CN117630683A - Multi-scale fusion GRU network-based automobile battery SOC multi-step prediction method and system - Google Patents
Multi-scale fusion GRU network-based automobile battery SOC multi-step prediction method and system Download PDFInfo
- Publication number
- CN117630683A CN117630683A CN202410106018.3A CN202410106018A CN117630683A CN 117630683 A CN117630683 A CN 117630683A CN 202410106018 A CN202410106018 A CN 202410106018A CN 117630683 A CN117630683 A CN 117630683A
- Authority
- CN
- China
- Prior art keywords
- data
- soc
- gru network
- time
- model
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 230000004927 fusion Effects 0.000 title claims abstract description 35
- 238000012549 training Methods 0.000 claims abstract description 48
- 238000012795 verification Methods 0.000 claims abstract description 19
- 230000007774 longterm Effects 0.000 claims abstract description 9
- 238000001514 detection method Methods 0.000 claims abstract description 3
- 238000012545 processing Methods 0.000 claims description 18
- 239000013598 vector Substances 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 6
- 230000002596 correlated effect Effects 0.000 claims description 4
- 238000009413 insulation Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000000717 retained effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/60—Other road transportation technologies with climate change mitigation effect
- Y02T10/70—Energy storage systems for electromobility, e.g. batteries
Abstract
The invention discloses a multi-step prediction method and a multi-step prediction system for automobile battery SOC based on a multi-scale fusion GRU network, wherein the method comprises the following steps: s1, taking each sensor detection item as a variable, collecting comprehensive data of a real vehicle sensor, and converting the comprehensive data into standard normal distribution; selecting variables highly related to the SOC to construct a sliding window; taking 80% of data of the sliding window as a training set and 20% of data as a verification set; s2, initializing parameters based on the multiscale fusion GRU network model, setting a plurality of key super parameters at the same time, and performing offline training on a training set to generate an SOC multi-step prediction result; the multi-step prediction method and system for the SOC of the automobile battery based on the multi-scale fusion GRU network can better capture short-term and long-term dependence in historical data, so that the behavior of the battery in diversified and unstable environments can be predicted more accurately, and higher accuracy and reliability can be realized in real-world application.
Description
Technical Field
The invention relates to the technical field of electric vehicle battery SOC prediction, in particular to an automobile battery SOC multi-step prediction method and system based on a multi-scale fusion GRU network.
Background
In the background of global climate change and increasingly severe environmental problems, electric vehicles are becoming an important option for sustainable traffic. In the technical construction of electric vehicles, a Battery Management System (BMS) plays a key role, and main tasks include charge and discharge management of a battery pack, and state monitoring. Among the tasks of the BMS, state of charge (SOC) prediction of the battery is certainly the most critical one.
SOC is used to describe the relative proportion of the remaining charge of a battery, and is typically expressed in percent form. Accurate SOC information can ensure running safety and avoid sudden exhaustion of the battery; and the optimization of battery charge and discharge is facilitated, the service life of the battery is prolonged, and the maintenance cost is reduced. Meanwhile, the economic benefit of the electric automobile can be improved, and the operation cost is reduced by optimizing the travel and the charging plan.
Current techniques for SOC estimation can be generally divided into three main categories, but each has limitations. First, although temperature models and OCV-SOC tables are used to improve accuracy based on defined methods such as Open Circuit Voltage (OCV) and ampere-hour integration, the accuracy of these methods is still limited for varying environments and conditions of use. Second, model-based methods such as kalman filtering and equivalent circuit models perform better in terms of accuracy, but at the cost of complexity of the model parameters. These parameters need to be adjusted in real time to accommodate different operating conditions and environmental factors, increasing the complexity of implementation. Third, data-driven methods such as machine learning and neural networks, while flexible, have dependencies on large amounts of training data, which can be a problem in practical applications.
