WO2022174601A1 - 电池状态预测模型的训练方法及相关装置 - Google Patents
电池状态预测模型的训练方法及相关装置 Download PDFInfo
- Publication number
- WO2022174601A1 WO2022174601A1 PCT/CN2021/124343 CN2021124343W WO2022174601A1 WO 2022174601 A1 WO2022174601 A1 WO 2022174601A1 CN 2021124343 W CN2021124343 W CN 2021124343W WO 2022174601 A1 WO2022174601 A1 WO 2022174601A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- battery
- information
- model
- prediction model
- training
- Prior art date
Links
- 238000012549 training Methods 0.000 title claims abstract description 188
- 238000000034 method Methods 0.000 title claims abstract description 129
- 239000013598 vector Substances 0.000 claims abstract description 149
- 238000000605 extraction Methods 0.000 claims abstract description 35
- 230000015654 memory Effects 0.000 claims description 65
- 238000004891 communication Methods 0.000 claims description 22
- 238000004590 computer program Methods 0.000 claims description 12
- 238000009413 insulation Methods 0.000 claims description 8
- 238000013145 classification model Methods 0.000 claims description 2
- 230000036541 health Effects 0.000 claims description 2
- 238000001453 impedance spectrum Methods 0.000 claims description 2
- 238000001514 detection method Methods 0.000 description 59
- 230000008569 process Effects 0.000 description 15
- 238000010586 diagram Methods 0.000 description 14
- 238000012545 processing Methods 0.000 description 14
- 230000006870 function Effects 0.000 description 13
- 238000013500 data storage Methods 0.000 description 10
- 239000011159 matrix material Substances 0.000 description 9
- 238000000157 electrochemical-induced impedance spectroscopy Methods 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 7
- 238000013473 artificial intelligence Methods 0.000 description 6
- 230000001360 synchronised effect Effects 0.000 description 5
- 238000010606 normalization Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 239000000872 buffer Substances 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- MHABMANUFPZXEB-UHFFFAOYSA-N O-demethyl-aloesaponarin I Natural products O=C1C2=CC=CC(O)=C2C(=O)C2=C1C=C(O)C(C(O)=O)=C2C MHABMANUFPZXEB-UHFFFAOYSA-N 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 239000000470 constituent Substances 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000037361 pathway Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- 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
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- 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
Definitions
- the present application relates to the field of artificial intelligence, and in particular, to a training method and related device for a battery state prediction model.
- Artificial intelligence is a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.
- artificial intelligence is a branch of computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that responds in a similar way to human intelligence.
- Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning and decision-making.
- Research in the field of artificial intelligence includes robotics, natural language processing, computer vision, decision-making and reasoning, human-computer interaction, recommendation and search, and basic AI theory.
- Electric vehicle (electric vehicle, EV) has become the main direction of contemporary automobile development due to its advantages of low pollution and high performance. At the same time, the battery state prediction of electric vehicles has also become a hot spot in the industry.
- the models used to predict various states of the battery are end-to-end prediction models, and they are all trained from a zero initial state. Quantity requirements are extremely high.
- the present application provides a training method and a related device for a battery state prediction model, so that the battery state of an EV can be predicted using real vehicle data, and the training cost of the battery state prediction model is reduced.
- the present application provides a training method for a battery state prediction model, the training method includes: acquiring first battery feature information of a target battery within a first period of time, and the battery feature information of the target battery includes one of the following or multiple kinds of information: current information of the target battery, voltage information of the target battery, temperature information of the target battery, and battery power indicator SOC of the target battery;
- the model performs self-supervised training, the first model includes a first pre-trained representation model and a first prediction model, the input of the first prediction model includes the output of the first pre-trained representation model, the first pre-trained
- the representation model is used to determine the representation vector of the input battery feature information, and the first prediction model is used to determine the target battery feature information corresponding to the input representation vector; obtain the second battery feature information and first label information of the target battery , the first label information is used to indicate the battery state information corresponding to the second battery feature information; the battery state prediction model is supervised and trained according to the second battery feature information and the first label information, and the The
- the parameters of the pre-trained representation model obtained by training can learn the correlation knowledge between battery feature information and battery state information, so labeled data is used.
- the battery state prediction model whose initialization parameters include the pre-trained representation model, the requirements for the labeled data of the battery state prediction model can be reduced.
- the battery state prediction model can also be trained using a small amount or stock of labeled data.
- each battery feature information in this method may be a battery raw data feature, for example, a battery raw data feature collected via a sensor.
- the battery characteristic information may include current information, voltage information, temperature information, SOC information, and the like.
- the structure of the second prediction model is the same as the structure of the first prediction model, and the second prediction model performs the initial
- the parameters include parameters of the first prediction model obtained after performing the self-supervised training on the first model.
- the structure of the second prediction model is the same as that of the first prediction model, and the initial parameters of the second prediction model include parameters of the first prediction model after self-supervised training is completed.
- the first prediction model After completing the self-supervised training, the first prediction model has learned the relationship between the representation vector of the battery feature information and the battery feature information, which is equivalent to pre-training the second prediction model before training it. Improves the training efficiency of the battery state prediction model.
- the first battery characteristic information includes N batteries of the target battery at N times within the first time period Feature information, N is a positive integer, the first pre-trained representation model includes a first encoder and a second encoder, the input of the first encoder includes battery feature information, and the input of the second encoder includes the output of the first encoder.
- the performing self-supervised training on the first model according to the first battery feature information includes: dividing the N battery feature information into n subsets, each of the n subsets contains at least one battery feature information, and any battery feature information in the i-th subset of the n subsets corresponds to a time earlier than the The time corresponding to any battery characteristic information in the i+1th subset of the n subsets in the first time, n is a positive integer and n is less than or equal to N, i is a positive integer and i is less than n;
- the battery feature information sequence obtained by arranging all the battery feature information in each subset of the n subsets according to the corresponding time sequence is input into the first encoder to obtain a representation vector corresponding to each subset;
- the battery characteristic information sequence obtained by arranging the n representation vectors corresponding to the subsets in the corresponding time sequence or in the order of the SOC values from low to high is input into the
- the N pieces of battery characteristic information in the first time period are divided into n subsets, and the battery characteristic information obtained by arranging all the battery characteristic information in each subset of the n subsets according to the corresponding chronological order
- the sequence is input to the first encoder to obtain the representation vector corresponding to each subset; the representation vector corresponding to each subset is input into the second encoder in chronological order or in the order of the SOC value from low to high to obtain the first battery
- the first representation vector of the feature information; the first representation vector is input into the first prediction model, and the target battery feature information corresponding to the first representation vector is obtained; the parameters of the first model are trained according to the target battery feature information and the first battery feature information to improve the The accuracy of the first model is improved and the training cost is reduced.
- the dividing the N pieces of battery characteristic information into n subsets includes: dividing the battery characteristic values into the N subsets according to the SOC value There are n subsets, and the SOC values corresponding to all battery characteristic information included in each of the n subsets are the same.
- the target battery is a battery pack
- the first battery characteristic information and/or the second battery The characteristic information also includes voltage information of each single cell in the battery pack.
- the first battery characteristic information includes M pieces of battery characteristic information of the target battery at M times within the first time period, where M is a positive integer
- the The first pretrained representation model includes a variational encoder
- the first prediction model includes a variational decoder
- the second prediction model includes a regression model or a classification model.
- the performing self-supervised training on the first model according to the first battery feature information includes: dividing the M battery feature information into m subsets, each of the m subsets contains at least one battery feature information, and any battery feature information in the jth subset of the m subsets corresponds to a time earlier than the The time corresponding to any battery feature information in the j+1th subset of the m subsets in the first time, m is a positive integer and m is less than or equal to M, j is a positive integer and j is less than m; according to the All the battery feature information in each of the m subsets determine the representation vector corresponding to each subset; calculate the average value of the m representation vectors corresponding to the m subsets; input the average value into the a pre-trained representation model to obtain a second representation vector of the first battery feature information; input the second representation vector into the first prediction model to obtain target battery feature information corresponding to the second representation
- the M pieces of battery feature information in the first time period are divided into m subsets, and a representation vector corresponding to each subset is determined according to all battery feature information in each of the m subsets; the m subsets are calculated.
- Corresponding target battery feature information; parameters of the first model are trained according to the target battery feature information and the first battery feature information, which improves the accuracy of the first model and reduces the training cost.
- the quantity of battery feature information in any one of the m subsets is equal to the number of battery feature information in any other subset of the m subsets.
- the target battery is a single cell
- the first battery characteristic information and/or the The characteristic information of the second battery includes the insulation resistance information of the single cell and/or the electrochemical AC impedance spectrum EIS of the single cell.
- the target battery characteristic information includes: an SOC sequence of the target battery in the first time period, the target battery The voltage sequence of the battery in the first period, the charging duration sequence of each single cell in the target battery in the first period, the charging mode of the target battery or the charging mode of the target battery in the The SOC value at the random mask in the first time, the charging mode includes a fast charging mode or a slow charging mode.
- the battery state information includes battery health index SOH information or fault information or remaining life information.
- the present application provides a battery state prediction method, the prediction method comprising: acquiring to-be-predicted battery feature information of a to-be-predicted battery; using a battery state prediction model to determine the to-be-predicted battery based on the to-be-predicted battery feature information Battery state information of the battery, where the battery state prediction model is a battery state prediction model obtained by training using the training method described in the first aspect or any one of the possible implementations.
- the feature information of the to-be-predicted battery of the to-be-predicted battery can be directly obtained from the data reported by the EV, so that the EV can predict the battery state according to the actual vehicle data.
- the present application provides an apparatus for training a battery state prediction model.
- the apparatus may include various modules for implementing the method in the first aspect, and these modules may be implemented in software and/or hardware.
- the present application provides an apparatus for predicting a battery state.
- the apparatus may include various modules for implementing the method in the second aspect, and these modules may be implemented in software and/or hardware.
- the present application provides a training device for a battery state prediction model.
- the apparatus may include a processor coupled to the memory.
- the memory is used for storing program codes
- the processor is used for executing the program codes in the memory, so as to implement the method in the first aspect or any one of the implementation manners.
- the apparatus may also include the memory.
- the present application provides a battery state prediction device.
- the apparatus may include a processor coupled to the memory.
- the memory is used for storing program codes
- the processor is used for executing the program codes in the memory, so as to implement the method in the second aspect.
- the apparatus may also include the memory.
- the present application provides a chip, comprising at least one processor and a communication interface, wherein the communication interface and the at least one processor are interconnected through a line, and the at least one processor is configured to run a computer program or instruction to execute The method according to the first aspect or the second aspect or any one of the possible implementations thereof.
- the present application provides a computer-readable medium storing program code for device execution, the program code including a computer-readable medium for executing the first aspect or the second aspect or any one of them. implement the method described.
- the present application provides a computer program product comprising instructions, which, when the computer program product is run on a computer, causes the computer to execute the method described in the first aspect or the second aspect or any one of the possible implementations thereof. method.
- the present application provides a computing device, comprising at least one processor and a communication interface, wherein the communication interface and the at least one processor are interconnected by a line, the communication interface communicates with a target system, and the at least one processing
- the computer is used to run a computer program or instructions to perform the method according to the first aspect or the second aspect or any one of the possible implementations thereof.
- the present application provides a computing system, comprising at least one processor and a communication interface, the communication interface and the at least one processor are interconnected by a line, the communication interface communicates with a target system, the at least one The processor is used to run a computer program or instructions to perform the method as described in the first aspect or the second aspect or any of the possible implementations thereof.
- the training method of the battery state prediction model provided by this application introduces the idea of embedding representation in deep learning and the idea of pre-training pipeline, and proposes to represent the state of the battery in any detection segment as an abstract battery representation vector. Make full use of the unlabeled data reported by the battery pack, and train the battery state prediction model through at least one pre-training task designed according to the electrochemical characteristics of the battery.
- the transfer learning method can be used to perform the end-to-end data-driven tasks of all subsequent batteries, or the pre-trained representation model in the battery state prediction model can be used to directly extract the representation vector of the battery.
- the subsequent battery state prediction task the dependence on label data that requires a lot of manual annotation is reduced, so that EV prediction of battery state only depends on the general data reported by EV, and there is no need to perform feature transformation on the general data reported by EV. , and also does not need to modify the EV hardware, reducing the cost.
- the training method of the battery state prediction model provided in this application can be adapted to EV battery packs of different manufacturers and different structures or single cells of different material systems through transfer learning.
- the pre-trained representation model in the battery state prediction model provided by this application can use the transformer encoder model.
- the samples of each time node can be fully utilized, and the time series of variable length can be processed and calculated small amount.
- FIG. 1 is a schematic diagram of a system architecture provided by an embodiment of the present application.
- FIG. 2 is a schematic structural diagram of a chip hardware provided by an embodiment of the present application.
- FIG. 3 is a schematic diagram of another system architecture provided by an embodiment of the present application.
- FIG. 4 is a schematic flowchart of a training method for a battery state prediction model according to an embodiment of the present application
- FIG. 5 is a schematic flowchart of a training method for an SOH prediction model of a battery pack according to an embodiment of the present application
- FIG. 6 is a schematic flowchart of a training method for a fault prediction model of a battery pack according to an embodiment of the present application
- FIG. 7 is a schematic flowchart of a training method for an SOH prediction model of a single cell according to an embodiment of the present application
- FIG. 8 is a schematic flowchart of a training method for a fault prediction model of a single cell according to an embodiment of the present application
- FIG. 9 is a schematic flowchart of a battery state prediction method according to an embodiment of the present application.
- FIG. 10 is a schematic structural diagram of a training device for a battery state prediction model according to an embodiment of the present application.
- FIG. 11 is a schematic structural diagram of an apparatus for predicting a battery state according to an embodiment of the present application.
- FIG. 12 is a schematic structural diagram of an apparatus provided by an embodiment of the present application.
- FIG. 1 is a schematic diagram of a system architecture provided by an embodiment of the present application.
- the data collection device 160 is used to collect battery characteristic information of the target battery and battery status label information corresponding to some battery characteristics and store them in the database 130 , wherein the battery characteristic information without corresponding battery status label information is called the first battery feature information, the battery feature information corresponding to the battery status label information is called the second battery feature information, and the battery status label information corresponding to the second battery feature information is called the first label information; the training device 120 is based on the first label information maintained in the database 130
- the battery state prediction model 101 is generated from the battery feature information, the second battery feature information, and the first label information, where the battery state model may also be referred to as a battery state prediction rule.
- the battery state prediction model 101 obtained by training the device 120 can be applied to different systems or devices, for example, to the execution device 110 .
- the execution device 110 is configured with an I/O interface 112 for data interaction with external devices, and the “user” can input the to-be-predicted battery characteristic information of the to-be-predicted battery to the I/O interface 112 through the client device 140 .
- the execution device 110 can call data, codes, etc. in the data storage system 150 , and can also store data, instructions, etc. in the data storage system 150 .
- the calculation module 111 uses the battery state prediction model 101 to process the to-be-predicted battery characteristic information of the to-be-predicted battery to obtain battery state information of the to-be-predicted battery.
- the I/O interface 112 returns the processing result to the client device 140 for provision to the user.
- the user can manually specify and input the battery characteristic information to be predicted in the execution device 110 , for example, operate in the interface provided by the I/O interface 112 .
- the client device 140 can automatically input the feature information of the battery to be predicted to the I/O interface 112 and obtain the battery status information. Corresponding permissions are set in the client device 140 .
- the user can view the battery status information output by the execution device 110 on the client device 140, and the specific presentation form can be a specific manner such as display, sound, and action.
- the client device 140 can also act as a data collection terminal to store the collected battery characteristic information and label information into the database 130 .
