WO2023226358A1

WO2023226358A1 - Prediction method for state of charge of energy storage battery pack

Info

Publication number: WO2023226358A1
Application number: PCT/CN2022/137079
Authority: WO
Inventors: 郭媛君; 周邦昱; 刘祥飞; 姚文娇; 杨之乐; 胡天宇
Original assignee: 深圳先进技术研究院
Priority date: 2022-05-27
Filing date: 2022-12-06
Publication date: 2023-11-30
Also published as: CN114966413B; CN114966413A

Abstract

A prediction method for a state of charge of an energy storage battery pack. A target prediction model is trained by combining a machine learning method and an input and output iteration method, and because the target prediction model uses a combination of a plurality of training methods, the precision is high, and the state of charge of the energy storage battery pack can be predicted online in real time, thereby solving the problem that existing measurement methods for the state of charge of a battery are difficult to satisfy the requirements for measurement time and measurement precision in a real driving environment.

Description

A state-of-charge prediction method for energy storage battery packs

Technical field

The present invention relates to the field of battery detection, and in particular to a method for predicting the state of charge of an energy storage battery pack.

Background technique

The battery state of charge is called SOC (State of Charge), which is used to reflect the actual number of charges (unit: ampere hours) existing in the energy storage medium during the electrochemical energy storage process in the energy storage medium corresponding to the rated energy storage capacity. The percentage of charge contained in ampere hours. At present, the main measurement methods of SOC include open circuit voltage method, ampere-hour integration method, etc. However, the open-circuit voltage method requires the battery to stand for a long time to achieve voltage stability, which usually takes several hours or even more than ten hours, and the measurement time cost is large; the ampere-hour integration method is easily affected by the current measurement accuracy and has a cumulative effect. error. At present, new energy electric vehicles have become popular. Accurately understanding the battery state of charge of new energy electric vehicles can help drivers make travel plans. However, the existing battery state of charge measurement methods cannot meet the requirements of measurement time and measurement in real driving environments. Accuracy requirements.

Therefore, the existing technology still needs to be improved and developed.

technical problem

The technical problem to be solved by the present invention is to provide a state-of-charge prediction method for an energy storage battery pack in view of the above-mentioned defects of the prior art, aiming to solve the problem that the existing battery state-of-charge measurement method is difficult to meet the needs of real driving environments. Regarding measurement time and measurement accuracy requirements.

Technical solutions

The technical solutions adopted by the present invention to solve the problem are as follows:

In a first aspect, embodiments of the present invention provide a method for predicting the state of charge of an energy storage battery pack, wherein the method includes:

Obtain the battery data corresponding to the first energy storage battery pack;

Obtain a target prediction model, input the battery data into the target prediction model, and obtain the state-of-charge data corresponding to the first energy storage battery pack;

The target prediction model is a pre-trained model, and the training process of the target prediction model includes:

Obtain a first training data set, wherein the first training data set includes a plurality of first training data, a plurality of the first training data respectively correspond to different time periods, and each of the first training data includes a first input data and The historical state of charge data corresponding to the first input data. The first input data is the historical battery data of the second energy storage battery pack in the time period corresponding to the first training data. The historical state of charge data is used When reflecting the state of charge of the second energy storage battery pack in the time period corresponding to the first training data, the battery pack types corresponding to the first energy storage battery pack and the second energy storage battery pack are the same;

Obtain a first prediction model, wherein the first prediction model is an untrained model;

The first prediction model is trained according to the first training data set, and after the training is completed, a second prediction model and a plurality of predicted state-of-charge data corresponding to the first training data are obtained;

A second training data set is determined according to the predicted state of charge data corresponding to the first training data set and a plurality of the first training data, wherein the second training data set includes a plurality of second training data, and a plurality of the second training data. The second training data has a one-to-one correspondence with a plurality of the first training data. Each second training data includes second input data and historical state-of-charge data corresponding to the second input data. The second input data The data includes first input data corresponding to the second training data and predicted state of charge data corresponding to the first input data, and historical state of charge data corresponding to the second input data and first first input data corresponding to the second input data. The historical state-of-charge data of the input data are the same;

Obtain a third prediction model, wherein the third prediction model has the same structure as the second prediction model;

The third prediction model is trained according to the second training data set, and the target prediction model is obtained after training.