Although SOC estimation has been a widely studied field, there is relatively little research on SOC prediction. Furthermore, most research is limited to laboratory environments, which often do not fully reflect the complexity and variability of real-world vehicle data. In the real world, there are significant limitations to definition-based methods and model-based methods due to the complex variability of the environment and conditions of use. Thus, data driven methods, particularly neural networks, may provide a more viable solution. Although these methods require a large amount of training data, their flexibility and adaptability make them particularly suitable for processing real-world complex and diverse data.
Disclosure of Invention
The invention provides a multi-step prediction method and a multi-step prediction system for the SOC of an automobile battery based on a multi-scale fusion GRU network, which can better capture short-term and long-term dependence in historical data, so as to more accurately predict the behaviors of the battery in diversified and unstable environments and realize higher accuracy and reliability in real world application.
In order to achieve the above purpose, the invention provides a multi-step prediction method for automobile battery SOC based on a multi-scale fusion GRU network, which comprises the following steps:
s1, collecting comprehensive data of a real vehicle sensor, regarding each sensor detection item as a variable, preprocessing the collected comprehensive data and constructing a sliding window; the data of 80% of the sliding window is used as a training set, and the data of 20% is used as a verification set;
s2, initializing parameters based on the multiscale fusion GRU network model, setting a plurality of key super parameters at the same time, and performing offline training on a training set to generate an SOC multi-step prediction result;
s3, applying a loss function, calculating the deviation between the predicted result and the actual result in the step S2, and updating GRU network model parameters by adopting a random gradient descent algorithm; carrying out one-time complete verification on the verification set by using the updated GRU network model, and calculating a loss function of the verification set; performing offline training by using the updated GRU network model to generate an SOC multi-step prediction model;
s4, repeating the step S3 until the minimum loss value is obtained on the verification set, and obtaining an optimal GRU network model; meanwhile, a grid searching method is adopted to determine the optimal value range of the key super-parameters, so that super-parameter setting of optimal performance of the model is realized;
s5, applying the optimal model to an actual vehicle, performing standardization processing on real-time data acquired by the vehicle-mounted sensor, selecting relevant variables selected in the step S1, inputting the relevant variables into the deployed optimal model, generating a multi-step prediction result of the real-time SOC, and inputting the prediction result into a battery management system for optimizing scheduling and management of an actual vehicle battery.
Preferably, the integrated data collected in the step S1 includes a vehicle speed, a total voltage and a total current of the battery pack, maximum and minimum voltages, maximum and minimum temperatures of the unit cells, an insulation resistance, a driving range, a driving state of the vehicle, and a charging state.
Preferably, preprocessing the collected comprehensive data and constructing a sliding window in the step S1 includes:
s1-1, importing the collected comprehensive data into a computer, and screening and processing abnormal values of the comprehensive data; carrying out standardized processing on the comprehensive data, and converting each data into standard normal distribution;
s1-2, evaluating the correlation between the SOC and each variable in the comprehensive data by adopting a Pearson correlation coefficient method, and selecting a variable with the Pearson correlation coefficient larger than 0.4 as a highly correlated variable;
s1-3, constructing a sliding window, wherein the size of the sliding window is 360 time steps, the prediction result is an SOC value of 6 time steps after the sliding window, and the sliding step length is 6 time steps.
Preferably, in step S2, the offline training process of the training set includes:
s2-1, super parameters comprise learning rate, batch processing size, the number of layers of GRU network and the number of hidden units; according to the preset batch processing size, randomly dividing the training set into a plurality of small batches, transmitting a plurality of batches of data into a multi-layer GRU network one by one, and finally forming a feature vector containing local information through iterative operation of the multi-layer GRU network；
S2-2, performing multi-layer GRU network calculation and simultaneously segmenting training set data in an overlapping manner to obtain dataData +.>Inputting the local information into another multi-layer GRU network, and extracting the local information from each segment;
s2-3, recombining the feature vectors obtained after segmentation into a new time sequence vector according to time sequenceThen the timing vector is->Inputting into another new multi-layer GRU network, further mining long-term dependence information, and forming feature vector +.>；
S2-4, feature vectorAnd->Respectively with initialized learnable weights omega 1 And omega 2 Multiplying, then adding to adaptively fuse the local information and the long-term dependence information, and further inputting the fused characteristics into a fully-connected network to perform multi-step SOC prediction.
