A method and system for diagnosing faults of energy storage batteries based on multi-source information fusion

Abstract

This invention provides a method and system for fault diagnosis of energy storage batteries based on multi-source information fusion. It involves performing staggered sampling on the voltage and current sampling sequence to obtain the voltage and current values ​​at previous times and the current value at the current time, which are used as feature inputs for a network model. Differential sampling is performed on the voltage sequence to calculate the voltage difference between the current time and the previous time, which serves as a label for the network model. Using the feature inputs and labels of healthy batteries, the network model is trained to obtain a network model of the current-voltage characteristics of healthy batteries. The current-voltage characteristic network model is used to predict the normal voltage increment at the current time. The residual between the actual voltage increment and the predicted voltage increment is calculated, and based on the residual, the location and start and end times of the fault are determined. This invention utilizes a deep learning model, employs staggered sampling, and integrates the current-voltage characteristics of individual batteries with correlation coefficients, enabling early diagnosis of minor faults in energy storage batteries with high sensitivity and high accuracy.

Description

Technical Field

[0001] This invention belongs to the field of battery fault diagnosis technology, and relates to a method and system for fault diagnosis of energy storage batteries based on multi-source information fusion. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Lithium-ion batteries play a crucial role in the rapidly growing field of new energy storage due to their high energy density, long cycle life, and environmental friendliness. Energy storage battery packs are composed of numerous individual cells connected in series and parallel. Internal short circuits or poor contact in any single cell can lead to performance degradation of the entire battery pack, or even major safety accidents such as fires and explosions. Therefore, researching effective methods for diagnosing early, minor faults will guide maintenance personnel to trace and repair faulty cells in their early stages, which is of significant practical importance for the safe and efficient operation of energy storage battery packs.

[0004] However, energy storage battery packs are large in scale and have complex topologies, making it easy for minor early-stage faults to be overlooked under complex operating conditions. Existing energy storage battery management systems generally set upper and lower thresholds for voltage, current, and temperature, and then detect faults by detecting values ​​exceeding these limits. Therefore, existing threshold diagnostic methods suffer from low sensitivity and poor accuracy, failing to detect minor early-stage faults, and when a fault occurs, they cannot accurately determine the location and type of the fault.

[0005] Chinese patents CN201911024438.2 and CN202110518346.0 use individual cell voltage information for fault diagnosis, but ignore fault current, which contains a large amount of fault information, thus reducing the accuracy of fault type identification. Chinese patents CN202011181863.5 and CN202110993005.9 use a data-driven method for fault diagnosis, but the sample data used includes information that cannot be accurately obtained, such as SOC and battery internal resistance. Applying this to online testing will reduce diagnostic performance and easily lead to misdiagnosis and missed diagnosis. Summary of the Invention

[0006] To address the aforementioned problems, this invention proposes a fault diagnosis method and system for energy storage batteries based on multi-source information fusion. This invention utilizes a deep learning model, employs a staggered sampling method, and integrates the volt-ampere characteristics of individual cells with the correlation coefficient method. This enables early diagnosis of minor faults in energy storage batteries, exhibiting high sensitivity and high accuracy.

[0007] According to some embodiments, the present invention adopts the following technical solution:

[0008] A fault diagnosis method for energy storage batteries based on multi-source information fusion includes the following steps:

[0009] The voltage and current sampling sequence is processed by staggered sampling to obtain the voltage and current values ​​at previous times and the current value at the current time, which are used as feature inputs for the network model.

[0010] Differential sampling is performed on the voltage sequence to calculate the voltage difference between the current time and the previous time, which is used as the label for the network model;

[0011] By using the feature inputs and labels of healthy batteries, a network model is trained to obtain a network model of the current-voltage characteristics of healthy batteries.

[0012] The current voltage increment is predicted using the current-voltage characteristic network model.

[0013] Calculate the residual between the actual voltage increment and the predicted voltage increment, and determine the location and start and end times of the fault based on the residual.