In one implementation, the first prediction model includes a plurality of first prediction models, and the plurality of first prediction models respectively correspond to different hyperparameter combinations. The first prediction model is configured according to the first training data set. Predictive models are trained, including:

Train a plurality of the first prediction models respectively according to the first training data set, and obtain a plurality of prediction models respectively corresponding to the first prediction models after the training is completed;

Obtain a test data set, wherein the test data set and the first training data set are generated based on the same data set;

According to the test data set, determine the prediction accuracy corresponding to several of the prediction models;

The second prediction model is determined based on the prediction model with the highest prediction accuracy.

In one implementation, several methods for determining the first prediction models include:

Get the preset hyperparameter value range;

Traverse the hyperparameter value range according to the preset step size to obtain several hyperparameter combinations;

One of the first prediction models is determined according to each of the hyperparameter combinations.

In one implementation, the preset step size includes several step sizes, and the hyperparameter value range is traversed according to the preset step size to obtain several hyperparameter combinations, including:

Obtain a plurality of step sizes, wherein each of the step sizes corresponds to different accuracy intervals, and the size of the accuracy interval corresponding to each step size is inversely proportional to the step size;

Determine the prediction accuracy of the hyperparameter combination corresponding to the previous round of search, and determine the target step size from several step sizes according to the prediction accuracy of the hyperparameter combination corresponding to the previous round of search, wherein the previous round of search corresponds to The prediction accuracy of the hyperparameter combination is located in the accuracy interval corresponding to the target step size;

Execute the current round of search according to the target step size, and obtain the hyperparameter combination corresponding to the current round of search;

Repeat the step of determining the prediction accuracy of the hyperparameter combination corresponding to the previous round of search until the hyperparameter value range is traversed.

In one implementation, training several first prediction models respectively according to the first training data set includes:

For each of the first prediction models, input the first input data in the first training data set into the first prediction model to obtain the predicted state of charge data corresponding to the first input data;

Determine the first loss function value corresponding to the first prediction model according to the predicted state of charge data and historical state of charge data corresponding to the first input data;

Adjust the model parameters of the first prediction model according to the first loss function value, and continue to perform the step of inputting the first input data in the first training data set into the first prediction model until the predetermined Set training conditions to obtain the prediction model corresponding to the first prediction model.

In one implementation, the third prediction model is trained according to the second training data set, and the target prediction model is obtained after the training is completed, including:

Enter the second input data in the second training data set into the third prediction model to obtain predicted state-of-charge data corresponding to the second input data;

Determine the second loss function value corresponding to the third prediction model according to the predicted state of charge data and historical state of charge data corresponding to the second input data;

Adjust the model parameters of the third prediction model according to the second loss function value, and continue to perform the step of inputting the second input data in the second training data set into the third prediction model until Meet the preset training conditions to obtain the target prediction model.

In one implementation, the third prediction model is the second prediction model or the second prediction model after initializing model parameters.

In a second aspect, embodiments of the present invention also provide a state-of-charge prediction device for an energy storage battery pack, wherein the device includes:

The acquisition module is used to acquire the battery data corresponding to the first energy storage battery pack;

A prediction module, used to obtain a target prediction model, input the battery data into the target prediction model, and obtain the state-of-charge data corresponding to the first energy storage battery pack;

Obtain a first training data set, wherein the first training data set includes a plurality of first training data, a plurality of the first training data respectively correspond to different time periods, and each of the first training data includes a first input data and The historical state of charge data corresponding to the first input data. The first input data is the historical battery data of the second energy storage battery pack in the time period corresponding to the first training data. The historical state of charge data is used The state of charge of the second energy storage battery pack within the time period corresponding to the first training data is reflected;

In a third aspect, embodiments of the present invention further provide a terminal, wherein the terminal includes a memory and one or more processors; the memory stores one or more programs; the program includes a program for executing: Instructions for the state-of-charge prediction method of any one of the above energy storage battery packs; the processor is used to execute the program.

In a fourth aspect, embodiments of the present invention further provide a computer-readable storage medium on which a plurality of instructions are stored, wherein the instructions are suitable for being loaded and executed by a processor to implement any of the above energy storage methods. Steps of battery pack state-of-charge prediction method.

beneficial effects

Beneficial effects of the present invention: Embodiments of the present invention obtain a target prediction model through training by combining machine learning methods and input-output iteration methods. Since the target prediction model combines a variety of training methods, it has high accuracy and can predict energy storage batteries online in real time. The state of charge of the package. This solves the problem that existing battery state-of-charge measurement methods are difficult to meet the measurement time and measurement accuracy requirements in real driving environments.

