B trained in the step 5 for testing, and realizing the prediction of the health state of the lithium battery.
In this embodiment, in the step 4, the three-layer BP neural network a is built and trained, the goal of training the BP neural network is 10 -6, the hidden layer activation function is poslin, the output layer activation function is purelin, the training function is trainlm, and the learning rate is 0.0001.
To verify the effect of the present invention, simulations were performed.
Fig. 1 is a block diagram of a BP neural network constructed in accordance with the present invention.
In this embodiment, first, a three-layer BP neural network a composed of an input layer, a hidden layer, and an output layer is constructed, where the input layer and the hidden layer are fully connected, and the hidden layer and the output layer are fully connected. The input layer comprises 4 input layer units, the output layer only comprises 1 output layer unit, and the hidden layer unit is obtained through training and testing.
In this embodiment, two different battery data sets, denoted as data set a and data set B, are used for simulation verification.
FIG. 3 is a comparison of simulation results of an embodiment of the present invention to FIG. 1. In this simulation, the data is taken from dataset a, preset value γ 1 =50%, in this embodiment the optimal number of hidden layer units h best1 =7. The abscissa of the graph is the charge-discharge cycle of data set a, and the ordinate is the lithium battery capacity in the charge-discharge cycle corresponding to data set a. The actual value curve is a curve drawn according to the original data of the data set A, the self-adaptive BP neural network curve is a curve drawn according to training and testing results of the prediction method, the neural network curve is a curve drawn according to training and testing results of a traditional BP neural network, and the traditional BP neural network is a BP neural network which is not provided with a self-adaptive hidden layer. As can be seen from fig. 3, the lithium battery health status obtained by the prediction method provided by the invention is closer to the actual value, i.e. the accuracy is higher.
FIG. 4 is a comparison of simulation results of an embodiment of the present invention to FIG. 2. In this simulation, the data is taken from the dataset B, the preset value γ 2 =50%, in this embodiment the optimal number of hidden layer units h best2 =9. The abscissa of the graph is the charge-discharge cycle of data set B, and the ordinate is the lithium battery capacity in the charge-discharge cycle corresponding to data set B. The actual value curve is a curve drawn according to the original data of the data set B, the self-adaptive BP neural network curve is a curve drawn according to training and testing results of the prediction method, the BP neural network curve is a curve drawn according to training and testing results of a traditional BP neural network, and the traditional BP neural network is a BP neural network which is not provided with a self-adaptive hidden layer. As can be seen from fig. 4, the lithium battery health status obtained by the prediction method provided by the present invention is closer to the actual value, i.e. the accuracy is higher.
Claims (2)
1. The lithium battery condition prediction method based on the self-adaptive hidden layer BP neural network is characterized by comprising the following steps of:
Step 1, a lithium battery is charged and discharged for n times, the lithium battery in the n times of charging and discharging cycles is sampled, and sampling data form a lithium battery charging and discharging data set;
The following lithium battery sampling parameters related to the lithium battery health condition are extracted from the charge-discharge data set and recorded as sampling health parameters: sampling end voltage in n charge-discharge cycles, sampling end current in n charge-discharge cycles, sampling lithium battery temperature in n charge-discharge cycles, sampling lithium battery charging time in n charge-discharge cycles and sampling lithium battery capacity in n charge-discharge cycles, so as to obtain five groups of sampling health parameters, and marking any one of the five groups as sampling health parameter groups j, j=1, 2,3,4 and 5; any one of the n charge-discharge cycles is denoted as cycle i, i=1, 2,..n; the sampled health parameter set j is used as a column vector to construct a health parameter data matrix, and the health parameter data matrix is recorded as a health matrix H n×5, and the expression is as follows:
Wherein V i is the sample terminal voltage measured in cycle i; i i is the sample side current measured in cycle I; t i is the sample lithium battery temperature measured in cycle i; t i is the sample lithium battery charge time measured in cycle i; c i is the sample lithium battery charge capacity measured in cycle i;
