CN114741963A - Lithium battery state-of-charge prediction method based on multi-scale attention mechanism - Google Patents
Lithium battery state-of-charge prediction method based on multi-scale attention mechanism Download PDFInfo
- Publication number
- CN114741963A CN114741963A CN202210398500.XA CN202210398500A CN114741963A CN 114741963 A CN114741963 A CN 114741963A CN 202210398500 A CN202210398500 A CN 202210398500A CN 114741963 A CN114741963 A CN 114741963A
- Authority
- CN
- China
- Prior art keywords
- scale
- model
- network
- node
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- WHXSMMKQMYFTQS-UHFFFAOYSA-N Lithium Chemical compound [Li] WHXSMMKQMYFTQS-UHFFFAOYSA-N 0.000 title claims abstract description 32
- 229910052744 lithium Inorganic materials 0.000 title claims abstract description 32
- 230000007246 mechanism Effects 0.000 title claims abstract description 32
- 238000003062 neural network model Methods 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims description 24
- 238000004364 calculation method Methods 0.000 claims description 14
- 238000013528 artificial neural network Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 7
- 238000007599 discharging Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 230000000644 propagated effect Effects 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 241000287196 Asthenes Species 0.000 claims description 3
- 108091026890 Coding region Proteins 0.000 claims description 3
- 238000001514 detection method Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000007774 longterm Effects 0.000 abstract description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- HBBGRARXTFLTSG-UHFFFAOYSA-N Lithium ion Chemical compound [Li+] HBBGRARXTFLTSG-UHFFFAOYSA-N 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 5
- 229910001416 lithium ion Inorganic materials 0.000 description 5
- 238000013459 approach Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 241000282414 Homo sapiens Species 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000004146 energy storage Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 208000013409 limited attention Diseases 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000006386 neutralization reaction Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- 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
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
-
- 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
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02E60/10—Energy storage using batteries
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Economics (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Geometry (AREA)
- Data Mining & Analysis (AREA)
- Computer Hardware Design (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Medical Informatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Secondary Cells (AREA)
Abstract
The invention provides a lithium battery state-of-charge prediction method based on a multi-scale attention mechanism, wherein a multi-scale hierarchical structure neural network model is constructed by the relation of each time node in the multi-scale attention mechanism, each non-leaf node is provided with C sub-nodes, each node only pays attention to a limited plurality of keys, multi-scale representation is established on an original time sequence by inter-scale connection, the node positioned at the minimum scale corresponds to the original point in the time sequence, and the node positioned at the larger scale corresponds to the characteristic expressed by the time sequence on the lower resolution, the invention has the beneficial effects that: a multi-scale attention mechanism is introduced into a lithium battery SOC prediction model, the model complexity is reduced while the long-term dependence capturing capacity is enhanced, and high-precision online lithium battery SOC prediction can be realized.
Description
[ technical field ] A method for producing a semiconductor device
The invention relates to the technical field of lithium battery SOC online prediction, in particular to a lithium battery state of charge prediction method based on a multi-scale attention mechanism.
[ background of the invention ]
The research and development of new energy technology become important measures of national strategy of carbon neutralization in China, and have important significance for realizing energy safety, environmental protection and industrial upgrading. With the rapid development of new Energy and electric vehicle industry in recent years, the role played by lithium ion batteries in Energy Storage Systems (ESS) is becoming more and more important. In order to ensure safe and reliable operation of the ESS, monitoring of the State of Charge (SOC) of the battery is essential. SOC is defined as the ratio of the available capacity remaining in the battery to the total capacity, but SOC cannot be directly sampled and observed, so accurate estimation of SOC remains a big problem for battery management systems.