Description of the drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only These are some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without exerting creative efforts.

Figure 1 is a schematic flowchart of a state-of-charge prediction method for an energy storage battery pack provided by an embodiment of the present invention.

FIG. 2 is a schematic module diagram of a state-of-charge prediction device for an energy storage battery pack provided by an embodiment of the present invention.

Figure 3 is a functional block diagram of a terminal provided by an embodiment of the present invention.

Embodiments of the invention

The present invention discloses a method for predicting the state of charge of an energy storage battery pack. In order to make the purpose, technical solutions and effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described here are only used to explain the present invention and are not intended to limit the present invention.

Those skilled in the art will understand that, unless expressly stated otherwise, the singular forms "a", "an", "the" and "the" used herein may also include the plural form. It should be further understood that the word "comprising" used in the description of the present invention refers to the presence of stated features, integers, steps, operations, elements and/or components, but does not exclude the presence or addition of one or more other features, Integers, steps, operations, elements, components and/or groups thereof. It will be understood that when we refer to an element being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Additionally, "connected" or "coupled" as used herein may include wireless connections or wireless couplings. As used herein, the term "and/or" includes all or any unit and all combinations of one or more of the associated listed items.

It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It should also be understood that terms, such as those defined in general dictionaries, are to be understood to have meanings consistent with their meaning in the context of the prior art, and are not to be used in an idealistic or overly descriptive manner unless specifically defined as here. to explain the formal meaning.

In view of the above defects of the prior art, the present invention provides a method for predicting the state of charge of an energy storage battery pack. The method includes: obtaining battery data corresponding to the first energy storage battery pack; obtaining a target prediction model, and converting the battery Data is input into the target prediction model to obtain the state-of-charge data corresponding to the first energy storage battery pack; the target prediction model is a pre-trained model, and the training process of the target prediction model includes: obtaining the first A training data set, wherein the first training data set includes a plurality of first training data, a plurality of the first training data respectively correspond to different time periods, and each of the first training data includes a first input data and the first input data. Historical state-of-charge data corresponding to an input data. The first input data is the historical battery data of the second energy storage battery pack within the time period corresponding to the first training data. The historical state-of-charge data is used to reflect the The state of charge of the second energy storage battery pack within the time period corresponding to the first training data; obtaining a first prediction model, wherein the first prediction model is an untrained model; according to the first training data The first prediction model is trained in a set, and after the training is completed, a second prediction model and predicted state-of-charge data corresponding to a plurality of the first training data are obtained; according to the first training data set and a plurality of the first training data, The predicted state of charge data corresponding to the training data respectively determines a second training data set, wherein the second training data set includes a plurality of second training data, a plurality of the second training data and a plurality of the first training data. Correspondingly, each second training data includes second input data and historical state-of-charge data corresponding to the second input data, and the second input data includes the first input data corresponding to the second training data and The predicted state of charge data corresponding to the first input data, the historical state of charge data corresponding to the second input data are the same as the historical state of charge data of the first input data corresponding to the second input data; obtaining the third prediction Model, wherein the third prediction model has the same structure as the second prediction model; the third prediction model is trained according to the second training data set, and the target prediction model is obtained after training. The present invention obtains a target prediction model by combining machine learning methods and input-output iteration method training. Since the target prediction model combines multiple training methods, it has high accuracy and can predict the state of charge of the energy storage battery pack online in real time. This solves the problem that existing battery state-of-charge measurement methods are difficult to meet the measurement time and measurement accuracy requirements in real driving environments.

As shown in Figure 1, the method includes the following steps:

Step S100: Obtain battery data corresponding to the first energy storage battery pack.

Specifically, the first energy storage battery pack in this embodiment can be any energy storage battery pack that needs to detect the state of charge. In order to obtain the current state of charge of the first energy storage battery pack, this embodiment needs to first obtain the battery data of the first energy storage battery pack. In one implementation, the battery data of the first energy storage battery pack includes one or more characteristic data such as voltage, current, and time.

As shown in Figure 1, the method also includes the following steps:

Step S200: Obtain a target prediction model, input the battery data into the target prediction model, and obtain state-of-charge data corresponding to the first energy storage battery pack.

