Battery charging time determination method based on dual-depth prediction model, electronic equipment and medium
By combining a dual-depth prediction model for battery components and a battery component prediction model with self-supervised learning and converter timing prediction, the accuracy and efficiency issues of battery remaining charging time prediction are solved, achieving higher accuracy and more efficient battery charging time prediction.
Patent Information
- Application Number
- CN202511740351.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-02
- Publication Date
- 2026-01-13
AI Technical Summary
Existing battery remaining charging time prediction models suffer from insufficient accuracy and low efficiency.
A dual-depth prediction model-based approach is adopted. By using the trained first and second battery component prediction models, the multi-frame prediction temperature and current of the battery component are predicted respectively. Combined with the battery component voltage, the remaining charging time of the battery is determined. Self-supervised learning and a large-scale converter timing prediction model are used to improve the generalization ability.
It improves the accuracy and efficiency of predicting the remaining battery charging time, simplifies the control process, and solves the problems of complexity and low accuracy of iterative prediction schemes in existing technologies.
Smart Images
Figure CN121324971A_ABST
Abstract
Description
Cross-reference to related applications
[0001] This application is a divisional application of Chinese invention patent application filed on September 2, 2025, with application number 202511250062.2 and entitled "Method for predicting charging time of energy storage battery, electronic device and medium". Technical Field
[0002] This disclosure relates to the field of battery technology, and in particular to a method, electronic device, and medium for determining battery charging time based on a dual-depth prediction model. Background Technology
[0003] Traditional methods for predicting remaining battery charging time typically employ approaches such as segmented lookup tables, model-based methods, and data-driven methods. The segmented lookup table approach involves establishing a mapping between the battery system's temperature range, state of charge (SOC) range, and cumulative charging time. The charging time within each range is calculated, and these are summed to obtain the remaining charging time. The model-based approach involves constructing a thermoelectric coupling model of the battery and iteratively calling the model from the initial state until voltage cutoff or full charge is achieved, thereby estimating the remaining charging time. The data-driven approach involves extracting features and establishing a mapping between these features and the remaining charging time to predict the remaining charging time. Summary of the Invention
[0004] This disclosure provides a method, electronic device, and medium for determining battery charging time based on a dual-depth prediction model, which at least helps to solve the problems of insufficient accuracy and low efficiency of existing battery remaining charging time prediction models.
[0005] According to some embodiments of this disclosure, a first aspect of this disclosure provides a method for determining battery charging time based on a dual-depth prediction model. The dual-depth prediction model includes a first battery component prediction model and a second battery component prediction model. The first battery component prediction model is used to determine the multi-frame predicted temperature and multi-frame predicted current of the first battery component in a second time period. The second battery component prediction model is used to determine the multi-frame predicted voltage of the second battery component in the second time period based on the multi-frame predicted temperature and multi-frame predicted current of the first battery component in the second time period and the voltage of the second battery component. The multi-frame predicted voltage of the second battery component in the second time period is used to determine the remaining charging time of the battery. The first battery component includes multiple battery parts, and the battery part with the highest voltage among the multiple battery parts constitutes the second battery component. The battery parts are one of the following: battery pack, battery module, or battery cell. The method includes: acquiring a first training dataset, which is multiple frames of battery charging data of a first battery component within a third time period; extracting features from the first training dataset to obtain a second training dataset; training a prediction model for the first battery component based on the second training dataset; and acquiring a fourth training dataset, which is multiple frames of battery charging data of a second battery component within the first battery component within a fourth time period, wherein each frame of battery charging data in the fourth training dataset includes at least one of the following: voltage, current, and temperature of the second battery component; extracting features from the fourth training dataset to obtain a fifth training dataset; and training a prediction model for the second battery component based on the fifth training dataset.
[0006] According to some embodiments of this disclosure, a second aspect of this disclosure provides an electronic device, including: At least one processor; and, A memory that is communicatively connected to at least one processor; wherein, The memory stores instructions that can be executed by at least one processor to enable the at least one processor to perform the method described in the first aspect.
[0007] According to some embodiments of the present disclosure, a third aspect of the present disclosure provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the method described in the first aspect.
[0008] In some embodiments, the first battery component prediction model includes a first convolutional neural network and a first conversion network, wherein the first conversion network includes a first encoder and a first decoder; Feature extraction is performed on the first training dataset to obtain the second training dataset, which includes: The first training dataset is input into the first convolutional neural network for feature extraction to obtain the second training dataset; The prediction model for the first battery component is trained based on the second training dataset, including: The third time period is divided into a first sub-time period and a second sub-time period, and the second training dataset is divided into a first training data subset and a second training data subset. The data in the first training data subset is the battery charging data within the first sub-time period, and the data in the second training data subset is the battery charging data within the second sub-time period. The second sub-time period is after the first sub-time period. The first training data subset is input into the first encoder, the third training dataset is input into the first decoder, and the second training data subset is used as the label to train the first battery component prediction model. The third training dataset is the dataset after replacing the data in the second training data subset in the second training dataset with dummy data.
[0009] In some embodiments, the duration of the second sub-time period is variable.
[0010] In some embodiments, each frame of battery charging data in the first training dataset includes at least one of the following: the highest voltage of the first battery component, the lowest voltage of the first battery component, the current of the first battery component, the temperature of the first battery component, the heating power of the first battery component, and the cooling power of the first battery component.
