A power battery fault analysis method for energy storage based on weighted fusion

Abstract

The application discloses a power battery fault analysis method based on weighted fusion, comprising the following steps: S1, collecting operation data of the power battery under multiple preset working conditions; S2, preprocessing the operation data and extracting key data features; S3, constructing a fault type identification model and outputting a fault type, if the fault type comprises a sub-fault type, executing step S4, otherwise, executing step S5; S4, constructing a sub-fault type identification model and outputting a sub-fault type; S5, verifying the fault type or the fault sub-type and outputting a verification conclusion and a maintenance suggestion. Through the two progressive models and the dynamically adjusted weights, the application realizes coarse positioning to accurate positioning of the battery fault, and after the working condition reproduction verification, the core fault type is output through the unique fault determination mechanism, so that the problems of qualitative ambiguity of the traditional diagnosis method, low efficiency of the maintenance decision and inaccurate fault positioning are solved.

Description

Technical Field

[0001] This invention relates to the field of power battery fault diagnosis technology, specifically to a power battery fault analysis method based on multi-dimensional data acquisition, weighted average, a three-level fault location model, and dual-screening output. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the safety and reliability of power batteries, as core components, directly affect the operational safety of the entire vehicle. During long-term use, power batteries are prone to various faults such as sampling chip failure, sensor aging, unstable communication, excessive voltage drop, and overcurrent / overdischarge. Furthermore, some major fault categories contain multiple sub-faults; for example, sampling chip failure can be further subdivided into voltage sampling abnormalities, temperature sampling abnormalities, etc. Failure to promptly and accurately diagnose major fault categories and precisely locate sub-faults may lead to battery performance degradation, thermal runaway, or even safety accidents.

[0003] Existing power battery fault diagnosis methods have significant drawbacks: Diagnosis is mostly qualitative, only identifying the presence of a fault but failing to quantify its probability of occurrence. This makes it difficult for maintenance personnel to prioritize faults, resulting in low maintenance efficiency. Fault matching logic is simplistic, relying heavily on single data features or fixed thresholds without dynamically adjusting judgment criteria based on operating conditions. This makes them susceptible to fluctuations in operating conditions, leading to a relatively high false positive rate. Furthermore, they lack sensitivity in identifying gradual faults such as NTC aging and lack a detailed fault location mechanism, only able to identify broad fault categories without accurately determining specific fault types and physical locations. This results in blind troubleshooting during maintenance, leading to low repair efficiency. Weighting coefficients are often manually set, resulting in strong subjectivity and inconsistent diagnostic results due to differences in settings by different personnel, affecting the method's universality and accuracy.

[0004] Therefore, there is an urgent need for a power battery fault analysis method that can quantify the probability of failure, dynamically adjust the judgment logic based on operating conditions, has the ability to accurately locate faults in detail, and determines the weight coefficients through algorithms, in order to overcome the shortcomings of existing technologies. Summary of the Invention

[0005] This invention addresses the problems in existing technologies by disclosing a power battery fault analysis method based on weighted fusion. This invention progressively locates fault types by setting up a fault identification model and a sub-fault identification model. The model uses dynamically adjusted weights to assist in the accurate location of battery faults. After verification through operating condition reproduction, the core fault type is output through a unique fault determination mechanism, which solves the problems of vague qualitative analysis, inefficient maintenance decisions, and inaccurate fault location in traditional diagnostic methods.

[0006] This invention is achieved through the following technical solution:

[0007] This invention first provides a power battery fault analysis method based on weighted fusion, comprising the following steps:

[0008] S1. Collect operating data of the power battery under various preset operating conditions;

[0009] S2. Preprocess the running data and extract key data features;

[0010] S3. Construct a fault type identification model and output the fault type. If the fault type includes sub-fault types, proceed to step S4; otherwise, proceed to step S5.

[0011] S4. Construct a sub-fault type identification model and output the sub-fault types;

[0012] S5. Verify the fault type or fault subtype, and output the verification conclusion and maintenance recommendations.

[0013] As a further embodiment, the operating data includes data collected by the BMS on individual cell voltage, individual cell temperature, total voltage, total current, insulation resistance, and differential pressure, as well as data acquisition channel identifiers.