Common SOC prediction methods can be roughly classified into four categories: a characterization parameter based approach, an ampere-hour integral approach, a model based approach, and a data-driven based approach. The method based on the characterization parameters needs to perform constant current discharge for a long time to determine the residual capacity, so that the method is only suitable for specific environments such as laboratories and the like. The ampere-hour integration method is classic and easy to use, but has the problems that the initial SOC accurate value of the battery is difficult to obtain, the requirement on the accuracy of a current sensor is high, the degradation of the static capacity of the battery can influence the calculation accuracy and the like, and has strong limitation. The method based on the model has higher dependency on the accuracy of the model, and has the problems of higher calculation cost and divergence of estimation results possibly caused by improper initial values. The method based on data driving is that a direct mapping relation model of data such as current, voltage and temperature of a lithium battery and the SOC of the battery is established and trained based on a large amount of historical offline data, has a particular advantage for solving the problem of strong nonlinearity, and is high in estimation accuracy.
In recent years, with the continuous development of deep learning technology, some deep learning models are gradually applied to the research of time series data. The deep learning model is a deep learning neural network model with a plurality of nonlinear mapping levels, and can abstract an input signal layer by layer, extract features and mine deeper potential rules. The lithium battery SOC prediction as a time series prediction problem can also be solved by utilizing a deep learning technology.
The Attention Mechanism (Attention Mechanism) of human is a means for human to rapidly screen out high-value information from a large amount of information by using limited Attention resources. The attention mechanism in deep learning uses the attention thinking mode of human beings for reference, is widely applied to various deep learning tasks of different types such as natural language processing, image classification, voice recognition and the like, and obtains remarkable results.
The lithium ion battery is a dynamic system with long-time dependency response, and the current SOC of the lithium ion battery is linked with the historical SOC, the voltage, the current and the temperature.
[ summary of the invention ]
The invention aims to solve the technical problems and provides a novel lithium battery state-of-charge prediction method based on a multi-scale attention mechanism, which has the advantages of enhancing long-term dependence capture capability, reducing model complexity, realizing high-precision online lithium battery SOC prediction and the like.
The invention is realized by the following technical scheme:
a multi-scale attention mechanism-based lithium battery state of charge prediction method is characterized in that a multi-scale hierarchical structure neural network model is built according to the relation of each time node in the multi-scale attention mechanism, each non-leaf node is provided with C sub-nodes, each node only focuses on a limited number of keys, multi-scale representation is built on an original time sequence through inter-scale connection, the node located at the minimum scale corresponds to the original point in the time sequence, and the node located at the larger scale corresponds to the feature of the time sequence expressed on the lower resolution.
Preferably, the multi-scale hierarchical structure is called a C-ary tree, and the multi-scale hierarchical structure model expression:
whereinDenoting the ith node on the S scale, S being 1, …, S in turn denoting the lowest to highest scale, each node may be associated with a three-scale node in a C-ary treePoint connections, which are respectively adjacent nodes A on the same scale, are marked asC child nodes C, noteThe parent node P in the C-ary tree is marked asThen a node is obtainedAttention of (c) can be reduced to:
therefore, the model can capture the dependency relationship in different long and short time ranges at the same time, the space and time complexity is reduced to O (L), the calculation time and the memory consumption can be effectively reduced, the model can process long time sequences, generally speaking, longer historical input can provide more information, and the prediction accuracy can be improved.
Preferably, the method further comprises a novel network structure, the novel network structure is received by the model at the same time when the observed quantity of the network input end and the parameter of the covariate related to the battery characteristic are received by the model, different network processing is not needed, and the training and optimization of the model are facilitated.
Preferably, the multi-scale hierarchical structure is implemented by a large-scale connection module in the network, the large-scale connection module aggregates the time sequences of the embedded codes on different scales to form a multi-level tree structure, and the scale level is from bottom to top for corresponding child nodesPerforming convolution operation, inputting a plurality of convolution layers with convolution kernel size and step length of C into the embedded coding sequence in sequence on the time dimension to obtain a sequence with length of L/Cs on the time scale sThe new sequences with different scales form a C-ary tree, and before the sequences are input into the stacked convolution layers and after the convolution is completed, the full-connection layers are used for reducing the number of parameters and the calculated amount, so that model overfitting is effectively avoided.