[0011] In some embodiments, the second battery component prediction model includes a second convolutional neural network and a second conversion network, wherein the second conversion network includes a second encoder and a second decoder; Feature extraction is performed on the fourth training dataset to obtain the fifth training dataset, which includes: The fourth training dataset is input into the second convolutional neural network for feature extraction to obtain the fifth training dataset; The second battery component prediction model was trained based on the fifth training dataset, including: The fourth time period is divided into the third sub-time period and the fourth sub-time period. The fifth training dataset is divided into the third training data subset and the fourth training data subset. The data in the third training data subset is the battery charging data within the third sub-time period, and the data in the fourth training data subset is the battery charging data within the fourth sub-time period. The fourth sub-time period is after the third sub-time period. The third training data subset is input into the second encoder, the sixth training dataset is input into the second decoder, and the fourth training data subset is used as the label to train the second battery component prediction model. The sixth training dataset is the dataset after replacing the data in the fourth training data subset of the fifth training dataset with invalid data.
[0012] In some embodiments, the length of the fourth sub-time period can be variable.
[0013] In some embodiments, the fourth time period and the third time period are the same time period.
[0014] In some embodiments, the method further includes: Obtain the first dataset, which consists of multiple frames of battery charging data for the first battery component within a first time period; The first dataset is processed based on a dual-depth prediction model to obtain the multi-frame predicted voltage of the second battery component in the second time period. The first data frame is determined as the first data frame in the second time period in which the predicted voltage of the second battery component is greater than or equal to the charging cutoff voltage. The remaining charging time of the battery is determined based on the first data frame.
[0015] In some embodiments, the first dataset is processed based on a dual-depth prediction model to obtain multi-frame predicted voltages of the second battery component within a second time period, including: The first dataset is input into the prediction model of the first battery component to obtain the multi-frame predicted temperature and multi-frame predicted current of the first battery component in the second time period. The predicted temperature and current of the first battery module in the second time period and the voltage of the second battery module are input into the prediction model of the second battery module to obtain the predicted voltage of the second battery module in the second time period.
[0016] In some embodiments, the first time period and the second time period are consecutive.
[0017] In some embodiments, the remaining battery charging time is the difference between the time corresponding to the first data frame and the start time of the second time period.
[0018] In some embodiments, the inter-frame time slots within the second time period are equal, the first data frame is the z-th frame within the second time period, where z is a natural number, and the remaining battery charging time satisfies: Treserved = zT; where Treserved is the remaining battery charging time, and T is the inter-frame time slot within the second time period.
[0019] In some embodiments, the first battery assembly and the second battery assembly are one of the following combinations: the first battery assembly is a battery pack and the second battery assembly is a battery module; the first battery assembly is a battery module and the second battery assembly is a battery cell; or, the first battery assembly is a battery pack and the second battery assembly is a battery cell.
[0020] The technical solution provided in this disclosure has at least the following advantages: Based on the technical solution provided in the disclosed embodiments, a trained dual-depth prediction model can be used to determine the second battery component with the highest voltage in the first battery component using a first dataset composed of multiple frames of battery charging data of the first battery component within a first time period. The predicted voltage of the second battery component within a second time period after the first time period is then determined using the first data frame in the multiple predicted voltage frames that is greater than or equal to the charging cutoff voltage, i.e., the first data frame. This can solve the problems of insufficient accuracy and low efficiency in the prediction of the remaining battery charging time in existing systems, thereby improving accuracy and efficiency.
[0021] Furthermore, the problem of estimating the remaining charging time can be transformed into a problem of predicting the future voltage curve, and the problem of complex control flow and low accuracy in iterative prediction schemes can be solved by using a variable-scale training mode. Furthermore, self-supervised learning and a large-scale converter timing prediction model are used to solve the voltage prediction problem to improve generalization ability; Furthermore, the remaining charging time prediction problem is deconstructed into two non-end-to-end model problems: one is the first battery module prediction model, which predicts future current and temperature based on the highest voltage, lowest voltage, temperature, current, real-time heating power, and real-time cooling power; the other is the second battery module prediction model, which predicts future terminal voltage based on cell temperature, current, and voltage. This transforms the remaining charging time problem into a voltage prediction problem, thereby simplifying the problem and improving prediction efficiency and accuracy. Attached Figure Description
[0022] One or more embodiments are illustrated by way of example with corresponding pictures in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Unless otherwise stated, the pictures in the accompanying drawings do not constitute a limitation on scale. In order to more clearly illustrate the technical solutions in the embodiments of this disclosure or the conventional technology, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0023] Figure 1 A flowchart illustrating a method for predicting the charging time of an energy storage battery according to an embodiment of this disclosure; Figure 2 This is a schematic diagram of the structure of a dual-depth prediction model provided in an embodiment of the present disclosure; Figure 3 A schematic flowchart of another battery charging prediction method provided in an embodiment of this disclosure; Figure 4 This is a schematic diagram illustrating a scenario for predicting the remaining charging time of a battery, provided as an embodiment of this disclosure. Figure 5 A flowchart illustrating a training method for a first battery component prediction model provided in an embodiment of this disclosure; Figure 6 A flowchart illustrating another method for training a first battery component prediction model provided in an embodiment of this disclosure; Figure 7 An example diagram of a training dataset for training a prediction model of a first battery component, provided as an embodiment of this disclosure; Figure 8 A flowchart illustrating a training method for a second battery component prediction model provided in an embodiment of this disclosure; Figure 9 A flowchart illustrating another method for training a second battery component prediction model provided in an embodiment of this disclosure; Figure 10 Example diagram of a dataset for training a prediction model of a second battery component provided in an embodiment of this disclosure; Figure 11 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. Detailed Implementation
[0024] As the background technology shows, segmented lookup table methods cannot adapt to complex operating conditions, and the accuracy of the calculation results is limited. Model-based solutions require calibration of the parameters of the relevant thermoelectric coupling model. Once the operating conditions change, recalibration is required, otherwise the accuracy will be affected and the generalization ability will be poor. Data-driven solutions generally directly establish the mapping relationship between features and remaining charging time. The accuracy and the coverage of operating conditions are directly related to the number of labeled samples. However, the process of constructing labels and samples is time-consuming and resource-intensive, resulting in low training efficiency. In other words, existing battery remaining charging time prediction models have the problems of insufficient accuracy and low efficiency.