- FIG. 1 is only a schematic diagram of a system architecture provided by an embodiment of the present application, and the positional relationship among the devices, devices, modules, etc. shown in the figure does not constitute any limitation.
- the data storage The system 150 is an external memory relative to the execution device 110 , and in other cases, the data storage system 150 may be located in the execution device 110 .
- FIG. 2 is a schematic structural diagram of a chip hardware provided by an embodiment of the present application.
- a neural-network processing unit NPU
- Host CPU host central processing unit
- the core part of the NPU is the operation circuit 20, and the controller 204 controls the operation circuit 203 to extract the data in the memory (the weight memory 202 and/or the input memory 201) and perform operations.
- the arithmetic circuit 203 includes multiple processing units (process engines, PEs). In some implementations, the arithmetic circuit 203 is a two-dimensional systolic array. The arithmetic circuit 203 may also be a one-dimensional systolic array or other electronic circuit capable of performing mathematical operations such as multiplication or addition. In some implementations, the arithmetic circuit 203 is a general-purpose matrix processor.
- the operation circuit fetches the data corresponding to the matrix B from the weight memory 202 and buffers it on each PE of the operation circuit 203 .
- the arithmetic circuit fetches the data of the matrix A from the input memory 201 and performs the matrix operation on the matrix B, and stores the partial result or the final result of the matrix in the accumulator 208.
- the vector calculation unit 207 can further process the output of the arithmetic circuit, such as vector multiplication, vector addition, exponential operation, logarithmic operation, size comparison and so on.
- the vector calculation unit 207 can be used for network calculation of non-convolutional/non-FC layers in the neural network, such as pooling (Pooling), batch normalization (Batch Normalization), local response normalization (Local Response Normalization), etc. .
- vector computation unit 207 may store the processed output vectors to unified buffer 206 .
- the vector calculation unit 207 may apply a nonlinear function to the output of the arithmetic circuit 203, such as a vector of accumulated values, to generate activation values.
- the vector computation unit 207 may generate normalized values, merged values, or both.
- the vector of processed outputs can be used as an activation input to the arithmetic circuit 203, eg, for use in subsequent layers in a neural network.
- Unified memory 206 is used to store input data and output data.
- the storage unit access controller (direct memory access controller, DMAC) 205 transfers the input data in the external memory to the input memory 201 and/or the unified memory 206, stores the weight data in the external memory into the weight memory 202, and transfers the unified memory.
- the data in 206 is stored in external memory.
- a bus interface unit (BIU) 210 is used to realize the interaction between the main CPU, the DMAC and the instruction fetch memory 209 through the bus.
- An instruction fetch buffer 209 connected to the controller 204 is used to store the instructions used by the controller 204.
- the controller 204 is used for invoking the instructions cached in the instruction fetch memory 209 to realize and control the working process of the operation accelerator.
- the unified memory 206 , the input memory 201 , the weight memory 202 and the instruction fetch memory 209 are all on-chip memories. External memory is private to the NPU hardware architecture.
- the chip shown in FIG. 2 may implement the method shown in any of FIGS. 4 to 8 to obtain a battery state prediction model.
- the method may be performed by the main CPU and the NPU in cooperation.
- the chip shown in FIG. 2 can implement the method shown in FIG. 9 to obtain battery state information of the battery to be predicted.
- the method may be performed by the main CPU and the NPU in cooperation.
- FIG. 3 is a schematic diagram of another system architecture provided by an embodiment of the present application.
- the computing device 310 is implemented by one or more servers, and optionally, it can cooperate with other computing devices.
- the computing device 310 can cooperate with devices such as data storage, routers, and load balancers; the computing device 310 can be arranged On one physical site, or distributed across multiple physical sites.
- the computing device 310 may use data in the data storage system 350 or invoke program code in the data storage system 150 to implement the method shown in any of FIGS. 4 to 8 .
- computing device 310 may be training device 120 in FIG. 1 .
- the computing device 310 may use the data in the data storage system 350 or invoke the program code in the data storage system 150 to implement the method shown in FIG. 9 .
- a user may operate respective user devices (eg, local device 301 and local device 302 ) to interact with computing device 310 .
- Each local device may represent any computing device, such as a personal computer, computer workstation, smartphone, tablet, smart camera, smart car or other type of cellular phone, media consumption device, wearable device, set-top box, gaming console, etc.
- Each user's local device may interact with computing device 310 through any communication mechanism or communication standard communication network, which may be a wide area network, a local area network, a point-to-point connection, etc., or any combination thereof.
- any communication mechanism or communication standard communication network which may be a wide area network, a local area network, a point-to-point connection, etc., or any combination thereof.
- computing device 310 may be implemented by each local device, eg, local device 301 may provide computing device 310 with local data or feedback calculation results.
- computing device 310 may also be implemented by the local device.
- local device 301 implements the functions of computing device 310 and provides services to its own users, or provides services to users of local device 302 .
- FIG. 4 is a schematic flowchart of a training method of a battery state prediction model according to an embodiment of the present application. As shown in FIG. 4 , the method includes at least S401 to S404 .
- S401 Acquire first battery characteristic information of a target battery within a first period of time.
- the target battery can be a battery pack containing multiple single cells, or a single cell.
- the first time period may be any time period, and the duration of the first time period may be predetermined.
- the first battery characteristic information of the target battery in the first time period may include battery characteristic information of the target battery at each time point in at least one time point in the first time period.
- the first battery characteristic information may include current information of the battery pack, voltage information of the battery pack, temperature information of the battery pack, voltage information of each single cell in the battery pack, and The battery power indicator (state of charge, SOC) of the battery pack, etc.
- the first battery characteristic information may include voltage information of the single cell, temperature information of the single cell, current information of the single cell, and information of the single cell.
- the first battery characteristic information of the target battery may be directly obtained from data reported by an original battery management system (battery management system, BMS) of the EV to which the target battery belongs.
- BMS battery management system
- S402. Perform self-supervised training on a first model according to the first battery feature information, where the first model includes a first pre-trained representation model and a first prediction model, the input of the first prediction model includes the output of the first pre-trained representation model, and the first model includes a first pre-trained representation model and a first prediction model.
- a pre-trained representation model is used to determine the representation vector of the input battery characteristic information
- the first prediction model is used to determine the target battery characteristic information corresponding to the input representation vector.
- the first pre-trained representation model includes a first encoder and a second encoder
- the input of the first encoder includes the first battery feature information
- the input of the second encoder includes the output of the first encoder
- the first prediction model includes a regressor or a classifier, etc.
- the first pre-trained representation model includes a variational auto-encoder (VAE), and the first prediction model includes a variational decoder.
- VAE variational auto-encoder
- the first encoder and the second encoder in the first pre-trained representation model may be a multi-layer inheritance transformer structure.
- the above-mentioned self-supervised training may include at least one of the following training tasks: predicting the SOC sequence of the target battery in the first period of time, that is, the characteristic information of the target battery is the SOC sequence of the target battery in the first period of time.
- the target battery characteristic information is the voltage sequence of the target battery in the first period
- predict the charging time of each SOC segment of the target battery in the first period that is, the target battery characteristics
- the information is the charging time of each SOC segment of the target battery in the first period
- the charging mode of the target battery in the first period is predicted, that is, the target battery characteristic information is whether the charging mode of the target battery in the first period is the fast charging mode or the Slow charging mode
- predicting the SOC value of the target battery at the random mask within the first period that is, the target battery characteristic information is the SOC value of the target battery at the random mask within the first period.
- an implementation manner of performing self-supervised training on the first model according to the characteristic information of the first battery may include steps 1 to 5.
- the first battery characteristic information includes N pieces of battery characteristic information of the target battery at N time points within the first time period, where N is a positive integer.
- Step 1 Divide the N battery characteristic information in the first time period into n subsets, each of the n subsets contains at least one battery characteristic information, and any battery characteristic information in the ith subset of the n subsets is in The time corresponding to the first time is earlier than the time corresponding to any battery characteristic information in the i+1th subset of the n subsets in the first time, n is a positive integer and n is less than or equal to N, i is a positive integer and i is less than n.
- the N battery characteristic values may be divided into n subsets according to the SOC values of the N battery characteristic information in the first time period, and each of the divided n subsets contains all battery characteristics
- the SOC values corresponding to the information are the same.
- the N battery characteristic values may be divided into n subsets according to the voltage values of the N battery characteristic information in the first time period, and all the batteries included in each subset in the divided n subsets The voltage values corresponding to the characteristic information are the same.
- Step 2 Input the battery feature information sequence obtained by arranging all the battery feature information in each of the n subsets according to the corresponding time sequence into the first encoder to obtain a representation vector corresponding to each subset.
- Step 3 Input the battery characteristic information sequence obtained by arranging the n representation vectors corresponding to the n subsets in the corresponding chronological order or in the order from low to high SOC value into the second encoder to obtain the first battery characteristic information.
- the first represents a vector.
- Step 4 Input the first representation vector into the first prediction model, and obtain characteristic information of the target battery corresponding to the first representation vector.
- Step 5 Train parameters of the first model according to the target battery feature information and the first battery feature information.
- another implementation manner of performing self-supervised training on the first model according to the characteristic information of the first battery may include steps 6 to 11 .
- the first battery characteristic information may include M pieces of battery characteristic information of the target battery at M times within the first time period, where M is a positive integer.
- Step 6 Divide the M battery characteristic information in the first time period into m subsets, each of the m subsets contains at least one battery characteristic information, and any battery characteristic information in the jth subset in the m subsets is in The time corresponding to the first time is earlier than the time corresponding to any battery characteristic information in the j+1th subset in the m subsets in the first time, m is a positive integer and m is less than or equal to M, j is a positive integer and j is less than m.
- the quantity of battery characteristic information in each of the m subsets may be preset, and the quantity of battery characteristic information in any one of the m subsets is equal to the quantity of the battery characteristic information in any other subset of the m subsets
- the number of battery characteristic information in the first time period is divided according to the preset number of battery characteristic information of each subset, if the battery characteristic information in any one of the m subsets is When the number of battery feature information does not reach the preset number of battery feature information in each subset, the missing battery feature information is recorded as 0.
- Step 7 Determine a representation vector corresponding to each subset according to all battery feature information in each of the m subsets.
- Step 8 Calculate the average value of the m representation vectors corresponding to the m subsets.
- Step 9 Input the average value into the first pre-trained representation model to obtain the second representation vector of the first battery characteristic information.
- Step 10 Input the second representation vector into the first prediction model, and obtain characteristic information of the target battery corresponding to the second representation vector.
- Step 11 Train parameters of the first model according to the target battery feature information and the first battery feature information.
- S403 Acquire second battery feature information and first label information, where the first label information is used to indicate battery state information corresponding to the second battery feature information.
- the target battery may be a battery pack
- the second battery characteristic information may include current information of the battery pack, voltage information of the battery pack, temperature information of the battery pack, voltage information of each single cell in the battery pack, and The SOC value of the battery pack, etc.
- the target battery may be a single cell
- the second battery characteristic information may include voltage information of the single cell, temperature information of the single cell, current information of the single cell, and information of the single cell. SOC information, insulation resistance information of a single cell or EIS of a single cell.
- the second battery characteristic information and the first battery characteristic information may be the same battery characteristic information, or may be different battery characteristic information.
- the battery state information corresponding to the second battery feature information may include SOH information, fault information, or remaining life information, etc. of the target battery.
- the combination of the second battery feature information and the first label information may be called training data.
- supervised training is performed on the battery state prediction model according to the second battery feature information and the first label information
- the battery state prediction model includes a feature vector extraction model and a second prediction model
- the input of the second prediction model includes the output of the feature vector extraction model
- the initial parameters of the feature vector extraction model before the supervised training includes the parameters of the first pre-trained representation model obtained after the first model is self-supervised training.
- all or part of the parameters in the feature vector extraction model in the battery state prediction model may be initialized by using the model parameters of the first pre-trained representation model obtained after self-supervised training in the first model, and then using The second battery feature information and the first label information further train the battery state prediction model of the feature vector extraction model to obtain the battery state prediction model.
- the first pre-trained representation model obtained after the first model performs self-supervised training may be used as the feature vector extraction model in the battery state prediction model, and the second battery feature information is input to the feature vector extraction model.
- the model a representation vector of the second battery feature information is obtained, and the second prediction model is trained by using the second battery feature information, the representation vector of the second battery feature information and the first label information to obtain a battery state prediction model.
- the battery state prediction model when the battery state information includes SOH information, the battery state prediction model may be called an SOH prediction model; when the battery state information includes fault information, the battery state prediction model may be called a failure prediction model.
- both the first battery feature information for self-supervised training of the first model and the second battery feature information for supervised training of the second model can be directly obtained from the data reported by the EV.
- the quantity of the first label information used for the supervised training of the second model is greatly reduced compared with the quantity of label data used for training the battery state prediction model in the prior art, which reduces the training cost of the battery state prediction model.
- the real vehicle data is used to train the pre-trained representation model of the battery state prediction model, so when using the battery state prediction model to predict the battery state of the EV, only the real vehicle can be used.
- the data makes predictions about the battery state of the EV.
- the first pre-trained representation model includes a SOC encoder and a SOC sequence encoder
- the first prediction model is a classifier or a regressor
- the battery state prediction model is an SOH prediction model as an example, the battery of the present application is introduced.
- FIG. 5 is a schematic flowchart of a training method for an SOH prediction model of a battery pack according to an embodiment of the present application. As shown in FIG. 5 , the method includes at least S501 to S508.
- S501 Acquire all first battery feature information of the battery pack within a first detection duration.
- the first battery characteristic information of the battery pack within the first detection duration may include current information of the battery pack, voltage information of the battery pack, temperature information of the battery pack, and voltage information of each single cell in the battery pack. And the SOC information of the battery pack, etc.
- the first battery characteristic information of the battery pack can be obtained based on the data reported by the original BMS of the EV.
- S502 Divide the first detection duration into K time segments according to the SOC value of the battery pack, the SOC values of the first battery characteristic information in each time segment are the same, and K is a positive integer.
- the first detection duration is divided into K time segments according to the SOC value of the battery pack in the first detection duration, each time segment corresponds to an SOC value, and the SOC value of the first battery characteristic information in each time segment is the same, and each time segment
- Each time segment includes at least one piece of first battery characteristic information, and the time segment may also be referred to as an SOC frame.
- the time corresponding to any first battery characteristic information in the xth time slice in the K time slices is earlier than any first battery characteristic information in the x+1th time slice in the K time slices within the first detection duration.
- x is a positive integer and x is less than K.
- S503 Input the first battery feature information in each of the K time segments into the SOC encoder to obtain a representation vector corresponding to the time segment.
- Each of the K time segments corresponds to a representation vector, and the representation vector is used to represent the first battery characteristic information of the corresponding time segment.
- the K time segments of the first detection duration correspond to K representation vectors.
- the representation vector corresponding to each time segment in the K time segments is sequentially input into the SOC sequence encoder according to the time sequence, and the representation vector corresponding to the first detection duration is obtained.
- the SOC sequence encoder obtains a representation vector corresponding to the first detection duration according to the representation vectors of the K time segments in the first detection duration, and the representation vector corresponding to the first detection duration is used to represent the battery pack within the first detection duration. Battery characteristic information.
- the pre-training task uses the pre-training task to perform self-supervised training on the first pre-trained representation model and the first prediction model, where the first pre-trained representation model includes a SOC encoder and a SOC sequence encoder, and the first prediction model includes a regressor or a classifier device.
- a variety of pre-training tasks can be designed for self-supervised training.
- the designed pre-training tasks can include predicting the SOC sequence of the battery pack during the first detection duration, predicting the voltage sequence of the battery pack during the first detection duration, and predicting the battery pack. Pack the charging time of each SOC segment within the first detection duration, predict the charging mode of the battery pack within the first detection duration, predict the SOC value of the battery pack at the random mask within the first detection duration, and so on.