[0014] As a further option, the preset operating conditions include the initial power-on condition, the continuous driving condition, the temperature rise condition, and the room temperature rest condition.

[0015] As a further solution, the initial power-on condition is the startup phase within 0 to 30 seconds after the vehicle is powered on, during which initial data is collected and abnormal sensor initial response conditions are captured.

[0016] As a further solution, the uninterrupted driving condition is that the vehicle travels at a constant speed of 30-60 km / h for 30-60 minutes. The speed and time parameters are obtained through stability verification under the constant speed condition. Real-time current and voltage data are collected to identify current jumps under the constant speed condition.

[0017] As a further approach, the temperature rise condition involves raising the battery pack temperature from the current room temperature to 40°C to 50°C through charging / discharging or heating in an ambient chamber. This temperature range is obtained through NTC sensor response characteristic tests, and temperature data is collected during the temperature rise process to verify the NTC sensor's response capability to temperature changes.

[0018] As a further solution, the room temperature static condition involves placing the battery pack in a room temperature environment of 20 to 25°C for more than 2 hours. After the internal temperature stabilizes, data is collected to eliminate interference from other operating conditions and determine the temperature error caused by NTC aging.

[0019] As a further option, key data characteristics include jump points and jump differences, fixed extreme values, unchanged values, null and invalid values, and deviations.

[0020] As a further approach, the jump point and jump difference refer to the numerical abrupt change of adjacent sampling points in the data sequence. The abrupt change difference is calculated by subtracting the sampling value from the sampling value at the previous time. If the difference is greater than 1.5 to 2 times the normal fluctuation range, it can be determined as a data jump point.

[0021] As a further approach, the fixed extreme value refers to the extreme value that remains unchanged over a long period of time in temperature sampling. The values ​​of 10 consecutive sampling points remain unchanged and are within the extreme value range of the sensor's measurement range. These 10 sampling points serve as the criterion for verifying the fit based on data stability.

[0022] As a further option, "unchanging value" refers to the NTC sampling value that does not change with temperature under temperature rise conditions, or data that remains unchanged throughout the driving process.

[0023] As a further solution, null and invalid values ​​refer to abnormal data where the data field is missing, has no value, or has three consecutive invalid sampling points. The three sampling points are obtained by fitting based on the accuracy verification of invalid value judgment.

[0024] As a further option, the deviation refers to the degree to which each parameter deviates from the normal threshold.

[0025] As a further embodiment, S3 includes:

[0026] S31. Construct a fault type library based on the types of power battery faults. The fault type library includes fault types and sub-fault characteristics.

[0027] S32. Determine the initial weight W0, frequency weight coefficient α, and deviation weight coefficient β of the fault through a dynamic algorithm;

[0028] S33. Obtain the final weight W of the fault and dynamically adjust the final weight according to the fault type;

[0029] S34. Construct the mapping logic of fault type, final weight, and fault characteristics, and output the fault type.

[0030] As a further embodiment, S32 includes:

[0031] S321. Obtain the initial weight W0:

[0032] ,(1);

[0033] Among them, k1 and k2 are the optimal coefficients, F represents the historical frequency of fault occurrence, which is determined based on historical fault data, and H represents the quantitative value of the fault severity, which is divided into 5 levels according to the degree of impact of the fault on equipment safety and operation.

[0034] Based on historical fault big data of similar power batteries, a sample set Y is constructed.

[0035] Y={( , , )}(4);

[0036] in, Let be the frequency of the fault occurrence history of the i-th sample. This is the quantification value of the fault severity for the i-th sample. Let the target initial weights be those for the i-th sample.

[0037] The optimal coefficients k1 and k2 are solved using the gradient descent algorithm. The objective function is to minimize the sum of squared errors between the theoretically calculated weights and the target initial weights. The objective function is:

[0038] (5);

[0039] in, Minloss is the objective function of gradient descent, which is the final goal of the iteration, and Loss is the sum of squared errors actually calculated in a single iteration.