A lithium battery state of charge prediction method based on a multi-scale attention mechanism comprises the following steps:
s1: constructing a multi-scale hierarchical structure neural network model, and setting training hyper-parameters, wherein the hyper-parameters comprise maximum iteration times, learning rate and the like;
s2: initializing the network parameters of the multi-scale hierarchical structure model in the step S1;
s3: preprocessing the model number input data, inputting the preprocessed model number input data into the network model initialized in the step S2, and performing forward propagation and network forward calculation by using a neural network;
s4: updating parameters after the model loss obtained by the network forward calculation in the step S3 is propagated reversely through a neural network;
s5: judging whether the accuracy of the updated network model in the step S4 meets the requirement or whether the training reaches the maximum iteration number; if not, returning to the step S3 to continue training; if so, finishing training to obtain a required model;
s6: and (5) predicting the state of charge of the lithium battery by using the network model obtained in the S5.
Preferably, after the neural network in step S3 performs forward propagation, the error function between the output value of the neural network and the real value is the L2 loss function, that is, the function is the L2 loss functionWherein T represents the sequence length selected for predicting the SOC of the battery; y istAndand respectively representing the real value and the predicted value of the SOC of the battery at the time t.
Preferably, in each iteration number in the step S4, the network performs back propagation, and an Adam optimizer is used to update the network weight and the bias.
Preferably, the model input data in step S3 is obtained by collecting observed values of voltage, current and temperature during charging and discharging to form a time sequence, and the covariate selects a characteristic quantity directly related to the SOC of the battery, and at the same time, the real SOC of the current battery is obtained by a technical detection means at each sampling as a label, and a training sample set is constructed by using the observed values and the covariate as characteristics.
The invention has the beneficial effects that:
(1) the multi-scale attention mechanism is introduced into the lithium battery SOC prediction model, the model complexity is reduced while the long-term dependence capturing capacity is enhanced, and high-precision online lithium battery SOC prediction can be realized;
(2) according to the multi-scale attention mechanism, the lithium battery SOC prediction model is introduced, dependency relationships in different long and short time ranges can be captured at the same time, the space and time complexity is reduced to O (L), the calculation time and the memory consumption can be effectively reduced, the model can process long-time sequences, generally, longer historical input can provide more information, and the prediction accuracy can be improved.
[ description of the drawings ]
FIG. 1 is a schematic diagram of the relationship between time nodes in the multi-scale attention mechanism of the present invention;
FIG. 2 is a schematic diagram of the novel network architecture of the present invention;
fig. 3 is a schematic flow chart of a lithium battery state of charge prediction method based on a multi-scale attention mechanism according to the present invention.
FIG. 4 is a graph illustrating the relationship between time nodes in the attention mechanism of the present invention;
FIG. 5 is a graph showing 5 sets of data and change curves selected according to the present invention;
FIG. 6 shows the predicted values and measured values of the model of the present invention.
[ detailed description ] embodiments
The invention is further described with reference to the following figures and embodiments:
it is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be noted that, for convenience of description, only the portions related to the present invention are shown in the drawings. In addition, the embodiments and features of the embodiments in the present application may be combined with each other without conflict.
Embodiment 1, as shown in fig. 1-2, a method for predicting a state of charge of a lithium battery based on a multi-scale attention system includes constructing a multi-scale hierarchical neural network model based on a relationship between time nodes in the multi-scale attention system, where each non-leaf node has C child nodes, each node only concerns a limited number of keys, and inter-scale connection establishes a multi-scale representation for an original time sequence, a node located at a minimum scale corresponds to an original point in the time sequence, the original point is a voltage/current value per minute, and a node located at a larger scale corresponds to a feature expressed by the time sequence at a lower resolution, and the expressed feature is a trend exhibited by temperature every hour, every day, and every week.
