A series-connected battery fault prediction method

Abstract

The present invention discloses a method for predicting series-connected battery faults, and the method is as follows: obtaining battery string data through experiments; preprocessing the experimental data, and obtaining the increased energy E of the battery string by combining the preprocessed battery string data with the operating characteristics of the series circuit n ; using the energy E n to establish a fault factor database, and selecting three different groups of data in the database as training data A, training data B, and training data C respectively; constructing a grey model GMⅠ and a grey model GMⅡ according to the training data A and the training data B respectively; adding the output values of the two GM models, and then taking the average value, and using the average value and the E in the training data C n as inputs, and obtaining an accurate combined prediction model through training the combined model; inputting the energy ordinal number through the combined prediction model to obtain an energy prediction value. If the energy prediction value is less than the threshold, it indicates that the battery is normal; otherwise, it indicates that the battery has a fault, and then the fault type discrimination and location are carried out.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery energy storage system design and control, and relates to a series battery fault prediction method. Background Art

[0002] As the secondary chemical power source with the highest energy density, lithium-ion batteries also have many advantages such as light weight, pollution-free, high safety and reliability, long life, high charge and discharge efficiency, etc. Since the capacity and voltage level of a single battery usually cannot meet the actual use requirements, hundreds of single batteries need to be connected in series and parallel to form a battery system to increase the capacity and output current of the battery system. However, with the increase in the number of batteries, the problem of inconsistency will inevitably arise. Once a certain battery fails, the system maintenance cost will be huge, and it may even lead to systemic risks and catastrophic consequences, seriously affecting the development and application of the battery system. Therefore, establishing an accurate series battery fault prediction method to accurately predict faults is crucial for its design, control and engineering applications.

[0003] Currently, research and patents on battery modeling at home and abroad mostly focus on single battery cells. There is not much literature on the fault diagnosis of series-connected batteries. Patent CN202211171127.0 invented a multi-parameter joint diagnosis method for electric vehicle battery faults. The method is as follows: obtain the time-series values of the characteristic parameters of each single battery in the power battery of the electric vehicle to be diagnosed, establish a time-series matrix of characteristic parameters, and calculate the variance of the characteristic parameters row by row to establish a variance matrix of characteristic parameters. According to a sliding window with a preset initial length, slide from top to bottom across the variance matrix of characteristic parameters, and screen for outliers in the variance within the sliding window until the variance matrix is traversed. If there is an abnormal variance within the sliding window, the moment corresponding to the initial abnormal variance is the occurrence time of the battery fault. At the same time, re-intercept the time-series matrix of characteristic parameters within a certain period of time before the occurrence time of the identified fault to form a sub-matrix, calculate the variance column by column for the sub-matrix, and use the outlier recognition algorithm to identify the abnormal variance and the corresponding serial number. The serial number of the abnormal variance is the position of the identified faulty cell, which can quickly locate the occurrence time and specific location of the battery fault. The advantage of this patent is that it summarizes all possible abnormal data by establishing a time-series matrix of characteristic parameters. The disadvantage is that it cannot predict potential faults and eliminate fault hidden dangers in advance to avoid the adverse effects caused. Patent application (CN202210723748.9) discloses a method for determining and processing faulty modules in a battery pack. The method is as follows: in the OCV state, obtain and record the voltage values of each battery module in the battery pack; calculate the average voltage of the battery pack by obtaining the voltage values of each battery module; compare the voltage value of each battery module with the average voltage of the battery pack to determine whether the corresponding battery module has a fault. The advantage of this patent application is that it avoids misjudgment of faults to a certain extent by collecting the average voltage. The disadvantage is that it cannot solve the temporarily occurring voltage abnormal faults. Patent (CN202210872840.1) obtains the voltage data of the power battery pack at set time intervals; constructs the voltage data into several first voltage time series, and compares the voltage of each first voltage time series with a preset voltage threshold to preliminarily diagnose whether the power battery pack has a fault; when it is preliminarily diagnosed that the power battery pack has no fault, perform phase space reconstruction on all first voltage time series to obtain several second voltage time series; use the fuzzy membership function to calculate the similarity between any two second voltage time series, and then calculate the fuzzy entropy value of each second voltage time series in turn; according to the comparison result between the fuzzy entropy value and the preset fuzzy entropy value threshold, finally diagnose the faults of the power battery pack and their occurrence time within each set time interval online. The advantage of this patent is that it improves the diagnosis efficiency through the method of phase space reconstruction. The disadvantage is that the whole process relies heavily on training data.According to the advantages and disadvantages of the above-mentioned patents and patent applications, the present invention extracts special fault factors and uses a small amount of training data. Through a combined prediction model, it can not only diagnose faults in series-connected batteries, but also predict potential faults, simplify some processes, and proposes a method for predicting faults in series-connected batteries. Summary of the Invention