Specifically, this embodiment pre-trains a target prediction model. Since the target prediction model is pre-trained based on a large amount of training data, it has learned the correlation between battery data with different characteristics and the state of charge. Therefore, after the battery data of the first energy storage battery pack is input into the target prediction model, the target prediction model can quickly predict the current state-of-charge data of the first energy storage battery pack based on the input battery data.

In one implementation, the training process of the target prediction model includes:

Step S10: Obtain a first training data set, wherein the first training data set includes a plurality of first training data, the plurality of first training data respectively correspond to different time periods, and each of the first training data includes a first training data set. The input data and the historical state-of-charge data corresponding to the first input data. The first input data is the historical battery data of the second energy storage battery pack within the time period corresponding to the first training data. The historical state-of-charge data is The status data is used to reflect the state of charge of the second energy storage battery pack in the time period corresponding to the first training data. The first energy storage battery pack and the second energy storage battery pack correspond to battery packs respectively. Same type;

Step S20: Obtain a first prediction model, wherein the first prediction model is an untrained model;

Step S30: Train the first prediction model according to the first training data set. After the training is completed, obtain the second prediction model and several predicted state-of-charge data corresponding to the first training data;

Step S40: Determine a second training data set according to the predicted state of charge data corresponding to the first training data set and a plurality of the first training data, wherein the second training data set includes a plurality of second training data. , a plurality of the second training data correspond to a plurality of the first training data, each of the second training data includes second input data and historical state-of-charge data corresponding to the second input data, the The second input data includes the first input data corresponding to the second training data and the predicted state of charge data corresponding to the first input data. The historical state of charge data corresponding to the second input data corresponds to the second input data. The historical state-of-charge data of the first input data are the same;

Step S50: Obtain a third prediction model, wherein the third prediction model has the same structure as the second prediction model;

Step S60: Train the third prediction model according to the second training data set, and obtain the target prediction model after training.

To put it simply, compared with ordinary machine learning, which only needs to use one training data set for iterative training, the target prediction model in this embodiment needs to use two training data sets for iterative training, so the model accuracy is higher. Specifically, this embodiment needs to predetermine a second energy storage battery pack of the same battery pack type as the first energy storage battery pack, and obtain its historical battery data and historical state-of-charge data in different time periods to form the first training A data set, wherein the first training data set includes a plurality of first training data, each first training data includes historical battery data for a time period, that is, the first input data, and also includes historical state-of-charge data for the time period. , equivalent to real label data. The untrained first prediction model is then iteratively trained based on the first training data set. The iterative training includes several rounds of training. The process of each round of training is: input the first input data corresponding to a first training data into the first prediction model of the current round, and obtain the output of the first prediction model based on the first input data. The predicted state of charge data, and the first prediction model of the current round is converged based on the predicted state of charge data and the historical state of charge data corresponding to the first input data. After the iterative training of the first prediction model is completed, the second prediction model and the predicted state of charge data corresponding to each first training data are obtained. Then a second training data set is reconstructed based on the predicted state of charge data corresponding to each first training data, where the second training data set includes a plurality of second training data, and each second training data is associated with a first training data One-to-one correspondence. For each second training data, the second training data consists of a second input data and a historical state of charge data, and the second input data corresponding to the second training data is composed of the first input corresponding to the second training data. The data is composed of the predicted state of charge data of the first input data, and the historical state of charge data corresponding to the second training data follows the historical state of charge data of the first training data corresponding to the second training data. Finally, the third prediction model is iteratively trained based on the second training data set, and the target prediction model is obtained after the training is completed.

In one implementation, the first prediction model is a lightgbm model.

Specifically, when the third prediction model is the second prediction model, it is equivalent to first using the first training data set for iterative training on the first prediction model to obtain the trained second prediction model, and then based on the second training data set Continue to iteratively train the second prediction model to obtain the trained target prediction model. Since the target prediction model has been iteratively trained with two training data sets, its prediction accuracy will be significantly improved; when the third prediction model is the initialization model The second prediction model after the parameters is equivalent to using the first training data set for iterative training on the first prediction model to obtain the trained second prediction model and the predicted state of charge data obtained in each round of iterative training. Construct a second training data set based on the first training data set and the predicted state of charge data obtained in each round, re-initialize the second prediction model, and finally iteratively train the initialized second prediction model based on the second training data set to obtain For the target prediction model, since the second training data set has undergone data amplification compared with the first training data set, the target prediction model is trained using the second training data set, and its prediction accuracy will also be significantly improved.