As a further description of the above technical solution: the multi-scale hierarchical structure is called a C-ary tree, and the multi-scale hierarchical structure model expression comprises:
whereinThe first node on the scale of S is represented, S is 1, …, S sequentially represents the lowest scale to the highest scale, each node can be connected with the nodes of three scales in the C-ary tree, and the nodes are respectively adjacent nodes a on the same scale, and are marked asC child nodes C, noteThe parent node P in the C-ary tree is marked asThen a node is obtainedAttention of (c) can be reduced to:therefore, the model can capture the dependency relationship in different long and short time ranges at the same time, the space and time complexity is reduced to O (L), the calculation time and the memory consumption can be effectively reduced, the model can process long time sequences, generally speaking, longer historical input can provide more information, and the prediction accuracy can be improved.
As a further description of the above technical solution: the novel network structure receives the observed quantities of current, voltage, temperature and the like at the input end of the network and the parameters of the battery type, the battery model, the manufacturer and the like of the covariate related to the battery characteristic at the same time, does not need to use different networks for processing, and facilitates the training and optimization of the model.
As a further description of the above technical solution: the multi-scale hierarchical structure is realized by a large-scale connection module in the network, the large-scale connection module converges time sequences embedded with codes on different scales to form a multi-level tree structure, and the scale level is from bottom to top for corresponding child nodesAnd (3) performing convolution operation, sequentially inputting a plurality of convolution layers with convolution kernels of which the sizes and the step lengths are C into the embedded coding sequence on a time dimension to obtain a sequence with the length of L/Cs on the time dimension s, forming a C-ary tree by using new sequences with different dimensions, and reducing the number of parameters and calculated amount by using a full-connection layer before the sequence is input into the stacked convolution layers and after the convolution is completed, thereby effectively avoiding model overfitting.
As shown in fig. 3, a lithium battery state of charge prediction method based on a multi-scale attention mechanism includes the following steps:
s1: constructing a multi-scale hierarchical structure neural network model, and setting training hyper-parameters, wherein the hyper-parameters comprise maximum iteration times, learning rate and the like;
s2: initializing the network parameters of the multi-scale hierarchical structure model in the step S1;
s3: preprocessing the model setting training hyper-parameter input data, inputting the preprocessed model setting training hyper-parameter input data into the network model which is initialized in the step S2, and performing forward propagation and network forward calculation by using a neural network;
s4: updating parameters after the model loss obtained by the network forward calculation in the step S3 is propagated reversely through the neural network;
s5: judging whether the accuracy of the updated network model in the step S4 meets the requirement or whether the training reaches the maximum iteration number; if not, returning to the step S3 to continue training; if so, finishing training to obtain a required model;
s6: and (5) predicting the state of charge of the lithium battery by using the network model obtained in the S5.
As a further description of the above technical solution: after the neural network in the step S3 is propagated forward, the error function between the network output value and the true value is an L2 loss function, that is, the function isWherein T represents the sequence length selected for predicting the SOC of the battery; y istAndand respectively representing the real value and the predicted value of the battery SOC at the time t.
As a further description of the above technical solution: in each iteration number in the step S4, the network performs back propagation, and an Adam optimizer is used to update the network weights and the bias.
As a further description of the above technical solution: the model input data in step S3 is obtained by collecting observed values of voltage, current, and temperature during charging and discharging to form a time sequence, and the covariate selects a characteristic quantity directly related to the battery SOC, and at the same time, the real SOC of the current battery is obtained by a technical detection means at each sampling as a label, and a training sample set is constructed by using the observed values and the covariate as characteristics.
Example 2, the lithium battery SOC prediction problem can be described in a formalized language as: the observed value (voltage, current, temperature) z of the known L step lengtht-L+1:tAnd associated covariates (e.g. type, model of battery) xt-L+1:tAnd corresponding SOC value yt-L+1:tPredicting the SOC value y of the future M stepst+1:t+M。
The existing neural networks such as LSTM, Transformer and the like need to process coding and decoding of sequences, the complexity of a calculation space is high, and an original calculation method of a self-attention mechanism is insensitive to local information, so that a model is easily influenced by abnormal points. The formalization of the original attention mechanism is expressed as follows:
wherein Q is a Query vector, K is a Key vector, V is a Value vector, and dk is the dimension of the Query vector.