[0025] To address this issue, this disclosure provides a method for predicting the remaining charging time of an energy storage battery. Based on a first dataset composed of multiple frames of battery charging data from a first battery module within a first time period, a trained dual-depth prediction model is used to determine the second battery module with the highest voltage among the first battery modules. The method then predicts the voltage of multiple frames within a second time period following the first time period. Finally, based on the first data frame (i.e., the first data frame) that is greater than or equal to the charging cutoff voltage from these multiple predicted voltage frames, the remaining charging time is accurately determined. This method solves the problems of insufficient accuracy and low efficiency in existing battery remaining charging time prediction methods, thereby improving both accuracy and efficiency.
[0026] In the description of the embodiments of this disclosure, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary or secondary relationship of the indicated technical features. In the description of the embodiments of this disclosure, "multiple" means two or more, unless otherwise explicitly defined. Similarly, "multiple sets" refers to two or more sets (including two sets), and "multiple pieces" refers to two or more pieces (including two pieces).
[0027] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this disclosure. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0028] In the description of the embodiments of this disclosure, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A exists, A and B exist simultaneously, and B exists. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.
[0029] In the description of the embodiments of this disclosure, when a component "includes" another component, other components are not excluded unless otherwise stated, and may be further included. The formation or placement of a second component above or on a first component, or on the surface of a first component, or on one side of a first component, may include embodiments where the first and second components are in direct contact, and may also include embodiments where an additional component may be placed between the first and second components, thereby preventing direct contact between the first and second components. For simplicity and clarity, various components may be drawn at different scales. In the drawings, some layers / components may be omitted for simplicity. Unless otherwise specified, the formation or placement of a second component on the surface of a first component refers to direct contact between the first and second components. The term "component" can refer to a layer, film, region, portion, structure, etc.
[0030] The terminology used in the description of the various embodiments herein is for the purpose of describing particular embodiments only and is not intended to be limiting.
[0031] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been provided in the embodiments of this disclosure to facilitate a better understanding of the disclosure. However, the technical solutions claimed in this disclosure can be implemented even without these technical details and various variations and modifications based on the following embodiments.
[0032] Figure 1 This is a flowchart illustrating a method for predicting the charging time of an energy storage battery according to an embodiment of this disclosure. Figure 1 As shown, the method includes: S101, Obtain the first dataset.
[0033] In some embodiments, the first dataset is multi-frame battery charging data of the first battery assembly within a first time period. The first battery assembly includes multiple battery components, and the battery component with the highest voltage among the multiple battery components constitutes the second battery assembly. The battery component is one of the following: a battery pack, a battery module, or a cell.
[0034] The first battery module and the second battery module are both components that make up the energy storage battery.
[0035] In some embodiments, the second battery assembly may be a component of the first battery assembly, meaning that one first battery assembly may include multiple second battery assemblies. For example, the first battery assembly may be a battery pack, and the second battery assembly may be a battery module; a battery pack may include one or more battery modules. As another example, the first battery assembly may be a battery module, and the second battery assembly may be a battery cell; a battery module may include one or more battery cells. Yet another example, the first battery assembly may be a battery pack, and the second battery assembly may be a battery cell; a battery pack may include one or more battery cells.
[0036] In other embodiments, the second battery assembly and the first battery assembly may also be battery components of the same type. In this case, the first battery assembly may be a collection of multiple battery components of the same type, and the second battery assembly may be the battery component in the collection that meets specific conditions (such as the highest voltage and / or the highest temperature). That is, the second battery assembly may be a battery component selected from the collection of the first battery assemblies. For example, the first battery assembly may be a collection of multiple battery modules, such as multiple battery modules belonging to the same battery pack, and the second battery assembly may be the battery module with the highest voltage in the collection. As another example, the first battery assembly may be a collection of multiple battery cells, such as multiple battery cells belonging to the same battery module or battery pack, and the second battery assembly may be the battery cell with the highest voltage in the collection.
[0037] It should be noted that a battery module can also be called a battery cluster, and a battery cell can also be called a single cell. That is, the names of the above-mentioned battery components are not limited in this embodiment.
