High-precision soc prediction method and device, equipment and storage medium

Abstract

The embodiment of the application provides a high-precision SOC prediction method, device, equipment and storage medium, the method comprises the following steps: collecting battery data of a target battery at a fixed time interval, constructing a sliding window sequence with a preset length based on the collected battery data; based on a plurality of battery SOC prediction models determined in advance, the sliding window sequence is analyzed respectively to obtain a multi-dimensional fusion feature sequence; a deep neural network based on an attention mechanism is used to deep encode and decode the multi-dimensional fusion feature sequence to obtain an SOC prediction value of the target battery. Through the combination of the multi-model fusion and the SOC prediction method based on the deep neural network of the attention mechanism, the dynamic load environment is adapted, and the accurate tracking of the SOC is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of battery management, and particularly relates to a high-precision SOC prediction method and device, equipment and a storage medium. BACKGROUND

[0002] The battery state of charge (SOC) is a key indicator for measuring the remaining energy of a battery, and is of great significance to the safe and stable operation of energy-intensive equipment such as electric vehicles, energy storage systems and unmanned aerial vehicles. The common SOC prediction methods currently include the coulomb counting method, the open-circuit voltage method, the model parameter prediction method, and the prediction method based on artificial intelligence. Among them: the coulomb counting method realizes SOC prediction through current integration, and is affected by current sensor bias and integration error accumulation, with significant long-term operation error; the open-circuit voltage method relies on measuring the open-circuit voltage of a static battery and cannot reflect the SOC under dynamic load conditions in real time; the equivalent circuit model method (such as Thevenin and RC model) requires accurate modeling and online parameter identification, and modeling error and working condition changes are easy to cause prediction deviation; neural networks, support vector machines and other intelligent algorithms have nonlinear modeling capabilities, but they are strongly dependent on training data and have limited generalization ability, especially in the case of different load changes and environmental temperature fluctuations, the prediction accuracy is unstable. In addition, in a dynamic environment (such as frequent start-stop, acceleration / deceleration, temperature fluctuation, etc.), the voltage and current of the battery show highly nonlinear and non-stationary characteristics, which makes it difficult for traditional methods to accurately capture the SOC change trajectory. SUMMARY

[0003] Therefore, the embodiments of the present application provide a high-precision SOC prediction method, device, equipment and storage medium, which combines a multi-model fusion and a deep neural network based on an attention mechanism to adapt to a dynamic load environment and realize accurate tracking of the SOC.

[0004] The embodiments of the present application provide a high-precision SOC prediction method, which comprises:

[0005] Collecting battery data of a target battery at fixed time intervals, and constructing a preset length of a sliding window sequence based on the collected battery data;

[0006] Analyzing the sliding window sequence based on a plurality of predetermined battery SOC prediction models to obtain a multi-dimensional fusion feature sequence;

[0007] Deeply encoding and decoding the multi-dimensional fusion feature sequence based on a deep neural network based on an attention mechanism to obtain an SOC prediction value of the target battery.

[0008] In an embodiment, the plurality of predetermined battery SOC prediction models comprises a combination of any of the following: an extended Kalman filter model, a time series network model, a physical model, or an electrochemical model.

[0009] In an embodiment, the plurality of predetermined battery SOC prediction models are respectively analyzed based on the sliding window sequence to obtain a multi-dimensional fusion feature sequence, comprising:

[0010] The sliding window sequence is respectively input into a plurality of battery SOC prediction models to obtain a plurality of initial SOC prediction sequences;

[0011] The plurality of initial SOC prediction sequences are spliced with the sliding window sequence to obtain the multi-dimensional fusion feature sequence.

[0012] In an embodiment, the attention mechanism-based deep neural network comprises: an input layer, an encoder, a decoder, and an output layer;

[0013] The attention mechanism-based deep neural network performs deep encoding and decoding on the multi-dimensional fusion feature sequence to obtain the SOC prediction value of the target battery, comprising:

[0014] In the input layer, the multi-dimensional fusion feature sequence is projected and mapped to a preset dimension, and each feature in the preset dimension feature sequence is respectively position encoded in a time step to obtain a target feature sequence;

[0015] In the encoder, the target feature sequence is analyzed to extract key time slice features related to SOC prediction;

[0016] In the decoder, the key time slice features are analyzed to predict a SOC change trend;

[0017] In the output layer, the SOC change trend is mapped to obtain the SOC prediction value.