[0040] The constraints are determined by statistical analysis of historical data. The constraints are that the sum of W0 must be 1.0, the initial weight W0 of a single fault type must fall within the interval [0.1, 0.3], the value range of k1 is [0.001, 0.005], and the value range of k2 is [0.02, 0.06].

[0041] The initial iteration values ​​of k1 and k2 are 0.003 and 0.04, respectively;

[0042] Calculate the value of each sample (Theoretically calculated initial weights of the i-th sample) and Loss (sum of squared errors in a single iteration), solve for the gradient:

[0043] (6); (7);

[0044] This represents the partial derivative of the objective function Loss with respect to the optimal coefficient k1, reflecting the degree to which a small change in k1 affects the sum of squared errors, Loss, and is used to guide the direction and magnitude of k1 updates. The same applies to k2.

[0045] The coefficients k1 and k2 are updated using an adaptive step size. The update formula is as follows:

[0046] (8);

[0047] (9);

[0048] k1^new and k2^new are the next update values ​​of k during the iteration process. With an initial adaptive step size of 0.05, repeat the above k1^new and k2^new update steps until Loss ≤ 0.001 or the number of iterations reaches 1000, and finally output the optimal value. , Substituting into formula (8), we obtain the initial weight W0;

[0049] S322, Weighting coefficient for the number of acquisitions:

[0050] The frequency weighting coefficient for each fault type is denoted as α, where α ∈ [0.05, 0.1].

[0051] α = 0.05 + 0.05 × (16)

[0052] in, The Pearson correlation coefficient represents the number of times a feature occurs (N) and the actual probability of a fault occurring (P_actual). ∈[-1,1];

[0053] S323. Obtain the deviation weighting coefficient:

[0054] The bias weighting coefficient is denoted as β:

[0055] β=0.1+0.1×(I / 1.0), (17);

[0056] Where I is the parameter impact factor, I = (frequency of safety accidents caused by parameter failure / total frequency of safety accidents) × 0.5 + (performance degradation caused by parameter failure / maximum allowable performance degradation) × 0.5, and the value range of I is [0.3, 0.7].

[0057] As a further embodiment, S33 includes:

[0058] S331. Obtain the final weight W:

[0059] (18);

[0060] in, Δ represents the initial weights after normalization, N is the number of times the data features match the fault characteristics, and Δ is the deviation between the data features and the normal threshold.

[0061] S332. Dynamically adjust the final weight according to the fault type. The methods for adjusting the final weight include:

[0062] Sampling chip failure: When the difference between the voltage, temperature, current, and insulation resistance sampling values ​​is greater than the preset threshold, or the number of invalid sampling values ​​exceeds 5, the final weight W will increase by 0.1-0.2 for each occurrence of the above features, with a maximum adjustment to 0.8.

[0063] Temperature sampling anomaly sub-fault: If the NTC is broken, short-circuited, or under temperature rise conditions, the values ​​of 20 consecutive sampling points of the NTC to be tested remain unchanged, and the sampling values ​​of other NTCs change normally, the final weight W will be directly adjusted to 1.0.

[0064] Current sampling anomaly sub-fault: Under continuous constant speed driving conditions, the total current shows a jump point corresponding to no change in operating conditions. Excluding the current change caused by normal operating conditions, the final weight W is directly adjusted to 1.0.

[0065] Communication failure: Null values ​​appeared in the collected running data. After excluding invalid values ​​and 0 values ​​caused by sampling anomalies, the final weight W was increased by 0.3-0.5.

[0066] Differential voltage fault: The maximum difference in individual unit voltage exceeds 0.2V and lasts for more than 5 seconds, while the total voltage data does not change. The final weight W increases by 0.2-0.3.

[0067] Overcurrent fault: The absolute value of the total current exceeds 1.2 times the rated current and the duration exceeds 2 seconds, and the final weight W is increased by 0.3-0.5.

[0068] Over-discharge fault, individual cell voltage below 2.5V and total current equal to discharge current, ultimately increase weight W by 0.3-0.5.