[0038] In some embodiments, the energy storage battery may include one or more battery packs, a battery pack may include one or more battery modules, and a battery module may include one or more battery cells. In other embodiments, the energy storage battery may not include battery modules, such as a cell-to-pack (CTP) battery, where the energy storage battery may include one or more battery packs, and a battery pack may include one or more battery cells. This disclosure does not limit the specific structure of the battery.
[0039] The first time period mentioned above can be a period of time before a certain moment in the energy storage battery charging process, such as from the start of charging to the current moment. The first dataset can be multiple frames of battery charging data collected within the first time period. Each frame of battery charging data can include at least one of the following: the highest voltage of the first battery component, the lowest voltage of the first battery component, the current of the first battery component, the temperature of the first battery component, the heating power of the first battery component, and the cooling power of the first battery component.
[0040] In some embodiments, the state of the first battery module can be monitored periodically, and charging data of the first battery module can be collected. For example, data on the state of the first battery module can be collected every 15 seconds, 30 seconds, and 60 seconds, and the collected data each time constitutes a frame of battery charging data. In other words, the data collection interval between two adjacent frames of battery charging data is the inter-frame time slot between the two frames of battery charging data. Therefore, charging-related time quantities, such as the time consumed to charge from 0 to 80% and the remaining charging time, can be calculated based on the frame number and inter-frame time slot of the battery charging data.
[0041] It should be noted that the inter-frame time slot can be fixed, such as 30 seconds / time or 15 seconds / time, or it can be dynamically adjusted, such as adaptively adjusting the inter-frame time slot based on the monitored temperature and voltage of the first battery component. For example, when the voltage or temperature in the first battery component is greater than or equal to a certain threshold, such as voltage ≥ 4V or temperature ≥ 35℃, the inter-frame time slot can be reduced to increase the sampling frequency, such as increasing the data collection from one frame every 30 seconds to one frame every 15 seconds, in order to collect more dense battery charging data. Based on the more dense battery charging data, more precise control can be exercised over the battery charging process, thereby improving charging efficiency and safety.
[0042] The collected multi-frame battery charging data can be used as a dataset, such as the first dataset, to monitor and evaluate the battery charging process and status, such as predicting the remaining charging time and generating control commands. The control commands are used to control the battery charging process, such as increasing or decreasing charging power, or increasing or decreasing cooling power.
[0043] S102, based on the dual-depth prediction model, the first dataset is processed to obtain the multi-frame predicted voltage of the second battery component in the second time period.
[0044] In some embodiments, the second time period is a period of time following the first time period.
[0045] The dual-depth prediction model can be a pre-trained artificial intelligence neural network model. Optionally, Figure 2 This is a schematic diagram of the structure of a dual-depth prediction model provided in an embodiment of this disclosure. Figure 2 As shown, the dual-depth prediction model includes a first battery component prediction model and a second battery component prediction model. The first battery component prediction model is used to determine the multi-frame predicted temperature and multi-frame predicted current of the first battery component in the second time period based on the first dataset. The second battery component prediction model is used to determine the multi-frame predicted voltage of the second battery component in the second time period based on the multi-frame predicted temperature and multi-frame predicted current of the first battery component in the second time period, as well as the voltage of the second battery component.
[0046] S103, determine the first data frame.
[0047] In some embodiments, the first data frame is the first data frame in which the predicted voltage of the second battery component is greater than or equal to the charging cutoff voltage during a second time period.
[0048] S104, determine the remaining charging time of the battery based on the first data frame.
[0049] In some embodiments, the remaining battery charging time is the difference between the time corresponding to the first data frame and the start time of the second time period.
[0050] Optionally, Figure 3 This is a schematic flowchart illustrating another battery charging prediction method provided in an embodiment of this disclosure. (In conjunction with...) Figure 1 ,like Figure 3 As shown, S102, based on the dual-depth prediction model, the first dataset is processed to obtain the multi-frame predicted voltage of the second battery component in the second time period, which may include: S1021, Input the first dataset into the prediction model of the first battery component to obtain the multi-frame predicted temperature and multi-frame predicted current of the first battery component in the second time period.
[0051] S1022, input the multi-frame predicted temperature and multi-frame predicted current of the first battery module in the second time period, and the voltage of the second battery module into the prediction model of the second battery module to obtain the multi-frame predicted voltage of the second battery module in the second time period.
[0052] The following is combined Figure 4 Further explanation of the specific implementation of S102-S104.
[0053] Optionally, Figure 4 This is a schematic diagram illustrating a scenario for predicting the remaining charging time of a battery, as provided in an embodiment of this disclosure. Figure 4 As shown, t is the charging time. The first time period includes m charging data frames, ending at t1. The second time period includes b charging data frames, starting at t2. The time slot between data frames is T. m and b are positive integers greater than 1, where b is much larger than m (e.g., b is more than 5 times or 10 times m). The frame numbers are 1, 2, 3, ..., b. The start times of each frame are t2, t2+T, t2+2T, t2+3T, ..., t2+zT, ..., t2+(b-1)T. The end times of each frame are t2+T, t2+2T, t2+3T, ..., t2+zT, ..., t2+bT. Therefore: In S1021, m charging data frames of the first battery component in the first time period are input into the prediction model of the first charging component to obtain the predicted temperature and multi-frame predicted current of b charging data frames of the first battery component in the second time period, and the battery component with the highest voltage is selected from the first battery component as the second battery component.