- different pre-training tasks may correspond to the same pre-training representation model and different first prediction models.
- a corresponding first prediction model can be designed for each pre-training task, and then the same pre-training task and the corresponding pre-training task can be designed for each pre-training task in turn.
- the first prediction model undergoes self-supervised training.
- a corresponding first prediction model can be designed for each pre-training task, that is, 5 first prediction models are finally designed; then the first prediction model of the 5 pre-training tasks is used.
- the training task trains the model composed of the pre-trained representation model and the first prediction model corresponding to the first pre-training task; then the second pre-training task is used to train the pre-trained representation model after training and the second prediction model.
- the model composed of the first prediction model corresponding to the pre-training task is subjected to self-supervised training; and so on, until the last pre-training task is used to compare the pre-trained representation model obtained by the fourth self-supervised training and the first training task corresponding to the last training task.
- the self-supervised training of the model composed of the prediction model is completed.
- S507 Acquire second battery characteristic information of the battery pack and SOH label information corresponding to the second battery characteristic information.
- the second battery characteristic information of the battery pack may include current information of the battery pack, voltage information of the battery pack, temperature information of the battery pack, voltage information of each single cell in the battery pack, and SOC information of the battery pack. Wait.
- the battery state prediction model includes a feature vector extraction model and a second prediction model.
- the parameters in the feature vector extraction model in the battery state prediction model are initialized using the model parameters of the first pre-trained representation model obtained after self-supervised training, and then the second battery feature information and The SOH label information corresponding to the second battery feature information further trains the feature vector extraction model and the second prediction model to obtain a battery state prediction model.
- the first pre-trained representation model obtained after self-supervised training is used as the feature vector extraction model in the battery state prediction model
- the second battery feature information is input into the feature vector extraction model to obtain
- the representation vector of the second battery feature information is used to train the second prediction model by using the second battery feature information, the representation vector of the second battery feature information and the SOH label information corresponding to the second battery feature information to obtain a battery state prediction model.
- the battery state prediction model obtained according to the training method shown in FIG. 5 can be used to predict the SOH information of the battery pack.
- all the first battery characteristic information of the battery pack within the first detection duration is divided into K time segments according to the SOC value of the battery pack, and each time segment in the K time segments is divided into K time segments.
- the battery feature information is input into the SOC encoder in the first pre-trained representation model, the representation vector corresponding to each time segment is obtained, and the representation vector corresponding to each time segment in the K time segments is sequentially input to the first time segment.
- a representation vector corresponding to the first detection duration is obtained, and the corresponding representation vector of the first detection duration is input into the first prediction model to obtain the battery pack within the first detection duration.
- Predict the output result use the pre-training task to perform self-supervised training on the first pre-training representation model and the first prediction model, and perform self-supervised training on the battery state prediction model according to the second battery feature information of the battery pack and the SOH label information corresponding to the second battery feature information.
- Supervised training is carried out, and the SOH prediction model of the battery pack is obtained.
- the first battery feature information and the second battery feature information used in the process of training the SOH prediction model of the battery pack can be directly obtained from the data reported by the EV, which reduces the training cost, and the EV can adjust the data according to the actual vehicle data.
- the SOH information of the battery pack is predicted.
- the first pre-training representation model includes a SOC encoder and a SOC sequence encoder
- the first prediction model is a classifier or a regressor
- the battery state prediction model is a fault prediction model as an example, the battery of the present application is introduced.
- FIG. 6 is a schematic flowchart of a training method for a fault prediction model of a battery pack according to an embodiment of the present application. As shown in FIG. 6 , the method includes at least S601 to S608.
- S601 Acquire all first battery characteristic information of the battery pack within a first detection period.
- S602 Divide the first detection duration into L time segments according to the SOC value of the battery pack, the SOC values of the first battery characteristic information in each time segment are the same, and L is a positive integer.
- S603 Input the first battery characteristic information in each of the L time segments into the SOC encoder to obtain a representation vector corresponding to the time segment.
- the representation vector corresponding to each time segment is sequentially input into the SOC sequence encoder according to the time sequence, to obtain the representation vector corresponding to the first detection duration.
- S605 Input the representation vector corresponding to the first detection duration into the regressor or the classifier to obtain a pre-training output result.
- the pre-training task uses the pre-training task to perform self-supervised training on the first pre-trained representation model and the first prediction model, where the first pre-trained representation model includes a SOC encoder and a SOC sequence encoder, and the first prediction model includes a regressor or a classifier.
- S607 Acquire second battery characteristic information of the battery pack and fault label information corresponding to the second battery characteristic information.
- the second battery characteristic information of the battery pack may include current information of the battery pack, voltage information of the battery pack, temperature information of the battery pack, voltage information of each single cell in the battery pack, and SOC information of the battery pack. Wait.
- the parameters in the feature vector extraction model in the battery state prediction model are initialized using the model parameters of the first pre-trained representation model obtained after self-supervised training, and then the second battery feature information and The fault label information corresponding to the second battery feature information further trains the feature vector extraction model and the second prediction model to obtain a battery state prediction model.
- the first pre-trained representation model obtained after self-supervised training is used as the feature vector extraction model in the battery state prediction model
- the second battery feature information is input into the feature vector extraction model to obtain
- the representation vector of the second battery feature information is used to train the second prediction model by using the second battery feature information, the representation vector of the second battery feature information and the fault label information corresponding to the second battery feature information to obtain a battery state prediction model.
- the battery state prediction model obtained according to the training method shown in Figure 6 can be used to predict the fault information of the battery pack.
- all the first battery characteristic information of the battery pack within the first detection duration is divided into L time segments according to the SOC value of the battery pack, and each time segment in the L time segments is divided into L time segments.
- the battery feature information is input into the SOC encoder in the first pre-trained representation model, the representation vector corresponding to each time segment is obtained, and the representation vector corresponding to each time segment in the L time segments is sequentially input to the first time segment.
- a representation vector corresponding to the first detection duration is obtained, and the corresponding representation vector of the first detection duration is input into the first prediction model to obtain the battery pack within the first detection duration.
- Predict the output result use the pre-training task to perform self-supervised training on the first pre-training representation model and the first prediction model, and perform self-supervised training on the battery state prediction model according to the second battery feature information of the battery pack and the fault label information corresponding to the second battery feature information.
- the fault prediction model of the battery pack is obtained.
- the first battery feature information and the second battery feature information used in the process of training the fault prediction model of the battery pack can be directly obtained from the data reported by the EV, reducing the training cost, and the EV can The failure information of the battery pack is predicted.
- the first pre-training representation model includes a variational encoder
- the first prediction model is a variational decoder
- the battery state prediction model is an SOH prediction model as an example to introduce the battery state prediction model of the present application.
- FIG. 7 is a schematic flowchart of a training method for an SOH prediction model of a single cell according to an embodiment of the present application. As shown in FIG. 7 , the method includes at least S701 to S708 .
- S701 Acquire all first battery feature information of a single battery cell within a first detection time period.
- the first battery characteristic information of the single cell within the first detection duration may include voltage information of the single cell, temperature information of the single cell, current information of the single cell, and information of the single cell. SOC information, insulation resistance information of a single cell, or EIS information of a single cell, etc.
- the first battery characteristic information of the battery cell may be obtained based on the data reported by the original BMS of the EV.
- S702 Divide the first detection duration into Q time window frames, all the first battery characteristic information in each time window frame form a representation vector corresponding to the time window frame, and Q is a positive integer.
- the quantity of the first battery characteristic information in each of the Q time window frames may be preset, and the first battery characteristic information in any one of the Q time window frames
- the quantity of information is equal to the quantity of the first battery characteristic information in any other time window frame in the Q time window frames, and the first battery characteristic information in the first detection duration is determined according to the first preset value of each sub-time window frame.
- the quantity of the battery characteristic information is divided, if the quantity of the first battery characteristic information in any one of the Q time window frames does not reach the preset quantity of the first battery characteristic information in each time window frame , the missing first battery feature information is marked as 0, and each time window frame includes at least one first battery feature information.
- the time corresponding to any first battery characteristic information in the qth time window frame in the Q time window frames is earlier than any first battery characteristic information in the q+1th time window frame in the Q time window frames within the first detection duration
- the time corresponding to the battery characteristic information within the first detection duration, and q is a positive integer smaller than Q.
- S703 Calculate the average value of the Q representation vectors corresponding to the Q time window frames.
- Each time window frame in the Q time window frames corresponds to a representation vector, and the representation vector is used to represent feature information of a single cell of the corresponding time window frame.
- S704 Input the average value of the calculated Q representation vectors into the variational encoder to obtain a representation vector corresponding to the first detection duration.
- the variational encoder obtains a representation vector corresponding to the first detection duration according to the average value of the Q representation vectors, and the representation vector corresponding to the first detection duration is used to characterize the single cell characteristics of the single cell within the first detection duration information.
- S705 Input the representation vector corresponding to the first detection duration into the variational decoder to obtain the predicted output result of the single cell within the first detection duration.
- the second battery feature information of the single cell within the first detection duration may include voltage information of the single cell, temperature information of the single cell, current information of the single cell, and information of the single cell. SOC information, insulation resistance information of a single cell, or EIS information of a single cell, etc.
- the battery state prediction model includes a feature vector extraction model and a second prediction model.
- the parameters in the feature vector extraction model in the battery state prediction model are initialized using the model parameters of the first pre-trained representation model obtained after self-supervised training, and then the second battery feature information and The SOH label information corresponding to the second battery feature information further trains the feature vector extraction model and the second prediction model to obtain a battery state prediction model.
- the first pre-trained representation model obtained after self-supervised training is used as the feature vector extraction model in the battery state prediction model
- the second battery feature information is input into the feature vector extraction model to obtain
- the representation vector of the second battery feature information is used to train the second prediction model by using the second battery feature information, the representation vector of the second battery feature information and the SOH label information corresponding to the second battery feature information to obtain a battery state prediction model.
- the battery state prediction model obtained according to the training method shown in FIG. 7 can be used to predict the SOH information of a single cell.
- all the first battery feature information of a single cell within the first detection duration is divided into Q time window frames, and all the first battery feature information in each time window frame constitutes the time The representation vector corresponding to the window frame, the average value of the Q representation vectors corresponding to the Q time window frames is calculated, and the average value is input into the first pre-trained representation model to obtain the representation vector corresponding to the first detection duration, and the first The representation vector corresponding to the detection duration is input into the first prediction model, the prediction output result of the single cell within the first detection duration is obtained, and the pre-training task is used to perform self-supervised training on the first pre-trained representation model and the first prediction model , supervise and train the battery state prediction model according to the second battery feature information of the single battery cell and the SOH label information corresponding to the second battery feature information, and obtain the SOH prediction model of the single battery cell.
- the first battery feature information and the second battery feature information used in the process of training the SOH prediction model of the single cell can be directly obtained from the data reported by the EV, which reduces the training cost, and the EV can The data predicts the SOH information of a single cell.
- the first pre-training representation model includes a variational encoder, the first prediction model is a variational decoder, and the battery state prediction model is a fault prediction model as an example, the battery state prediction model of the present application is introduced.
- FIG. 8 is a schematic flowchart of a training method for a fault prediction model of a single cell according to an embodiment of the present application. As shown in FIG. 8 , the method includes at least S801 to S808 .
- S801 Acquire all first battery feature information of a single battery cell within a first detection time period.
- S802 Divide the first detection duration into W time window frames, and all the first battery characteristic information in each time window frame forms a representation vector corresponding to the time window frame.
- S804 Input the average value of the W representation vectors obtained by calculation into the variational encoder to obtain a representation vector corresponding to the first detection duration.
- S805 Input the representation vector corresponding to the first detection duration into the variational decoder to obtain the predicted target battery feature information.
- S807 Acquire second battery feature information of a single cell and fault label information corresponding to the second battery feature information.
- the second battery feature information of the single cell within the first detection duration may include voltage information of the single cell, temperature information of the single cell, current information of the single cell, and information of the single cell. SOC information, insulation resistance information of a single cell, or EIS information of a single cell, etc.
- the parameters in the feature vector extraction model in the battery state prediction model are initialized using the model parameters of the first pre-trained representation model obtained after self-supervised training, and then the second battery feature information and The fault label information corresponding to the second battery feature information further trains the feature vector extraction model and the second prediction model to obtain a battery state prediction model.
- the first pre-trained representation model obtained after self-supervised training is used as the feature vector extraction model in the battery state prediction model
- the second battery feature information is input into the feature vector extraction model to obtain
- the representation vector of the second battery feature information is used to train the second prediction model by using the second battery feature information, the representation vector of the second battery feature information and the fault label information corresponding to the second battery feature information to obtain a battery state prediction model.
- the battery state prediction model obtained according to the training method shown in FIG. 8 can be used to predict the failure information of a single cell.
- all the first battery feature information of a single cell within the first detection duration is divided into W time window frames, and all the first battery feature information in each time window frame constitutes the time window
- the representation vector corresponding to the window frame, the average value of the W representation vectors corresponding to the W time window frames is calculated, and the average value is input into the first pre-trained representation model to obtain the representation vector corresponding to the first detection duration, and the first
- the representation vector corresponding to the detection duration is input into the first prediction model, the prediction output result of the single cell within the first detection duration is obtained, and the pre-training task is used to perform self-supervised training on the first pre-trained representation model and the first prediction model , supervise and train the battery state prediction model according to the second battery feature information of the single cell and the fault label information corresponding to the second battery feature information, and obtain the failure prediction model of the single cell.
- the first battery feature information and the second battery feature information used in the process of training the fault prediction model of the single cell can be directly obtained from the data reported by the
- FIG. 9 is a schematic flowchart of a battery state prediction method according to an embodiment of the present application. As shown in FIG. 9 , the method includes at least S901 to S902 .
- the battery to be predicted may be a battery pack containing a plurality of single cells, or may be a single cell.
- the feature information of the battery to be predicted may include current information of the battery pack, voltage information of the battery pack, temperature information of the battery pack, and voltage information of each single cell in the battery pack And the SOC information of the battery pack, etc.
- the feature information of the battery to be predicted may include voltage information of the single cell, temperature information of the single cell, current information of the single cell, The SOC information of the cell, the insulation resistance information of the single cell, or the EIS information of the single cell, etc.
- the to-be-predicted battery characteristic information of the to-be-predicted battery may be directly obtained from data reported by the original BMS of the EV to which the to-be-predicted battery belongs.
- the battery state information of the battery to be predicted can be obtained, and the state information of the battery to be predicted may include SOH information and/or fault information and/or remaining life information of the battery to be predicted.
- the battery state prediction model can be obtained by training according to the training methods shown in FIG. 4 to FIG. 8 .
- the battery state prediction model obtained by the training method shown in FIG. 5 is used for prediction, and the SOH information of the battery pack can be obtained;
- the prediction model makes predictions, and the fault information of the battery pack can be obtained.
- the battery to be predicted is a single cell
- the obtained battery state prediction model is used for prediction, and the fault information of the single cell can be obtained.
- the characteristic information of the to-be-predicted battery of the to-be-predicted battery is input into the battery state prediction model trained according to the training method described in any one of the embodiments in Figs. Battery status information.
- the to-be-predicted battery characteristic information of the to-be-predicted battery can be directly obtained from the data reported by the EV, so that the EV can predict the battery state according to the actual vehicle data.
- FIG. 10 is a schematic structural diagram of a training device for a battery state prediction model according to an embodiment of the present application.
- the apparatus 1000 may include an acquisition module 1001 and a training module 1002 .
- Any module in the acquisition module and the training module in the embodiments of the present application may be implemented in whole or in part by software and/or hardware.