[0014] As an alternative implementation, the voltage and current sampling sequence is subjected to staggered sampling processing to obtain the voltage and current values ​​at previous times and the current value at the current time. The specific process of using these values ​​as feature inputs to the network model includes: using a FIFO queue to receive the current and voltage sampling values ​​at each time moment, and the FIFO queue forming feature inputs based on the earliest multiple voltage and current values ​​and the latest current value.

[0015] As an alternative implementation, the tag is obtained by subtracting the voltage from the previous time from the current voltage.

[0016] As an alternative implementation, the network model includes an LSTM network followed by an FC fully connected layer.

[0017] As an alternative implementation, the network model outputs a predicted label after each input feature. During training, the predicted label is used to subtract from the original label to optimize the parameters of the network model.

[0018] As an alternative implementation method, a fault is considered to have occurred when the residual is greater than the fault threshold.

[0019] As an alternative implementation, when distinguishing fault types, the Pearson correlation coefficient method is introduced to calculate the type coefficient. The Pearson correlation coefficient is calculated based on the differential current vector and the residual vector. When the type coefficient is greater than or equal to a set value, it is judged as a poor contact fault, and when the type coefficient is less than the set value, it is judged as a short circuit fault.

[0020] A fault diagnosis system for energy storage batteries based on out-of-step differential sampling includes:

[0021] The feature calculation module is configured to perform staggered sampling processing on the voltage and current sampling sequence to obtain the voltage and current values ​​at previous times and the current value at the current time, which are used as feature inputs for the network model.

[0022] The label module is configured to perform differential sampling on the voltage sequence, calculate the voltage difference between the current time and the previous time, and use it as a label for the network model.

[0023] The model training module is configured to use the feature inputs and labels of healthy batteries to train the network model and obtain the network model of the current-voltage characteristics of healthy batteries.

[0024] The fault diagnosis module is configured to use the current voltage characteristic network model to predict the normal voltage increment at the current moment, calculate the residual between the actual voltage increment and the predicted voltage increment, and determine the location and start and end time of the fault based on the residual.

[0025] A battery management system storing multiple instructions adapted for loading and execution of steps in the method by a processor of a terminal device.

[0026] A terminal device includes a processor and a computer-readable storage medium, the processor being configured to implement instructions; the computer-readable storage medium being configured to store a plurality of instructions adapted to be loaded by the processor and executed in accordance with the steps of the method described therein.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0028] This invention has the advantages of high sensitivity and high accuracy.

[0029] This invention can not only detect the type, location, start and end time of minor faults, but also maintain good performance in the fully charged state of the same type of battery.

[0030] The diagnostic process of this invention only requires two directly measurable analog quantities: voltage and current, which has high practical value. Attached Figure Description

[0031] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0032] Figure 1 This is a block diagram illustrating the principle of the fault diagnosis method in this embodiment;

[0033] Figure 2 This is a comparison chart of fault diagnosis methods for battery type 2 and type 1.

[0034] Figure 3This is a comparison chart of battery fault diagnosis methods based on timing 3 and timing 1.

[0035] Figure 4 This is the result of the fault diagnosis. Detailed Implementation

[0036] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0037] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0038] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0039] During the operation of an energy storage battery pack, the battery management system can easily obtain the series current in the series-connected battery pack and the voltage of each individual cell. For example... Figure 1 As shown, by processing the voltage and current sampling sequence with staggered sampling, the voltage and current values ​​at the previous N time steps and the current value at the current time step can be obtained as feature inputs to the network model. By differentially sampling the voltage sequence, voltage increments that can eliminate the autocorrelation effect of LSTM can be obtained and used as labels for the network model. For healthy batteries, deep learning network training using the above feature inputs and labels can yield a network model of the current-voltage characteristics of healthy batteries, aiming to predict the normal voltage increment at the current time step using the voltage and current values ​​at the previous few time steps and the current value at the current time step. When applied to online fault diagnosis, the location and start and end times of the fault can be identified by calculating the residual between the actual voltage increment and the predicted voltage increment. After identifying the fault, the current differential sequence and voltage increment residual sequence of the fault stage are retrieved, and the correlation coefficient algorithm is fused to accurately distinguish between short-circuit faults and poor contact faults.