The relationship between time nodes under this attention mechanism is shown in fig. 4: it can be seen that each time node needs to be interconnected, and thus the spatial and temporal complexity is O (L)2) And L is the time series length.
In embodiment 3, the test is performed by using 10 groups of LG Chem 18650 lithium ion batteries, the rated capacity of the lithium ion battery is 2.1Ah, and the normal working voltage is 3.2-4.2V. To obtain battery discharge data for training and validation of the model, the battery was charged in a 4.2V/2.9A charge mode, indicating that the battery was fully charged if the battery voltage remained at 4.2V and the charge current dropped to 50 mA. After charging, standing the battery for 1h at room temperature, connecting a 0.5 omega resistor to discharge the battery at constant resistance, and recording the discharged electric quantity; if the voltage of the battery is reduced to 2.5V, the battery power is completely discharged; standing the battery for 1h again, performing the charge and discharge experiment again, and recording data including voltage, current, temperature, SOC and sampling time; and repeating charging and discharging for many times under the room temperature environment to obtain 950 charging and discharging curves. From these 5 sets of data were selected, the curves are shown in FIG. 5:
the embodiment realizes data enhancement (data augmentation) on original sample data by introducing Gaussian white noise. Specifically, according to an error range in the original sample data testing process, Gaussian white noise with the average value of 0 and the amplitude of 1% -2% of the original data amplitude is introduced, and according to inherent errors which cannot be eliminated by voltage, current and temperature sensors, plus or minus 0.1V, plus or minus 0.005A and plus or minus 2.5 ℃ offset quantities are additionally set for the voltage, the current and the temperature respectively, and finally the original sample data is expanded by 65 times.
And (4) preprocessing data. Through appropriate data preprocessing, the training data of the network can be made more efficient and robust. The observed values of current, voltage and temperature are normalized, and the formula is as follows:and processing covariates such as the battery type, the battery model and the like into data which can be directly processed by the neural network by using one-hot coding.
The historical data time window is set to 672, and the prediction time window is set to 96, namely the SOC value of 96 future time points is predicted by the historical data with the length of 672. And performing windowing processing on the data according to the setting to form a training set and a test set. However, the networks such as LSTM, Transformer, etc. cannot effectively extract the dependency relationship in such a long time window.
The hyper-parameter settings of the experimental model are shown in the following table:
at the network input end, parameters such as battery type, battery model and manufacturer are received by the model as covariates, so that the model can comprehensively consider the influence of battery characteristics, battery initial state and battery charging and discharging process historical data on the battery SOC, and better accuracy and robustness are obtained.
Error function is L2 lossFunction, i.e.Wherein T represents the sequence length selected for predicting the SOC of the battery; y istAndand respectively representing the real value and the predicted value of the SOC of the battery at the time t. In each iteration, the network will be back-propagated, updating the network weights and biases using an Adam optimizer. The initial learning rate is set to 10-4, the learning rate decay rate is set to 0.1 per epoch, and the number of epochs for training is set to 5.
The trained model verifies the actually measured data of the battery with the serial number of 8 in the test set, the predicted value and the actually measured value of the model are shown in figure 6, the root-mean-square error, the average absolute error and the maximum absolute error of the obtained prediction result are respectively 0.986%, 0.455% and 1.96%, and the deviation degree of the predicted value and the actual value is very small, so that the lithium battery SOC prediction model based on the multi-scale attention mechanism can be determined to have high prediction precision.
Appropriate changes and modifications to the embodiments described above will become apparent to those skilled in the art from the disclosure and teachings of the foregoing description. Therefore, the present invention is not limited to the specific embodiments disclosed and described above, and modifications and variations of the present invention are also intended to fall within the scope of the appended claims. Furthermore, although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.