[0018] In an embodiment, the analysis of the target feature sequence in the encoder to extract key time slice features related to SOC prediction comprises:

[0019] In the encoder, the attention value between any two time steps in the target feature sequence is calculated, and each time step feature is locally nonlinearly mapped to extract key time slice features related to SOC prediction.

[0020] In an embodiment, the calculation of the attention weight between any two time steps in the target feature sequence in the encoder, and the local nonlinear mapping of each time step embedding feature to extract key time slice features related to SOC prediction, comprises:

[0021] calculate attention weights between any two time steps in the target feature sequence based on a multi-head self-attention mechanism, extract features corresponding to time steps with attention weights greater than a preset value;

[0022] perform local nonlinear mapping on the embedding features of each time step, and extract features corresponding to time steps where mutations occur;

[0023] use the features corresponding to the time steps with attention weights greater than the preset value and the features corresponding to the time steps where mutations occur as key time segment features related to SOC prediction.

[0024] In an embodiment, the analysis of the key time segment features in the decoder to predict the SOC change trend comprises:

[0025] analyze the key time segment features and the global features output by the encoder in the decoder, fuse the analysis results, and input the fused time segment features into a trend prediction module to predict the SOC change trend.

[0026] The second aspect of the embodiments of the present application provides a high-precision SOC prediction device, comprising:

[0027] a collection module configured to collect battery data of a target battery at fixed time intervals, and construct a sliding window sequence of a preset length based on the collected battery data;

[0028] an analysis module configured to analyze the sliding window sequence based on a plurality of pre-determined battery SOC prediction models to obtain a multi-dimensional fusion feature sequence;

[0029] a coding and decoding module configured to perform deep coding and decoding on the multi-dimensional fusion feature sequence based on a deep neural network of an attention mechanism to obtain an SOC prediction value of the target battery.

[0030] In an embodiment, the plurality of pre-determined battery SOC prediction models comprise a combination of any of the following models: an extended Kalman filter model, a time series network model, a physical model, or an electrochemical model.

[0031] In an embodiment, the analysis module is specifically configured to:

[0032] input the sliding window sequence into a plurality of battery SOC prediction models to obtain a plurality of initial SOC prediction sequences;

[0033] splice the plurality of initial SOC prediction sequences and the sliding window sequence to obtain the multi-dimensional fusion feature sequence.

[0034] In an embodiment, the attention mechanism-based deep neural network comprises an input layer, an encoder, a decoder, and an output layer.

[0035] The codec module comprises:

[0036] The first obtaining unit is configured to project and map the multi-dimensional fusion feature sequence to a preset dimension in the input layer, and perform position encoding on each feature in the preset dimension feature sequence respectively to obtain a target feature sequence.

[0037] The extraction unit is configured to analyze the target feature sequence in the encoder to extract SOC prediction-related key time segment features.

[0038] The prediction unit is configured to analyze the key time segment features in the decoder to predict an SOC change trend.

[0039] The second obtaining unit is configured to map the SOC change trend in the output layer to obtain an SOC prediction value.

[0040] In an embodiment, the extraction unit is specifically configured to:

[0041] In the encoder, the attention value between any two time steps in the target feature sequence is calculated, and the features of each time step are locally nonlinearly mapped to extract SOC prediction-related key time segment features.

[0042] In an embodiment, the extraction unit comprises:

[0043] The first extraction subunit is configured to calculate the attention weight between any two time steps in the target feature sequence based on a multi-head self-attention mechanism, and extract features corresponding to time steps with an attention weight greater than a preset value.

[0044] The second extraction subunit is configured to locally nonlinearly map the embedding features of each time step to extract features corresponding to time steps with mutations; and the features corresponding to time steps with the attention weight greater than the preset value and the features corresponding to time steps with mutations are taken as the SOC prediction-related key time segment features.