[0054] In S1022, the predicted temperature of the above b charging data frames, the predicted current of multiple frames, and the voltage of the second battery module are input into the prediction model of the second battery module to obtain the predicted voltage of the second battery module in the b frames during the second time period.
[0055] In S103, the first data frame with a predicted voltage greater than or equal to the charging cutoff voltage among the above b charging data frames is determined as the first data frame, such as the z-th data frame. That is, the first data frame is the first data frame with a predicted voltage greater than or equal to the charging cutoff voltage among the b charging data frames.
[0056] In S104, the remaining charging time Treserved is calculated based on the start time t2 of the first data frame and the second time period, i.e., Treserved = t3 - t2, where t3 is the time corresponding to the first data frame.
[0057] Further, the inter-frame time slots T within the second time period are equal, for example, all 30 seconds per time, to simplify the calculation and improve the prediction efficiency. Assume that the first data frame is the z-th frame within the second time period, where z is a natural number and 0 ≤ z < b. Then the remaining charging time of the battery satisfies: Treserved = zT, where Treserved is the remaining charging time of the battery and T is the inter-frame time slot within the second time period.
[0058] In some embodiments, please continue to refer to Figure 4 , the first time period and the second time period may not be continuous, that is, t2 > t1, or may be continuous, that is, t2 = t1. In other words, the battery charging data as close as possible in time can be used to predict the remaining battery charging time, so as to make full use of the temporal correlation of the battery charging data and further improve the prediction accuracy of the remaining battery charging time.
[0059] Based on the technical solution provided by the embodiments of the present disclosure, the second battery component with the highest voltage in the first battery component can be determined using the trained double-depth prediction model according to the first data set composed of multiple frames of battery charging data of the first battery component within the first time period, and the predicted voltages of multiple frames within the second time period after the first time period. Then, according to the first data frame (the first data frame) in the multiple frames of predicted voltages that is greater than or equal to the charging cut-off voltage for the first time, the remaining battery charging time can be accurately determined, which can solve the problem that the existing battery remaining charging time prediction scheme cannot兼顾 both accuracy and efficiency.
[0060] Moreover, the remaining charging time estimation problem can be transformed into a future voltage curve prediction problem, and the variable-scale training mode can be used to solve the problems of complex control flow and low accuracy existing in the iterative prediction scheme; Further, self-supervised learning and the Transformer time series prediction large model are used to solve the voltage prediction problem to improve the generalization ability; In addition, the remaining charging time prediction problem is deconstructed into a non-end-to-end two-model problem: one is the first battery component prediction model, which predicts the future current and temperature from the highest voltage, lowest voltage, temperature, current, heating real-time power, cooling real-time power, etc.; the other is the second battery component prediction model, which predicts the future terminal voltage from the cell temperature, current, and voltage, transforming the remaining charging time problem into a voltage prediction problem, thereby simplifying the problem and improving the prediction efficiency and accuracy.
[0061] Next, in combination with Figures 5-8 , the training method of the double-depth prediction model will be described in detail.
[0062] It should be noted that the first battery component prediction model and the second battery component prediction model in the double-depth prediction model can be trained separately or together, and the embodiments of the present disclosure do not limit this.
[0063] When the prediction model for the first battery module and the prediction model for the second battery module are trained separately, training datasets can be collected and training can be completed separately according to the training requirements of the two models, without affecting each other, thereby improving the flexibility and efficiency of training.
[0064] When the first battery component prediction model and the second battery component prediction model are trained together, they can be trained using the same training dataset. For example, the training dataset can be input into the first battery component prediction model for training, and the training results of the first battery component prediction model can be used to train the second battery component prediction model. Since the two models are trained using the same training dataset, they have a strong temporal correlation, which can improve the accuracy of the models.
[0065] In practical applications, the first battery component prediction model and the second battery component prediction model can be trained separately using their respective training datasets, and then the cascaded first battery component prediction model and the second battery component prediction model can be trained using the same training dataset, so as to balance training accuracy and efficiency.
[0066] Figure 5 This is a flowchart illustrating a training method for a first battery component prediction model provided in one embodiment of this disclosure. Figure 5 As shown, the method includes: S501, Obtain the first training dataset.
[0067] The first training dataset consists of multiple frames of battery charging data for the first battery component during the third time period.
[0068] In some embodiments, each frame of battery charging data in the first training dataset includes at least one of the following: the highest voltage of the first battery component, the lowest voltage of the first battery component, the current of the first battery component, the temperature of the first battery component, the heating power of the first battery component, and the cooling power of the first battery component.
[0069] The third time period is a historical time period. The first training dataset includes multi-frame battery charging data of the first battery component collected in the third time period. The content and collection method of the first training dataset can be referred to the first dataset, which will not be repeated here.
[0070] S502, extract features from the first training dataset to obtain the second training dataset.
[0071] For example, the first training dataset can be input into a convolutional neural network for feature extraction, such as extracting the time-related features of each frame of battery charging data in the first training dataset to obtain the second training dataset.
[0072] S503, trains the prediction model for the first battery component based on the second training dataset.