Preferably, the calculation process of the multi-layer GRU network is as follows:
；
；
；
；
wherein the method comprises the steps ofFor the current time steptInput of->In time steps for resetting the gatetStatus of->In time steps for updating doorstStatus of->Is a time steptCandidate hidden state of->Is a time steptIs hidden in->、/>、/>Representing a weight matrix and bias terms associated with a reset gate,/->、/>、/>Representing the weight matrix and bias associated with the update gate,、/>、/>representing a weight matrix and bias terms associated with candidate hidden states, < ->Representing hyperbolic tangent activation function, ">Representing a sigmoid activation function,/->The Hadamard product, i.e., the product of elements by elements, is represented.
Preferably, the data in the step S2-2XThe segment length was 60 and the non-overlapping portion length was 12 when the segmentation was performed.
Preferably, in the step S2-2, all the segments share the same network and share network parameters when performing the multi-layer GRU network calculation, so as to reduce the calculation complexity and enhance the model consistency.
Preferably, the loss function of the prediction model is calculated in the step S3 as follows:
；
wherein the method comprises the steps ofFor training sample size, ++>For prediction step size +.>Is thatt+jTrue value of time of day +.>Is thatt+jPredicted value of time.
The system established by the multi-step prediction method for the SOC of the automobile battery based on the multi-scale fusion GRU network comprises the following components:
the data acquisition module is used for acquiring comprehensive data of the running vehicles in the real world;
the data preprocessing module is used for acquiring variables highly related to the SOC and constructing a data set required by offline training;
the off-line training module divides the data set into a training set and a verification set, and performs off-line training by using the training set and the verification set to obtain an optimal multi-scale fusion GRU network model;
the real vehicle deployment module deploys the obtained optimal multi-scale fusion GRU network model into a real vehicle, performs standardization processing on real-time data acquired by the vehicle-mounted sensor, selects related variables selected by the data preprocessing module, inputs the related variables into the deployed optimal multi-scale fusion GRU network model, generates a multi-step prediction result of the real-time SOC, and inputs the prediction result into the battery management system for optimizing the dispatching and management of the actual vehicle battery.
Therefore, the automobile battery SOC multi-step prediction method and system based on the multi-scale fusion GRU network have the following technical effects:
(1) According to the invention, the dual-stage segmented GRU is combined with the conventional GRU, so that the SOC multi-scale characteristics are successfully captured, the local characteristics are recognized, and the prediction capability of the model in a long-term context relationship is enhanced, so that the prediction accuracy is remarkably improved;
(2) According to the invention, the learnable weights are used for carrying out self-adaptive fusion on the characteristics of different levels in the GRU network model, and the importance of each characteristic is adjusted independently according to the data, so that the prediction accuracy is improved, and the robustness of the model is enhanced;
(3) The invention utilizes the vehicle-mounted sensor to collect real vehicle data, screens out the most relevant variable based on the collected historical data, trains and determines the optimal SOC prediction model in an offline environment by means of a super-parameter grid search technology;
(4) According to the invention, the offline model is subjected to real-vehicle deployment, the data can be acquired in real time by using the vehicle-mounted sensor, and the multi-step prediction of the SOC can be performed on line. The model can adapt to complex real-world environments, and the continuous safety monitoring and risk early warning capacity of the battery system is enhanced, so that the safety of the battery system of the electric automobile is comprehensively improved.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. It will be apparent to those of ordinary skill in the art that the drawings in the following description are exemplary only and that other implementations can be obtained from the extensions of the drawings provided without inventive effort.
FIG. 1 is a schematic diagram of a multi-scale fusion GRU network model according to an embodiment;
FIG. 2 is a flow chart of an embodiment of a method for multi-step prediction of the SOC of an electric vehicle battery based on a multi-scale fusion GRU network;
FIG. 3 is a specific training process for offline training to successfully generate an optimal model according to an embodiment;
fig. 4 is a diagram comparing the prediction result of the prediction model of real vehicle deployment with the actual measurement value.