step 2, carrying out normalization processing on the 5 groups of column vectors to obtain the following data after normalization processing:
n normalized lithium battery terminal voltages, and recording the lithium battery terminal voltage in the cycle i as terminal voltage N normalized end currents, and the end current in the cycle i is recorded as end current/>N normalized lithium battery temperatures, and recording the lithium battery temperature in cycle i as lithium battery temperature/>N normalized lithium battery charging times, and recording the lithium battery charging time in cycle i as lithium battery charging time/>N normalized lithium battery charge capacities, and recording the lithium battery charge capacity in cycle i as lithium battery charge capacity/>
Voltage of n terminalsN terminal currents/>N lithium battery temperatures/>And n lithium battery charging times/>Composition of input dataset X n×4, n lithium Battery Charge Capacity/>Composing the output dataset Y n×1;
Step 3, dividing the input data set X n×4 into two groups, namely an input training set X train and an input testing set X test in sequence;
Dividing the output data set Y n×1 into two groups, namely an output training set Y train and an output testing set Y test in sequence;
Wherein, the proportion of the two training sets to the original data set is given preset value gamma, and gamma=50-90%;
Setting an output training set Y train to include m pieces of data in total, and recording any one of the m pieces of data as an output training value Y f, f=1, 2, & gt, m, m < n;
Step 4, establishing a three-layer BP neural network A and training;
Step 4.1, the structure of the three-layer BP neural network A comprises an input layer, a hidden layer and an output layer; the input layer comprises 4 input layer units, and the output layer comprises only 1 output layer unit; the input layer is fully connected with the hidden layer, and the hidden layer is fully connected with the output layer;
the number of hidden layer units is determined by an empirical formula and an adaptive adjustment formula, wherein the empirical formula is used for estimating the maximum value of the number of hidden layer units, and the empirical formula is as follows:
wherein h is the number of hidden layer units, h is an integer, r is the number of input layer units, s is the number of output layer units, a is a constant less than or equal to 10, i.e. the number of hidden layer units h is less than or equal to 12;
Step 4.2, setting 12 training hidden layer unit numbers h 'in a mode of adding 1 unit at a time, wherein the 1 st training hidden layer unit number h' is 1; substituting the number h 'of the 12 training hidden layer units into the BP neural network A established in the step 4.1 successively to obtain 12 training hidden layers, and then training until the number h' of the training hidden layer units is 12; the process of one training is as follows:
Inputting the input training set X train into BP neural network A comprising training hidden layer to train to obtain a set of output predicted values Wherein/>To output the predicted values, a set of predicted values/>Consistent with the Y train dimension of the output training set, the output predicted value/>Substituting the training error MAE with an output training value y f into an average absolute value error formula to calculate a training error MAE, wherein the average absolute value error formula has the following expression:
After the 12 times of training are finished, the number h 'of training hidden layer units in the 12 th time of training is recorded as a new hidden layer unit number h new, a training error MAE corresponding to the new hidden layer unit number h new is recorded as a new error MAE new, the minimum value in the first 11 training error MAEs is recorded as a minimum error MAE pre, and the number h' of training hidden layer units corresponding to the minimum error MAE pre is recorded as a minimum error hidden layer unit number h pre;
The optimal hidden layer unit number h best is determined according to an adaptive adjustment formula, wherein the expression of the adaptive adjustment formula is as follows:
substituting the optimal hidden layer unit number h best as the hidden layer unit number into the BP neural network A obtained in the step 4.1, and finishing the construction of the three-layer BP neural network A;
Step 5, adding the input training set X train and the output training set Y train obtained in the step 3 into the BP neural network A established in the step 4, training the weight and the threshold of the BP neural network A, completing the final training of the BP neural network A, and marking as BP neural network B;
and 6, adding the input test set X test and the output test set Y test obtained in the step 3 into the BP neural network B trained in the step 5 for testing, and realizing the prediction of the health state of the lithium battery.