[0073] In some embodiments, the first battery component prediction model includes a first convolutional neural network and a first conversion network, the first conversion network including a first encoder and a first decoder.
[0074] Optionally, Figure 6 This is a schematic flowchart illustrating a training method for another first battery component prediction model provided in an embodiment of this disclosure. (Combined with...) Figure 5 ,like Figure 6 As shown, the method includes: In S502, feature extraction is performed on the first training dataset to obtain the second training dataset, which includes: S5021, Input the first training dataset into the first convolutional neural network for feature extraction to obtain the second training dataset.
[0075] In S503, a prediction model for the first battery component is trained based on the second training dataset, including: S5031, the third time period is divided into the first sub-time period and the second sub-time period, and the second training dataset is divided into the first training data subset and the second training data subset.
[0076] In some embodiments, the data in the first training data subset is battery charging data within a first sub-time period, and the data in the second training data subset is battery charging data within a second sub-time period, wherein the second sub-time period is after the first sub-time period.
[0077] S5032, the first training data subset is input into the first encoder, the third training dataset is input into the first decoder, and the second training data subset is used as the label to train the first battery component prediction model.
[0078] In some embodiments, the third training dataset is a dataset in which data from the second training data subset of the second training dataset is replaced with dummy data.
[0079] Optionally, Figure 7 This is an example diagram of a dataset provided for training a prediction model for a first battery component, according to an embodiment of this disclosure. (Combined with...) Figure 6 ,like Figure 7 As shown, m1+n1 data frames correspond to the aforementioned third time period, where m1 data frames correspond to the first sub-time period within the third time period, and n1 data frames correspond to the second sub-time period within the third time period. The first training dataset consists of m1+n1 original data frames collected during the third time period, and the second training dataset consists of m1+n1 feature data frames obtained after inputting the m1+n1 original data frames into the first convolutional neural network for feature extraction. Figure 7 The first training data subset, labeled A10, includes m1 feature data frames. Figure 7 The middle part is labeled A11), and the second training data subset consists of n1 feature data frames ( Figure 7 (The data is labeled A12). Then, A11 is input into the first encoder, and the A12 portion in A10 is replaced with dummy data and input into the first decoder. A12 is used as a label to compare with the forward training results of the first encoder and the first decoder to generate output error and gradient, and to update the weights of the first battery component prediction model during the reverse training process, thereby completing the training.
[0080] It should be noted that in this embodiment of the disclosure, "dumb data" can also be called blank data, invalid data, or padding data. It refers to data that is filled or replaced to meet a specific data format and does not participate in subsequent calculations or have no impact on the calculation results.
[0081] As can be seen, the training data includes at least six charging data points, such as maximum voltage, minimum voltage, current, temperature, real-time heating power, and real-time cooling power. This allows the first battery component prediction model to learn information such as the battery's thermal management strategy and charging strategy. Furthermore, since the training data does not need to distinguish between charging, discharging, or resting scenarios, it can be used as both input and output to the model. The large data volume can improve the training accuracy of the first battery component prediction model, thereby improving the prediction accuracy of the remaining battery charging time. For example, the training dataset can be divided as follows: 98% for the training set, 1% for the validation set, and 1% for the test set.
[0082] In some embodiments, please refer to Figure 7 The length of the second sub-time period is variable, such as n1, which can be random or configurable between [a1, b1], where a1>0 and b1 is much larger than m1, such as b1 being more than 5 times or 10 times greater than m1. That is, the present invention adopts a variable scale mode for training to solve the problems of complex process and low accuracy of repeated iterative prediction, so as to further improve efficiency and accuracy.
[0083] Optionally, Figure 8 This is a flowchart illustrating a training method for a second battery module prediction model provided in one embodiment of this disclosure. Figure 8 As shown, the method includes: S801, Obtain the fourth training dataset.
[0084] In some embodiments, the fourth training dataset is multi-frame battery charging data of the second battery component in the first battery component during a fourth time period.
[0085] In some embodiments, each frame of battery charging data in the fourth training dataset includes at least one of the following: the voltage, current, and temperature of the second battery component.
[0086] S802, feature extraction is performed on the fourth training dataset to obtain the fifth training dataset.
[0087] For example, the fourth training dataset can be input into a convolutional neural network for feature extraction, such as extracting the time-related features of each frame of battery charging data in the fourth training dataset to obtain the fifth training dataset.
[0088] S803 trains a prediction model for the second battery component based on the fifth training dataset.
[0089] In some embodiments, the second battery component prediction model includes a second convolutional neural network and a second conversion network, wherein the second conversion network includes a second encoder and a second decoder.
[0090] Optionally, Figure 9 This is a flowchart illustrating another method for training a second battery component prediction model according to an embodiment of this disclosure. (Combined with...) Figure 8 ,like Figure 9 As shown, the method includes: In S802, feature extraction is performed on the fourth training dataset to obtain the fifth training dataset, which includes: S8021, input the fourth training dataset into the second convolutional neural network for feature extraction to obtain the fifth training dataset.
[0091] In S803, a prediction model for the second battery component is trained based on the fifth training dataset, including: S8031 divides the fourth time period into the third sub-time period and the fourth sub-time period, and divides the fifth training dataset into the third training data subset and the fourth training data subset.
