A battery residual capacity real-time prediction method, a terminal device, and a storage medium

Abstract

The application relates to a battery residual capacity real-time prediction method, a terminal device and a storage medium, the method comprising the following steps: S1: collecting discharge data of a battery, and extracting feature data required for model training to form a training set through the collected discharge data; S2: constructing a double-input-layer stacked LSTM network model, and training the model through the training set; the double-input-layer stacked LSTM network model comprises a first input channel and a second input channel, the input of the first input channel is signal change characteristics, and the input of the second input channel is accumulated discharge capacity; the network structure of the first input channel comprises an LSTM network and two full connection layers, and the network structure of the second input channel is one full connection layer; the output of the first input channel and the output of the second input channel are normalized through a sigmoid function, and then residual capacity SOC is obtained through aggregation; and S3: the residual capacity SOC of the battery is predicted in real time through the trained model. The application improves the prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of battery, and particularly relates to a battery residual capacity real-time prediction method, a terminal device and a storage medium. BACKGROUND

[0002] With the rapid development of economy and society, energy problems have become the focus of global attention. Large-scale battery energy storage systems (ESS) have been widely used in various industrial, commercial and residential scenarios due to their safety, cleanliness, high energy density, low cost, long cycle life, fast response speed and other characteristics. In uninterruptible power supply systems and energy storage systems, battery packs are important energy storage devices and are closely related to the reliable operation of power supply systems. At present, the mainstream power batteries on the market are divided into lead-acid batteries and lithium batteries. However, even if the same batch of batteries produced by the production line of the same manufacturer, there will be differences due to complex environment, specific operating conditions and other factors. Overcharging, over-discharging and battery aging may all lead to a decrease in battery health, thereby affecting the normal power supply of the energy storage system. This difference in battery monomers leads to the Leipzig minimum factor effect in large-scale battery grouping, that is, the short board effect of the barrel, which greatly damages the cycle life, safety, reliability and effective capacity of the battery energy storage system.

[0003] At present, there is no method for online detection of real-time residual capacity of batteries in industrial applications. Precise prediction of the real-time residual capacity of single batteries can facilitate real-time monitoring of the health status of each battery in the battery pack by the staff, the energy consumption, and the early adoption of safety and economic measures. It is an important research direction with economic, safety and energy-saving environmental benefits.

[0004] The research on the problem of online monitoring the real-time residual capacity of the battery has the following difficulties: 1. compatible with diversified discharge scenarios. The battery discharge scenarios under actual working conditions usually include multiple rates, and it is an important problem that needs to be considered in model design to be compatible with multiple discharge scenarios. 2. compatible with diversified health batteries. The batteries used in actual working conditions are of different ages, and the model needs to have the prediction ability of compatible multiple capacity batteries. 3. real-time of SOC prediction. Domestic and foreign researches focus on observing the battery discharge data for a long time for feature engineering processing, or using the charging process curve of the battery before discharge to predict the capacity. In the actual discharge scenario, it is difficult to obtain the charging process data of the battery, and the prediction of a longer discharge time cannot meet the industrial demand. Therefore, how to obtain accurate prediction results with shorter observation time is a difficulty of the scheme. 4. sensor signal processing. In actual working conditions, the sensor data of the battery discharge has a lot of noise and measurement error, and the sampling interval is different. How to clean the data through feature construction so that the model learns the real physical law based on good data is one of the problems that need to be considered. SUMMARY

[0005] In order to solve the above problems, the application provides a battery residual capacity real-time prediction method, a terminal device and a storage medium.

[0006] The specific scheme is as follows:

[0007] A battery residual capacity real-time prediction method, comprising the following steps:

[0008] S1: collecting the discharge data of the battery under different discharge rates and different health degrees, and extracting the feature data required for model training through the collected discharge data, the feature data including signal change features, cumulative discharge capacity and residual capacity SOC, and the extracted feature data forming a training set;

[0009] S2: constructing a double-input layer stacked LSTM network model, and training the model through the training set;

[0010] The double-input layer stacked LSTM network model includes a first input channel and a second input channel, the input of the first input channel being signal change features, and the input of the second input channel being cumulative discharge capacity; the network structure of the first input channel includes an LSTM network and two fully connected layers, and the network structure of the second input channel is one fully connected layer; the output of the first input channel and the output of the second input channel are normalized through a sigmoid function, and then aggregated to obtain the residual capacity SOC;

[0011] S3: real-time prediction of the residual capacity SOC of the battery through the trained model.