In one implementation, the first prediction model includes a plurality of first prediction models, and the plurality of first prediction models respectively correspond to different hyperparameter combinations. The step S30 specifically includes the following steps:

Step S31: Train a plurality of the first prediction models respectively according to the first training data set, and obtain a plurality of prediction models respectively corresponding to the first prediction models after the training is completed;

Step S32: Obtain a test data set, wherein the test data set and the first training data set are generated based on the same data set;

Step S33: Determine the prediction accuracy corresponding to several of the prediction models according to the test data set;

Step S34: Determine the second prediction model based on the prediction model with the highest prediction accuracy.

To put it simply, in this embodiment, when the first training data set is used for model training, the process of hyperparameter tuning of the model is also included. Specifically, the data set is divided into a training data set and a test data set according to a preset ratio in advance, and then for each hyperparameter combination, the hyperparameters of the model are set based on the hyperparameter combination, and the model is trained based on the first training data set. Develop a prediction model and then test the prediction accuracy of the prediction model based on the test data set. Finally, the prediction model with the highest prediction accuracy among all prediction models is used as the second prediction model to complete the two processes of hyperparameter tuning and network parameter optimization.

In one implementation, after obtaining the data set, feature engineering, data cleaning and other operations are first performed on the data in the data set, and then divided into a training data set and a test data set.

Step S01: Obtain the preset hyperparameter value range;

Step S02: Traverse the hyperparameter value range according to the preset step size to obtain several hyperparameter combinations;

Step S03: Determine one of the first prediction models according to each of the hyperparameter combinations.

Specifically, a hyperparameter value range is preset to represent the parameter range corresponding to the hyperparameter combination. Then it traverses the hyperparameter value range sequentially according to the preset step size to obtain all possible hyperparameter combinations, and then sets the hyperparameters of the model based on each hyperparameter combination to obtain the first value corresponding to each hyperparameter combination. Predictive model.

In one implementation, the preset step size includes several step sizes, and step S02 specifically includes the following steps:

Step S021: Obtain a plurality of step sizes, wherein each of the step sizes corresponds to different accuracy intervals, and the size of the accuracy interval corresponding to each step size is inversely proportional to the step size;

Step S022: Determine the prediction accuracy of the hyperparameter combination corresponding to the previous round of search, and determine the target step size from a number of the step sizes according to the prediction accuracy of the hyperparameter combination corresponding to the previous round of search, wherein the previous step size is determined. The prediction accuracy of the hyperparameter combination corresponding to the round search is located in the accuracy interval corresponding to the target step size;

Step S023: Execute the current round of search according to the target step size, and obtain the hyperparameter combination corresponding to the current round of search;

Step S024: Repeat the step of determining the prediction accuracy of the hyperparameter combination corresponding to the previous round of search until the hyperparameter value range is traversed.

To put it simply, in order to speed up the time of hyperparameter tuning and model training, this embodiment does not use a unified step size throughout the process, but adjusts the current round of search based on the prediction accuracy of the hyperparameter combination searched in the previous round. The step size to take. In other words, use a larger step size traversal in the early stage to quickly reach the vicinity of the optimal hyperparameter combination; use a smaller step size traversal in the later stage to accurately find the optimal hyperparameter combination.

In one implementation, step S31 specifically includes:

Step S311: For each first prediction model, input the first input data in the first training data set into the first prediction model, and obtain the predicted state of charge data corresponding to the first input data;

Step S312: Determine the first loss function value corresponding to the first prediction model based on the predicted state of charge data and historical state of charge data corresponding to the first input data;

Step S313: Adjust the model parameters of the first prediction model according to the first loss function value, and continue to perform the step of inputting the first input data in the first training data set into the first prediction model, Until the preset training conditions are met, a prediction model corresponding to the first prediction model is obtained.

Simply put, the training process corresponding to the first training data set is equivalent to using the battery data of the second energy storage battery pack obtained in different time periods as the input data of the model. Specifically, during each round of training, the first input data corresponding to the current round is input into the first prediction model to obtain the predicted state of charge data of the current round. According to the predicted state of charge data and the historical charge corresponding to the first input data, The state data calculates the first loss function value of the current round. Since the first loss function value can reflect the gap between the model output and the real label, the module model parameters are adjusted based on the first loss function value.