Claims (8)
1. A lithium battery state of charge prediction method based on a multi-scale attention mechanism is characterized in that: the relation of each time node in the multi-scale attention mechanism constructs a multi-scale hierarchical structure neural network model, each non-leaf node has C sub-nodes, each node only pays attention to a limited plurality of keys, the inter-scale connection establishes multi-scale representation for an original time sequence, the node at the minimum scale corresponds to the original point in the time sequence, and the node at the larger scale corresponds to the characteristic of the time sequence expressed at the lower resolution.
2. The multi-scale attention mechanism-based lithium battery state of charge prediction method of claim 1, wherein: the multi-scale hierarchical structure is called a C-ary tree, and the multi-scale hierarchical structure model expression comprises:
whereinThe first node on the scale S is represented, S is 1, …, S sequentially represents the lowest to highest scale, each node can be connected with the nodes of three scales in the C-ary tree, and is respectively the adjacent node a on the same scale, and is marked asC child nodes C, noteThe parent node P in the C-ary tree is marked asThen a node is obtainedThe attention of (a) can be reduced to:
therefore, the model can capture the dependency relationship in different long and short time ranges at the same time, the space and time complexity is reduced to O (L), the calculation time and the memory consumption can be effectively reduced, the model can process long time sequences, and generally, the longer the historical input can be improvedThe prediction accuracy can be improved by providing more information.
3. The lithium battery state-of-charge prediction method based on the multi-scale attention mechanism as claimed in claim 1, wherein: the novel network structure is used for receiving the observed quantity of the network input end and the parameter of the covariate related to the battery characteristic by the model at the same time, does not need to use different network processing, and facilitates the training and optimization of the model.
4. The multi-scale attention mechanism-based lithium battery state of charge prediction method of claim 2, wherein: the multi-scale hierarchical structure is realized by a large-scale connection module in a network, the large-scale connection module converges time sequences embedded with codes on different scales to form a multi-level tree structure, and the scale level is from bottom to top to corresponding child nodesAnd (3) performing convolution operation, sequentially inputting a plurality of convolution layers with convolution kernels of which the sizes and the step lengths are C into the embedded coding sequence on a time dimension to obtain a sequence with the length of L/Cs on the time dimension s, forming a C-ary tree by using new sequences with different dimensions, and reducing the number of parameters and calculated amount by using a full-connection layer before the sequence is input into the stacked convolution layers and after the convolution is completed, thereby effectively avoiding model overfitting.
5. The multi-scale attention mechanism-based lithium battery state of charge prediction method of claim 1, characterized in that the method comprises the following steps:
s1: constructing a multi-scale hierarchical structure neural network model and setting a training hyper-parameter;
s2: initializing the network parameters of the multi-scale hierarchical structure model in the step S1;
s3: preprocessing the model number input data, inputting the preprocessed model number input data into the network model initialized in the step S2, and performing forward propagation and network forward calculation by using a neural network;
s4: updating parameters after the model loss obtained by the network forward calculation in the step S3 is propagated reversely through a neural network;
s5: judging whether the accuracy of the updated network model in the step S4 meets the requirement or whether the training reaches the maximum iteration number; if not, returning to the step S3 to continue training; if so, finishing training to obtain a required model;
s6: and (5) predicting the state of charge of the lithium battery by using the network model obtained in the S5.
6. The multi-scale attention mechanism-based lithium battery state of charge prediction method of claim 5, wherein: after the neural network in the step S3 is propagated forward, the error function between the network output value and the true value is an L2 loss function, that is, the function isWherein T represents the sequence length selected for predicting the SOC of the battery; y istAndand respectively representing the real value and the predicted value of the SOC of the battery at the time t.
7. The lithium battery state-of-charge prediction method based on the multi-scale attention mechanism as claimed in claim 5, wherein: in each iteration number in the step S4, the network performs back propagation, and an Adam optimizer is used to update the network weights and the bias.