Detailed Description
The technical scheme of the invention is further described below through the attached drawings and the embodiments.
Unless defined otherwise, technical or scientific terms used herein should be given the ordinary meaning as understood by one of ordinary skill in the art to which this invention belongs.
Example 1
A multi-step prediction method for the SOC of an automobile battery based on a multi-scale fusion GRU network is provided, and a structural schematic diagram of the multi-scale fusion GRU network model is shown in figure 1. Firstly, various data are collected through a real vehicle sensor, then data cleaning and standardization processing are carried out, and factors related to the SOC height are further analyzed. These preprocessed key data are then used for offline training to successfully generate the optimal model. Finally, the model is implemented for deployment for real-time multi-step prediction of real vehicle SOC. The implementation flow is shown in fig. 2, and the specific training flow for offline training to successfully generate the optimal model is shown in fig. 3.
The specific implementation steps are as follows:
s1, data acquisition and preprocessing
The method comprises the steps of deploying a vehicle-mounted sensor on an electric vehicle in running to acquire key data, wherein the acquisition interval is 10 seconds, and the key data comprise vehicle speed, total voltage and total current of a battery pack, maximum and minimum voltage, maximum and minimum temperature, insulation resistance and running mileage of a single battery. At the same time, the running state and the charging state of the vehicle are recorded. The collected data is then transmitted to a computer for processing. This stage of processing involves outlier screening and normalization of the data to construct a structured offline data set.
S2, correlation analysis
The pearson correlation coefficient is used to analyze the correlation between the SOC and the various data variables collected. Variables with a correlation coefficient exceeding 0.4 are selected as highly correlated variables, and these highly correlated variables include the total voltage and total current of the battery pack, the maximum and minimum voltages of the unit cells, and the maximum and minimum temperatures of the cells, for a total of six variables.
S3, constructing an offline training and verification data set
First, non-critical influence factor data is eliminated, and only data identified as critical influence factors is retained. On this basis, the retained data is divided into two parts: the first 80% is used as a training set for learning and adjusting the model; the latter 20% was used as a validation set to evaluate the efficacy and accuracy of the model. On both data sets we each constructed a sliding window containing 360 time steps. Each window is spaced apart by 6 time steps to achieve continuous prediction of 6 time steps into the future.
S4, offline training of prediction model
Parameters based on the multiscale fusion GRU network model are initialized, and meanwhile, a plurality of key super parameters are set, including learning rate, batch processing size, the number of layers of the GRU network and the number of hidden units. The training set is then randomly divided into a plurality of small batches according to a predetermined batch size, and the batch data are fed into the model one by one for processing. The various modules process the data in accordance with the computational process illustrated in fig. 1 and generate corresponding results. The method specifically comprises the following steps:
s4-1, transmitting the training set data to a multi-layer GRU network, and finally forming a feature vector containing local information through iterative operation of the multi-layer GRU network；
S4-2, performing multi-layer GRU network calculation and simultaneously segmenting training set data in an overlapping manner to obtain data. For dataXThe segment length was 60 and the non-overlapping portion length was 12 when the segmentation was performed. All segments share the same network and share network parameters to reduce computational complexity and enhance model consistency. Data +.>Inputting the local information into another multi-layer GRU network, and extracting the local information from each segment;
s4-3, recombining the feature vectors obtained after segmentation into a new time sequence vector according to time sequenceThen the timing vector is->Inputting into another new multi-layer GRU network, further mining long-term dependence information, and forming feature vector +.>；
S4-4, feature vectorAnd->Respectively with initialized learnable weights omega 1 And omega 2 Multiplying, then adding to adaptively fuse the local information and the long-term dependence information, and further inputting the fused characteristics into a fully-connected network to perform multi-step SOC prediction.