In one implementation, step S60 specifically includes the following steps:

Step S61: Enter the second input data in the second training data set into the third prediction model to obtain predicted state-of-charge data corresponding to the second input data;

Step S62: Determine the second loss function value corresponding to the third prediction model based on the predicted state of charge data and historical state of charge data corresponding to the second input data;

Step S63: Adjust the model parameters of the third prediction model according to the second loss function value, and continue to perform the step of inputting the second input data in the second training data set into the third prediction model. Steps until the preset training conditions are met to obtain the target prediction model.

Specifically, simply speaking, the training process corresponding to the second training data set is equivalent to using the battery data of the second energy storage battery pack obtained in different time periods and the predicted state-of-charge data corresponding to the battery data as the input data of the model. Specifically, during each round of training, the second input data corresponding to the current round is input into the first prediction model to obtain the predicted state of charge data of the current round. According to the predicted state of charge data and the historical charge corresponding to the second input data, The state data calculates the second loss function value of the current round. Since the second loss function value can reflect the gap between the model output and the real label, the module model parameters are adjusted based on the second loss function value.

Based on the above embodiments, the present invention also provides a device for predicting the state of charge of an energy storage battery pack. As shown in Figure 2, the device includes:

Acquisition module 01 is used to obtain battery data corresponding to the first energy storage battery pack;

The prediction module 02 is used to obtain a target prediction model, input the battery data into the target prediction model, and obtain the state-of-charge data corresponding to the first energy storage battery pack;

Based on the above embodiments, the present invention also provides a terminal, the functional block diagram of which can be shown in Figure 3 . The terminal includes a processor, memory, network interface, and display screen connected through a system bus. Among them, the processor of the terminal is used to provide computing and control capabilities. The memory of the terminal includes non-volatile storage media and internal memory. The non-volatile storage medium stores operating systems and computer programs. This internal memory provides an environment for the execution of operating systems and computer programs in non-volatile storage media. The network interface of the terminal is used to communicate with external terminals through a network connection. The computer program is executed by a processor to implement a state-of-charge prediction method for an energy storage battery pack. The terminal's display screen may be a liquid crystal display or an electronic ink display.

Those skilled in the art can understand that the principle block diagram shown in Figure 3 is only a block diagram of a partial structure related to the solution of the present invention, and does not constitute a limitation on the terminals to which the solution of the present invention is applied. Specific terminals may include There may be more or fewer parts than shown, or certain parts may be combined, or may have a different arrangement of parts.

In one implementation, one or more programs are stored in the memory of the terminal, and are configured to be executed by one or more processors, including performing an energy storage battery pack operation. Instructions for the state of charge prediction method.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage. In the media, when executed, the computer program may include the processes of the above method embodiments. Any reference to memory, storage, database or other media used in the various embodiments provided by the present invention may include non-volatile and/or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Synchlink DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

To sum up, the present invention discloses a method for predicting the state of charge of an energy storage battery pack. The method includes: obtaining battery data corresponding to the first energy storage battery pack; obtaining a target prediction model, and inputting the battery data. The target prediction model obtains the state-of-charge data corresponding to the first energy storage battery pack; the target prediction model is a pre-trained model, and the training process of the target prediction model includes: obtaining the first training data Set, wherein the first training data set includes a plurality of first training data, a plurality of the first training data respectively correspond to different time periods, and each of the first training data includes first input data and the first input The historical state of charge data corresponding to the data, the first input data is the historical battery data of the second energy storage battery pack in the time period corresponding to the first training data, and the historical state of charge data is used to reflect the first The state of charge of the second energy storage battery pack in the time period corresponding to the training data; obtain a first prediction model, wherein the first prediction model is an untrained model; and calculate the state of charge of the second energy storage battery pack according to the first training data set The first prediction model is trained, and after the training is completed, a second prediction model and predicted state-of-charge data corresponding to a plurality of the first training data are obtained; according to the first training data set and a plurality of the first training data Respectively corresponding predicted state of charge data determines a second training data set, wherein the second training data set includes a plurality of second training data, and a plurality of the second training data corresponds to a plurality of the first training data in a one-to-one manner. , each second training data includes second input data and historical state-of-charge data corresponding to the second input data, and the second input data includes the first input data corresponding to the second training data and the first input data. Predicted state-of-charge data corresponding to an input data, historical state-of-charge data corresponding to the second input data being the same as historical state-of-charge data corresponding to the first input data corresponding to the second input data; obtaining a third prediction model, Wherein, the third prediction model has the same structure as the second prediction model; the third prediction model is trained according to the second training data set, and the target prediction model is obtained after training. The present invention obtains a target prediction model by combining machine learning methods and input-output iteration method training. Since the target prediction model combines multiple training methods, it has high accuracy and can predict the state of charge of the energy storage battery pack online in real time. This solves the problem that existing battery state-of-charge measurement methods are difficult to meet the measurement time and measurement accuracy requirements in real driving environments.