[0069] As a further embodiment, S34 includes:

[0070] S341. Match fault types using data collected from the host computer;

[0071] S342. Adjust the weights corresponding to the fault types. After all fault weights are calculated, filter the faults by calculating the fault probability P-value. The fault probability is:

[0072] P = (Final weight of the fault W / Sum of final weights of all faults) × 100% Quantifies the probability of a fault occurring, determines the final weight W of the fault, calculates the proportion of W to the sum of final weights of all faults, and multiplies this proportion by 100% to obtain 0-100%.

[0073] When the failure probability P > 50%, the failure type is output as a candidate failure type. If all failure probabilities P ≤ 50%, the 1-2 failure types with the highest probabilities are selected as candidate failure types for output.

[0074] As a further embodiment, S4 includes:

[0075] S41. Construct a fault feature library, which includes fault types, sub-fault types, and features of fault types.

[0076] S42. Determine the sub-fault type according to the sub-fault type filtering rules;

[0077] Sub-fault type filtering rules include:

[0078] Set the feature matching degree M:

[0079] M = (Number of features matching the subdivision / Total number of fault features in this subdivision) × + (Cumulative deviation of subdivided features / Maximum permissible deviation of subdivided features) × Where ω1 is the feature number weight, and the total number of features = 1. =0.4, total characteristic number =2, =0.45, Total characteristic number =3 =0.55, The cumulative weight for the deviation is 1- ;

[0080] When the feature matching degree M of a certain sub-fault is greater than or equal to 0.7, the sub-fault is output as a candidate sub-fault. If there are multiple sub-faults with a feature matching degree M greater than or equal to 0.7, the sub-faults are output in order of priority according to the value of M.

[0081] As a further embodiment, S5 includes:

[0082] S51. Verify the correlation between fault types and fault characteristics.

[0083] S52. Establish physical location mapping rules.

[0084] S53. Verify the candidate fault types by reproducing the operating conditions, and output the verification conclusions and maintenance recommendations;

[0085] Preferably, the method for verifying the operating condition reproduction of candidate fault types includes:

[0086] Communication failure: The system needs to be restarted to reproduce the power-on and driving conditions. Observe whether the null values ​​in the data are reproduced. If they are reproduced, calculate the final weight after reproduction. If they are not reproduced, output the weight before reproduction and output a reproduction failure message.

[0087] Temperature sampling anomaly: The temperature rise condition or room temperature static condition needs to be repeated to verify whether the characteristic of no temperature change or excessive error continues. If it is reproduced, the final weight after reproduction is calculated. If it is not reproduced, the weight before reproduction is output and a reproduction failure prompt is output.

[0088] Current sampling anomaly: The continuous driving condition needs to be repeated to observe whether the current jump characteristics are reproduced. If they are reproduced, the final weight after reproduction is calculated. If they are not reproduced, the weight before reproduction is output and a reproduction failure prompt is output.

[0089] Differential pressure fault: After standing for 2 hours, remeasure the individual unit voltage to verify whether the differential pressure is still excessive. If it is reproduced, calculate the final weight after reproduction. If it is not reproduced, output according to the weight before reproduction and output a reproduction failure prompt.

[0090] For overcurrent faults, over-discharge faults, and sampling chip faults, the corresponding triggering conditions need to be reproduced to verify whether the fault characteristics persist. If they are reproduced, the final weight after reproduction is calculated. If they are not reproduced, the weight before reproduction is output, and a reproduction failure message is output.

[0091] After reproduction, the final weight and probability of the fault are recalculated. For fault types that still meet the output conditions after reproduction, the comprehensive priority score S is calculated, and the fault type with the highest S is taken as the unique output fault type. If S is equal, the fault with more reproduction times is selected as the output fault type. The output results include fault type or fault subtype, fault probability, judgment basis, priority, unique fault judgment score, fault physical location and maintenance suggestions.

[0092] Preferably, the maintenance recommendations include:

[0093] Voltage sampling abnormality: The physical location is the corresponding voltage sampling channel and chip pin. It is recommended to replace the sampling chip or related sampling module.

[0094] Temperature sampling abnormality: If the NTC sensor is abnormal, the physical location is the corresponding NTC installation location. It is recommended to replace the faulty NTC sensor.