The calculation process of the multi-layer GRU network is as follows:
；
；
；
；
wherein the method comprises the steps ofFor the current time steptInput of->In time steps for resetting the gatetStatus of->In time steps for updating doorstStatus of->Is a time steptCandidate hidden state of->Is a time steptIs hidden in->、/>、/>Representing a weight matrix and bias terms associated with a reset gate,/->、/>、/>Representing the weight matrix and bias associated with the update gate,、/>、/>representing a weight matrix and bias terms associated with candidate hidden states, < ->Representing hyperbolic tangent activation function, ">Representing a sigmoid activation function,/->The Hadamard product, i.e., the product of elements by elements, is represented.
And S5, further, applying a loss function, calculating the deviation between the predicted result and the actual result, and updating the model parameters by adopting a random gradient descent algorithm. This process aims to bring the model predictions closer to true values. After the training data of one round is processed, the updated model is used for carrying out one-time complete verification on the verification set, and the loss value of the verification set is calculated. The training set is then trained again until a minimum loss value is obtained on the validation set. Meanwhile, a grid search method is adopted to determine the optimal value range of the key super-parameters, so that super-parameter setting of optimal performance of the model is realized.
The loss function is as follows:
；
wherein the method comprises the steps ofFor training sample size, ++>For prediction step size +.>Is thatt+jTrue value of time of day +.>Is thatt+jPredicted value of time.
Real vehicle deployment prediction model
After offline training, deploying the optimized model into a control system of the electric automobile, generating a multi-step prediction result of the real-time SOC, and inputting the prediction result into a battery management system for optimizing the dispatching and management of the actual vehicle battery
The on-board sensor collects key data including total voltage and total current of the battery pack, maximum and minimum voltages of the unit cells, and maximum and minimum temperatures of the cells every 10 seconds. These data cover the last 1 hour (corresponding to 360 time steps) of vehicle operation. The data collected in real time is then cleaned and normalized and then input into the model to predict the battery SOC change for the next 1 minute (i.e., 6 time steps). The above prediction step was repeated every 1 minute, and continuous test was performed for 20 minutes. During this test, the model continuously receives new data, predicts the battery SOC, and compares the predicted result with the actual measured value, the comparison result being shown in fig. 4.
Therefore, the automobile battery SOC multi-step prediction method and system based on the multi-scale fusion GRU network are adopted, the real automobile data are collected by using the vehicle-mounted sensor, the SOC multi-step prediction is carried out through the GRU network model, the self-adaptive fusion is carried out on the characteristics of different levels by using the learnable weights in the GRU network model, the importance of each characteristic is adjusted independently according to the data, the prediction accuracy is improved, and the robustness of the model is enhanced; and the offline model is subjected to real-vehicle deployment, data can be acquired in real time by using the vehicle-mounted sensor, and multi-step prediction of the SOC is performed on line. The model can adapt to complex real-world environments, and the continuous safety monitoring and risk early warning capacity of the battery system is enhanced, so that the safety of the battery system of the electric automobile is comprehensively improved.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention and not for limiting it, and although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that: the technical scheme of the invention can be modified or replaced by the same, and the modified technical scheme cannot deviate from the spirit and scope of the technical scheme of the invention.
Claims (9)
1. The multi-step prediction method for the SOC of the automobile battery based on the multi-scale fusion GRU network is characterized by comprising the following steps of:
s1, collecting comprehensive data of a real vehicle sensor, regarding each sensor detection item as a variable, preprocessing the collected comprehensive data and constructing a sliding window; the data of 80% of the sliding window is used as a training set, and the data of 20% is used as a verification set;
s2, initializing parameters based on the multiscale fusion GRU network model, setting a plurality of key super parameters at the same time, and performing offline training on a training set to generate an SOC multi-step prediction result;
s3, applying a loss function, calculating the deviation between the predicted result and the actual result in the step S2, and updating GRU network model parameters by adopting a random gradient descent algorithm; carrying out one-time complete verification on the verification set by using the updated GRU network model, and calculating a loss function of the verification set; performing offline training by using the updated GRU network model to generate an SOC multi-step prediction model;
s4, repeating the step S3 until the minimum loss value is obtained on the verification set, and obtaining an optimal GRU network model; meanwhile, a grid searching method is adopted to determine the optimal value range of the key super-parameters, so that super-parameter setting of optimal performance of the model is realized;
s5, applying the optimal model to an actual vehicle, performing standardization processing on real-time data acquired by the vehicle-mounted sensor, selecting relevant variables selected in the step S1, inputting the relevant variables into the deployed optimal model, generating a multi-step prediction result of the real-time SOC, and inputting the prediction result into a battery management system for optimizing scheduling and management of an actual vehicle battery.