8. The multi-scale attention mechanism-based lithium battery state of charge prediction method of claim 5, wherein: the model input data in step S3 is obtained by collecting observed values of voltage, current, and temperature during charging and discharging to form a time sequence, and the covariate selects a characteristic quantity directly related to the battery SOC, and at the same time, the real SOC of the current battery is obtained by a technical detection means at each sampling as a label, and a training sample set is constructed by using the observed values and the covariate as characteristics.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210398500.XA CN114741963A (en) | 2022-04-15 | 2022-04-15 | Lithium battery state-of-charge prediction method based on multi-scale attention mechanism |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210398500.XA CN114741963A (en) | 2022-04-15 | 2022-04-15 | Lithium battery state-of-charge prediction method based on multi-scale attention mechanism |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114741963A true CN114741963A (en) | 2022-07-12 |
Family
ID=82281325
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210398500.XA Pending CN114741963A (en) | 2022-04-15 | 2022-04-15 | Lithium battery state-of-charge prediction method based on multi-scale attention mechanism |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114741963A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117930028A (en) * | 2024-03-21 | 2024-04-26 | 成都赛力斯科技有限公司 | Method, system, equipment and medium for predicting thermal failure of new energy vehicle battery |
-
2022
- 2022-04-15 CN CN202210398500.XA patent/CN114741963A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117930028A (en) * | 2024-03-21 | 2024-04-26 | 成都赛力斯科技有限公司 | Method, system, equipment and medium for predicting thermal failure of new energy vehicle battery |
CN117930028B (en) * | 2024-03-21 | 2024-05-17 | 成都赛力斯科技有限公司 | Method, system, equipment and medium for predicting thermal failure of new energy vehicle battery |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Xiao et al. | Accurate state-of-charge estimation approach for lithium-ion batteries by gated recurrent unit with ensemble optimizer | |
Bian et al. | Stacked bidirectional long short-term memory networks for state-of-charge estimation of lithium-ion batteries | |
CN110888059B (en) | Charge state estimation algorithm based on improved random forest combined volume Kalman | |
Fan et al. | A novel machine learning method based approach for Li-ion battery prognostic and health management | |
CN113030764B (en) | Battery pack health state estimation method and system | |
CN112379269B (en) | Battery abnormality detection model training and detection method and device thereof | |
Qin et al. | State of health prediction for lithium-ion battery using a gradient boosting-based data-driven method | |
CN113917334B (en) | Battery health state estimation method based on evolution LSTM self-encoder | |
CN115201686B (en) | Lithium ion battery health state assessment method under incomplete charge and discharge data | |
CN112731183B (en) | Improved ELM-based lithium ion battery life prediction method | |
CN112834927A (en) | Lithium battery residual life prediction method, system, device and medium | |
CN114966436A (en) | Lithium battery state of charge prediction method, device, equipment and readable storage medium | |
CN114594378A (en) | Lithium battery SOH model training method, prediction method, device and equipment | |
CN115389946A (en) | Lithium battery health state estimation method based on isobaric rise energy and improved GRU | |
CN117031310A (en) | Method for predicting residual service life of power battery of electric automobile | |
CN115963407A (en) | ICGWO (intensive care unit) optimization ELM (element-based robust model) based lithium battery SOC estimation method | |
CN114741963A (en) | Lithium battery state-of-charge prediction method based on multi-scale attention mechanism | |
CN114578234A (en) | Lithium ion battery degradation and capacity prediction model considering causality characteristics | |
CN113203953B (en) | Lithium battery residual service life prediction method based on improved extreme learning machine | |
CN114460481A (en) | Energy storage battery thermal runaway early warning method based on Bi-LSTM and attention mechanism | |
CN116774086B (en) | Lithium battery health state estimation method based on multi-sensor data fusion | |
Song et al. | Capacity estimation method of lithium-ion batteries based on deep convolution neural network | |
Lyu et al. | State-of-charge estimation of lithium-ion batteries based on deep neural network | |
CN116840724A (en) | Lithium ion battery health condition estimation and residual service life prediction method | |
CN111337833A (en) | Lithium battery capacity integrated prediction method based on dynamic time-varying weight |
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 |