[0095] Current sampling abnormality: The physical location is the current sensor or its wiring; the current sensor needs to be replaced.

[0096] Insulation resistance sampling abnormality: The physical location is the insulation resistance sampling sensor, which needs to be calibrated or replaced.

[0097] Communication failure: The physical location is the connection harness or interface between the battery pack and the BMS. It is recommended to check whether the harness is loose or the interface is oxidized or corroded, clean the interface or tighten the harness again.

[0098] Differential voltage fault: The physical location is a single cell or module with abnormal voltage. It is recommended to perform cell balancing or replace the cells.

[0099] Overcurrent fault: The physical location is the charging / discharging circuit or current sensor. It is necessary to check whether the circuit is short-circuited, whether the sensor is abnormal, and to eliminate potential safety hazards. Over-discharge fault: The physical location is a single cell with too low voltage. It is necessary to check the health status of the cell and replace it if necessary.

[0100] Over-discharge fault: The physical location is the BMS and voltage sensor. It is necessary to check whether the BMS over-discharge protection parameter settings are reasonable or whether the voltage sensor error is too large. Recalibrate the voltage sensor or modify the BMS over-discharge protection parameters.

[0101] The features and beneficial effects of this invention are as follows:

[0102] (1) This invention classifies the causes of battery failures according to the subordinate relationship in the first-level fault identification model. First, the major categories of fault types are determined, and then the fault types are further subdivided into sub-faults. The second-level sub-fault identification model is used for further classification, thereby accurately locating the fault type. At the same time, the location results are verified to ensure the accuracy of the results. In addition, this method also sets the priority order of faults according to the special circumstances of battery use, effectively ensuring that emergency treatment is carried out first, and then stability is sought. In the battery verification stage, other factors affecting battery verification are set. This method solves the problems of qualitative ambiguity, inefficient maintenance decision-making, and inaccurate fault location in traditional diagnostic methods. It effectively improves the accuracy, precision and practicality of power battery fault diagnosis. It can accurately identify faults such as sampling chip faults, excessive voltage difference faults, overcurrent faults, over-discharge faults, and unstable communication, quantify the fault probability and focus on high-priority faults.

[0103] (2) This invention verifies the output results of the model by reproducing the fault type or sub-fault type input by two models. It not only verifies whether the model is useful and whether the accuracy is correct, but also achieves accurate positioning by combining the correlation of working condition data. The correlation verification requires analyzing the correlation between fault characteristics and other operating parameters to avoid interference from external factors. At the same time, it outputs corresponding maintenance suggestions according to the fault type, which is beneficial for staff to directly refer to the suggestions according to the fault type, greatly improving work efficiency and maintenance accuracy of faults.

[0104] (3) By dynamically adjusting the weights, the weights can be dynamically modified according to the real-time status and environmental changes. This makes the decision-making system upgrade from "static and rigid" to "adaptive and highly flexible", always conforming to the core needs of the real scenario. The dynamic weight adjustment process and effect data can be accumulated into scenario-based decision-making experience, and at the same time provide quantitative basis for model iteration. Attached Figure Description

[0105] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0106] Figure 1This is a flowchart of a power battery fault analysis method based on weighted fusion, as described in an embodiment of the present invention. Detailed Implementation

[0107] To facilitate understanding of the present invention, a more comprehensive description of the present invention will be given below, and embodiments of the present invention will be provided, but this does not limit the scope of the present invention.