[0092] In some embodiments, the data in the third training data subset is battery charging data within the third sub-time period, and the data in the fourth training data subset is battery charging data within the fourth sub-time period, which is after the third sub-time period.
[0093] S8032, input the third training data subset into the second encoder, input the sixth training dataset into the second decoder, and use the fourth training data subset as labels to train the second battery component prediction model.
[0094] In some embodiments, the sixth training dataset is the dataset after replacing the data in the fourth training data subset of the fifth training dataset with invalid data.
[0095] Optionally, Figure 10This is an example diagram of a dataset provided for training a prediction model for a second battery component, as shown in one embodiment of this disclosure. (Combined with...) Figure 9 ,like Figure 10 As shown, m² + n² data frames correspond to the fourth time period mentioned above, where m² data frames correspond to the third sub-time period within the fourth time period, and n² data frames correspond to the fourth sub-time period within the fourth time period. The fourth training dataset consists of m² + n² raw data frames collected during the fourth time period, and the fifth training dataset consists of m² + n² feature data frames obtained after inputting the m² + n² raw data frames into the second convolutional neural network for feature extraction. Figure 10 The third training data subset, labeled A20, includes m2 feature data frames. Figure 10 The fourth training data subset, marked as A21, consists of n2 feature data frames. Figure 10 (The data is labeled A22). Then, A21 is input into the second encoder, and the A22 portion in A20 is replaced with dummy data and input into the second decoder. A22 is used as a label to compare with the forward training results of the second encoder and the second decoder to generate output error and gradient, and to update the weights of the second battery component prediction model during the backward training process, thereby completing the training.
[0096] As can be seen, the training data includes at least three charging data points for the second battery module, such as voltage, current, and temperature. This allows the prediction model for the second battery module to learn information such as the battery's thermal management strategy and charging strategy. Furthermore, since the training data does not need to distinguish between charging, discharging, or resting scenarios, it can be used as both input and output to the model. The large amount of data can improve the training accuracy of the prediction model for the second battery module, thereby improving the accuracy of predicting the remaining battery charging time. For example, the training dataset can be divided as follows: 98% for the training set, 1% for the validation set, and 1% for the test set.
[0097] In some embodiments, please refer to Figure 10 The length of the fourth sub-time period is variable, such as n2, which can be random or configurable between [a2, b2], where a2>0 and b2 is much larger than m2, such as b2 being more than 5 times or 10 times greater than m2. That is, the present invention adopts a variable scale mode for training to solve the problems of complex process and low accuracy of repeated iterative prediction, so as to further improve efficiency and accuracy.
[0098] In some embodiments, the fourth time period and the third time period are the same time period, such as m1=m2, n1=n2, a1=a2, b1=b2. In this case, A10=A20, A11=A21, A12=A22. That is, the same raw data can be used when training the first battery component prediction model and the second battery component model. For example, the raw charging data collected in the same time period can be used simultaneously or sequentially for training the first battery component prediction model and the second battery component model, so as to take advantage of the inherent time correlation of the raw charging data collected in the same time period and improve the accuracy and efficiency of model training.
[0099] In some embodiments, the first battery assembly and the second battery assembly satisfy one of the following: The first battery component can be a battery pack, and the second battery component can be a battery module; or... The first battery assembly can be a battery module, and the second battery assembly can be a battery cell; or... The first battery component can be a battery pack, and the second battery component can be a battery cell.
[0100] Figure 11 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this disclosure. Figure 11 As shown, the electronic device includes: at least one processor 1101; and a memory 1102 communicatively connected to at least one processor 1101; wherein the memory 1102 stores instructions executable by at least one processor 1101, the instructions being executed by at least one processor 1101 to enable at least one processor 1101 to perform the method described in the above method embodiments.
[0101] The memory 1102 and processor 1101 can be connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors 1101 and memory 1102. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface can provide an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 1101 is transmitted over a wireless medium via an antenna, which further receives data and transmits it to processor 1101.
[0102] Processor 1101 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory 1102 can be used to store data used by processor 1101 during operation.
[0103] In some embodiments, the electronic device may be an energy storage device, such as an energy storage cabinet. This energy storage device can be part of a larger energy storage system, or it can be a stand-alone energy storage device, such as a small household energy storage device. It can also be a sub-device or energy storage battery component or module in other devices, such as a power battery pack installed in an electric vehicle. In other embodiments, the electronic device may also be a charging chip or chip system installed within the aforementioned devices.
[0104] One embodiment of this disclosure provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the method described in the first aspect.
[0105] Those skilled in the art will understand that the above embodiments are specific examples of implementing this disclosure, and in practical applications, various changes in form and detail may be made without departing from the spirit and scope of this disclosure. Any person skilled in the art can make various alterations and modifications without departing from the spirit and scope of this disclosure; therefore, the scope of protection of this disclosure should be determined by the scope defined in the claims.