[0045] In an embodiment, the prediction unit is specifically configured to:

[0046] In the decoder, the key time segment features and the global features output by the encoder are analyzed, and the analysis results are fused; and the fused time segment features are input into a trend prediction module to predict the SOC change trend.

[0047] A third aspect of an embodiment of the present application provides a high-precision SOC prediction device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor; when the processor executes the computer program, the steps of the method described in the first aspect above are implemented.

[0048] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the method described in the first aspect above are implemented.

[0049] The beneficial effects of the embodiments of the present application include: collecting battery data from a target battery at fixed time intervals, constructing a sliding window sequence of preset length based on the collected battery data; analyzing the sliding window sequence based on multiple predetermined battery SOC prediction models to obtain a multidimensional fused feature sequence; and performing deep encoding and decoding of the multidimensional fused feature sequence using a deep neural network based on an attention mechanism to obtain a predicted SOC value for the target battery. By combining multi-model fusion with an attention-based deep neural network SOC prediction method, the method adapts to dynamic load environments and achieves accurate SOC tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0051] Figure 1 A flowchart of a high-precision SOC prediction method provided in one embodiment of the present application;

[0052] Figure 2 for Figure 1 Schematic diagram of the specific implementation process of S130;

[0053] Figure 3 A schematic diagram of a high-precision SOC prediction device provided in one embodiment of the present application;

[0054] Figure 4 A schematic diagram of a high-precision SOC prediction device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0055] The following embodiments of the technical solution of the present application will be described in detail with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present application and are therefore only examples and are not intended to limit the scope of protection of the present application.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of this application; the use of the terms "including," "comprising," or "having" and variations thereof herein is intended to be broad and encompass the terms "consisting of" and "consisting essentially of" and variations thereof. Unless otherwise noted, the terms "including" and "comprising" are open-ended and do not exclude the presence of unrecited elements or limitations.

[0057] In the description of the embodiments of the application, the technical terms "first", "second" and the like are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the technical features indicated. In the description of the embodiments of the application, the meaning of "a plurality of" is two or more, unless otherwise explicitly and specifically limited.

[0058] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily independent or alternative embodiments to each other. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0059] In the description of the embodiments of the application, the term "and / or" is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone. In addition, the character " / " herein generally represents that the front and rear associated objects have an "or" relationship.

[0060] In the description of the embodiments of the application, the term "a plurality of frames" refers to two or more (including two).

[0061] In the description of the embodiments of the application, the technical terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the embodiments of the application and simplifying the description, and therefore cannot be understood as indicating or implying that the devices or elements indicated must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the embodiments of the application.

[0062] The embodiments of the application provide a high-precision SOC prediction method, which can adapt to a dynamic load environment by combining a multi-model fusion and a SOC prediction method based on an attention mechanism deep neural network, and realize accurate tracking of SOC.

[0063] Referring to Figure 1 as shown, Figure 1 A flowchart of a high-precision SOC prediction method provided by an embodiment of the present application. The high-precision SOC prediction method is applied to an energy storage wireless BMS system and is implemented by a high-precision SOC prediction device. The high-precision SOC prediction method comprises the following steps. Figure 1 It can be seen that the high-precision SOC prediction method comprises the following steps:

[0064] S110: Collect battery data at a fixed time interval, and construct a sliding window sequence of a preset length based on the collected battery data.

[0065] Battery data is collected every preset time interval, such as 1s or 5s. Exemplarily, the battery data includes voltage, current, temperature, current or previous SOC, etc. The battery data collected at the past N time points is organized into a time sequence matrix. The time sequence matrix is updated forward every time new battery data is collected, forming a new time sequence matrix. The time sequence matrix updated with the collection time point is referred to as a sliding window sequence.

[0066] S120: Analyze the sliding window sequence based on a plurality of predetermined battery SOC prediction models respectively, to obtain a multi-dimensional fusion feature sequence.

[0067] The plurality of predetermined battery SOC prediction models include any combination of the following models: an extended Kalman filter model, a time sequence network model, a physical model, or an electrochemical model.