[0108] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0109] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0110] Existing methods for diagnosing power battery faults mostly rely on manual judgment. Since there are many factors that can cause battery faults, manual judgment is affected by work experience and subjective factors, resulting in a high rate of misjudgment. Furthermore, the diagnostic results of different personnel are often inconsistent, making it impossible to accurately determine the specific fault type and physical location. This leads to blind troubleshooting during maintenance and low repair efficiency. Therefore, this invention provides a power battery fault analysis method based on weighted fusion. The method classifies the causes of battery faults according to their hierarchical relationships using a first-level fault identification model. First, it determines the major categories of fault types, then further subdivides each type into sub-faults. A second-level sub-fault identification model is used for further classification, enabling precise fault location. The location results are verified to ensure accuracy. To ensure the accuracy of the classification process, dynamically adjustable weights are used to assist in adjusting the method based on actual battery operation, making it more practical. Furthermore, this method sets a fault priority order based on the specific circumstances of battery use, effectively ensuring that emergency repairs are addressed first, followed by stability maintenance. In the battery verification stage, other factors affecting battery verification are considered. This method solves the problems of qualitative ambiguity, inefficient maintenance decisions, and inaccurate fault location in traditional diagnostic methods, effectively improving the accuracy, precision, and practicality of power battery fault diagnosis. It can accurately identify faults such as sampling chip faults, excessive voltage difference faults, overcurrent faults, over-discharge faults, and unstable communication, quantifying fault probabilities and focusing on high-priority faults.

[0111] A power battery fault analysis method based on weighted fusion includes the following steps:

[0112] S1. Establish a communication connection with the power battery BMS through the host computer software, and collect the operating data of the power battery under various preset working conditions.

[0113] The operational data includes, but is not limited to, data that can be collected by the BMS, such as individual cell voltage, individual cell temperature, total voltage, total current, insulation resistance, and differential pressure. The data acquisition channel identifier is also recorded for subsequent physical location mapping.

[0114] Preset operating conditions include, but are not limited to, initial power-on operating conditions, continuous driving operating conditions, temperature rise operating conditions, and room temperature static operating conditions.

[0115] In some embodiments, the initial power-on condition is the startup phase within 0 to 30 seconds after the vehicle is powered on, during which initial data is collected and abnormal conditions of the sensor's initial response are captured, such as chip power-on sampling delay.

[0116] In some embodiments, the uninterrupted driving condition is when the vehicle travels at a constant speed of 30-60 km / h for 30-60 minutes. The speed and time parameters are obtained through stability verification of the constant speed condition. Real-time data such as current and voltage are collected to identify anomalies such as current jumps under the constant speed condition.

[0117] In some embodiments, the temperature rise condition is to raise the battery pack temperature from the current room temperature to 40°C to 50°C through charging / discharging or heating in an ambient chamber. This temperature range is obtained through NTC sensor response characteristic tests, and temperature data is collected during the temperature rise process to verify the NTC sensor's response capability to temperature changes.

[0118] In some embodiments, the room temperature resting condition involves placing the battery pack in a room temperature environment of 20 to 25°C for more than 2 hours. After the internal temperature stabilizes, data is collected to eliminate interference from other operating conditions and determine the temperature error caused by NTC aging.

[0119] S2. Preprocess the running data and extract key data features;

[0120] In some embodiments, preprocessing includes averaging five consecutive sampling points using a moving average method to remove significant noise.

[0121] In some embodiments, key data features include jump points and jump differences, fixed extreme values, unchanged values, null and invalid values, and deviations.

[0122] Among them, the jump point and the jump difference refer to the numerical change of adjacent sampling points in the data sequence. The jump difference is calculated by subtracting the sampling value of the previous time from the current sampling value. If the difference is greater than 1.5 to 2 times the normal fluctuation range, it can be determined as a data jump point.

[0123] Fixed extreme values ​​refer to extreme values ​​that remain unchanged over a long period of time in temperature sampling. The values ​​of 10 consecutive sampling points remain unchanged and are within the extreme value range of the sensor's measurement range. These 10 sampling points are the criteria for judging the fit based on data stability verification.

[0124] "Unchanging value" refers to the NTC sampling value that does not change with temperature under temperature rise conditions, or data that remains unchanged throughout the driving process.

[0125] Null and invalid values ​​refer to abnormal data where the data field is missing, has no value, or has three consecutive invalid sampling points. The three sampling points are obtained by fitting based on the accuracy verification of invalid value judgment.

[0126] Deviation refers to the degree to which each parameter deviates from its normal threshold, such as temperature error or voltage deviation. Data cleaning ensures the accuracy of features, providing a reliable basis for subsequent fault matching, weight adjustment, and three-level fault location.