[0004] The problem to be solved by the present invention is to provide a method for predicting faults in series-connected batteries. On the one hand, it solves the problem of inaccurate characterization of fault factors in series-connected batteries, and can not only diagnose existing faults in series-connected batteries, but also predict potential faults in series-connected batteries; on the other hand, it simplifies the fault location process, classifies fault types, and improves the efficiency and effectiveness of fault prediction.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] The present invention provides a method for predicting faults in series-connected batteries. The battery string is composed of m battery monomers connected in series, where m is a natural number greater than 1, and its structural diagram is as Figure 1 shown.

[0007] A method for predicting faults in series-connected batteries is as follows:

[0008] Obtain the current I of the battery string parameters through experiments. Combining the working characteristics of the series circuit, when the voltage of the battery monomer to be predicted in the battery string reaches the terminal voltage U of constant current charging H , the time T H required for other battery monomers to reach the voltage U i (i = 1, 2, 3... m - 1) is statistically analyzed to obtain m - 1 time values. Preprocess the data, eliminate the battery monomers with incomplete voltage or time data, and record the remarks. From the preprocessed battery monomer data (including voltage, current, time T i ), according to the formula ΔE i = U H IT i (i = 1, 2, 3... m - 1), calculate the energy increase of each battery within the time T i , and then according to the formula accumulate the ΔE i values to obtain the energy increase E n of the battery string during each charge (n represents the number of times, n = 1, 2, 3...). The schematic diagram of the energy increase of the series-connected battery is as Figure 2 shown.

[0009] Taking E nTake numerical values as fault factors, establish a fault factor database, and select three different groups of data from the database as training data A, training data B, and training data C respectively. Among them, the number of fault factors in training data A and training data B is maintained between 4 and 10, and the number of fault factors in training data C is maintained between 10 and 200.

[0010] According to training data A and training data B, construct grey model GMⅠ and grey model GMⅡ respectively. Add the output values of the two GM models and then calculate the average value as the input of the training combined model. Use the average value and E in training data C n as the input, and through the training combined model, obtain an accurate combined prediction model. The grey model GMⅠ is a GM(1,1) model, and its modeling process is as follows:

[0011] Construct the original non - negative data sequence x (0) =(x (0) (1),x (0) (2),...x (0) (n)), where n represents the length of the original data sequence, and n≥4.

[0012] According to k = 1, 2, 3,..., n.

[0013] Obtain the new accumulated data sequence x (1) =(x (1) (1),x (1) (2),...x (1) (n)).

[0014] Define the first - order linear grey differential equation of the GM(1,1) model as:

[0015]

[0016] The model parameter vector is:

[0017]

[0018] Among them,

[0019] Matrix

[0020] Y = [x (0) (2),x (0) (3),...x (0) (n)] T ,

[0021] Matrix

[0022]

[0023] Z (1)= (Z (1) (2), Z (1) (3),... Z (1) (n)),

[0024] wherein,

[0025] is the background value, k = 1, 2, 3,..., n - 1.

[0026] The discrete solution of the original differential equation is:

[0027]

[0028] wherein, k = 1, 2, 3,..., n.

[0029] When the initial condition the restoration formula of the original sequence can be obtained as:

[0030]

[0031] wherein, k = 1, 2, 3,..., n.

[0032] The described grey model GMⅡ is an improvement of the GM(1,1) model. In the specific modeling process, data conversion is performed on the original data, and the background value construction formula of the GM(1,1) model is optimized by using combined interpolation. Its background value construction formula is:

[0033]

[0034] wherein, k = 1, 2, 3,..., n - 2.

[0035] Its solution process is the same as that of GM(1,1) modeling.

[0036] As Figure 3 shown, the described training combined model forms a more accurate combined prediction model by training SVM, together with the grey model GMⅠ and the grey model GMⅡ.

[0037] Taking the ordinal number n to be predicted as the input, through the combined prediction model, the corresponding energy prediction value E m is obtained. When E m is less than the set threshold, it indicates that the battery is normal; otherwise, it indicates that the battery has a fault. Then, the fault type determination is carried out, and finally the fault location is carried out. Regarding the setting of the threshold E p , it is related to the number of battery monomers m in the battery string, the constant current charging time t of the normal battery, and the battery parameters. It can be set according to the formula E p = 0.2U H It(m - 1).