[0068] Analyzing the sliding window sequence based on a plurality of predetermined battery SOC prediction models respectively to obtain a multi-dimensional fusion feature sequence comprises: inputting the sliding window sequence into the plurality of battery SOC prediction models respectively to obtain a plurality of initial SOC prediction sequences; and splicing the plurality of initial SOC prediction sequences with the sliding window sequence to obtain a multi-dimensional fusion feature sequence.

[0069] For example, in this embodiment, the sliding window sequence is input into multiple battery SOC prediction models to obtain multiple initial SOC prediction sequences; the multiple initial SOC prediction sequences are spliced with the sliding window sequence to obtain a multi-dimensional fusion feature sequence, including: inputting the sliding window sequence into a Kalman filter model, using the voltage, current and temperature in the sliding window as input in the Kalman filter model, combining the internal battery equivalent circuit parameters (such as resistance, capacitance) for state estimation, outputting the predicted SOC values of the current to future H time steps, to obtain a first initial SOC prediction sequence; inputting the sliding window sequence into a time series network model, analyzing based on multiple neural network layers and fully connected layers in the time series network model, outputting a feature vector of the SOC change trajectory, taking the feature vector of the SOC change trajectory as a second initial SOC prediction sequence; inputting the sliding window sequence into a physical model (such as an RC model), based on the transient response of voltage and current, using state equations for estimation, and according to the data in the sliding window, the charge state is inversely solved, to obtain an SOC prediction sequence as a third initial SOC prediction sequence; inputting the sliding window sequence into an electrochemical model, simulating and calculating the current change of lithium ion concentration inside the battery, and deriving a fourth initial SOC prediction sequence according to the change of lithium ion concentration inside the battery. The initial SOC prediction sequence output by each model is spliced with the sliding window sequence respectively, and a new feature sequence is formed at each time step of the sliding window; since the SOC prediction value is added to each time step of the sliding window, a multi-dimensional fusion feature sequence is finally formed. The spliced multi-dimensional fusion feature sequence includes features extracted by each model, which do not interfere with each other, and can improve the prediction accuracy of the subsequent prediction, and is used to output the complete SOC trajectory change trend.

[0070] S130: A deep neural network based on an attention mechanism encodes and decodes the multi-dimensional fusion feature sequence to obtain the SOC prediction value of the target battery.

[0071] The deep neural network based on the attention mechanism includes an input layer, an encoder, a decoder, and an output layer. By using the deep network model based on the attention mechanism to learn the input feature sequence, long-time dependence, key time segments, and nonlinear change rules are captured to output SOC prediction results with better physical consistency and time evolution trend. Illustratively, the input layer is used for feature mapping and position encoding, the encoder is used for extracting key time segment features, the decoder is used for trend modeling and fusion analysis of the key time segment features and global features, predicting the future change trend of SOC, and the output is used for mapping the change trend to the predicted SOC value at the current time point or the SOC prediction sequence in the future.

[0072] Please refer to Figure 2 , Figure 2 To Figure 1 the specific implementation process of S130.Figure 2 It can be known that, in the embodiment, S130 comprises S131 to S134. Details are as follows:

[0073] S131: Projecting and mapping the multi-dimensional fusion feature sequence to a preset dimension in the input layer, and respectively performing position coding of each feature in the preset dimension feature sequence in a time step to obtain a target feature sequence.

[0074] In the embodiment, the preset dimension refers to an embedding dimension, and projecting and mapping the multi-dimensional fusion feature sequence to the preset dimension in the input layer refers to linearly mapping the multi-dimensional fusion feature sequence to an embedding dimension space in the input layer, which can be expressed as: ; wherein X represents the multi-dimensional fusion feature sequence, the dimension of which is , represents T time steps, and each time step has d-dimensional features; represents an embedding weight matrix, the dimension of which is , is used to map the original feature of each time step from the dimension d to the unified ; is a bias term, the dimension of which is , is used to add a fixed bias to each time step; The embedding result after linear mapping has a dimension of , which means that each time step has features for subsequent processing; d is the dimension of the original input feature; is the encoding dimension of the current model.

[0075] The deep network model based on the attention mechanism has no sequence perception ability, so "position" information needs to be added to tell the model which time step the current feature belongs to.