2. The method for predicting the lithium battery condition based on the self-adaptive hidden layer BP neural network according to claim 1, wherein in the step 4, the three-layer BP neural network A is built and trained, the goal of training the BP neural network is 10 -6, the hidden layer activation function is poslin, the output layer activation function is purelin, the training function is trainlm, and the learning rate is 0.0001.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210554879.9A CN115078999B (en) | 2022-05-20 | 2022-05-20 | Lithium battery condition prediction method based on self-adaptive hidden layer BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210554879.9A CN115078999B (en) | 2022-05-20 | 2022-05-20 | Lithium battery condition prediction method based on self-adaptive hidden layer BP neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115078999A CN115078999A (en) | 2022-09-20 |
| CN115078999B true CN115078999B (en) | 2024-04-26 |
Family
ID=83249920
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210554879.9A Active CN115078999B (en) | 2022-05-20 | 2022-05-20 | Lithium battery condition prediction method based on self-adaptive hidden layer BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115078999B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112881914A (en) * | 2021-01-12 | 2021-06-01 | 常州大学 | Lithium battery health state prediction method |
| JPWO2022080377A1 (en) * | 2020-10-15 | 2022-04-21 |
-
2022
- 2022-05-20 CN CN202210554879.9A patent/CN115078999B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPWO2022080377A1 (en) * | 2020-10-15 | 2022-04-21 | ||
| CN112881914A (en) * | 2021-01-12 | 2021-06-01 | 常州大学 | Lithium battery health state prediction method |
Non-Patent Citations (2)
| Title |
|---|
| BP神经网络预估锂离子电池SOC训练数据选择;封进;;电源技术;20160220(02);全文 * |
| 基于SA-BP神经网络算法的电池SOH预测;徐元中;曹翰林;吴铁洲;;电源技术;20200320(03);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115078999A (en) | 2022-09-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tian et al. | Consistency evaluation and cluster analysis for lithium-ion battery pack in electric vehicles | |
| CN110398697B (en) | Lithium ion health state estimation method based on charging process | |
| Liu et al. | An advanced Lithium-ion battery optimal charging strategy based on a coupled thermoelectric model | |
| CN110525269B (en) | SOC battery pack balance control method | |
| CN111007401A (en) | Electric vehicle battery fault diagnosis method and device based on artificial intelligence | |
| Zhang et al. | Intelligent state of charge estimation of battery pack based on particle swarm optimization algorithm improved radical basis function neural network | |
| CN112557907A (en) | SOC estimation method of electric vehicle lithium ion battery based on GRU-RNN | |
| CN114861527A (en) | Lithium battery life prediction method based on time series characteristics | |
| CN114839536B (en) | Lithium ion battery health state estimation method based on multiple health factors | |
| CN113740736A (en) | Electric vehicle lithium battery SOH estimation method based on deep network self-adaptation | |
| CN112163372B (en) | SOC estimation method of power battery | |
| Sun et al. | Study of parameters identification method of Li-ion battery model for EV power profile based on transient characteristics data | |
| CN115308606B (en) | Lithium ion battery health state estimation method based on adjacent characteristics | |
| CN115639481B (en) | Battery data preprocessing system and method based on big data prediction SOC | |
| Zhao et al. | Estimation of the SOC of energy-storage lithium batteries based on the voltage increment | |
| KR20210000207A (en) | Method of detecting internal short-circuit cell | |
| CN115166561A (en) | Lithium battery life prediction method based on CNN-GRU combined neural network | |
| CN116298936A (en) | Intelligent lithium ion battery health state prediction method in incomplete voltage range | |
| CN115389946A (en) | Lithium battery health state estimation method based on isobaric rise energy and improved GRU | |
| CN114545275A (en) | Indirect prediction method for remaining service life of lithium ion battery | |
| CN115078999B (en) | Lithium battery condition prediction method based on self-adaptive hidden layer BP neural network | |
| CN117087495A (en) | Method for estimating driving range of electric automobile based on battery aging level | |
| CN116774045A (en) | Lithium battery health state prediction method based on HHO-SVR | |
| CN116679213A (en) | SOH estimation method for electric vehicle power battery based on integrated deep learning | |
| CN114545279A (en) | Lithium battery health state estimation method based on neural network ordinary differential equation |
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 |