- the part implemented by software may run on the processor to implement corresponding functions, and the part implemented by hardware may be a constituent part of the processor.
- the apparatus 1000 may be used to implement the method shown in FIG. 4 above.
- the acquisition module 1001 is used to implement S401 and S403
- the training module 1002 is used to implement S402 and S404.
- the apparatus 1000 may be used to implement the method shown in FIG. 5 above.
- the acquisition module 1001 is used to implement S501 and S507
- the training module 1002 is used to implement S506 and S508.
- the apparatus 1000 may be used to implement the method shown in FIG. 6 above.
- the acquisition module 1001 is used to implement S601 and S607
- the training module 1002 is used to implement S606 and S608.
- the apparatus 1000 may be used to implement the method shown in FIG. 7 above.
- the acquisition module 1001 is used to implement S701 and S707
- the training module 1002 is used to implement S706 and S708.
- the apparatus 1000 may be used to implement the method shown in FIG. 8 above.
- the acquisition module 1001 is used to implement S801 and S807
- the training module 1002 is used to implement S806 and S808.
- FIG. 11 is a schematic structural diagram of an apparatus for predicting a battery state according to an embodiment of the present application.
- the apparatus 1100 may include an acquisition module 1101 and a processing module 1102 .
- Any module in the acquisition module and the processing module in the embodiments of the present application may be implemented in whole or in part by software and/or hardware.
- the part implemented by software may run on the processor to implement corresponding functions, and the part implemented by hardware may be a constituent part of the processor.
- the apparatus 1100 may be used to implement the method shown in FIG. 9 above.
- the acquiring module 1101 is used to implement S901
- the processing module 1102 is used to implement S902.
- FIG. 12 is a schematic structural diagram of an apparatus provided by an embodiment of the present application.
- the apparatus 1200 shown in FIG. 12 may be used to perform the method described in any one of the foregoing embodiments.
- the apparatus 1200 in this embodiment includes: a memory 1201 , a processor 1202 , a communication interface 1203 , and a bus 1204 .
- the memory 1201 , the processor 1202 , and the communication interface 1203 are connected to each other through the bus 1204 for communication.
- the memory 1201 may be a read only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM).
- the memory 1201 may store programs, and when the programs stored in the memory 1201 are executed by the processor 1202, the processor 1202 may be used to perform various steps of the methods shown in FIGS. 4 to 9 .
- the processor 1202 can use a general-purpose central processing unit (CPU), a microprocessor, an application specific integrated circuit (ASIC), or one or more integrated circuits for executing related programs to The training method of the battery state prediction model and the battery state prediction method of the method embodiment of the present application are realized.
- CPU central processing unit
- ASIC application specific integrated circuit
- the processor 1202 may also be an integrated circuit chip with signal processing capability. In the implementation process, each step of the method in the various embodiments of the present application may be completed by an integrated logic circuit of hardware in the processor 1202 or an instruction in the form of software.
- the above-mentioned processor 1202 can also be a general-purpose processor, a digital signal processor (digital signal processing, DSP), an application specific integrated circuit (ASIC), an off-the-shelf programmable gate array (field programmable gate array, FPGA) or other programmable logic devices, Discrete gate or transistor logic devices, discrete hardware components.
- DSP digital signal processor
- ASIC application specific integrated circuit
- FPGA field programmable gate array
- a general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
- the steps of the methods disclosed in conjunction with the embodiments of the present application may be directly embodied as executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor.
- the software module may be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other storage media mature in the art.
- the storage medium is located in the memory 1201, and the processor 1202 reads the information in the memory 1201, and completes the functions required to be performed by each method in the embodiments of the present application in combination with its hardware. For example, each of the embodiments shown in FIG. 4 to FIG. 9 can be executed. steps/functions.
- the communication interface 1203 can use, but is not limited to, a transceiver such as a transceiver to implement communication between the device 1200 and other devices or a communication network.
- a transceiver such as a transceiver to implement communication between the device 1200 and other devices or a communication network.
- the bus 1204 may include a pathway for communicating information between the various components of the apparatus 1200 (eg, the memory 1201, the processor 1202, the communication interface 1203).
- the apparatus 1200 shown in this embodiment of the present application may be an electronic device, or may also be a chip configured in the electronic device.
- the processor in the embodiment of the present application may be a central processing unit (central processing unit, CPU), and the processor may also be other general-purpose processors, digital signal processors (digital signal processors, DSP), application-specific integrated circuits (application specific integrated circuit, ASIC), off-the-shelf programmable gate array (field programmable gate array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
- a general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
- the memory in the embodiments of the present application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.
- the non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically programmable Erase programmable read-only memory (electrically EPROM, EEPROM) or flash memory.
- Volatile memory may be random access memory (RAM), which acts as an external cache.
- RAM random access memory
- SRAM static random access memory
- DRAM dynamic random access memory
- DRAM synchronous dynamic random access memory
- SDRAM synchronous dynamic random access memory
- DDR SDRAM double data rate synchronous dynamic random access memory
- enhanced SDRAM enhanced synchronous dynamic random access memory
- SLDRAM synchronous connection dynamic random access memory Fetch memory
- direct memory bus random access memory direct rambus RAM, DR RAM
- the above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination.
- the above-described embodiments may be implemented in whole or in part in the form of a computer program product.
- the computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated.
- the computer may be a general purpose computer, special purpose computer, computer network, or other programmable device.
- the computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be downloaded from a website site, computer, server, or data center Transmission to another website site, computer, server or data center by wire (eg, infrared, wireless, microwave, etc.).
- the computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, a data center, or the like containing one or more sets of available media.
- the usable media may be magnetic media (eg, floppy disks, hard disks, magnetic tapes), optical media (eg, DVDs), or semiconductor media.
- the semiconductor medium may be a solid state drive.
- At least one means one or more, and “plurality” means two or more.
- At least one item(s) below” or similar expressions thereof refer to any combination of these items, including any combination of single item(s) or plural items(s).
- at least one item (a) of a, b, or c can represent: a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, c may be single or multiple .
- the size of the sequence numbers of the above-mentioned processes does not mean the sequence of execution, and the execution sequence of each process should be determined by its functions and internal logic, and should not be dealt with in the embodiments of the present application. implementation constitutes any limitation.