[0076] Exemplarily, each feature in the preset dimension feature sequence is respectively subjected to position coding of a time step to obtain a target feature sequence, which is expressed as:

[0077] ; wherein P represents a position coding matrix, which is used to give each time step a unique position vector for representing the sequence order; Target feature after adding position coding.

[0078] S132: Analyzing the target feature sequence in the encoder to extract key time segment features related to SOC prediction.

[0079] In the encoder, the target feature sequence is analyzed to extract key time segment features related to SOC prediction, including: calculating the attention value between any two time steps in the target feature sequence in the encoder, and performing local nonlinear mapping on the features of each time step to extract key time segment features related to SOC prediction.

[0080] In the encoder, the attention weight between any two time steps in the target feature sequence is calculated, and the embedding features of each time step are locally nonlinearly mapped to extract the key time segment features related to SOC prediction, including: calculating the attention weight between any two time steps in the target feature sequence based on the multi-head self-attention mechanism, and extracting the features corresponding to the time steps with attention weight greater than a preset value; locally nonlinearly mapping the embedding features of each time step to extract the features corresponding to the time steps with mutations; and taking the features corresponding to the time steps with attention weight greater than the preset value and the features corresponding to the time steps with mutations as the key time segment features related to SOC prediction.

[0081] In this embodiment, the higher the attention weight between two time steps, the greater the mutual influence between the two time steps. By screening the time steps with attention weight greater than a preset value and detecting the time steps with mutation points (such as change rate exceeding a set threshold) of embedding features, the key time segments are obtained to capture the global or local time sequence dependency related to SOC, thereby improving the accuracy of SOC prediction results.

[0082] Specifically, the position-encoded target feature sequence is input into the encoder. The encoder adopts a multi-head self-attention mechanism and combines a feedforward neural network for nonlinear transformation to extract the key time segment features related to SOC prediction.

[0083] In the attention calculation, for any two time steps in the sequence, the attention weight calculation formula is:

[0084] ; wherein Q, K, and V are query, key, and value matrices obtained by linear mapping.

[0085] For each time step, by comparing whether the attention weight is greater than a preset value and detecting whether the feature has a mutation, representative key time segment features are extracted. At the same time, the global features of the entire encoder output can be retained.

[0086] S133: Analyzing the key time segment features in the decoder to predict the SOC change trend.

[0087] Analyzing the key time segment features in the decoder to predict the SOC change trend, including: analyzing the key time segment features and the global features output by the encoder in the decoder, fusing the analysis results, inputting the fused time segment features into a trend prediction module, and predicting the SOC change trend.

[0088] In this embodiment, in order to prevent future information leakage, the key time segment is analyzed by a masked multi-head self-attention mechanism. The masked multi-head self-attention mechanism only allows attention to itself and previous time steps, and is suitable for time series prediction tasks.

[0089] Specifically, the key time segment feature is input as a query (Q) in the decoder, and all time segment features output by the encoder (i.e., saved global features) are input as a key (K) and a value (V), and are fused by a cross-attention mechanism. The calculation process is as follows:

[0090] ; wherein, represents the key time segment feature matrix extracted by the encoder, represents all time step features output by the encoder, are learnable linear mapping matrices for mapping the inputs to Q, K, and V, respectively.

[0091] The fused time segment feature is obtained by the cross-attention mechanism, and is represented as:

[0092] ; wherein, represents the similarity between Q and K; represents a scaling factor to avoid too sharp attention distribution, (.) represents the weight distribution of each Q vector to all K, represents the fused time segment feature, which contains not only the feature of the key segment itself, but also the global context information.

[0093] The fused time segment feature is input into a trend prediction module to predict the change trend of the time series, and the SOC change trend of the future time step, such as the SOC growth rate or decay rate per unit time, is obtained.

[0094] S134: Map the SOC change trend in the output layer to obtain the SOC prediction value.

[0095] In the output layer, the SOC change trend is mapped based on a preset mapping function to obtain the SOC prediction value. The preset mapping function includes but is not limited to a slope superposition mapping function, a nonlinear mapping function, or a first value mapping function.