2. The multi-step prediction method for SOC of an automotive battery based on a multi-scale fusion GRU network according to claim 1, wherein the integrated data collected in step S1 includes a vehicle speed, a total voltage and a total current of a battery pack, a maximum and a minimum voltage of a single battery, a maximum and a minimum temperature, an insulation resistance, a driving mileage, a driving state of a vehicle, and a charging state.
3. The multi-step prediction method for SOC of an automotive battery based on a multi-scale fusion GRU network according to claim 1, wherein the preprocessing of the collected comprehensive data and constructing the sliding window in step S1 includes:
s1-1, importing the collected comprehensive data into a computer, and screening and processing abnormal values of the comprehensive data; carrying out standardized processing on the comprehensive data, and converting each data into standard normal distribution;
s1-2, evaluating the correlation between the SOC and each variable in the comprehensive data by adopting a Pearson correlation coefficient method, and selecting a variable with the Pearson correlation coefficient larger than 0.4 as a highly correlated variable;
s1-3, constructing a sliding window, wherein the size of the sliding window is 360 time steps, the prediction result is an SOC value of 6 time steps after the sliding window, and the sliding step length is 6 time steps.
4. The multi-step prediction method of the SOC of the automotive battery based on the multi-scale fusion GRU network according to claim 1, wherein in step S2, the offline training process of the training set includes:
s2-1, super parameters comprise learning rate, batch processing size, the number of layers of GRU network and the number of hidden units; according to the preset batch processing size, randomly dividing the training set into a plurality of small batches, transmitting a plurality of batches of data into a multi-layer GRU network one by one, and finally forming a feature vector containing local information through iterative operation of the multi-layer GRU network；
S2-2, performing multi-layer GRU network calculation and simultaneously segmenting training set data in an overlapping manner to obtain dataData +.>Inputting the local information into another multi-layer GRU network, and extracting the local information from each segment;
s2-3, recombining the feature vectors obtained after segmentation into a new time sequence vector according to time sequenceThen the timing vector is->Input into another new multi-layer GRU network, and further excavate long-term dependence informationInformation and form feature vectors；
S2-4, feature vectorAnd->Respectively with initialized learnable weights omega 1 And omega 2 Multiplying, then adding to adaptively fuse the local information and the long-term dependence information, and further inputting the fused characteristics into a fully-connected network to perform multi-step SOC prediction.
5. The multi-scale fusion GRU network-based multi-step prediction method for automobile battery SOC according to claim 4, wherein the multi-layer GRU network is calculated as follows:
；
；
；
；
wherein the method comprises the steps ofFor the current time steptInput of->In time steps for resetting the gatetIs used for the control of the state of (a),/>in time steps for updating doorstStatus of->Is a time steptCandidate hidden state of->Is a time steptIs hidden in->、/>、/>Representing a weight matrix and bias terms associated with a reset gate,/->、/>、/>Representing a weight matrix and bias associated with the update gate, < ->、/>、Representing a weight matrix and bias terms associated with candidate hidden states, < ->Representing hyperbolic tangent activation function, ">Representing a sigmoid activation function,/->The Hadamard product, i.e., the product of elements by elements, is represented.
6. The multi-step prediction method for SOC of automotive battery based on multi-scale fusion GRU network of claim 4, wherein the data in step S2-2XThe segment length was 60 and the non-overlapping portion length was 12 when the segmentation was performed.
7. The multi-step prediction method of SOC of an automotive battery based on a multi-scale fusion GRU network according to claim 4, wherein all segments share the same network and share network parameters when the multi-layer GRU network calculation is performed in step S2-2, so as to reduce the calculation complexity and enhance the model consistency.