[0040] The fault diagnosis method proposed in this embodiment does not rely on the battery equivalent model and does not require indirect quantities such as SOC and internal resistance obtained from secondary calculations. It only uses the current and voltage quantities that can be directly measured by sensors. At the same time, through special sampling processing of the current and voltage sequences and information fusion after the fault, the sensitivity of minor fault diagnosis and the accuracy of fault type identification are ensured.

[0041] Specifically, for a single battery cell, for a given sampling sequence of length N, the sampling sequence M can be expressed as M = {S} n |n=1,2,...,N}, where S n =(I n U n This represents the sampled current and voltage values ​​at a specific moment. The sampled current and voltage values ​​at each moment are sent to a FIFO queue. The data M stored in the FIFO (First In First Out) is... s (t) is shown below:

[0042] M s (t)=[S t ,S t+1 ,...,S t+m-1 ],t=1,2,...,N-m+1 (1)

[0043] Where m is the FIFO depth, set to 4. At each time step, the FIFO will generate a feature vector and a label. Therefore, for a sequence of length N, N-3 feature vectors and labels can be generated.

[0044] This invention provides a staggered sampling method. Assuming n analog quantities can be sampled at time t, these n analog quantities are arranged in a fixed order as a single sampling point. M consecutive sampling points are considered a sampling sequence, forming an n x m matrix where the column growth direction is consistent with the sampling time growth direction. A staggered sampling mechanism systematically discards sampling points from the matrix, with the set of discarded points being {ni, m-(i-1) | i = 1, 2, ..., n-1}. This causes the number of samples of different analog quantities to decrease sequentially within the matrix. Specifically, in this embodiment, within a FIFO, a voltage-current imbalance sequence is generated using the staggered sampling mechanism. Specifically, the voltage and current from the previous 3 seconds and the current at the current moment are used as a new sequence, discarding the current voltage, thus creating a sampling method or sequence generation method with "staggered," "misaligned," and "unbalanced" characteristics.

[0045] The FIFO generates an unbalanced combination of voltage and current through a step-out mechanism, which is used as a feature vector F. t It consists of the earliest three voltage and current values ​​in the FIFO and the latest current value, and the sequence is expressed as follows:

[0046] F t =[I t-3 U t-3 ,I t-2 U t-2 ,I t-1 U t-1 ,It ],t=4,5,...,N-m+1 (2)

[0047] At the same time, a label L is generated by FIFO. t It is obtained by subtracting the voltage from the voltage at the previous moment from the current voltage, and the expression is as follows:

[0048] L t =ΔU t =U t -U t-1 ,t=4,5,...,N-m+1 (3)

[0049] In summary, the feature matrix M fed into the network for training f The label matrix L is shown in equation (4), where M f Each row in L represents a feature vector, and each row in L represents a corresponding label.

[0050]

[0051] The network model consists of one LSTM layer followed by one fully connected (FC) layer. The LSTM has an input size of 7, a hidden layer size of 10, and a loss function of MSE. The FC layer has an input size of 10 and an output size of 1. The network model outputs a predicted label after each input feature vector. During training Used with L t By combining network parameter optimization, during training, the network parameters are continuously optimized through gradient descent and backpropagation to minimize the loss function that takes the label and predicted label as input. For example... Figure 1 As indicated by the dashed arrow. During the fault diagnosis process... Used with L t Calculate residuals To determine the fault status, when |ΔL| is greater than the fault threshold F (Th) At that time, it is considered that a fault has occurred.