[0096] ​​​​It can be known through the above analysis that the high-precision SOC prediction method provided in the embodiment comprises: collecting battery data of a target battery at a fixed time interval, constructing a sliding window sequence of a preset length based on the collected battery data, respectively analyzing the sliding window sequence based on a plurality of pre-determined battery SOC prediction models, obtaining a multi-dimensional fusion feature sequence, and performing deep encoding and decoding on the multi-dimensional fusion feature sequence based on a deep neural network of an attention mechanism to obtain an SOC prediction value of the target battery. Through the combination of the multi-model fusion and the SOC prediction method based on the deep neural network of the attention mechanism, the dynamic load environment is adapted, and accurate tracking of the SOC is realized.

[0097] Please refer to Figure 3 , Figure 3 The high-precision SOC prediction device provided in the embodiment is shown in a schematic diagram. Each module or unit included in the high-precision SOC prediction device is configured to perform each step in the corresponding embodiment. For details, please refer to the related description in the corresponding embodiment. For the sake of illustration, only the parts related to the present embodiment are shown. Please refer to Figure 1 or Figure 2 The high-precision SOC prediction device 300 comprises: Figure 1 Figure 2 The collection module 310 is configured to collect battery data of a target battery at a fixed time interval, and construct a sliding window sequence of a preset length based on the collected battery data. Figure 3 The analysis module 320 is configured to respectively analyze the sliding window sequence based on a plurality of pre-determined battery SOC prediction models, and obtain a multi-dimensional fusion feature sequence.

[0098] The encoding and decoding module 330 is configured to perform deep encoding and decoding on the multi-dimensional fusion feature sequence based on a deep neural network of an attention mechanism, and obtain an SOC prediction value of the target battery.

[0099] In an embodiment, the plurality of pre-determined battery SOC prediction models comprises a combination of any of the following models: an extended Kalman filter model, a time series network model, a physical model, or an electrochemical model.

[0100] In an embodiment, the analysis module 320 is specifically configured to:

[0101] input the sliding window sequence into a plurality of battery SOC prediction models respectively, and obtain a plurality of initial SOC prediction sequences;

[0102] splice the plurality of initial SOC prediction sequences and the sliding window sequence, and obtain the multi-dimensional fusion feature sequence.

[0103]

[0104]

[0105] ​​​In an embodiment, the attention mechanism based deep neural network comprises an input layer, an encoder, a decoder, and an output layer.

[0106] The codec module 330 comprises:

[0107] The first obtaining unit is configured to project and map the multi-dimensional fusion feature sequence to a preset dimension in the input layer, and perform position encoding on each feature in the preset dimension feature sequence respectively to obtain a target feature sequence.

[0108] The extraction unit is configured to analyze the target feature sequence in the encoder to extract SOC prediction related key time segment features.

[0109] The prediction unit is configured to analyze the key time segment features in the decoder to predict an SOC change trend.

[0110] The second obtaining unit is configured to map the SOC change trend in the output layer to obtain an SOC prediction value.

[0111] In an embodiment, the extraction unit is specifically configured to:

[0112] In the encoder, attention values between any two time steps in the target feature sequence are calculated, and local nonlinear mapping is performed on the features of each time step to extract SOC prediction related key time segment features.

[0113] In an embodiment, the extraction unit comprises:

[0114] The first extraction subunit is configured to calculate attention weights between any two time steps in the target feature sequence based on a multi-head self-attention mechanism, and extract features corresponding to time steps with attention weights greater than a preset value.

[0115] The second extraction subunit is configured to perform local nonlinear mapping on the embedded features of each time step to extract features corresponding to time steps with mutations; and the features corresponding to the time steps with attention weights greater than the preset value and the features corresponding to the time steps with mutations are taken as the SOC prediction related key time segment features.

[0116] In an embodiment, the prediction unit is specifically configured to:

[0117] In the decoder, the key time segment features and global features output by the encoder are analyzed, and the analysis results are fused; and the fused time segment features are input into a trend prediction module to predict the SOC change trend.