8. The multi-step prediction method of the SOC of the automotive battery based on the multi-scale fusion GRU network according to claim 1, wherein the loss function of the prediction model is calculated in the step S3 as follows:
；
wherein the method comprises the steps ofFor training sample size, ++>For prediction step size +.>Is thatt+jTrue value of time of day +.>Is thatt+jPredicted value of time.
9. A system established by the multi-step prediction method for SOC of an automotive battery based on a multi-scale fusion GRU network according to any one of claims 1 to 8, comprising:
the data acquisition module is used for acquiring comprehensive data of the running vehicles in the real world;
the data preprocessing module is used for acquiring variables highly related to the SOC and constructing a data set required by offline training;
the off-line training module divides the data set into a training set and a verification set, and performs off-line training by using the training set and the verification set to obtain an optimal multi-scale fusion GRU network model;
the real vehicle deployment module deploys the obtained optimal multi-scale fusion GRU network model into a real vehicle, performs standardization processing on real-time data acquired by the vehicle-mounted sensor, selects related variables selected by the data preprocessing module, inputs the related variables into the deployed optimal multi-scale fusion GRU network model, generates a multi-step prediction result of the real-time SOC, and inputs the prediction result into the battery management system for optimizing the dispatching and management of the actual vehicle battery.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410106018.3A CN117630683B (en) | 2024-01-25 | 2024-01-25 | Multi-scale fusion GRU network-based automobile battery SOC multi-step prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410106018.3A CN117630683B (en) | 2024-01-25 | 2024-01-25 | Multi-scale fusion GRU network-based automobile battery SOC multi-step prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117630683A true CN117630683A (en) | 2024-03-01 |
| CN117630683B CN117630683B (en) | 2024-03-29 |
Family
ID=90030774
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410106018.3A Active CN117630683B (en) | 2024-01-25 | 2024-01-25 | Multi-scale fusion GRU network-based automobile battery SOC multi-step prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117630683B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112557907A (en) * | 2020-12-17 | 2021-03-26 | 杭州六纪科技有限公司 | SOC estimation method of electric vehicle lithium ion battery based on GRU-RNN |
| US10969436B1 (en) * | 2019-10-25 | 2021-04-06 | The Florida International University Board Of Trustees | Systems and methods for forecasting battery state of charge |
| CN115656831A (en) * | 2022-10-08 | 2023-01-31 | 昆明理工大学 | Multi-step advanced prediction and fault diagnosis method for single battery voltage |
| CN115718263A (en) * | 2023-01-09 | 2023-02-28 | 北京科技大学 | Attention-based lithium ion battery calendar aging prediction model and method |
| CN116482539A (en) * | 2023-03-27 | 2023-07-25 | 西安交通大学 | Multi-dimensional data driving-based battery state-of-charge estimation method, system, equipment and medium |
| CN116500444A (en) * | 2023-04-06 | 2023-07-28 | 重庆大学 | Multi-state joint estimation method for battery of electric flying car facing operation safety |
| CN117406100A (en) * | 2023-11-30 | 2024-01-16 | 江南大学 | Lithium ion battery remaining life prediction method and system |
-
2024
- 2024-01-25 CN CN202410106018.3A patent/CN117630683B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10969436B1 (en) * | 2019-10-25 | 2021-04-06 | The Florida International University Board Of Trustees | Systems and methods for forecasting battery state of charge |
| CN112557907A (en) * | 2020-12-17 | 2021-03-26 | 杭州六纪科技有限公司 | SOC estimation method of electric vehicle lithium ion battery based on GRU-RNN |
| CN115656831A (en) * | 2022-10-08 | 2023-01-31 | 昆明理工大学 | Multi-step advanced prediction and fault diagnosis method for single battery voltage |
| CN115718263A (en) * | 2023-01-09 | 2023-02-28 | 北京科技大学 | Attention-based lithium ion battery calendar aging prediction model and method |
| CN116482539A (en) * | 2023-03-27 | 2023-07-25 | 西安交通大学 | Multi-dimensional data driving-based battery state-of-charge estimation method, system, equipment and medium |