[0052] After a fault occurs, to distinguish the fault type, the Pearson correlation coefficient method is introduced to calculate the type coefficient. c Its expression is as follows:

[0053]

[0054] Where t is the next moment after the fault is detected, and n is the sampling length used to calculate the type coefficient, preferably 4. X and Y are the ΔL residual vector and the differential current vector IV, respectively. t The expression is as follows:

[0055] X={ΔL i|i=t,t+1,...,t+n-1} (6)

[0056] Y = {(I i -I i-1 )|i=t,t+1,...,t+n-1} (7)

[0057] like Figure 1 As shown, information fusion refers to fusing voltage increment residuals and type coefficients to obtain the location, start and end times, and type of the fault. Specifically, the voltage residual refers to the residual between the actual voltage increment and the predicted voltage increment, provided by LSTM. The type coefficient is obtained using the correlation coefficient method, which calculates the correlation coefficient between the voltage increment residual sequence and the current difference sequence during the fault phase.

[0058] As circuit principles show, a short-circuit fault begins with a negative ΔL and ends with a positive ΔL, and the magnitude of ΔL is independent of the current value. However, when a poor contact fault occurs, ΔL is negative during the charging phase and positive during the discharging phase, exhibiting a strong correlation with the differential current. Theoretically, the Type of poor contact fault... c It should approximate 1. Therefore, when Type c A value greater than or equal to 0.8 indicates a poor contact fault, while when Type c A value less than 0.8 indicates a short circuit fault.

[0059] In summary, after training the network model for a healthy battery, the fault location and start / end time can be diagnosed using the current and voltage values ​​from the previous three time points and the current value at the current time point. Furthermore, the fault type can be distinguished using a small sample size of current and voltage data during the fault period.

[0060] As an application example:

[0061] Using lithium iron phosphate batteries with capacities of 2000mAh, nominal voltages, charging cut-off voltages of 3.2V, charging cut-off voltages of 3.65V, and discharging cut-off voltages of 2V as healthy batteries, the system operated for 11,500 seconds under UDDS conditions, which are more complex than those for energy storage batteries, and the SOC value decreased from 100% to 10%. Voltage and current were sampled during the operation at a sampling frequency of 1Hz to establish a healthy battery network model.

[0062] Three batteries were connected in series, designated as Battery 1, Battery 2, and Battery 3. The series-connected battery pack was operated under UDDS conditions for 11500 seconds, during which the SOC value decreased from 100% to 10%. Voltage and current were sampled during operation at a frequency of 1Hz. A 50mΩ resistor was used to simulate a poor contact fault in Battery 2, and a 2Ω resistor was used to simulate a short circuit fault in Battery 3. Battery 1 served as a control group and did not trigger any faults. The triggering timing and duration of each fault are shown in Table 1.

[0063] Table 1 Fault Triggering Timing

[0064]

[0065] The invention method was used to diagnose timing 1 of battery 2 and timing 1 of battery 3, and the sample entropy method was used for comparison. Figure 2 and Figure 3 As shown, the shaded area represents the fault triggering time region. It can be seen that the invented method can significantly and accurately reflect the fault start and end time, while the sample entropy method shows the phenomenon of missed diagnosis and misdiagnosis.

[0066] Simultaneously, the inventive method is used to diagnose timing 2 of battery 2 and timing 2 of battery 3, and the diagnostic results are as follows: Figure 4 As shown, this method can accurately reflect the start and end times of a fault.

[0067] To set a suitable fault threshold F (Th) The |ΔL| value was statistically analyzed when all three batteries were functioning correctly. The results showed that 99.937% of |ΔL| values ​​were less than 0.004V, and the remaining 0.063% fell between 0.004V and 0.0064V. Therefore, 0.008V was selected as the fault threshold F. (Th) It can ensure both extremely high diagnostic sensitivity and reliably prevent misdiagnosis.

[0068] After the fault occurs, the sampling sequence is substituted into equation (5) to calculate the type coefficient, and the results are shown in Table 2. It can be seen that by introducing the correlation coefficient algorithm between the differential current sequence and the tag residual, short-circuit faults and poor contact faults can be significantly distinguished.