[0118] See Figure 4 ,Figure 4 A schematic diagram of a high-precision SOC prediction device provided for an embodiment of the present application is shown. The high-precision SOC prediction device 400 includes a processor 410, a memory 420, and a computer program 430 stored in the memory 420 and executable on the processor 410. The processor 410 implements the steps in the above-mentioned high-precision SOC prediction method embodiments when executing the computer program 430, such as steps S110-S130 shown in FIG. 1. Figure 4 Alternatively, the processor 410 implements the functions of the modules / units in the above-mentioned device embodiments when executing the computer program 430, such as the functions of the modules 310-330 shown in FIG. 3. Figure 1 Figure 3 For example, the computer program 430 can be divided into one or more modules / units, one or more of which are stored in the memory 420 and executed by the processor 410 to complete the present application. One or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program 430 in the high-precision SOC prediction device. For example, the computer program 430 can be divided into an acquisition module, an analysis module, and a coding and decoding module.

[0119] The high-precision SOC prediction device provided in this embodiment can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the high-precision SOC prediction device provided in this embodiment is only an example and does not constitute a limitation on the high-precision SOC prediction device, and can include more or fewer components than shown, or combine certain components, or different components, for example, the high-precision SOC prediction device can also include an input / output device, a network access device, a bus, etc.

[0120] The processor 410 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. Figure 4

[0121]

[0122] ​​​The memory 420 can be an internal storage unit of the high-precision SOC prediction device, such as a hard disk or a memory of the high-precision SOC prediction device. The memory 420 can also be an external storage device of the high-precision SOC prediction device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, and the like equipped on the high-precision SOC prediction device. Further, the high-precision SOC prediction device can also include both an internal storage unit and an external storage device of the high-precision SOC prediction device. The memory 420 is used to store computer programs and other programs and data required by the high-precision SOC prediction device. The memory 420 can also be used to temporarily store data that has been output or will be output.

[0123] It should be noted that the information interaction, execution process, and the like between the above-described apparatuses / units, since based on the same concept as the method embodiments of the present application, specific functions and brought technical effects can be referred to the method embodiments part, and will not be described here.

[0124] The embodiments of the present application also provide a network device, which comprises at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, and the processor implements the steps in any of the above method embodiments when executing the computer program.

[0125] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the steps in any of the above method embodiments.

[0126] The embodiments of the present application provide a computer program product, which, when running on a mobile terminal, enables the mobile terminal to implement the steps in any of the above method embodiments.

[0127] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the present application can implement all or part of the processes in the above-mentioned embodiment methods through a computer program to instruct relevant hardware to complete, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of each method embodiment described above can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer readable medium can at least include any entity or device capable of carrying the computer program code to the photographing device / terminal equipment, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium. For example, U disk, mobile hard disk, magnetic disk or optical disk, etc. In some jurisdictions, according to legislation and patent practice, the computer readable medium can not be an electrical carrier signal and a telecommunication signal.

[0128] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0129] Those skilled in the art can appreciate that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0130] In the embodiments provided in the present application, it should be understood that the disclosed apparatus / network device and method can be implemented in other ways. For example, the apparatus / network device embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection between each other can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0131] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may also be distributed to multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiment scheme according to actual needs.

[0132] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A high-precision SOC prediction method, characterized in that: include: Collecting battery data of a target battery at fixed time intervals, and constructing a sliding window sequence of a preset length based on the collected battery data; The sliding window sequence is input into the extended Kalman filter model, the time series network model, the physical model and the electrochemical model respectively; the voltage, current and temperature in the sliding window are used as input in the Kalman filter model, and the state is estimated in combination with the internal battery equivalent circuit parameters to obtain a first initial SOC prediction sequence; in the time series network model, analysis is performed based on the multi-layer neural network layer and the fully connected layer, and a second initial SOC prediction sequence representing the characteristic vector of the SOC change trajectory is output; in the physical model, based on the transient response of the voltage and current, the state equation is used for estimation, and the charge state is inversely solved according to the data in the sliding window to obtain a third initial SOC prediction sequence; in the electrochemical model, the change of the lithium ion concentration inside the current battery is simulated and calculated, and the fourth initial SOC prediction sequence is derived according to the change of the lithium ion concentration inside the battery; the initial SOC prediction sequence output by each model is spliced ​​with the sliding window sequence respectively, and a new feature sequence is formed at each time step of the sliding window to obtain a multi-dimensional fusion feature sequence; The multidimensional fusion feature sequence is input into a deep neural network based on the attention mechanism, the multidimensional fusion feature sequence is projected and mapped to a preset dimension in the input layer of the deep neural network based on the attention mechanism, and the position encoding of each feature in the feature sequence of the preset dimension is performed on the time step to obtain a target feature sequence, the attention weight between any two time steps in the target feature sequence is calculated in the encoder of the deep neural network based on the attention mechanism, and the embedded features of each time step are locally nonlinearly mapped to extract key time segment features related to SOC prediction, the key time segment features are used as queries in the decoder of the deep neural network based on the attention mechanism, and are analyzed with the global features output by the encoder, and the analysis results are fused, and the fused time segment features are input into the trend prediction module to predict the SOC change trend.