[0038] After predicting a fault in a battery cell, the serial number of the corresponding battery cell in the output database is output, and the specific location of the faulty battery cell is determined according to the serial number to achieve the fault location function. For example Figure 4 As shown, when the fault analysis module determines a fault, the serial number of the battery cell is traced according to the predicted value of the combined model; using other stable battery cells in the battery string as reference objects, a data set containing the historical T i value of the faulty cell is established; by re - establishing the GM(1,1) model, a simple prediction is directly made on its T i value; when both the T i value in the data set and the predicted value tend to 0, it is determined that the battery cell has an emergency fault, is in a severe degradation state or in a short - circuit state, and requires immediate repair and replacement by maintenance personnel; when the predicted value shows a slow increase, it is determined that the battery cell is in a non - emergency fault state and may develop into an emergency fault in the future, and maintenance personnel need to select a reasonable solution.

[0039] The fault prediction method described in the present invention is applicable to both lithium batteries and lead - acid batteries and nickel - cadmium batteries.

[0040] The fault prediction method described in the present invention is applicable to the series connection grouping methods of battery cells, battery modules, and battery packs. Brief Description of the Drawings

[0041] Figure 1 is a schematic diagram of the structure of a series - type battery system;

[0042] Figure 2 is a schematic diagram of the energy increase of a series - type battery;

[0043] Figure 3 is a block diagram of the fault prediction method for a series - type battery;

[0044] Figure 4 is a schematic diagram of fault type diagnosis. Detailed Embodiment

[0045] The following further elaborates on the present invention with specific examples, which is an explanation rather than a limitation of the present invention.

[0046] The series battery system introduced in the specific example is composed of 10 battery cells connected in series, numbered #1, #2, #3, #4, #5, #6, #7, #8, #9 and #10. The rated voltage of each battery cell is 3.7V, the rated capacity is 860mAh, and the discharge cut-off voltage is 3V. Charging with a constant current of 250mA, at the end of constant current charging, the voltage of battery cell #3 first reaches 3.7V and the voltage is stable. At this time, the other 9 cells are timed, and the time required for their voltage to reach 3.7V is 522s, 533s, 545s, 553s, 567s, 559s, 547s, 531s, and 511s respectively. Because the battery data is complete, the data preprocessing process does not need to eliminate battery cells with incomplete voltage or time data.

[0047] Calculate the battery energy added by the 9 batteries in the corresponding time.

[0048] According to the following formula

[0049] ΔE i =U H IT i

[0050] Calculate the battery energy ΔE of each battery within the time Ti i . After calculation, ΔE i The values ​​are 482.85, 493.025, 504.125, 511.525, 524.465, 517.075, 505.975, 491.175, and 472.675 respectively.

[0051] According to the following formula

[0052]

[0053] Calculate the battery energy E added by the battery string within time Ti in each charge n (n represents the number of times, n = 1, 2, 3...) The 9 values ​​are accumulated to get the energy change value of the charging battery string this time, which is 4502.89, which can be recorded as E 1 =4165.173. In the next charging process, similarly, we can get E 2 . By calculating this method, 20 E n Numeric value.

[0054] According to 20 E n Numerical values, establish a fault factor database, select three different groups of data from the database as training data A (including E 1 To E 10 , 10 fault factors), training data B (including E 3 To E 12, 10 failure factors) and training data C (including E 13 to E 32 , 20 failure factors).

[0055] According to training data A and training data B, GM model I and GM model II are respectively constructed, and the predicted values corresponding to the ordinals in training data C are respectively output (prediction results of 2 groups of E 13 to E 32 );

[0056] Add the predicted result values of the 2 groups of E 13 to E 32 numerically, and find the average value as the input of the training combined model. Then, use the values of E 13 to E 32 in training data C as the input to train the combined model to obtain an accurate combined prediction model;

[0057] Threshold setting of the fault analysis module.