It should be understood that the application of the present invention is not limited to the above examples. Those of ordinary skill in the art can make improvements or changes based on the above descriptions. All these improvements and changes should fall within the protection scope of the appended claims of the present invention.

Claims

A state-of-charge prediction method for an energy storage battery pack, characterized in that the method includes:

Obtain the battery data corresponding to the first energy storage battery pack;

Obtain a target prediction model, input the battery data into the target prediction model, and obtain the state-of-charge data corresponding to the first energy storage battery pack;

The target prediction model is a pre-trained model, and the training process of the target prediction model includes:

Obtain a first training data set, wherein the first training data set includes a plurality of first training data, a plurality of the first training data respectively correspond to different time periods, and each of the first training data includes a first input data and The historical state of charge data corresponding to the first input data. The first input data is the historical battery data of the second energy storage battery pack in the time period corresponding to the first training data. The historical state of charge data is used When reflecting the state of charge of the second energy storage battery pack in the time period corresponding to the first training data, the battery pack types corresponding to the first energy storage battery pack and the second energy storage battery pack are the same;

Obtain a first prediction model, wherein the first prediction model is an untrained model;

The first prediction model is trained according to the first training data set, and after the training is completed, a second prediction model and a plurality of predicted state-of-charge data corresponding to the first training data are obtained;

A second training data set is determined according to the predicted state of charge data corresponding to the first training data set and a plurality of the first training data, wherein the second training data set includes a plurality of second training data, and a plurality of the second training data. The second training data has a one-to-one correspondence with a plurality of the first training data. Each second training data includes second input data and historical state-of-charge data corresponding to the second input data. The second input data The data includes first input data corresponding to the second training data and predicted state of charge data corresponding to the first input data, and historical state of charge data corresponding to the second input data and first first input data corresponding to the second input data. The historical state-of-charge data of the input data are the same;

Obtain a third prediction model, wherein the third prediction model has the same structure as the second prediction model;

The third prediction model is trained according to the second training data set, and the target prediction model is obtained after training.
The method for predicting the state of charge of an energy storage battery pack according to claim 1, wherein the first prediction model includes a plurality of first prediction models, and the plurality of first prediction models respectively correspond to different hyperparameter combinations, The training of the first prediction model according to the first training data set includes:

Train a plurality of the first prediction models respectively according to the first training data set, and obtain a plurality of prediction models respectively corresponding to the first prediction models after the training is completed;

Obtain a test data set, wherein the test data set and the first training data set are generated based on the same data set;

According to the test data set, determine the prediction accuracy corresponding to several of the prediction models;

The second prediction model is determined based on the prediction model with the highest prediction accuracy.
The state-of-charge prediction method of an energy storage battery pack according to claim 2, characterized in that the determination methods of several of the first prediction models include:

Get the preset hyperparameter value range;

Traverse the hyperparameter value range according to the preset step size to obtain several hyperparameter combinations;

One of the first prediction models is determined according to each of the hyperparameter combinations.