[0127] This method preprocesses raw data, transforming it from "low-quality, highly redundant, and irregular" data into "high-quality, low-dimensional, and strongly correlated" usable data assets. This provides reliable input for subsequent data analysis, machine learning, intelligent operation and maintenance, and business optimization, avoiding the problem of "garbage input, garbage output." Key data features are extracted from the preprocessed raw data, providing core indicators that accurately reflect operational status and business patterns. This lowers the technical threshold for data application, allowing non-data professionals to understand operational status through key features and promoting the implementation of data-driven business practices.

[0128] S3. Construct a fault type identification model;

[0129] S31. Construct a fault type library based on the types of power battery faults.

[0130] The fault type library includes fault types and sub-fault characteristics.

[0131] The fault types include sampling chip fault, communication fault, differential voltage fault, overcurrent fault, and over-discharge fault. Among them, only sampling chip fault has a subtype, while the other faults do not have subtypes.

[0132] The fault characteristics of sampling chip failures include abnormal voltage sampling, abnormal temperature sampling, abnormal current sampling, and abnormal insulation resistance sampling.

[0133] Communication failures manifest as data transmission interruptions, loose connections, data null values ​​caused by oxidation and corrosion of connectors, transmission delays, and other unstable communication phenomena.

[0134] The differential pressure fault manifests as an imbalance in the voltage of individual cells, with the maximum difference exceeding the threshold and the duration reaching the standard, indicating an excessively large differential pressure.

[0135] An overcurrent fault manifests as a charging or discharging current exceeding the safe range, i.e., 1.2 times the rated current, for a duration exceeding 2 seconds.

[0136] Over-discharge faults manifest as a single-cell over-discharge phenomenon where the individual cell voltage is lower than the discharge cutoff voltage (2.5V) and the total current is equal to the discharge current. See Table 1 for a detailed fault type list.

[0137] Table 1 Fault Type Library

[0138]

[0139] S32. Determine the initial weight W0, frequency weight coefficient α, and deviation weight coefficient β of the fault through algorithm fitting.

[0140] S321. Obtain initial weights:

[0141] The initial weights W0 are determined through fitting. The target initial weights are set by combining industry fault maintenance experience and risk priority logic. The target initial weights are ideal weight values ​​set according to risk priority. The sum of the target initial weights for all faults is 1, which is the benchmark for subsequent fitting. The initial weights W0 are the actual weights obtained through fitting, and the sum of all initial weights W0 is 1.0, providing a target benchmark for fitting.

[0142] Use the formula: ,(1);

[0143] Among them, k1 and k2 are the optimal coefficients, F represents the historical frequency of fault occurrence, which is determined based on historical fault data, and H represents the quantitative value of the fault severity, which is divided into 5 levels according to the degree of impact of the fault on equipment safety and operation.

[0144] Two constraints are set, which are determined by statistical fitting of historical data. The constraints are: First, the sum of the initial weights W0 of all faults is 1.0 to ensure the normalization of weight allocation; Second, the value of the initial weight W0 of each fault is in the range of 0.1 to 0.3 to avoid misjudgment or omission of key faults due to excessively high initial weights of a single fault or excessively low initial weights. This range is obtained by fitting the weight allocation balance verification.

[0145] The method for verifying two constraints is as follows:

[0146] Let W0^init be the initial weight value calculated based on historical fault data; normalize the initial weight value to obtain the intermediate weight W0^temp, which is calculated using the following formula:

[0147] W0^temp=W0^init; (2);

[0148] W0^init represents the initial weights calculated based on historical fault data. W0^temp represents the intermediate weight variables before normalization, with the same value as W0^init, and is only used for subsequent calculations. When the sum of all intermediate weights is not 1, the normalized initial weights are calculated using the following normalization formula. :

[0149] ; (3);

[0150] This represents the initial weight values ​​after normalization.

[0151] Based on historical fault big data of similar power batteries, a sample set Y is constructed.

[0152] Y={( , , )} (4);

[0153] in, denoted as the historical frequency of failures for the i-th sample, expressed in times per 1000 hours, and obtained based on long-term historical failure statistics. This is the quantification value of the fault severity for the i-th sample.