- the disclosed system, apparatus and method may be implemented in other manners.
- the apparatus embodiments described above are only illustrative.
- the division of the units is only a logical function division. In actual implementation, there may be other division methods.
- multiple units or components may be combined or Can be integrated into another system, or some features can be ignored, or not implemented.
- the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be in electrical, mechanical or other forms.
- the units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.
- each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
- the functions, if implemented in the form of software functional units and sold or used as independent products, may be stored in a computer-readable storage medium.
- the technical solution of the present application can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution.
- the computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.
- the aforementioned storage medium includes: a U disk, a removable hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk and other media that can store program codes.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Neurology (AREA)
- Charge And Discharge Circuits For Batteries Or The Like (AREA)
Abstract
一种电池状态预测模型的训练方法及相关装置,所述方法根据目标电池的第一电池特征信息对第一模型进行自监督训练,第一模型包括第一预训练表示模型和第一预测模型;根据目标电池的第二电池特征信息和第一标签信息对电池状态预测模型进行监督训练,电池状态预测模型包括特征向量提取模型和第二预测模型,特征向量提取模型进行监督训练之前的初始参数包括对第一模型进行自监督训练后得到的第一预训练表示模型的参数。能够使用实车数据预测实车的电池状态,并且降低了电池状态预测模型的训练成本。
Description
本申请要求于2021年02月22日提交中国国家知识产权局、申请号为202110197414.8、申请名称为“电池状态预测模型的训练方法及相关装置”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请涉及人工智能领域,尤其涉及电池状态预测模型的训练方法及相关装置。
人工智能(artificial intelligence,AI)是利用数字计算机或者数字计算机控制的机器模拟、延伸和扩展人的智能、感知环境、获取知识并使用知识获得最佳结果的理论、方法、技术及应用系统。换句话说,人工智能是计算机科学的一个分支,它企图了解智能的实质,并生产出一种新的能以人类智能相似的方式作出反应的智能机器。人工智能也就是研究各种智能机器的设计原理与实现方法,使机器具有感知、推理与决策的功能。人工智能领域的研究包括机器人、自然语言处理、计算机视觉、决策与推理、人机交互、推荐与搜索、AI基础理论等。
电动汽车(electric vehicle,EV)因具有低污染和高性能的优点,成为了当代汽车发展的主要方向。同时,电动汽车的电池状态预测也成为了行业的关注热点。
目前,用于预测电池的各种状态(例如SOC信息、SOH信息、故障信息或剩余寿命信息)的模型为端到端的预测模型,且都是从零初始状态开始训练,因此对标签数据的数据量要求极高。
发明内容
本申请提供了一种电池状态预测模型的训练方法及相关装置,使得能够使用实车数据预测EV的电池状态,并且降低了电池状态预测模型的训练成本。
第一方面,本申请提供一种电池状态预测模型的训练方法,所述训练方法包括:获取目标电池在第一时段内的第一电池特征信息,所述目标电池的电池特征信息包括以下一种或多种信息:所述目标电池的电流信息、所述目标电池的电压信息、所述目标电池的温度信息和所述目标电池的电池电量指标SOC;根据所述第一电池特征信息对第一模型进行自监督训练,所述第一模型包括第一预训练表示模型和第一预测模型,所述第一预测模型的输入包括所述第一预训练表示模型的输出,所述第一预训练表示模型用于确定输入的电池特征信息的表示向量,所述第一预测模型用于确定输入的表示向量对应的目标电池特征信息;获取所述目标电池的第二电池特征信息和第一标签信息,所述第一标签信息用于指示所述第二电池特征信息对应的电池状态信息;根据所述第二电池特征信息和所述第一标签信息对所述电池状态预测模型进行监督训练,所述电池状态预测模型包括特征向量提取模型和第二预测模型,所述第二预测模型的输入包括所述特征向量提取模型的输出,所述特征向量提取模型的初始参数包括对所 述第一模型进行所述自监督训练后得到的所述第一预训练表示模型的参数。
本方法中,因为使用无标签数据对第一模型进行自监督训练之后,训练得到的预训练表示模型的参数中能够学习到电池特征信息与电池状态信息之间的关联知识,所以使用带标签数据对初始化参数包含了该预训练表示模型训练后的电池状态预测模型训练时,可以降低该电池状态预测模型对带标签数据的要求。换句话说,使用少量或存量带标签数据也能训练得到电池状态预测模型。
可选地,本方法中的各个电池特征信息可以是电池原始数据特征,例如可以是经由传感器采集到的电池原始数据特征。例如,电池特征信息可以包括电流信息、电压信息、温度信息和SOC信息等。这样可以使得训练得到的电池状态模型能够直接根据电池原始数据进行状态预测,即可以基于电池实时数据实现电池状态预测。
结合第一方面,在第一种可能的实现方式中,所述第二预测模型的结构与所述第一预测模型的结构相同,并且,所述第二预测模型进行所述监督训练之前的初始参数包括对所述第一模型进行所述自监督训练后得到的所述第一预测模型的参数。
该实现方式中,第二预测模型的结构与第一预测模型的结构相同,并且第二预测模型的初始参数包括完成自监督训练后的第一预测模型的参数。完成自监督训练后的第一预测模型已经学习到了电池特征信息的表示向量与电池特征信息之间的关联关系,这等同于在对第二预测模型进行训练之前,已经对其进行了预训练,提高了电池状态预测模型的训练效率。
结合第一方面或第一种可能的实现方式,在第二种可能的实现方式中,所述第一电池特征信息包括所述目标电池在所述第一时段内的N个时间的N个电池特征信息,N为正整数,所述第一预训练表示模型中包括第一编码器和第二编码器,所述第一编码器的输入包括电池特征信息,所述第二编码器的输入包括所述第一编码器的输出。
结合第二种可能的实现方式,在第三种可能的实现方式中,所述根据所述第一电池特征信息对第一模型进行自监督训练,包括:将所述N个电池特征信息划分为n个子集合,所述n个子集合中每个子集合包含至少一个电池特征信息,所述n个子集合中第i个子集合中的任意电池特征信息在所述第一时间内对应的时间早于所述n个子集合中第i+1个子集合中的任意电池特征信息在所述第一时间内对应的时间,n为正整数且n小于或等于N,i为正整数且i小于n;将所述n个子集合中的每个子集合中的所有电池特征信息按照对应的时间先后顺序排列得到的电池特征信息序列输入所述第一编码器,得到所述每个子集合对应的表示向量;将所述n个子集合对应的n个表示向量按照对应的时间先后顺序或按照SOC值从低到高的顺序排列后得到的电池特征信息序列输入所述第二编码器,得到所述第一电池特征信息的第一表示向量;将所述第一表示向量输入所述第一预测模型,得到所述第一表示向量对应的目标电池特征信息;根据所述目标电池特征信息和所述第一电池特征信息调整所述第一模型的参数。
该实现方式中,将第一时间段内的N个电池特征信息划分为n个子集合,将n个子集合中的每个子集合中的所有电池特征信息按照对应的时间先后顺序排列得到的电池特征信息序列输入至第一编码器,得到每个子集合对应的表示向量;将每个子集合对应的表示向量按照时间先后顺序或按照SOC值从低到高的顺序依次输入第二编码器,得到第一电池特征信息的第一表示向量;将第一表示向量输入第一预测模型,得 到第一表示向量对应的目标电池特征信息;根据目标电池特征信息和第一电池特征信息训练第一模型的参数,提高了第一模型的准确度,降低了训练成本。
结合第三种可能的实现方式,在第四种可能的实现方式中,所述将所述N个电池特征信息划分为n个子集合,包括:按照SOC值将所述电池特征值划分为所述n个子集合,所述n个子集合中的每个子集合包含的所有电池特征信息对应的SOC值相同。
结合第二种或第三种或第四种可能的实现方式,在第五种可能的实现方式中,所述目标电池为电池包,所述第一电池特征信息和/或所述第二电池特征信息还包括所述电池包中每个单体电芯的电压信息。
结合第一方面,在第六种可能的实现方式中,所述第一电池特征信息包括所述目标电池在第一时段内的M个时间的M个电池特征信息,M为正整数,所述第一预训练表示模型包括变分编码器,所述第一预测模型包括变分解码器,所述第二预测模型包括回归模型或分类模型。
结合第六种可能的实现方式,在第七种可能的实现方式中,所述根据所述第一电池特征信息对第一模型进行自监督训练,包括:将所述M个电池特征信息划分为m个子集合,所述m个子集合中每个子集合包含至少一个电池特征信息,所述m个子集合中第j个子集合中的任意电池特征信息在所述第一时间内对应的时间早于所述m个子集合中第j+1个子集合中的任意电池特征信息在所述第一时间内对应的时间,m为正整数且m小于或等于M,j为正整数且j小于m;根据所述m个子集合中的每个子集合中的所有电池特征信息确定所述每个子集合对应的表示向量;计算所述m个子集合对应的m个表示向量的平均值;将所述平均值输入所述第一预训练表示模型,得到所述第一电池特征信息的第二表示向量;将所述第二表示向量输入所述第一预测模型,得到所述第二表示向量对应的目标电池特征信息;根据所述目标电池特征信息和所述第一电池特征信息调整所述第一模型的参数。
该实现方式中,将第一时间段内的M个电池特征信息划分为m个子集合,根据m个子集合中的每个子集合中的所有电池特征信息确定每个子集合对应的表示向量;计算m个子集合对应的m个表示向量的平均值;将平均值输入第一预训练表示模型,得到第一电池特征信息的第二表示向量;将第二表示向量输入第一预测模型,得到第二表示向量对应的目标电池特征信息;根据目标电池特征信息和第一电池特征信息训练第一模型的参数,提高了第一模型的准确度,降低了训练成本。
结合第七种可能的实现方式,在第八种可能的实现方式中,所述m个子集合中任意一个子集合中的电池特征信息的数量等于所述m个子集合中其它任意一个子集合中的电池特征信息的数量。
结合第六种或第七种或第八种可能的实现方式,在第九种可能的实现方式中,所述目标电池为单体电芯,所述第一电池特征信息和/或所述第二电池特征信息包括所述单体电芯的绝缘电阻信息和/或所述单体电芯的电化学交流阻抗谱EIS。
结合第一方面或上述任意一种可能的实现方式,在第十种可能的实现方式中,所述目标电池特征信息包括:所述目标电池在所述第一时段内的SOC序列,所述目标电池在所述第一时段内的电压序列、所述目标电池内每个单体电芯在所述第一时段内的充电时长序列、所述目标电池的充电模式或所述目标电池在所述第一时间内的随机掩 码处的SOC值,所述充电模式包括快充模型或慢充模式。
结合第一方面或上述任意一种可能的实现方式,在第十一种可能的实现方式中,所述电池状态信息包括电池健康度指标SOH信息或故障信息或剩余寿命信息。
第二方面,本申请提供一种电池状态的预测方法,所述预测方法包括:获取待预测电池的待预测电池特征信息;使用电池状态预测模型基于所述待预测电池特征信息确定所述待预测电池的电池状态信息,所述电池状态预测模型为使用第一方面或其中任意一种可能的实现方式所述的训练方法训练得到的电池状态预测模型。
本方法中,待预测电池的待预测电池特征信息可以直接从EV上报的数据中获取,使得EV可以根据实车数据预测电池状态。
第三方面,本申请提供一种电池状态预测模型的训练装置,所述装置可以包括用于实现第一方面中的方法的各个模块,这些模块可以通过软件和/或硬件的方式实现。
第四方面,本申请提供一种电池状态的预测装置,所述装置可以包括用于实现第二方面中的方法的各个模块,这些模块可以通过软件和/或硬件的方式实现。
第五方面,本申请提供一种电池状态预测模型的训练装置。该装置可以包括与存储器耦合的处理器。其中,该存储器用于存储程序代码,该处理器用于执行该存储器中的程序代码,以实现第一方面或其中任意一种实现方式中的方法。
可选地,该装置还可以包括该存储器。
第六方面,本申请提供一种电池状态的预测装置。该装置可以包括与存储器耦合的处理器。其中,该存储器用于存储程序代码,该处理器用于执行该存储器中的程序代码,以实现第二方面中的方法。
可选地,该装置还可以包括该存储器。
第七方面,本申请提供一种芯片,包括至少一个处理器和通信接口,所述通信接口和所述至少一个处理器通过线路互联,所述至少一个处理器用于运行计算机程序或指令,以执行如第一方面或第二方面或其中任意一种可能的实现方式所述的方法。
第八方面,本申请提供一种计算机可读介质,该计算机可读介质存储用于设备执行的程序代码,该程序代码包括用于执行如第一方面或第二方面或其中任意一种可能的实现方式所述的方法。
第九方面,本申请提供一种包含指令的计算机程序产品,当该计算机程序产品在计算机上运行时,使得计算机执行如第一方面或第二方面或其中任意一种可能的实现方式所述的方法。
第十方面,本申请提供一种计算设备,包括至少一个处理器和通信接口,所述通信接口和所述至少一个处理器通过线路互联,所述通信接口与目标系统通信,所述至少一个处理器用于运行计算机程序或指令,以执行如第一方面或第二方面或其中任意一种可能的实现方式所述的方法。
第十一方面,本申请提供一种计算系统,包括至少一个处理器和通信接口,所述通信接口和所述至少一个处理器通过线路互联,所述通信接口与目标系统通信,所述至少一个处理器用于运行计算机程序或指令,以执行如第一方面或第二方面或其中任意一种可能的实现方式所述的方法。
本申请提供的电池状态预测模型的训练方法,引入了深度学习中的嵌入式表示 (embedding representation)思想及预训练流水线思想,提出将电池在任意一个检测段的状态表示为抽象的电池表示向量,充分利用电池包上报的无标签数据,通过根据电池的电化学特性设计的至少一个预训练任务对电池状态预测模型进行训练。
针对训练完成后的电池状态预测模型,既可以利用迁移学习的方法执行后续所有电池的端到端的数据驱动任务,也可以使用电池状态预测模型中的预训练表示模型直接提取电池的表示向量,用于后续的电池状态预测任务,减少了对需要大量人工标注的标签数据的依赖,使得EV对电池状态的预测只依赖于EV上报的一般性数据,不需要对EV上报的一般性数据进行特征转化,也不需要对EV的硬件进行改造,降低了成本。
本申请提供的电池状态预测模型的训练方法可以通过迁移学习适配不同厂家不同结构的EV电池包或者不同材料体系的单体电芯。
本申请提供的电池状态预测模型中的预训练表示模型可以使用transformer编码器模型,通过设计两层继承编码器模型,充分利用每个时间节点的样本,可以处理可变长度的时间序列,并且计算量较小。
当EV上报的实车数据量足够大时,可以覆盖电压、温度和电流多种特性,可通过扩大电池状态预测模型的规模,充分挖掘输入数据间的非线性关系,达到覆盖复杂工况和复杂电池包内部结构的效果。
图1为本申请的实施例提供的一种系统架构的示意图;
图2为本申请的实施例提供的一种芯片硬件的结构示意图;
图3为本申请的实施例提供的另一种系统架构的示意图;
图4为本申请的实施例的一种电池状态预测模型的训练方法的流程示意图;
图5为本申请一个实施例的电池包的SOH预测模型的训练方法的流程示意图;
图6为本申请一个实施例的电池包的故障预测模型的训练方法的流程示意图;
图7为本申请一个实施例的单体电芯的SOH预测模型的训练方法的流程示意图;
图8为本申请一个实施例的单体电芯的故障预测模型的训练方法的流程示意图;
图9为本申请的实施例的一种电池状态预测方法的流程示意图;
图10为本申请一个实施例的电池状态预测模型的训练装置的示意性结构图;
图11为本申请一个实施例的电池状态的预测装置的示意性结构图;
图12为本申请一个实施例提供的装置的结构示意图。
下面将结合本申请的实施例中的附图,对本申请实施例中的技术方案进行描述,显然,所描述的实施例仅仅是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本申请保护的范围。
图1为本申请的实施例提供的一种系统架构的示意图。参见图1,数据采集设备160用于采集目标电池的电池特征信息和部分电池特征对应的电池状态标签信息并存 入数据库130,其中,没有对应电池状态标签信息的电池特征信息称为第一电池特征信息,有对应电池状态标签信息的电池特征信息称为第二电池特征信息,第二电池特征信息对应的电池状态标签信息称为第一标签信息;训练设备120基于数据库130中维护的第一电池特征信息、第二电池特征信息和第一标签信息生成电池状态预测模型101,其中,电池状态模型也可以称为电池状态预测规则。
训练设备120基于第一电池特征信息、第二电池特征信息和第一标签信息获取电池状态预测模型101的方法可以参见图4至图8中任意图所示的实施例。
训练设备120得到的电池状态预测模型101可以应用不同的系统或设备中,例如应用到执行设备110中。
执行设备110配置有I/O接口112,与外部设备进行数据交互,“用户”可以通过客户设备140向I/O接口112输入待预测电池的待预测电池特征信息。
执行设备110可以调用数据存储系统150中的数据、代码等,也可以将数据、指令等存入数据存储系统150中。
计算模块111使用电池状态预测模型101对待预测电池的待预测电池特征信息进行处理,从而得到待预测电池的电池状态信息。
最后,I/O接口112将处理结果返回给客户设备140,提供给用户。
在图1所示的情况下,用户可以手动指定输入执行设备110中的待预测电池特征信息,例如,在I/O接口112提供的界面中操作。另一种情况下,客户设备140可以自动地向I/O接口112输入待预测电池特征信息并获得电池状态信息,如果客户设备140自动输入待预测电池特征信息需要获得用户的授权,用户可以在客户设备140中设置相应权限。用户可以在客户设备140查看执行设备110输出的电池状态信息,具体的呈现形式可以是显示、声音、动作等具体方式。客户设备140也可以作为数据采集端将采集到电池特征信息和标签信息存入数据库130。