| CN116500444A (en) * | 2023-04-06 | 2023-07-28 | 重庆大学 | Multi-state joint estimation method for battery of electric flying car facing operation safety |
| CN117406100A (en) * | 2023-11-30 | 2024-01-16 | 江南大学 | Lithium ion battery remaining life prediction method and system |
Non-Patent Citations (4)
| Title |
|---|
| TIANYU HU ETAL.: "Electrochemical-Theory-Guided Modeling of the Conditional Generative Adversarial Network for Battery Calendar Aging Forecast", IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, vol. 11, no. 1, 28 February 2023 (2023-02-28), pages 67 - 77, XP011933553, DOI: 10.1109/JESTPE.2022.3154785 * |
| TIANYU HU ETAL.: "Lithium-Ion Battery Calendar Health Prognostics Based on Knowledge-Data-Driven Attention", IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, vol. 70, no. 1, 31 January 2023 (2023-01-31), pages 8191 - 8202 * |
| TIANYU HU ETAL.: "Probabilistic Electric Vehicle Charging Demand Forecast Based on Deep Learning and Machine Theory of Mind", 2021 IEEE TRANSPORTATION ELECTRIFICATION CONFERENCE & EXPO (ITEC), vol. 2021, no. 1, 2 August 2021 (2021-08-02), pages 795 - 799 * |
| 陈治铭等: "一种对抗性的权重注意力机制Seq2Seq模型估算SOC", 电工技术学报, 20 November 2023 (2023-11-20), pages 1 - 13 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117630683B (en) | 2024-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Cadini et al. | State-of-life prognosis and diagnosis of lithium-ion batteries by data-driven particle filters | |
| Khan et al. | Batteries state of health estimation via efficient neural networks with multiple channel charging profiles | |
| Nguyen et al. | A new dynamic predictive maintenance framework using deep learning for failure prognostics | |
| Duong et al. | Heuristic Kalman optimized particle filter for remaining useful life prediction of lithium-ion battery | |
| Tao et al. | A novel support vector regression method for online reliability prediction under multi-state varying operating conditions | |
| CN102208028B (en) | Fault predicting and diagnosing method suitable for dynamic complex system | |
| Kara | A data-driven approach based on deep neural networks for lithium-ion battery prognostics | |
| CN110321940B (en) | Aircraft telemetry data feature extraction and classification method and device | |
| Zhang et al. | Cost-effective Lebesgue sampling long short-term memory networks for lithium-ion batteries diagnosis and prognosis | |
| CN107329088A (en) | The health status diagnostic device and method of battery | |
| Chen et al. | A health index construction framework for prognostics based on feature fusion and constrained optimization | |
| KR102215107B1 (en) | Vehicle state predicting system and method based on driving data | |
| Chehade et al. | A multioutput convolved Gaussian process for capacity forecasting of li-ion battery cells | |
| Li et al. | Accurate state of charge prediction for real-world battery systems using a novel dual-dropout-based neural network | |
| Liu et al. | Lithium-ion battery remaining useful life prediction with long short-term memory recurrent neural network | |
| Heinrich et al. | Unsupervised data-preprocessing for long short-term memory based battery model under electric vehicle operation | |
| Ta et al. | Adaptive staged remaining useful life prediction method based on multi-sensor and multi-feature fusion | |
| Li et al. | An optimal stacking ensemble for remaining useful life estimation of systems under multi-operating conditions | |
| Huang et al. | An adversarial learning approach for machine prognostic health management | |
| Cornelius et al. | Estimating and leveraging uncertainties in deep learning for remaining useful life prediction in mechanical systems | |
| CN114325397A (en) | Method and apparatus for training artificial intelligence based state models | |
| Bak et al. | Accurate estimation of battery SOH and RUL based on a progressive lstm with a time compensated entropy index | |
| CN117630683B (en) | Multi-scale fusion GRU network-based automobile battery SOC multi-step prediction method and system | |
| CN117310500A (en) | Battery state classification model construction method and battery state classification method | |
| Tirovolas et al. | Introducing fuzzy cognitive map for predicting engine’s health status |
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 |