Claims
1. A method for determining battery charging time based on a dual-depth prediction model, characterized in that, The dual-depth prediction model includes a first battery component prediction model and a second battery component prediction model. The first battery component prediction model is used to determine the multi-frame predicted temperature and multi-frame predicted current of the first battery component within a second time period. The second battery component prediction model is used to determine the multi-frame predicted voltage of the second battery component within the second time period based on the multi-frame predicted temperature and multi-frame predicted current of the first battery component within the second time period, and the voltage of the second battery component. The multi-frame predicted voltage of the second battery component within the second time period is used to determine the remaining charging time of the battery. The first battery component includes multiple battery parts, and the battery part with the highest voltage among the multiple battery parts constitutes the second battery component. The battery part is one of the following: a battery pack, a battery module, or a battery cell. The method includes: Obtain a first training dataset, which is multi-frame battery charging data of the first battery component within a third time period; Feature extraction is performed on the first training dataset to obtain the second training dataset; The first battery component prediction model is trained based on the second training dataset; and Obtain a fourth training dataset, which is a multi-frame battery charging data of the second battery component in the first battery component within a fourth time period. Each frame of battery charging data in the fourth training dataset includes at least one of the following: the voltage, current and temperature of the second battery component. Feature extraction is performed on the fourth training dataset to obtain the fifth training dataset; The second battery component prediction model is trained based on the fifth training dataset.
2. The method according to claim 1, characterized in that, The first battery component prediction model includes a first convolutional neural network and a first conversion network, wherein the first conversion network includes a first encoder and a first decoder; The step of extracting features from the first training dataset to obtain the second training dataset includes: The first training dataset is input into the first convolutional neural network for feature extraction to obtain the second training dataset; The step of training the first battery component prediction model based on the second training dataset includes: The third time period is divided into a first sub-time period and a second sub-time period, and the second training dataset is divided into a first training data subset and a second training data subset. The data in the first training data subset is the battery charging data within the first sub-time period, and the data in the second training data subset is the battery charging data within the second sub-time period. The second sub-time period is after the first sub-time period. The first training data subset is input into the first encoder, the third training dataset is input into the first decoder, and the second training data subset is used as the label to train the first battery component prediction model. The third training dataset is the dataset after replacing the data in the second training data subset in the second training dataset with dummy data.
3. The method according to claim 2, characterized in that, The length of the second sub-time period is variable.
4. The method according to claim 1, characterized in that, In the first training dataset, each frame of battery charging data includes at least one of the following: the highest voltage of the first battery component, the lowest voltage of the first battery component, the current of the first battery component, the temperature of the first battery component, the heating power of the first battery component, and the cooling power of the first battery component.
5. The method according to claim 4, characterized in that, The second battery component prediction model includes a second convolutional neural network and a second conversion network, wherein the second conversion network includes a second encoder and a second decoder; The step of extracting features from the fourth training dataset to obtain the fifth training dataset includes: The fourth training dataset is input into the second convolutional neural network for feature extraction to obtain the fifth training dataset; The step of training the second battery component prediction model based on the fifth training dataset includes: The fourth time period is divided into a third sub-time period and a fourth sub-time period, and the fifth training dataset is divided into a third training data subset and a fourth training data subset. The data in the third training data subset is the battery charging data within the third sub-time period, and the data in the fourth training data subset is the battery charging data within the fourth sub-time period. The fourth sub-time period is after the third sub-time period. The third training data subset is input into the second encoder, the sixth training dataset is input into the second decoder, and the fourth training data subset is used as the label to train the second battery component prediction model. The sixth training dataset is the dataset after replacing the data in the fourth training data subset in the fifth training dataset with invalid data.
6. The method according to claim 5, characterized in that, The length of the fourth sub-time period is variable.
7. The method according to claim 1, characterized in that, The fourth time period and the third time period are the same time period.
8. The method according to claim 1, characterized in that, Also includes: Obtain the first dataset, which is multi-frame battery charging data of the first battery component within a first time period; The first dataset is processed based on the dual-depth prediction model to obtain the multi-frame predicted voltage of the second battery component in the second time period. A first data frame is determined, wherein the first data frame is the first data frame in which the predicted voltage of the second battery component is greater than or equal to the charging cutoff voltage during the second time period; The remaining charging time of the battery is determined based on the first data frame.
9. The method according to claim 8, characterized in that, The process of processing the first dataset based on a dual-depth prediction model to obtain the multi-frame predicted voltage of the second battery component within a second time period includes: Input the first dataset into the first battery component prediction model to obtain the multi-frame predicted temperature and multi-frame predicted current of the first battery component in the second time period. The predicted temperature and current of the first battery module in the second time period and the voltage of the second battery module are input into the prediction model of the second battery module to obtain the predicted voltage of the second battery module in the second time period.
10. The method according to claim 8, characterized in that, The first time period and the second time period are consecutive.
11. The method according to claim 8, characterized in that, The remaining battery charging time is the difference between the time corresponding to the first data frame and the start time of the second time period.
12. The method according to claim 11, characterized in that, The inter-frame time slots within the second time period are equal, the first data frame is the z-th frame within the second time period, where z is a natural number, and the remaining battery charging time satisfies: Treserved = zT; Where Treserved represents the remaining charging time of the battery, and T represents the inter-frame time slot within the second time period.
13. The method according to any one of claims 1-12, characterized in that, The first battery assembly and the second battery assembly are one of the following combinations: The first battery assembly is a battery pack, and the second battery assembly is a battery module; or... The first battery assembly is a battery module, and the second battery assembly is a battery cell; or, The first battery component is a battery pack, and the second battery component is a battery cell.
14. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1-13.
15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the method of any one of claims 1-13.