2. The high-precision SOC prediction method according to claim 1, wherein: The initial SOC prediction sequence output by each model is spliced ​​with the sliding window sequence respectively, forming a new feature sequence at each time step of the sliding window, and obtaining a multi-dimensional fusion feature sequence, including: The initial SOC prediction sequence output by each model is spliced ​​with the sliding window sequence respectively, and the voltage, current and temperature corresponding to each time step of the sliding window are combined with the initial SOC prediction value output by each model to obtain a multidimensional fusion feature sequence.

3. The high-precision SOC prediction method according to claim 1, wherein: The encoder calculates the attention weight between any two time steps in the target feature sequence, performs local nonlinear mapping on the embedded features of each time step, and extracts key time segment features related to SOC prediction, including: In the encoder, the attention weight between any two time steps in the target feature sequence is calculated based on the multi-head self-attention mechanism, and the features corresponding to the time steps with attention weights greater than a preset value are extracted; the embedded features of each time step are locally nonlinearly mapped, and the features corresponding to the time steps where mutations occur are extracted, and the features corresponding to the time steps where the attention weights are greater than the preset value and the features corresponding to the time steps where mutations occur are used as the key time segment features related to the SOC prediction.

4. A high-precision SOC prediction device, characterized in that: include: A construction module, configured to collect battery data of a target battery at fixed time intervals, and construct a sliding window sequence of a preset length based on the collected battery data; An analysis module is configured to input the sliding window sequence into an extended Kalman filter model, a time series network model, a physical model, and an electrochemical model, respectively; in the Kalman filter model, the voltage, current, and temperature in the sliding window are used as inputs, and state estimation is performed in combination with the internal battery equivalent circuit parameters to obtain a first initial SOC prediction sequence; in the time series network model, analysis is performed based on a multi-layer neural network layer and a fully connected layer, and a second initial SOC prediction sequence representing a characteristic vector of the SOC change trajectory is output; in the physical model, based on the transient response of voltage and current, an estimation is performed using a state equation, and the charge state is inversely solved according to the data in the sliding window to obtain a third initial SOC prediction sequence; in the electrochemical model, a simulation is performed to calculate the change in lithium ion concentration inside the current battery, and a fourth initial SOC prediction sequence is derived according to the change in lithium ion concentration inside the battery; the initial SOC prediction sequence output by each model is spliced ​​with the sliding window sequence, and a new feature sequence is formed at each time step of the sliding window to obtain a multi-dimensional fusion feature sequence; The encoding and decoding module is used to input the multidimensional fusion feature sequence into a deep neural network based on the attention mechanism, project the multidimensional fusion feature sequence to a preset dimension in the input layer of the deep neural network based on the attention mechanism, and perform time step position encoding on each feature in the feature sequence of the preset dimension to obtain a target feature sequence, calculate the attention weight between any two time steps in the target feature sequence in the encoder of the deep neural network based on the attention mechanism, perform local nonlinear mapping on the embedded features of each time step, extract key time segment features related to SOC prediction, use the key time segment features as queries in the decoder of the deep neural network based on the attention mechanism, analyze them with the global features output by the encoder, and fuse the analysis results, input the fused time segment features into the trend prediction module to predict the SOC change trend.

5. A high-precision SOC prediction device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and executable on the processor; When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images