[0012] Further, before extracting the feature data in step S1, the missing values and outliers of the collected discharge data are processed.

[0013] Further, the signal change features include voltage, current, current change amount, voltage change amount and temperature change amount.

[0014] A battery residual capacity real-time prediction terminal device, comprising a processor, a memory and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method of the above-mentioned embodiment of the application when executing the computer program.

[0015] A computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to implement the steps of the method of the above-mentioned embodiment of the application.

[0016] The application adopts the above technical solution, and proposes a double-input layer stacked LSTM network model on the basis of an LSTM neural network, so that the double-input layer stacked LSTM network model is more suitable for related prediction research work in the battery field, and the prediction accuracy is improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 Fig. 1 shows a flowchart of an embodiment of the application.

[0018] Figure 2 Fig. 2 shows a structure diagram of a double-input layer stacked LSTM network model in the embodiment. DETAILED DESCRIPTION

[0019] To further illustrate the embodiments, the application provides drawings. These drawings are part of the disclosure of the application, mainly used to illustrate the embodiments, and can be used to explain the operating principle of the embodiments in conjunction with the related description of the specification. Those skilled in the art should understand other possible implementation manners and advantages of the application by referring to these contents.

[0020] The application will be further described in conjunction with the drawings and specific embodiments.

[0021] Embodiment one:

[0022] The embodiment of the application provides a battery residual capacity real-time prediction method, which comprises the following steps: Figure 1 As shown in the figure, the method comprises the following steps:

[0023] S1: collecting discharge data of the battery under different discharge rates and different health degrees, and extracting feature data required to input the model from the collected discharge data, wherein the feature data comprises signal change features, cumulative discharge amount and residual capacity SOC, and the training set is composed of the extracted feature data.

[0024] The data collected in this embodiment is the discharge data collected through lead-acid battery experiments under a certain actual discharge scenario.

[0025] The state of health (SOH) of the battery represents the difference between the actual capacity, performance state, etc. of the battery and the performance of a new battery. The SOH is generally classified according to grades. Numerically, the SOH is defined as the ratio of the rated capacity of the battery to the actual capacity. According to the percentage range in which the calculated value is located, the SOH can also be summarized as a health label. For example, a battery with an actual SOH value of 60%-80% is generally evaluated as having a "good" state of health. The calculation formula is:

[0026]

[0027] In the formula, C0 is the rated capacity of the battery, and C is the actual capacity of the battery. 实

[0028] In this embodiment, different health degrees are set, including the health states of: damaged, general, good, and healthy.

[0029] The discharge rate is used to measure the speed of battery discharge. It refers to the current value used by the battery to accumulate discharge to the rated capacity within a certain time. It is usually used to describe the discharge scenario of the battery. In this embodiment, the discharge rate is set to be in the range of 0.3C-1.8C.

[0030] The state of charge (SOC) of the battery, i.e., the state of charge, is used to reflect the remaining capacity of the battery. Numerically, the SOC is defined as the ratio of the remaining capacity to the rated capacity of the battery. The value of the SOC ranges from 0% to 100%. When SOC = 0%, it indicates that the battery is completely discharged. When SOC = 100%, it indicates that the battery is fully charged. The calculation formula is:

[0031]

[0032] In the formula, C0 is the rated capacity of the battery, and C is the current actual capacity of the battery.

[0033] In this embodiment, the multi-cycle discharge data of the single battery with a health degree of damaged, general, good, and healthy under a discharge rate of 0.3C-1.8C is collected. Each discharge cycle is a csv data set. In the data set, there are sensor data of multiple sampling specifications, and the sampling interval is 10S-20S.

[0034] Further, in this embodiment, the missing values and abnormal values in the collected discharge data are processed.