值得注意的,图1仅是本申请实施例提供的一种系统架构的示意图,图中所示设备、器件、模块等之间的位置关系不构成任何限制,例如,在图1中,数据存储系统150相对执行设备110是外部存储器,在其它情况下,也可以将数据存储系统150置于执行设备110中。
图2为本申请的实施例提供的一种芯片硬件的结构示意图。参见图2,神经网络处理器(neural-network processing unit,NPU)作为协处理器挂载到主中央处理器(host central processing unit,Host CPU)上,由Host CPU分配任务。NPU的核心部分为运算电路20,控制器204控制运算电路203提取存储器(权重存储器202和/或输入存储器201)中的数据并进行运算。
在一些实现方式中,运算电路203内部包括多个处理单元(process engine,PE)。在一些实现方式中,运算电路203是二维脉动阵列。运算电路203还可以是一维脉动阵列或者能够执行例如乘法或加法这样的数学运算的其它电子线路。在一些实现方式中,运算电路203是通用的矩阵处理器。
举例来说,假设有输入矩阵A,权重矩阵B,输出矩阵C。运算电路从权重存储器202中取矩阵B相应的数据,并缓存在运算电路203的中每一个PE上。运算电路从输入存储器201中取矩阵A数据与矩阵B进行矩阵运算,得到的矩阵的部分结果或 最终结果,保存在累加器(accumulator)208中。
向量计算单元207可以对运算电路的输出做进一步处理,如向量乘,向量加,指数运算,对数运算,大小比较等等。例如,向量计算单元207可以用于神经网络中非卷积/非FC层的网络计算,如池化(Pooling),批归一化(Batch Normalization),局部响应归一化(Local Response Normalization)等。
在一些实现中,向量计算单元207可以将经处理的输出的向量存储到统一缓存器206。例如,向量计算单元207可以将非线性函数应用到运算电路203的输出,例如累加值的向量,用以生成激活值。在一些实现方式中,向量计算单元207可以生成归一化的值、合并值,或二者均有。在一些实现方式中,处理过的输出的向量能够用作运算电路203的激活输入,例如用于在神经网络中的后续层中的使用。
统一存储器206用于存放输入数据以及输出数据。
存储单元访问控制器(direct memory access controller,DMAC)205将外部存储器中的输入数据搬运到输入存储器201和/或统一存储器206,将外部存储器中的权重数据存入权重存储器202,以及将统一存储器206中的数据存入外部存储器。
总线接口单元(bus interface unit,BIU)210,用于通过总线实现主CPU、DMAC和取指存储器209之间的交互。
与控制器204连接的取指存储器(instruction fetch buffer)209,用于存储控制器204使用的指令。
控制器204,用于调用取指存储器209中缓存的指令,实现控制该运算加速器的工作过程。统一存储器206、输入存储器201、权重存储器202以及取指存储器209均为片上(On-Chip)存储器。外部存储器私有于该NPU硬件架构。
在一些实现方式中,图2所示的芯片可以实现图4至图8中任意图所示的方法,以获得电池状态预测模型。作为一种示例,该方法可以是由主CPU和NPU共同配合完成的。
在另一些实现方式中,图2所示的芯片可以实现图9所示的方法,以获得待预测电池的电池状态信息。作为一种示例,该方法可以是由主CPU和NPU共同配合完成的。
图3为本申请的实施例提供的另一种系统架构的示意图。参见图3,计算设备310由一个或多个服务器实现,可选的,可以与其它计算设备配合,例如,计算设备310可以与数据存储、路由器和负载均衡器等设备配合;计算设备310可以布置在一个物理站点上,或者分布在多个物理站点上。
在图3所示的系统中,计算设备310可以使用数据存储系统350中的数据,或者调用数据存储系统150中的程序代码实现图4至图8中任意图所示的方法。作为一种示例,计算设备310可以是图1中的训练设备120。
或者,在图3所示的系统中,计算设备310可以使用数据存储系统350中的数据,或者调用数据存储系统150中的程序代码实现图9所示的方法。
用户可以操作各自的用户设备(例如本地设备301和本地设备302)与计算设备310进行交互。每个本地设备可以表示任何计算设备,例如个人计算机、计算机工作站、智能手机、平板电脑、智能摄像头、智能汽车或其他类型蜂窝电话、媒体消费设 备、可穿戴设备、机顶盒、游戏机等。
每个用户的本地设备可以通过任何通信机制或通信标准的通信网络与计算设备310进行交互,通信网络可以是广域网、局域网、点对点连接等方式或它们的任意组合。
在另一种实现中,计算设备310的一个方面或多个方面可以由每个本地设备实现,例如,本地设备301可以为计算设备310提供本地数据或反馈计算结果。
需要注意的,计算设备310的所有功能也可以由本地设备实现。例如,本地设备301实现计算设备310的功能并为自己的用户提供服务,或者为本地设备302的用户提供服务。
图4为本申请的实施例的一种电池状态预测模型的训练方法的流程示意图,如图4所示,该方法至少包括S401至S404。
S401,获取目标电池在第一时段内的第一电池特征信息。
目标电池可以是包含多个单体电芯的电池包,也可以是单体电芯。第一时段可以为任意时段,第一时段的时长可以是预先规定好的。
可以理解的是,目标电池在第一时段内的第一电池特征信息可以包括目标电池在第一时段内至少一个时间点中每个时间点的电池特征信息。
作为一种示例,目标电池为电池包时,第一电池特征信息可以包括电池包的电流信息、电池包的电压信息、电池包的温度信息、电池包中每个单体电芯的电压信息和电池包的电池电量指标(state of charge,SOC)等。
作为另一种示例,目标电池为单体电芯时,第一电池特征信息可以包括单体电芯的电压信息,单体电芯的温度信息,单体电芯的电流信息,单体电芯的SOC信息,单体电芯的绝缘电阻信息或单体电芯的电化学交流阻抗谱(electrochemical impedance spectroscopy,EIS)等。
在一些实现方式中,目标电池的第一电池特征信息可以从目标电池所属的EV的原始电池管理系统(battery management system,BMS)上报的数据中直接获取。
S402,根据第一电池特征信息对第一模型进行自监督训练,第一模型包括第一预训练表示模型和第一预测模型,第一预测模型的输入包括第一预训练表示模型的输出,第一预训练表示模型用于确定输入的电池特征信息的表示向量,第一预测模型用于确定输入的表示向量对应的目标电池特征信息。
在一些实现方式中,第一预训练表示模型中包括第一编码器和第二编码器,第一编码器的输入包括第一电池特征信息,第二编码器的输入包括第一编码器的输出,第一预测模型包括回归器或分类器等。
在另一些实现方式中,第一预训练表示模型中包括变分编码器(variational auto-encoder,VAE),第一预测模型包括变分解码器。
第一预训练表示模型包括两级编码器时,示例性的,第一预训练表示模型中的第一编码器和第二编码器可以为多层继承转换器(transformer)结构。
本实施例的一些实现方式中,上述自监督训练可以包括以下训练任务中至少一项任务:预测目标电池在第一时段内的SOC序列,即目标电池特征信息为目标电池在第一时段内的SOC序列;预测目标电池在第一时段的电压序列,即目标电池特征信息为 目标电池在第一时段的电压序列;预测目标电池在第一时段内每个SOC片段的充电时间,即目标电池特征信息为目标电池在第一时段内每个SOC片段的充电时间;预测目标电池在第一时段内的充电模式,即目标电池特征信息为目标电池在第一时段内的充电模式为快充模式还是慢充模式;或,预测目标电池在第一时段内的随机掩码处的SOC值,即目标电池特征信息为目标电池在第一时段内的随机掩码处的SOC值。
本实施例中,根据第一电池特征信息对第一模型进行自监督训练的一种实现方式可以包括步骤1至步骤5。这种实现方式中,第一电池特征信息包括目标电池在第一时段内的N个时间点处的N个电池特征信息,N为正整数。
步骤1,将第一时间段内的N个电池特征信息划分为n个子集合,n个子集合中每个子集合包含至少一个电池特征信息,n个子集合中第i个子集合中的任意电池特征信息在第一时间内对应的时间早于n个子集合中第i+1个子集合中的任意电池特征信息在第一时间内对应的时间,n为正整数且n小于或等于N,i为正整数且i小于n。
作为一种示例,可以根据第一时间段内的N个电池特征信息的SOC值,将N个电池特征值划分为n个子集合,划分后的n个子集合中的每个子集合包含的所有电池特征信息对应的SOC值相同。
作为另一种示例,可以根据第一时间段内的N个电池特征信息的电压值,将N个电池特征值划分为n个子集合,划分后的n个子集合中的每个子集合包含的所有电池特征信息对应的电压值相同。
步骤2,将n个子集合中的每个子集合中的所有电池特征信息按照对应的时间先后顺序排列得到的电池特征信息序列输入第一编码器,得到每个子集合对应的表示向量。
步骤3,将n个子集合对应的n个表示向量按照对应的时间先后顺序或按照SOC值从低到高的顺序排列后得到的电池特征信息序列输入第二编码器,得到第一电池特征信息的第一表示向量。
步骤4,将第一表示向量输入至第一预测模型,得到第一表示向量对应的目标电池特征信息。
步骤5,根据目标电池特征信息和第一电池特征信息训练第一模型的参数。
本实施例中,根据第一电池特征信息对第一模型进行自监督训练的另一种实现方式可以包括步骤6至步骤11。其中,第一电池特征信息可以包括目标电池在第一时段内的M个时间的M个电池特征信息,M为正整数。
步骤6,将第一时间段内的M个电池特征信息划分为m个子集合,m个子集合中每个子集合包含至少一个电池特征信息,m个子集合中第j个子集合中的任意电池特征信息在第一时间内对应的时间早于m个子集合中第j+1个子集合中的任意电池特征信息在第一时间内对应的时间,m为正整数且m小于或等于M,j为正整数且j小于m。
作为一种示例,可以预设m个子集合中每个子集合中的电池特征信息的数量,且m个子集合中任意一个子集合中的电池特征信息的数量等于m个子集合中其他任意一个子集合中的电池特征信息的数量,将第一时间段内的M个电池特征信息根据预设的每个子集合的电池特征信息的数量进行划分,若m个子集合中的任意一个子集合中的电池特征信息的数量未达到预设的每个子集合中的电池特征信息的数量时,将缺少的 电池特征信息记为0。
步骤7,根据m个子集合中的每个子集合中的所有电池特征信息确定每个子集合对应的表示向量。
步骤8,计算m个子集合对应的m个表示向量的平均值。
步骤9,将平均值输入第一预训练表示模型,得到第一电池特征信息的第二表示向量。
步骤10,将第二表示向量输入第一预测模型,得到第二表示向量对应的目标电池特征信息。
步骤11,根据目标电池特征信息和第一电池特征信息训练第一模型的参数。
S403,获取第二电池特征信息和第一标签信息,第一标签信息用于指示第二电池特征信息对应的电池状态信息。
作为一种示例,目标电池可以为电池包,第二电池特征信息可以包括电池包的电流信息、电池包的电压信息、电池包的温度信息、电池包中每个单体电芯的电压信息和电池包的SOC值等。
作为另一种示例,目标电池可以为单体电芯,第二电池特征信息可以包括单体电芯的电压信息,单体电芯的温度信息,单体电芯的电流信息,单体电芯的SOC信息,单体电芯的绝缘电阻信息或单体电芯的EIS。
本实施例中,第二电池特征信息与第一电池特征信息可以是同一个电池特征信息,也可以是不同的电池特征信息。
本实施例的一些实现方式中,第二电池特征信息对应的电池状态信息可以包括目标电池的SOH信息、故障信息或剩余寿命信息等。
本实施例中,第二电池特征信息和第一标签信息合起来可以称为训练数据。
S404,根据第二电池特征信息和第一标签信息对电池状态预测模型进行监督训练,电池状态预测模型包括特征向量提取模型和第二预测模型,第二预测模型的输入包括特征向量提取模型的输出,特征向量提取模型进行监督训练之前的初始参数包括对第一模型进行自监督训练后得到的第一预训练表示模型的参数。
在一些实现方式中,可以使用第一模型中进行自监督训练后得到的第一预训练表示模型的模型参数对电池状态预测模型中的特征向量提取模型中的全部或部分参数进行初始化,再使用第二电池特征信息和第一标签信息对特征向量提取模型电池状态预测模型进行进一步训练,得到电池状态预测模型。
在另一些可能的实现方式中,可以将第一模型进行自监督训练后得到的第一预训练表示模型作为电池状态预测模型中的特征向量提取模型,将第二电池特征信息输入至特征向量提取模型中,得到该第二电池特征信息的表示向量,使用第二电池特征信息、第二电池特征信息的表示向量和第一标签信息对第二预测模型进行训练,得到电池状态预测模型。
本实施例中,电池状态信息包括SOH信息时,电池状态预测模型可以称为SOH预测模型;电池状态信息包括故障信息时,电池状态预测模型可以称为故障预测模型。
本申请提出的技术方案中,对第一模型进行自监督训练的第一电池特征信息和对第二模型进行监督训练的第二电池特征信息均可以直接从EV上报的数据中直接获取, 对第二模型进行监督训练所使用的第一标签信息的数量与现有技术中对电池状态预测模型进行训练所使用的标签数据的数量相比大大减少,降低了对电池状态预测模型的训练成本。
因为在电池状态预测模型的训练过程中,仅使用实车数据对电池状态预测模型的预训练表示模型进行训练,因此在使用电池状态预测模型对EV的电池状态进行预测时,可以仅使用实车数据对EV的电池状态进行预测。
下面以目标电池为电池包,第一预训练表示模型包括SOC编码器和SOC序列编码器,第一预测模型为分类器或回归器,电池状态预测模型为SOH预测模型为例,介绍本申请电池状态预测模型的一种示例性训练方法。
图5为本申请一个实施例的电池包的SOH预测模型的训练方法的流程示意图。如图5所示,该方法至少包括S501至S508。
S501,获取电池包在第一检测时长内的所有第一电池特征信息。
可选的,电池包在第一检测时长内的第一电池特征信息可以包括电池包的电流信息、电池包的电压信息、电池包的温度信息、电池包中每个单体电芯的电压信息和电池包的SOC信息等。
电池包的第一电池特征信息可以基于EV的原始BMS上报的数据获取得到。
S502,根据电池包的SOC值将第一检测时长划分为K个时间片段,每个时间片段中的第一电池特征信息的SOC值相同,K为正整数。
根据电池包在第一检测时长内的SOC值将第一检测时长划分为K个时间片段,每个时间片段对应一个SOC值,每个时间片段中的第一电池特征信息的SOC值相同,每个时间片段中至少包括一个第一电池特征信息,时间片段也可以称为SOC帧。
K个时间片段中第x个时间片段中的任意第一电池特征信息在第一检测时长内对应的时间早于K个时间片段中第x+1个时间片段中的任意第一电池特征信息在第一检测时长内对应的时间,x为正整数且x小于K。
S503,将K个时间片段中的每个时间片段中的第一电池特征信息输入至SOC编码器中,得到该时间片段对应的表示向量。
K个时间片段中的每个时间片段对应一个表示向量,该表示向量用于表征对应的时间片段的第一电池特征信息。
第一检测时长的K个时间片段对应K个表示向量。
S504,将K个时间片段中的每个时间片段对应的表示向量按照时间顺序依次输入至SOC序列编码器中,得到第一检测时长对应的表示向量。
SOC序列编码器根据第一检测时长中的K个时间片段的表示向量,得到第一检测时长对应的表示向量,该第一检测时长对应的表示向量用于表征电池包在第一检测时长内的电池特征信息。
S505,将第一检测时长对应的表示向量输入至回归器或分类器中,得到电池包在第一检测时长内的预测输出结果。
S506,使用预训练任务对第一预训练表示模型和第一预测模型进行自监督训练,其中,第一预训练表示模型包括SOC编码器和SOC序列编码器,第一预测模型包括回归器或分类器。
可选的,可以设计多种预训练任务来进行自监督训练,设计的预训练任务可以包括预测电池包在第一检测时长的SOC序列、预测电池包在第一检测时长的电压序列、预测电池包在第一检测时长内每个SOC片段的充电时间,预测电池包在第一检测时长内的充电模式和预测电池包在第一检测时长内的随机掩码处的SOC值等。
可以理解的是,基于多种预训练任务进行自监督训练时,不同的预训练任务可以对应同一个预训练表示模型和不同的第一预测模型。或者说,设计了多个预训练任务的情况下,可以为每个预训练任务设计对应的第一预测模型,然后依次针对每个预训练任务对同一个预训练任务和该预训练任务对应的第一预测模型进行自监督训练。
例如,设计了5个预训练任务时,可以为每个预训练任务设计对应的第一预测模型,即最终设计5个第一预测模型;然后使用这5个预训练任务中的第一个预训练任务对由预训练表示模型和第一个预训练任务对应的第一预测模型构成的模型进行训练;接下来再使用第二个预训练任务对由训练后的预训练表示模型和第二个预训练任务对应的第一预测模型构成的模型进行自监督训练;依次类推,直到使用最后一个预训练任务对由第四次自监督训练得到的预训练表示模型和最后一个训练任务对应的第一预测模型构成的模型自监督训练完毕。
S507,获取电池包的第二电池特征信息和第二电池特征信息对应的SOH标签信息。
可选的,电池包的第二电池特征信息可以包括电池包的电流信息、电池包的电压信息、电池包的温度信息、电池包中每个单体电芯的电压信息和电池包的SOC信息等。
S508,根据电池包的第二电池特征信息和第二电池特征信息对应的SOH标签信息对电池状态预测模型进行监督训练,电池状态预测模型包括特征向量提取模型和第二预测模型。
在一些可能的实现方式中,使用进行自监督训练后得到的第一预训练表示模型的模型参数对电池状态预测模型中的特征向量提取模型中的参数进行初始化,再使用第二电池特征信息和第二电池特征信息对应的SOH标签信息对特征向量提取模型和第二预测模型进行进一步训练,得到电池状态预测模型。
在另一些可能的实现方式中,将进行自监督训练后得到的第一预训练表示模型作为电池状态预测模型中的特征向量提取模型,将第二电池特征信息输入至特征向量提取模型中,得到该第二电池特征信息的表示向量,使用第二电池特征信息、第二电池特征信息的表示向量和第二电池特征信息对应的SOH标签信息对第二预测模型进行训练,得到电池状态预测模型。
根据图5所示的训练方法得到的电池状态预测模型可以用来预测电池包的SOH信息。
本申请提出的技术方案中,将电池包在第一检测时长内的所有第一电池特征信息根据电池包的SOC值划分为K个时间片段,将K个时间片段中的每个时间片段中的电池特征信息输入至第一预训练表示模型中的SOC编码器中,得到每个时间片段对应的表示向量,将K个时间片段中的每个时间片段对应的表示向量按照时间顺序依次输入至第一预训练表示模型中的SOC序列编码器中,得到第一检测时长对应的表示向量,将第一检测时长的对应的表示向量输入至第一预测模型,得到电池包在第一检测时长内的预测输出结果,使用预训练任务对第一预训练表示模型和第一预测模型进行自监 督训练,根据电池包的第二电池特征信息和第二电池特征信息对应的SOH标签信息对电池状态预测模型进行监督训练,得到了电池包的SOH预测模型。在对电池包的SOH预测模型进行训练的过程中所使用的第一电池特征信息和第二电池特征信息可以从EV上报的数据中直接获取,降低了训练成本,并且EV可以根据实车数据对电池包的SOH信息进行预测。
下面以目标电池为电池包,第一预训练表示模型包括SOC编码器和SOC序列编码器、第一预测模型为分类器或回归器,电池状态预测模型为故障预测模型为例,介绍本申请电池状态预测模型的一种示例性训练方法。
图6为本申请一个实施例的电池包的故障预测模型的训练方法的流程示意图。如图6所示,该方法至少包括S601至S608。
S601,获取电池包在第一检测时长内的所有第一电池特征信息。
S602,根据电池包的SOC值将第一检测时长划分为L个时间片段,每个时间片段中的第一电池特征信息的SOC值相同,L为正整数。
S603,将L个时间片段中的每个时间片段中的第一电池特征信息输入至SOC编码器中,得到该时间片段对应的表示向量。