[0069] Table 2 Fault Type Coefficients

[0070]

[0071] In summary, the method of the present invention can achieve highly sensitive and accurate diagnosis of minor faults by using only directly measurable current and voltage values. It can not only detect the type, location, and start and end time of the fault, but also maintain good performance under the full charge state of the same type of battery, and has high practical value.

[0072] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0073] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0074] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0075] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0077] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A fault diagnosis method for energy storage batteries based on multi-source information fusion, characterized in that, Includes the following steps: The voltage and current sampling sequence is processed by staggered sampling to obtain the voltage and current values ​​at previous times and the current value at the current time, which are used as feature inputs for the network model. Differential sampling is performed on the voltage sequence to calculate the voltage difference between the current time and the previous time, which is used as the label for the network model; By using the feature inputs and labels of healthy batteries, a network model is trained to obtain a network model of the current-voltage characteristics of healthy batteries. The current voltage increment is predicted using the current-voltage characteristic network model. Calculate the residual between the actual voltage increment and the predicted voltage increment, and determine the location and start and end times of the fault based on the residual.

2. The energy storage battery fault diagnosis method based on multi-source information fusion as described in claim 1, characterized in that, The specific process of performing staggered sampling on the voltage and current sampling sequence to obtain the voltage and current values ​​at previous times and the current value at the current time, which are used as feature inputs for the network model, includes: using a first-in-first-out (FIFO) queue to receive the current and voltage sampling values ​​at each time step, and the FIFO queue forming the feature input based on the earliest voltage and current values ​​and the latest current value.

3. The method for fault diagnosis of energy storage batteries based on multi-source information fusion as described in claim 1, characterized in that, The label is obtained by subtracting the voltage from the previous voltage at the current moment.

4. The energy storage battery fault diagnosis method based on multi-source information fusion as described in claim 1, characterized in that, The network model consists of an LSTM network followed by a fully connected FC layer.

5. A method for fault diagnosis of energy storage batteries based on multi-source information fusion as described in claim 1 or 4, characterized in that, The network model outputs a predicted label after each input feature. During training, the predicted label is used to subtract from the original label to optimize the network model's parameters.

6. The energy storage battery fault diagnosis method based on multi-source information fusion as described in claim 1, characterized in that, A fault is considered to have occurred when the residual exceeds the fault threshold.

7. A method for fault diagnosis of energy storage batteries based on multi-source information fusion as described in claim 1 or 6, characterized in that, in When distinguishing fault types, the Pearson correlation coefficient method is introduced to calculate the type coefficient. The Pearson correlation coefficient is calculated based on the differential current vector and the residual vector. When the type coefficient is greater than or equal to a set value, it is judged as a poor contact fault, and when the type coefficient is less than the set value, it is judged as a short circuit fault.

8. A fault diagnosis system for energy storage batteries based on staggered differential sampling, characterized in that, include: The feature calculation module is configured to perform staggered sampling processing on the voltage and current sampling sequence to obtain the voltage and current values ​​at previous times and the current value at the current time, which are used as feature inputs for the network model. The label module is configured to perform differential sampling on the voltage sequence, calculate the voltage difference between the current time and the previous time, and use it as a label for the network model. The model training module is configured to use the feature inputs and labels of healthy batteries to train the network model and obtain the network model of the current-voltage characteristics of healthy batteries. The fault diagnosis module is configured to use the current voltage characteristic network model to predict the normal voltage increment at the current moment, calculate the residual between the actual voltage increment and the predicted voltage increment, and determine the location and start and end time of the fault based on the residual.

9. A battery management system, characterized in that, It stores multiple instructions adapted for loading by the processor of a terminal device and executing the steps of the method according to any one of claims 1-7.

10. A terminal device, characterized in that, It includes a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store a plurality of instructions adapted to be loaded by the processor and executed as steps in the method of any one of claims 1-7.

Images