[0035] ​The missing value processing is mainly aimed at the original internal resistance sensor data. Through preliminary observation of the data, it is found that the resistance values of some batteries are missing. The delete method is used to discard the part of the battery.

[0036] The abnormal value of the data represents that the value significantly deviates from the rest of the observations of the sample. In the sample, it needs to be removed, otherwise it will affect the feature learning of the model. In the collected data, the first few voltage sampling points at the initial moment are still in the floating state voltage value during charging, which deviates from the normal discharge stable voltage value, and is removed.

[0037] For the case that different types of battery sensor signals have different sampling intervals in the actual scene, the linear interpolation method is used to down-sample the health degree battery data set in this embodiment, so that all data are on the same sampling distribution.

[0038] In the selection of feature data, the Pearson correlation analysis is used to select the high correlation feature combination as the clustering vector, evaluate the clustering effect, and select the final feature combination method. When the correlation coefficient is 0, X and Y two variables have no linear correlation, the greater the positive correlation between X and Y two variables, the correlation coefficient tends to 1, the greater the negative correlation between X and Y two variables, the correlation coefficient tends to-1, and the formula is expressed as:

[0039]

[0040] According to the analysis result, the first four features that most affect the battery SOC prediction are the battery discharge voltage V, the current I, the cumulative discharge quantity Qd at the current moment and the discharge scene temperature T. Therefore, the feature data selected in this embodiment includes signal change feature, cumulative discharge quantity and remaining capacity SOC, wherein the signal change feature specifically includes: voltage, current, current change, voltage change, temperature change, wherein the current change, voltage change and temperature change are all changes relative to the initial moment.

[0041] In the feature data extraction process, in view of the different data distribution of different discharge rates in the battery data, the provided sensor observation features voltage, current and temperature are transformed, and the change relative to the initial discharge moment is constructed, so that the model can capture the discharge change rule under different rates, and the influence of the rate itself is reduced, and the generalization ability of the model is enhanced.

[0042] At the same time, the voltage feature recorded by the sensor has an unstable trend of fluctuating upward. On the basis of the traditional feature transformation, the change algorithm is improved, and the change relative to the first discharge time window is constructed, so that the model feature data are all of the same sign, the convergence speed of the model is accelerated, and the fitting of the model to the true law is improved.

[0043] S2: Construct a double-input layer stacked LSTM network model, and train the model through the training set.

[0044] The prediction related to time series usually adopts a recurrent neural network, but if the input sequence is too long, the information at the beginning of the sequence will gradually decrease and disappear with the circulation, resulting in the long-term dependent information required for the prediction task being forgotten. As an improved recurrent neural network, LSTM adds gating technology and memory cell state to the hidden layer of the recurrent neural network, and improves the long-term dependence problem of the recurrent neural network. It solves the problem of gradient disappearance or gradient explosion when processing long sequence data, and has more effective long-term prediction ability.

[0045] The ampere-hour integration method is a classic SOC prediction method. The idea is to estimate the SOC of the battery based on the initial SOC0. The ampere-hour integration method calculates the percentage of electric quantity, i.e. the percentage of consumed electric quantity, by calculating the charging and discharging current and the corresponding integral within a certain time. Finally, by calculating the difference between the initial time and the cumulative discharge capacity, the required remaining battery capacity is obtained. As shown in the following formula:

[0046]

[0047] Due to the error in the measurement of the ampere-hour integration method, the error of the result obtained by the ampere-hour integration method will become larger and larger with the increase of the use time. Inspired by the traditional ampere-hour integration method, the cumulative discharge capacity has a linear effect on the final SOC prediction, and the multi-layer nonlinear transformation of the neural network will affect the expression ability of this feature on the final prediction result. In this embodiment, the ResNet network residual layer idea is borrowed, and the cumulative discharge capacity original signal is taken alone, the LSTM unit layer transformation is skipped, and the sum is applied to the LSTM hidden layer output result.