S604,将每个时间片段对应的表示向量按照时间顺序依次输入至SOC序列编码器中,得到第一检测时长对应的表示向量。
S605,将第一检测时长对应的表示向量输入至回归器或分类器中,得到预训练输出结果。
S606,使用预训练任务对第一预训练表示模型和第一预测模型进行自监督训练,第一预训练表示模型包括SOC编码器和SOC序列编码器,第一预测模型包括回归器或分类器。
需要说明的是,S601至S606可以参照S501至S506,此处不再进行赘述。
S607,获取电池包的第二电池特征信息和第二电池特征信息对应的故障标签信息。
可选的,电池包的第二电池特征信息可以包括电池包的电流信息、电池包的电压信息、电池包的温度信息、电池包中每个单体电芯的电压信息和电池包的SOC信息等。
S608,根据电池包的第二电池特征信息和第二电池特征信息对应的故障标签信息对电池状态预测模型进行监督训练,电池状态预测模型包括特征向量提取模型和第二预测模型。
在一些可能的实现方式中,使用进行自监督训练后得到的第一预训练表示模型的模型参数对电池状态预测模型中的特征向量提取模型中的参数进行初始化,再使用第二电池特征信息和第二电池特征信息对应的故障标签信息对特征向量提取模型和第二预测模型进行进一步训练,得到电池状态预测模型。
在另一些可能的实现方式中,将进行自监督训练后得到的第一预训练表示模型作为电池状态预测模型中的特征向量提取模型,将第二电池特征信息输入至特征向量提取模型中,得到该第二电池特征信息的表示向量,使用第二电池特征信息、第二电池特征信息的表示向量和第二电池特征信息对应的故障标签信息对第二预测模型进行训练,得到电池状态预测模型。
根据图6所示的训练方法得到的电池状态预测模型可以用来预测电池包的故障信 息。
本申请提出的技术方案中,将电池包在第一检测时长内的所有第一电池特征信息根据电池包的SOC值划分为L个时间片段,将L个时间片段中的每个时间片段中的电池特征信息输入至第一预训练表示模型中的SOC编码器中,得到每个时间片段对应的表示向量,将L个时间片段中的每个时间片段对应的表示向量按照时间顺序依次输入至第一预训练表示模型中的SOC序列编码器中,得到第一检测时长对应的表示向量,将第一检测时长的对应的表示向量输入至第一预测模型,得到电池包在第一检测时长内的预测输出结果,使用预训练任务对第一预训练表示模型和第一预测模型进行自监督训练,根据电池包的第二电池特征信息和第二电池特征信息对应的故障标签信息对电池状态预测模型进行监督训练,得到了电池包的故障预测模型。在对电池包的故障预测模型进行训练的过程中所使用的第一电池特征信息和第二电池特征信息可以从EV上报的数据中直接获取,降低了训练成本,并且EV可以根据实车数据对电池包的故障信息进行预测。
下面以目标电池为单体电芯,第一预训练表示模型包括变分编码器、第一预测模型为变分解码器,电池状态预测模型为SOH预测模型为例,介绍本申请电池状态预测模型的一种示例性训练方法。
图7为本申请一个实施例的单体电芯的SOH预测模型的训练方法的流程示意图,如图7所示,该方法至少包括S701至S708。
S701,获取单体电芯在第一检测时长内的所有第一电池特征信息。
可选的,单体电芯在第一检测时长内的第一电池特征信息可以包括单体电芯的电压信息,单体电芯的温度信息,单体电芯的电流信息,单体电芯的SOC信息,单体电芯的绝缘电阻信息或单体电芯的EIS信息等。
电体电芯的第一电池特征信息可以基于EV的原始BMS上报的数据获取。
S702,将第一检测时长划分为Q个时间窗口帧,每个时间窗口帧中的所有第一电池特征信息组成该时间窗口帧对应的表示向量,Q为正整数。
在一些可能的实现方式中,可以预设Q个时间窗口帧中每个时间窗口帧中的第一电池特征信息的数量,且Q个时间窗口帧中任意一个时间窗口帧中的第一电池特征信息的数量等于Q个时间窗口帧中其他任意一个时间窗口帧中的第一电池特征信息的数量,将第一检测时长内的第一电池特征信息根据预设的每个子时间窗口帧的第一电池特征信息的数量进行划分,若Q个时间窗口帧中的任意一个时间窗口帧中的第一电池特征信息的数量未达到预设的每个时间窗口帧中的第一电池特征信息的数量时,将缺少的第一电池特征信息记为0,且每个时间窗口帧中包含至少一个第一电池特征信息。
Q个时间窗口帧中第q个时间窗口帧中的任意第一电池特征信息在第一检测时长内对应的时间早于Q个时间窗口帧中第q+1个时间窗口帧中的任意第一电池特征信息在第一检测时长内对应的时间,q为小于Q的正整数。
S703,计算Q个时间窗口帧对应的Q个表示向量的平均值。
Q个时间窗口帧中的每个时间窗口帧对应一个表示向量,该表示向量用于表征对应的时间窗口帧的单体电芯特征信息。
S704,将计算得到的Q个表示向量的平均值输入至变分编码器中,得到第一检测 时长对应的表示向量。
变分编码器根据Q个表示向量的平均值得到第一检测时长对应的表示向量,该第一检测时长对应的表示向量用于表征单体电芯在第一检测时长内的单体电芯特征信息。
S705,将第一检测时长对应的表示向量输入至变分解码器中,得到单体电芯在第一检测时长内的预测输出结果。
S706,使用预训练任务对第一预训练表示模型和第一预测模型进行自监督训练,第一预训练表示模型包括变分编码器,第一预测模型包括变分解码器。
该步骤可以参考S506,此处不再赘述。
S707,获取单体电芯的第二电池特征信息和第二电池特征信息对应的SOH标签信息。
可选的,单体电芯在第一检测时长内的第二电池特征信息可以包括单体电芯的电压信息,单体电芯的温度信息,单体电芯的电流信息,单体电芯的SOC信息,单体电芯的绝缘电阻信息或单体电芯的EIS信息等。
S708,根据单体电芯的第二电池特征信息和第二电池特征信息对应的SOH标签信息对电池状态预测模型进行监督训练,电池状态预测模型包括特征向量提取模型和第二预测模型。
在一些可能的实现方式中,使用进行自监督训练后得到的第一预训练表示模型的模型参数对电池状态预测模型中的特征向量提取模型中的参数进行初始化,再使用第二电池特征信息和第二电池特征信息对应的SOH标签信息对特征向量提取模型和第二预测模型进行进一步训练,得到电池状态预测模型。
在另一些可能的实现方式中,将进行自监督训练后得到的第一预训练表示模型作为电池状态预测模型中的特征向量提取模型,将第二电池特征信息输入至特征向量提取模型中,得到该第二电池特征信息的表示向量,使用第二电池特征信息、第二电池特征信息的表示向量和第二电池特征信息对应的SOH标签信息对第二预测模型进行训练,得到电池状态预测模型。
根据图7所示的训练方法得到的电池状态预测模型可以用来预测单体电芯的SOH信息。
本申请提出的技术方案中,将单体电芯在第一检测时长内的所有第一电池特征信息划分为Q个时间窗口帧,每个时间窗口帧中的所有第一电池特征信息组成该时间窗口帧对应的表示向量,计算Q个时间窗口帧对应的Q个表示向量的平均值,将该平均值输入至第一预训练表示模型中,得到第一检测时长对应的表示向量,将第一检测时长对应的表示向量输入至第一预测模型中,得到单体电芯在第一检测时长内的预测输出结果,使用预训练任务对第一预训练表示模型和第一预测模型进行自监督训练,根据单体电芯的第二电池特征信息和第二电池特征信息对应的SOH标签信息对电池状态预测模型进行监督训练,得到单体电芯的SOH预测模型。在对单体电芯的SOH预测模型进行训练的过程中所使用的第一电池特征信息和第二电池特征信息可以从EV上报的数据中直接获取,降低了训练成本,并且EV可以根据实车数据对单体电芯的SOH信息进行预测。
下面以目标电池为单体电芯,第一预训练表示模型包括变分编码器,第一预测模 型为变分解码器,电池状态预测模型为故障预测模型为例,介绍本申请电池状态预测模型的一种示例性训练方法。
图8为本申请一个实施例的单体电芯的故障预测模型的训练方法的流程示意图,如图8所示,该方法至少包括S801至S808。
S801,获取单体电芯在第一检测时长内的所有第一电池特征信息。
S802,将第一检测时长划分为W个时间窗口帧,每个时间窗口帧中的所有第一电池特征信息组成该时间窗口帧对应的表示向量。
S803,计算W个时间窗口帧对应的W个表示向量的平均值。
S804,将计算得到的W个表示向量的平均值输入至变分编码器中,得到第一检测时长对应的表示向量。
S805,将第一检测时长对应的表示向量输入至变分解码器中,得到预测的目标电池特征信息。
S806,使用预训练任务对第一预训练表示模型和第一预测模型进行自监督训练,第一预训练表示模型包括变分编码器,第一预测模型包括变分解码器。
需要说明的是,S801至S806可以参照S701至S706,此处不再进行赘述。
S807,获取单体电芯的第二电池特征信息和第二电池特征信息对应的故障标签信息。
可选的,单体电芯在第一检测时长内的第二电池特征信息可以包括单体电芯的电压信息,单体电芯的温度信息,单体电芯的电流信息,单体电芯的SOC信息,单体电芯的绝缘电阻信息或单体电芯的EIS信息等。
S808,根据单体电芯的第二电池特征信息和第二电池特征信息对应的故障标签信息对电池状态预测模型进行监督训练,电池状态预测模型包括特征向量提取模型和第二预测模型。
在一些可能的实现方式中,使用进行自监督训练后得到的第一预训练表示模型的模型参数对电池状态预测模型中的特征向量提取模型中的参数进行初始化,再使用第二电池特征信息和第二电池特征信息对应的故障标签信息对特征向量提取模型和第二预测模型进行进一步训练,得到电池状态预测模型。
在另一些可能的实现方式中,将进行自监督训练后得到的第一预训练表示模型作为电池状态预测模型中的特征向量提取模型,将第二电池特征信息输入至特征向量提取模型中,得到该第二电池特征信息的表示向量,使用第二电池特征信息、第二电池特征信息的表示向量和第二电池特征信息对应的故障标签信息对第二预测模型进行训练,得到电池状态预测模型。
根据图8所示的训练方法得到的电池状态预测模型可以用来预测单体电芯的故障信息。
本申请提出的技术方案中,将单体电芯在第一检测时长内的所有第一电池特征信息划分为W个时间窗口帧,每个时间窗口帧中的所有第一电池特征信息组成该时间窗口帧对应的表示向量,计算W个时间窗口帧对应的W个表示向量的平均值,将该平均值输入至第一预训练表示模型中,得到第一检测时长对应的表示向量,将第一检测时长对应的表示向量输入至第一预测模型中,得到单体电芯在第一检测时长内的预测 输出结果,使用预训练任务对第一预训练表示模型和第一预测模型进行自监督训练,根据单体电芯的第二电池特征信息和第二电池特征信息对应的故障标签信息对电池状态预测模型进行监督训练,得到单体电芯的故障预测模型。在对单体电芯的故障预测模型进行训练的过程中所使用的第一电池特征信息和第二电池特征信息可以从EV上报的数据中直接获取,降低了训练成本,并且EV可以根据实车数据对单体电芯的故障信息进行预测。
图9为本申请的实施例的一种电池状态预测方法的流程示意图,如图9所示,该方法至少包括S901至S902。
S901,获取待预测电池的待预测电池特征信息。
待预测电池可以是包含多个单体电芯的电池包,也可以是单体电芯。
作为一种示例,待预测电池为电池包时,待预测电池特征信息可以包括电池包的电流信息、电池包的电压信息、电池包的温度信息、电池包中每个单体电芯的电压信息和电池包的SOC信息等。
作为另一种示例,待预测电池为单体电芯时,待预测电池特征信息可以包括单体电芯的电压信息,单体电芯的温度信息,单体电芯的电流信息,单体电芯的SOC信息,单体电芯的绝缘电阻信息或单体电芯的EIS信息等。
在一些实现方式中,待预测电池的待预测电池特征信息可以从待预测电池所属的EV的原始BMS上报的数据中直接获取。
S902,使用电池状态预测模型基于待预测电池特征信息确定待预测电池的电池状态信息。
将待预测电池特征信息输入至电池状态预测模型,可以得到待预测电池的电池状态信息,待预测电池的状态信息可以包括待预测电池的SOH信息和/或故障信息和/或剩余寿命信息等。
可以理解的是,电池状态预测模型可以根据图4至图8所示的训练方法进行训练得到。
示例性的,当待预测电池为电池包时,使用图5所示的训练方法得到的电池状态预测模型进行预测,可以得到电池包的SOH信息;使用图6所示的训练方法得到的电池状态预测模型进行预测,可以得到电池包的故障信息。
示例性的,当待预测电池为单体电芯时,使用图7所示的训练方法得到的电池状态预测模型进行预测,可以得到单体电芯的SOH信息;使用图8所示的训练方法得到的电池状态预测模型进行预测,可以得到单体电芯的故障信息。
本申请提出的技术方案中,将待预测电池的待预测电池特征信息输入至根据图4至图8中任意一个实施例所述的训练方法训练得到的电池状态预测模型中,得到待预测电池的电池状态信息。其中,待预测电池的待预测电池特征信息可以从EV上报的数据中直接获取,使得EV可以根据实车数据预测电池状态。
图10为本申请一个实施例的电池状态预测模型的训练装置的示意性结构图。如图10所示,装置1000可以包括获取模块1001和训练模块1002。
本申请实施例中的获取模块和训练模块中任意模块可以全部或部分通过软件和/硬件方式实现。其中,通过软件实现的部分可以在处理器上运行以实现相应的功能, 通过硬件方式实现的部分可以是处理器的构成部分。
在一种实现方式中,装置1000可以用于实现上述图4所示的方法。例如,获取模块1001用于实现S401和S403,训练模块1002用于实现S402和S404。
在另一种实现方式中,装置1000可以用于实现上述图5所示的方法。例如,获取模块1001用于实现S501和S507,训练模块1002用于实现S506和S508。
在又一种实现方式中,装置1000可以用于实现上述图6所示的方法。例如,获取模块1001用于实现S601和S607,训练模块1002用于实现S606和S608。
在再一种实现方式中,装置1000可以用于实现上述图7所示的方法。例如,获取模块1001用于实现S701和S707,训练模块1002用于实现S706和S708。
在再一种实现方式中,装置1000可以用于实现上述图8所示的方法。例如,获取模块1001用于实现S801和S807,训练模块1002用于实现S806和S808。
图11为本申请一个实施例的电池状态的预测装置的示意性结构图。如图11所示,装置1100可以包括获取模块1101和处理模块1102。
本申请实施例中的获取模块和处理模块中任意模块可以全部或部分通过软件和/硬件方式实现。其中,通过软件实现的部分可以在处理器上运行以实现相应的功能,通过硬件方式实现的部分可以是处理器的构成部分。
在一种实现方式中,装置1100可以用于实现上述图9所示的方法。例如,获取模块1101用于实现S901,处理模块1102用于实现S902。
图12为本申请一个实施例提供的装置的结构示意图。图12所示的装置1200可以用于执行前述任意一个实施例所述的方法。
如图12所示,本实施例的装置1200包括:存储器1201、处理器1202、通信接口1203以及总线1204。其中,存储器1201、处理器1202、通信接口1203通过总线1204实现彼此之间的通信连接。
存储器1201可以是只读存储器(read only memory,ROM),静态存储设备,动态存储设备或者随机存取存储器(random access memory,RAM)。存储器1201可以存储程序,当存储器1201中存储的程序被处理器1202执行时,处理器1202可以用于执行图4至图9所示的方法的各个步骤。
处理器1202可以采用通用的中央处理器(central processing unit,CPU),微处理器,应用专用集成电路(application specific integrated circuit,ASIC),或者一个或多个集成电路,用于执行相关程序,以实现本申请方法实施例的电池状态预测模型的训练方法和电池状态的预测方法。
处理器1202还可以是一种集成电路芯片,具有信号的处理能力。在实现过程中,本申请各个实施例的方法的各个步骤可以通过处理器1202中的硬件的集成逻辑电路或者软件形式的指令完成。
上述处理器1202还可以是通用处理器、数字信号处理器(digital signal processing,DSP)、专用集成电路(ASIC)、现成可编程门阵列(field programmable gate array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。可以实现或者执行本申请实施例中的公开的各方法、步骤及逻辑框图。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
结合本申请实施例所公开的方法的步骤可以直接体现为硬件译码处理器执行完成,或者用译码处理器中的硬件及软件模块组合执行完成。软件模块可以位于随机存储器,闪存、只读存储器,可编程只读存储器或者电可擦写可编程存储器、寄存器等本领域成熟的存储介质中。该存储介质位于存储器1201,处理器1202读取存储器1201中的信息,结合其硬件完成本申请实施例中各个方法所需执行的功能,例如,可以执行图4至图9所示实施例的各个步骤/功能。
通信接口1203可以使用但不限于收发器一类的收发装置,来实现装置1200与其他设备或通信网络之间的通信。
总线1204可以包括在装置1200各个部件(例如,存储器1201、处理器1202、通信接口1203)之间传送信息的通路。
应理解,本申请实施例所示的装置1200可以是电子设备,或者,也可以是配置于电子设备中的芯片。
应理解,本申请实施例中的处理器可以为中央处理单元(central processing unit,CPU),该处理器还可以是其他通用处理器、数字信号处理器(digital signal processor,DSP)、专用集成电路(application specific integrated circuit,ASIC)、现成可编程门阵列(field programmable gate array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
还应理解,本申请实施例中的存储器可以是易失性存储器或非易失性存储器,或可包括易失性和非易失性存储器两者。其中,非易失性存储器可以是只读存储器(read-only memory,ROM)、可编程只读存储器(programmable ROM,PROM)、可擦除可编程只读存储器(erasable PROM,EPROM)、电可擦除可编程只读存储器(electrically EPROM,EEPROM)或闪存。易失性存储器可以是随机存取存储器(random access memory,RAM),其用作外部高速缓存。通过示例性但不是限制性说明,许多形式的随机存取存储器(random access memory,RAM)可用,例如静态随机存取存储器(static RAM,SRAM)、动态随机存取存储器(DRAM)、同步动态随机存取存储器(synchronous DRAM,SDRAM)、双倍数据速率同步动态随机存取存储器(double data rate SDRAM,DDR SDRAM)、增强型同步动态随机存取存储器(enhanced SDRAM,ESDRAM)、同步连接动态随机存取存储器(synchlink DRAM,SLDRAM)和直接内存总线随机存取存储器(direct rambus RAM,DR RAM)。
上述实施例,可以全部或部分地通过软件、硬件、固件或其他任意组合来实现。当使用软件实现时,上述实施例可以全部或部分地以计算机程序产品的形式实现。所述计算机程序产品包括一个或多个计算机指令或计算机程序。在计算机上加载或执行所述计算机指令或计算机程序时,全部或部分地产生按照本申请实施例所述的流程或功能。所述计算机可以为通用计算机、专用计算机、计算机网络、或者其他可编程装置。所述计算机指令可以存储在计算机可读存储介质中,或者从一个计算机可读存储介质向另一个计算机可读存储介质传输,例如,所述计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机 能够存取的任何可用介质或者是包含一个或多个可用介质集合的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质(例如,软盘、硬盘、磁带)、光介质(例如,DVD)、或者半导体介质。半导体介质可以是固态硬盘。