The state-of-charge prediction method of an energy storage battery pack according to claim 2, characterized in that the determination methods of several of the first prediction models include:

Get the preset hyperparameter value range;

Traverse the hyperparameter value range according to the preset step size to obtain several hyperparameter combinations;

One of the first prediction models is determined according to each of the hyperparameter combinations.
The method for predicting the state of charge of an energy storage battery pack according to claim 3, wherein the preset step size includes several step sizes, and the hyperparameter value range is traversed according to the preset step size, Several hyperparameter combinations are obtained, including:

Obtain a plurality of step sizes, wherein each of the step sizes corresponds to different accuracy intervals, and the size of the accuracy interval corresponding to each step size is inversely proportional to the step size;

Determine the prediction accuracy of the hyperparameter combination corresponding to the previous round of search, and determine the target step size from several step sizes according to the prediction accuracy of the hyperparameter combination corresponding to the previous round of search, wherein the previous round of search corresponds to The prediction accuracy of the hyperparameter combination is located in the accuracy interval corresponding to the target step size;

Execute the current round of search according to the target step size, and obtain the hyperparameter combination corresponding to the current round of search;

Repeat the step of determining the prediction accuracy of the hyperparameter combination corresponding to the previous round of search until the hyperparameter value range is traversed.
The method for predicting the state of charge of an energy storage battery pack according to claim 2, wherein the step of training a plurality of the first prediction models according to the first training data set includes:

For each of the first prediction models, input the first input data in the first training data set into the first prediction model to obtain the predicted state of charge data corresponding to the first input data;

Determine the first loss function value corresponding to the first prediction model based on the predicted state of charge data and historical state of charge data corresponding to the first input data;

Adjust the model parameters of the first prediction model according to the first loss function value, and continue to perform the step of inputting the first input data in the first training data set into the first prediction model until the predetermined Set training conditions to obtain the prediction model corresponding to the first prediction model.
The method for predicting the state of charge of an energy storage battery pack according to claim 1, wherein the third prediction model is trained according to the second training data set, and the target prediction is obtained after training. Models, including:

Enter the second input data in the second training data set into the third prediction model to obtain predicted state-of-charge data corresponding to the second input data;

Determine the second loss function value corresponding to the third prediction model according to the predicted state of charge data and historical state of charge data corresponding to the second input data;

Adjust the model parameters of the third prediction model according to the second loss function value, and continue to perform the step of inputting the second input data in the second training data set into the third prediction model until Meet the preset training conditions to obtain the target prediction model.
The method for predicting the state of charge of an energy storage battery pack according to claim 1, wherein the third prediction model is the second prediction model or the second prediction model after initializing model parameters.
A state-of-charge prediction device for an energy storage battery pack, characterized in that the device includes:

The acquisition module is used to acquire the battery data corresponding to the first energy storage battery pack;

A prediction module, used to obtain a target prediction model, input the battery data into the target prediction model, and obtain the state-of-charge data corresponding to the first energy storage battery pack;

The target prediction model is a pre-trained model, and the training process of the target prediction model includes:

Obtain a first training data set, wherein the first training data set includes a plurality of first training data, a plurality of the first training data respectively correspond to different time periods, and each of the first training data includes a first input data and The historical state of charge data corresponding to the first input data. The first input data is the historical battery data of the second energy storage battery pack in the time period corresponding to the first training data. The historical state of charge data is used The state of charge of the second energy storage battery pack during the time period corresponding to the first training data;

Obtain a first prediction model, wherein the first prediction model is an untrained model;

The first prediction model is trained according to the first training data set, and after the training is completed, a second prediction model and a plurality of predicted state-of-charge data corresponding to the first training data are obtained;

A second training data set is determined according to the predicted state of charge data corresponding to the first training data set and a plurality of the first training data, wherein the second training data set includes a plurality of second training data, and a plurality of the second training data. The second training data has a one-to-one correspondence with a plurality of the first training data. Each second training data includes second input data and historical state-of-charge data corresponding to the second input data. The second input data The data includes first input data corresponding to the second training data and predicted state of charge data corresponding to the first input data, and historical state of charge data corresponding to the second input data and first first input data corresponding to the second input data. The historical state-of-charge data of the input data are the same;

Obtain a third prediction model, wherein the third prediction model has the same structure as the second prediction model;

The third prediction model is trained according to the second training data set, and the target prediction model is obtained after training.
A terminal, characterized in that the terminal includes a memory and one or more processors; the memory stores one or more programs; the program includes a program for executing any one of claims 1-7 Instructions for the state-of-charge prediction method of the energy storage battery pack; the processor is used to execute the program.
A computer-readable storage medium on which a plurality of instructions are stored, characterized in that the instructions are adapted to be loaded and executed by a processor to implement the energy storage battery pack described in any one of claims 1-7. Steps in the state of charge prediction method.