[0048] As shown in Figure 2 The specific structure of the double-input layer stacked LSTM network model includes a first input channel and a second input channel. The input of the first input channel is the signal change feature, and the input of the second input channel is the cumulative discharge capacity. The network structure of the first input channel includes an LSTM network and two fully connected layers, and the network structure of the second input channel is one fully connected layer. The output of the first input channel and the output of the second input channel are normalized by a sigmoid function, and the remaining capacity SOC is obtained by aggregation.

[0049] S4: Real-time prediction of the remaining capacity SOC of the battery through the trained model.

[0050] The embodiment of the present application improves the traditional form of LSTM network around the characteristics of cumulative discharge amount from the actual significance of discharge parameters. Different features are spliced and constructed, and a double-input layer stacked LSTM network model is learned respectively, and acts on the regression prediction result. The cumulative discharge amount feature is separated out, and less nonlinear transformation is performed, and the original feature form is directly used for the prediction result.

[0051] Embodiment two:

[0052] The present application also provides a battery remaining capacity real-time prediction terminal device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the above method embodiments of the first embodiment of the present application when executing the computer program.

[0053] Further, as an executable solution, the battery remaining capacity real-time prediction terminal device can be a desktop computer, a notebook computer, a palm computer and a cloud server, etc. The battery remaining capacity real-time prediction terminal device can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above-mentioned composition structure of the battery remaining capacity real-time prediction terminal device is only an example of the battery remaining capacity real-time prediction terminal device, and does not constitute a limitation on the battery remaining capacity real-time prediction terminal device, and can include more or less components than the above, or combine certain components, or different components, for example, the battery remaining capacity real-time prediction terminal device can also include an input / output device, a network access device, a bus, etc., and the present application does not limit this.

[0054] Further, as an executable solution, the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the battery remaining capacity real-time prediction terminal device, and connects each part of the battery remaining capacity real-time prediction terminal device through various interfaces and lines.

[0055] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the terminal device for real-time prediction of the remaining capacity of the battery by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and application programs required by at least one function; and the data storage area can store data created according to the use of the mobile phone and the like. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory such as a hard disk, a memory card, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory device.

[0056] The application further provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to realize the steps of the method of the above-mentioned embodiments of the application.

[0057] The modules / units integrated in the terminal device for real-time prediction of the remaining capacity of the battery can be stored in a computer readable storage medium if they are realized in the form of software function units and sold or used as independent products. Based on such understanding, all or part of the processes in the above-mentioned embodiments of the method of the application can also be completed by a computer program instructing related hardware. The computer program can be stored in a computer readable storage medium. The computer program is executed by a processor to realize the steps of the above-mentioned various method embodiments. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM) and software distribution medium, etc.

[0058] Although the application is specifically shown and described in connection with the preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the spirit and scope of the application as defined in the appended claims.

Claims

1. A method for real-time prediction of remaining battery capacity, characterized in that: The following steps are involved: S1: Collect discharge data of batteries at different discharge rates and different health levels, and extract feature data required for model training from the collected discharge data. Feature data includes signal change characteristics, cumulative discharge capacity, and remaining capacity (SOC). The extracted feature data form a training set. S2: Build a dual-input layer stacked LSTM network model and train the model using the training set; The dual-input stacked LSTM network model includes a first input channel and a second input channel. The input of the first input channel is the signal change feature, and the input of the second input channel is the accumulated discharge amount. The network structure of the first input channel includes an LSTM network and two fully connected layers, while the network structure of the second input channel is a single fully connected layer. The outputs of the first and second input channels are normalized by the sigmoid function and aggregated to obtain the remaining capacity (SOC). S3: Use the trained model to predict the remaining capacity (SOC) of the battery in real time.

2. The method for real-time prediction of remaining battery capacity according to claim 1, wherein: Before extracting the characteristic data in step S1, the process also includes processing missing values ​​and abnormal values ​​of the collected discharge data.

3. The method for real-time prediction of remaining battery capacity according to claim 1, wherein: Signal change characteristics include: voltage, current, current change, voltage change and temperature change.

4. A terminal device for real-time prediction of remaining battery capacity, characterized by: The method comprises a processor, a memory, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of the method according to any one of claims 1 to 3 when executing the computer program.

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

Images