[0058] According to the following formula

[0059] E p = 0.2U H It(m - 1)

[0060] A preliminary value of 11988 can be obtained. Make a small - range floating of the value according to the actual situation of the battery system. Here, the threshold can be set to 11000;

[0061] It is planned to make sampling predictions for the next 30 times of battery cell #3. Input the ordinals 33, 51, 72, and after prediction by the combined prediction model, 3 predicted values E m (m = 33, 51, 72);

[0062] When the values of E 33 , E 51 , E 72 are less than the set threshold, the fault analysis module outputs that the predicted battery is normal; when there is one or more values in E 33 , E 51 , E 72 less than the set threshold, trace the battery cell number according to the combined model prediction value. Take other stable battery cells in the battery string as the reference object to establish a dataset containing the historical T i values. By re - establishing the GM(1,1) model, directly make a simple prediction for its T i . When the T iBoth the measured value and the predicted value tend to 0, which indicates that the battery cell has an emergency fault, is in a severe degradation state or a short-circuit state, and requires immediate repair and replacement by maintenance personnel; when the predicted value shows a slow increase, it is determined that the battery cell is in a non-emergency fault state and may develop into an emergency fault in the future, and maintenance personnel need to select a reasonable solution. Here, the battery cell serial number #3 corresponding to the output database is used to determine the specific location of the faulty battery cell in the battery system according to the serial number #3, realizing the function of fault location.

Claims

1. A method for predicting faults in series-connected batteries, where the series-connected batteries are formed by connecting m battery cells in series, and m is a natural number greater than 1; The method includes the following steps: Step 1: Obtain the current I of the battery string parameters through experiments. Combining with the working characteristics of the series circuit, when the voltage of a certain battery cell in the battery string reaches the end voltage U of constant current charging H the time T required for other battery cells to reach the voltage U H is statistically analyzed for i = 1, 2, 3... m - 1 i ​ ； Preprocess the data to eliminate battery cells with incomplete voltage or time data; Step 2: Preprocessed battery cell data, including voltage, current, time T i , according to the formula △E i =U H IT i Calculate and obtain the time T of each battery i Increased battery energy △E i , △E i The value is accumulated to get the energy E added by each charging battery string n , n represents the number of times, n = 1, 2, 3...; Step 3: Using E n as the fault factor, establish a fault factor database, and select three different sets of data from the database as training data A, training data B, and training data C respectively; Step 4: Construct Grey Model GMⅠ and Grey Model GMⅡ based on Training Data A and Training Data B respectively; Step 5: Add the output values of the two GM models and then calculate the average value, which is used as the input for training the combined model. The average value and E in the training data C n are used as inputs. Through the trained combined model, an accurate combined prediction model is obtained; Step 6: Use the ordinal number n to be predicted as the input, and through the combined prediction model, obtain the corresponding energy prediction value E m ; Step 7: When E m is less than the set threshold value, it indicates that the battery is normal; otherwise, it indicates that the battery has a fault. Then, the fault type is determined, and finally, the fault location is carried out.

2. A method for predicting faults in series-connected batteries according to claim 1, characterized in that, The increased energy E of the battery string n is generated as follows: (1) When the voltage of a certain battery cell in the battery string reaches U H measure the time T H required for the other battery cells to reach the voltage U i , where i = 1, 2, 3... m - 1; (2) According to the following formula △E i = U H IT i Calculate and obtain the battery energy ΔE increased by each battery within time T i where i = 1, 2, 3... m - 1; i ​ (3) According to the following formula Calculate the battery energy E increased by the battery string during each charge at time T i where n represents the number of times, n = 1, 2, 3... n where n represents the number of times, n = 1, 2, 3...

3. A method for predicting faults in series-connected batteries according to claim 1, characterized in that, The steps for generating the fault type are as follows: (1) After the fault analysis module determines a fault, trace the battery cell number according to the combined model prediction value; (2) Using other stable battery cells in the battery string as reference objects, establish a data set containing the historical T i values; (3) By re - establishing the GM(1,1) model, directly perform a simple prediction on its T i ; (4) When both the T i value and the predicted value in the data set tend to 0, it is determined that the battery cell has an emergency fault, is in a severely degraded state or in a short - circuit state, and requires immediate repair and replacement by maintenance personnel; When the prediction value shows a slow increase, it is determined that the battery cell is in a non-emergency fault state and may develop into an emergency fault in the future, and maintenance personnel need to select a reasonable solution.

4. A method for predicting faults in series-connected batteries according to claim 1, characterized in that, The fault location step is: when a fault in the predicted battery cell is detected, output the battery cell number corresponding to the database, and determine the specific location of the faulty battery cell according to the number to achieve the fault location function.

5. A method for predicting faults in series-connected batteries according to claim 1, characterized in that, The battery is a lithium battery, a lead-acid battery or a nickel-cadmium battery.

Images