应理解,本文中术语“和/或”,仅仅是一种描述关联对象的关联关系,表示可以存在三种关系,例如,A和/或B,可以表示:单独存在A,同时存在A和B,单独存在B这三种情况,其中A,B可以是单数或者复数。另外,本文中字符“/”,一般表示前后关联对象是一种“或”的关系,但也可能表示的是一种“和/或”的关系,具体可参考前后文进行理解。
本申请中,“至少一个”是指一个或者多个,“多个”是指两个或两个以上。“以下至少一项(个)”或其类似表达,是指的这些项中的任意组合,包括单项(个)或复数项(个)的任意组合。例如,a,b,或c中的至少一项(个),可以表示:a,b,c,a-b,a-c,b-c,或a-b-c,其中a,b,c可以是单个,也可以是多个。
应理解,在本申请的各种实施例中,上述各过程的序号的大小并不意味着执行顺序的先后,各过程的执行顺序应以其功能和内在逻辑确定,而不应对本申请实施例的实施过程构成任何限定。
本领域普通技术人员可以意识到,结合本文中所公开的实施例描述的各示例的单元及算法步骤,能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行,取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能,但是这种实现不应认为超出本申请的范围。
所属领域的技术人员可以清楚地了解到,为描述的方便和简洁,上述描述的系统、装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
在本申请所提供的几个实施例中,应该理解到,所揭露的系统、装置和方法,可以通过其它的方式实现。例如,以上所描述的装置实施例仅仅是示意性的,例如,所述单元的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如多个单元或组件可以结合或者可以集成到另一个系统,或一些特征可以忽略,或不执行。另一点,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口,装置或单元的间接耦合或通信连接,可以是电性,机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本申请各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本申请各个实施例所述方法 的全部或部分步骤。而前述的存储介质包括:U盘、移动硬盘、只读存储器、随机存取存储器、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述,仅为本申请的具体实施方式,但本申请的保护范围并不局限于此,任何熟悉本技术领域的技术人员在本申请揭露的技术范围内,可轻易想到变化或替换,都应涵盖在本申请的保护范围之内。因此,本申请的保护范围应以所述权利要求的保护范围为准。
Claims (20)
- 一种电池状态预测模型的训练方法,其特征在于,所述训练方法包括:获取目标电池在第一时段内的第一电池特征信息,所述目标电池的电池特征信息包括以下一种或多种信息:所述目标电池的电流信息、所述目标电池的电压信息、所述目标电池的温度信息和所述目标电池的电池电量指标SOC;根据所述第一电池特征信息对第一模型进行自监督训练,所述第一模型包括第一预训练表示模型和第一预测模型,所述第一预测模型的输入包括所述第一预训练表示模型的输出,所述第一预训练表示模型用于确定输入的电池特征信息的表示向量,所述第一预测模型用于确定输入的表示向量对应的目标电池特征信息;获取所述目标电池的第二电池特征信息和第一标签信息,所述第一标签信息用于指示所述第二电池特征信息对应的电池状态信息;根据所述第二电池特征信息和所述第一标签信息对所述电池状态预测模型进行监督训练,所述电池状态预测模型包括特征向量提取模型和第二预测模型,所述第二预测模型的输入包括所述特征向量提取模型的输出,所述特征向量提取模型的初始参数包括对所述第一模型进行所述自监督训练后得到的所述第一预训练表示模型的参数。
- 根据权利要求1所述的训练方法,其特征在于,所述第二预测模型的结构与所述第一预测模型的结构相同,并且,所述第二预测模型进行所述监督训练之前的初始参数包括对所述第一模型进行所述自监督训练后得到的所述第一预测模型的参数。
- 根据权利要求1或2所述的训练方法,其特征在于,所述第一电池特征信息包括所述目标电池在所述第一时段内的N个时间的N个电池特征信息,N为正整数,所述第一预训练表示模型中包括第一编码器和第二编码器,所述第一编码器的输入包括电池特征信息,所述第二编码器的输入包括所述第一编码器的输出。
- 根据权利要求3所述的训练方法,其特征在于,所述根据所述第一电池特征信息对第一模型进行自监督训练,包括:将所述N个电池特征信息划分为n个子集合,所述n个子集合中每个子集合包含至少一个电池特征信息,所述n个子集合中第i个子集合中的任意电池特征信息在所述第一时间内对应的时间早于所述n个子集合中第i+1个子集合中的任意电池特征信息在所述第一时间内对应的时间,n为正整数且n小于或等于N,i为正整数且i小于n;将所述n个子集合中的每个子集合中的所有电池特征信息按照对应的时间先后顺序排列得到的电池特征信息序列输入所述第一编码器,得到所述每个子集合对应的表示向量;将所述n个子集合对应的n个表示向量按照对应的时间先后顺序或按照SOC值从低到高的顺序排列后得到的电池特征信息序列输入所述第二编码器,得到所述第一电池特征信息的第一表示向量;将所述第一表示向量输入所述第一预测模型,得到所述第一表示向量对应的目标电池特征信息;根据所述目标电池特征信息和所述第一电池特征信息调整所述第一模型的参数。
- 根据权利要求4所述的训练方法,其特征在于,所述将所述N个电池特征信 息划分为n个子集合,包括:按照SOC值将所述电池特征值划分为所述n个子集合,所述n个子集合中的每个子集合包含的所有电池特征信息对应的SOC值相同。
- 根据权利要求3至5中任一项所述的训练方法,其特征在于,所述目标电池为电池包,所述第一电池特征信息和/或所述第二电池特征信息还包括:所述电池包中每个单体电芯的电压信息。
- 根据权利要求1所述的训练方法,其特征在于,所述第一电池特征信息包括所述目标电池在第一时段内的M个时间的M个电池特征信息,M为正整数,所述第一预训练表示模型包括变分编码器,所述第一预测模型包括变分解码器,所述第二预测模型包括回归模型或分类模型。
- 根据权利要求7所述的训练方法,其特征在于,所述根据所述第一电池特征信息对第一模型进行自监督训练,包括:将所述M个电池特征信息划分为m个子集合,所述m个子集合中每个子集合包含至少一个电池特征信息,所述m个子集合中第j个子集合中的任意电池特征信息在所述第一时间内对应的时间早于所述m个子集合中第j+1个子集合中的任意电池特征信息在所述第一时间内对应的时间,m为正整数且m小于或等于M,j为正整数且j小于m;根据所述m个子集合中的每个子集合中的所有电池特征信息确定所述每个子集合对应的表示向量;计算所述m个子集合对应的m个表示向量的平均值;将所述平均值输入所述第一预训练表示模型,得到所述第一电池特征信息的第二表示向量;将所述第二表示向量输入所述第一预测模型,得到所述第二表示向量对应的目标电池特征信息;根据所述目标电池特征信息和所述第一电池特征信息调整所述第一模型的参数。
- 根据权利要求8所述的训练方法,其特征在于,所述m个子集合中任意一个子集合中的电池特征信息的数量等于所述m个子集合中其它任意一个子集合中的电池特征信息的数量。
- 根据权利要求7至9中任一项所述的训练方法,其特征在于,所述目标电池为单体电芯,所述第一电池特征信息和/或所述第二电池特征信息还包括:所述单体电芯的绝缘电阻信息和/或所述单体电芯的电化学交流阻抗谱EIS。
- 根据权利要求1至10中任一项所述的训练方法,其特征在于,所述目标电池特征信息包括:所述目标电池在所述第一时段内的SOC序列,所述目标电池在所述第一时段内的电压序列、所述目标电池内每个单体电芯在所述第一时段内的充电时长序列、所述目标电池的充电模式或所述目标电池在所述第一时间内的随机掩码处的SOC值,所述充电模式包括快充模型或慢充模式。
- 根据权利要求1至11中任一项所述的方法,其特征在于,所述电池状态信息包括电池健康度指标SOH信息或故障信息或剩余寿命信息。
- 一种电池状态的预测方法,其特征在于,包括:获取待预测电池的待预测电池特征信息;使用电池状态预测模型基于所述待预测电池特征信息确定所述待预测电池的电池状态信息,所述电池状态预测模型为使用如权利要求1至12中任一项所述的训练方法训练得到的电池状态预测模型。
- 一种电池状态预测模型的训练装置,其特征在于,包括用于实现权利要求1至12中任一项所述的方法的各个功能模块。
- 一种电池状态的预测装置,其特征在于,包括用于实现权利要求13所述的方法的各个功能模块。
- 一种电池状态预测模型的训练装置,其特征在于,包括:存储器和处理器;所述存储器用于存储程序指令;所述处理器用于调用所述存储器中的程序指令执行如权利要求1至12中任一项所述的方法。
- 一种电池状态的预测装置,其特征在于,包括:存储器和处理器;所述存储器用于存储程序指令;所述处理器用于调用所述存储器中的程序指令执行如权利要求13所述的方法。
- 一种芯片,其特征在于,包括至少一个处理器和通信接口,所述通信接口和所述至少一个处理器通过线路互联,所述至少一个处理器用于运行计算机程序或指令,以执行如权利要求1至12中任一项所述的方法或权利要求13中所述的方法。
- 一种计算机可读介质,其特征在于,所述计算机可读介质存储用于计算机执行的程序代码,该程序代码包括用于执行如权利要求1至12中任一项所述的方法的指令或权利要求13中所述的方法的指令。
- 一种计算机程序产品,其特征在于,所述计算机程序产品包括指令,当所述指令被执行时,使得计算机执行权利要求1至12中任一项所述的方法或权利要求13所述的方法。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110197414.8A CN114970841A (zh) | 2021-02-22 | 2021-02-22 | 电池状态预测模型的训练方法及相关装置 |
CN202110197414.8 | 2021-02-22 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022174601A1 true WO2022174601A1 (zh) | 2022-08-25 |
Family
ID=82931972
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2021/124343 WO2022174601A1 (zh) | 2021-02-22 | 2021-10-18 | 电池状态预测模型的训练方法及相关装置 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN114970841A (zh) |
WO (1) | WO2022174601A1 (zh) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115169756B (zh) * | 2022-09-07 | 2022-12-09 | 深圳市信润富联数字科技有限公司 | 电芯水份预测方法、装置、设备及存储介质 |
CN115563704B (zh) * | 2022-09-23 | 2023-12-08 | 四川新能源汽车创新中心有限公司 | 电池状态预测模型的优化方法、容量预测方法及相关装置 |
CN116224091B (zh) * | 2022-12-01 | 2024-02-02 | 伏瓦科技(苏州)有限公司 | 电池的电芯故障检测方法、装置、电子设备及存储介质 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108519556A (zh) * | 2018-04-13 | 2018-09-11 | 重庆邮电大学 | 一种基于循环神经网络的锂离子电池soc预测方法 |
US20190113577A1 (en) * | 2017-10-17 | 2019-04-18 | The Board Of Trustees Of The Leland Stanford Junior University | Data-driven Model for Lithium-ion Battery Capacity Fade and Lifetime Prediction |
CN112092675A (zh) * | 2020-08-31 | 2020-12-18 | 长城汽车股份有限公司 | 一种电池热失控预警方法、系统及服务器 |
CN112379269A (zh) * | 2020-10-14 | 2021-02-19 | 武汉蔚来能源有限公司 | 电池异常检测模型训练及其检测方法、装置 |
-
2021
- 2021-02-22 CN CN202110197414.8A patent/CN114970841A/zh active Pending
- 2021-10-18 WO PCT/CN2021/124343 patent/WO2022174601A1/zh active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190113577A1 (en) * | 2017-10-17 | 2019-04-18 | The Board Of Trustees Of The Leland Stanford Junior University | Data-driven Model for Lithium-ion Battery Capacity Fade and Lifetime Prediction |
CN108519556A (zh) * | 2018-04-13 | 2018-09-11 | 重庆邮电大学 | 一种基于循环神经网络的锂离子电池soc预测方法 |
CN112092675A (zh) * | 2020-08-31 | 2020-12-18 | 长城汽车股份有限公司 | 一种电池热失控预警方法、系统及服务器 |
CN112379269A (zh) * | 2020-10-14 | 2021-02-19 | 武汉蔚来能源有限公司 | 电池异常检测模型训练及其检测方法、装置 |
Also Published As
Publication number | Publication date |
---|---|
CN114970841A (zh) | 2022-08-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2022174601A1 (zh) | 电池状态预测模型的训练方法及相关装置 | |
Bian et al. | Stacked bidirectional long short-term memory networks for state-of-charge estimation of lithium-ion batteries | |
Bian et al. | State-of-charge sequence estimation of lithium-ion battery based on bidirectional long short-term memory encoder-decoder architecture | |
Weerakody et al. | A review of irregular time series data handling with gated recurrent neural networks | |
Dahiwade et al. | Designing disease prediction model using machine learning approach | |
WO2021120719A1 (zh) | 神经网络模型更新方法、图像处理方法及装置 | |
CN111368993B (zh) | 一种数据处理方法及相关设备 | |
Zhou et al. | Battery health prognosis using improved temporal convolutional network modeling | |
EP4145351A1 (en) | Neural network construction method and system | |
CN112651511A (zh) | 一种训练模型的方法、数据处理的方法以及装置 | |
US20220215159A1 (en) | Sentence paraphrase method and apparatus, and method and apparatus for training sentence paraphrase model | |
JP2023522468A (ja) | バッテリ検出方法及び装置 | |
WO2021208799A1 (zh) | 训练迁移模型的方法、故障检测的方法以及装置 | |
EP4235506A1 (en) | Neural network model training method, image processing method, and apparatus | |
CN111612215A (zh) | 训练时间序列预测模型的方法、时间序列预测方法及装置 | |
Zheng et al. | State of health estimation for lithium battery random charging process based on CNN-GRU method | |
WO2021136058A1 (zh) | 一种处理视频的方法及装置 | |
US20240135174A1 (en) | Data processing method, and neural network model training method and apparatus | |
EP4273754A1 (en) | Neural network training method and related device | |
JP2023520970A (ja) | リチウム電池のsoc推定方法、装置及びコンピュータ読み取り可能な記憶媒体 | |
Bai et al. | Correlative channel-aware fusion for multi-view time series classification | |
EP4401007A1 (en) | Neural network acquisition method, data processing method and related device | |
CN115238909A (zh) | 一种基于联邦学习的数据价值评估方法及其相关设备 | |
CN111522926A (zh) | 文本匹配方法、装置、服务器和存储介质 | |
CN114997036A (zh) | 基于深度学习的网络拓扑重构方法、装置和设备 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21926315 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 21926315 Country of ref document: EP Kind code of ref document: A1 |