[0154] The impact of fault types on equipment safety and operation is divided into 5 levels, represented by the letter H. For example, level 1 is represented by H=1. The higher the level, the greater the impact. The specific levels are shown below:

[0155] Overcurrent fault and over-discharge fault are H=5. Temperature sampling abnormality and current sampling abnormality in sampling chip faults are H=3. Communication fault and differential voltage fault are H=2. Voltage sampling abnormality is H=4. Insulation resistance sampling abnormality is H=3.

[0156] Let be the target initial weight for the i-th sample. Based on the historical fault handling results, the higher the fault handling accuracy, the closer the target initial weight is to the actual weight, and the sum of the target initial weights for all samples is 1.0.

[0157] The optimal coefficients k1 and k2 are solved using the gradient descent algorithm. The objective function is to minimize the sum of squared errors between the theoretically calculated weights and the target initial weights. The objective function is:

[0158] (5);

[0159] Where W0^calc represents the theoretically calculated initial weight of the i-th sample. Minloss is the gradient descent objective function, which is the final goal of the iteration, and Loss is the sum of squared errors actually calculated during word iteration.

[0160] Based on the above constraints, the final value range of k1 is [0.001, 0.005], and the value range of k2 is [0.02, 0.06].

[0161] The iterative steps are as follows: Based on the statistical mean of historical sample data, the initial iteration values ​​of k1 and k2 are determined to be 0.003 and 0.04, respectively. The theoretical initial weight of the i-th sample in each sample is calculated. Given the sum of squared errors (Loss) from a single iteration, calculate the gradient:

[0162] (6); (7);

[0163] It represents the partial derivative of the objective function Loss with respect to the optimal coefficient k1, reflecting the degree of influence of small changes in k1 on the sum of squared errors Loss, and is used to guide the direction and magnitude of k1 updates.

[0164] It represents the partial derivative of the objective function Loss with respect to the optimal coefficient k2, reflecting the degree of influence of small changes in k2 on the sum of squared errors Loss, and is used to guide the direction and magnitude of k2 updates.

[0165] The coefficients k1 and k2 are updated according to the adaptive step size, which is dynamically calculated based on the rate of change of the Loss. The update formula is as follows:

[0166] (8);

[0167] (9);

[0168] k1^new and k2^new are the next update values ​​of k during the iteration process. The above update steps of k1^new and k2^new are repeated. The adaptive step size is initially set to 0.05, and is maintained until Loss ≤ 0.001 or the number of iterations reaches 1000. 0.001 is the threshold obtained by verifying the model's convergence. The final output is the optimal value. , .

[0169] S322, Weighting coefficient for the number of acquisitions:

[0170] The frequency weighting coefficient α reflects the contribution of the frequency of occurrence of data features to the weight adjustment and is related to the feature sensitivity of fault types.

[0171] For each fault type, m valid samples of the fault occurrence from historical data are selected, where m ≥ 50. The value of m is determined based on sample representativeness verification. For each sample group,

[0172] Calculate the Pearson correlation coefficient between the number of occurrences of a feature N and the actual probability of failure P_actual. The specific steps are as follows:

[0173] 1. First, calculate the mean of the two variables.

[0174] The mean frequency of feature occurrences is represented as follows:

[0175] (10);

[0176] Where m represents the total number of samples involved in the statistical calculation, and j represents the sample number. This represents the number of times the corresponding fault feature appears in the j-th sample.

[0177] The mean of the actual probability of a fault occurring is expressed as follows:

[0178] (11);

[0179] in, This represents the actual probability of the corresponding fault occurring in the j-th sample. This represents the average probability of the actual occurrence of the fault across all samples.

[0180] 2. Calculate the covariance and the variance of the variables.

[0181] covariance = (12);

[0182] Variance N of the frequency of feature occurrence (13);

[0183] Variance of the actual probability of failure (14);

[0184] Substitute into the formula to calculate the Pearson coefficient:

[0185] (15);

[0186] ∈[-1,1], the closer to 1, the stronger the correlation. The frequency weight coefficient α of this fault type.

[0187] α = 0.05 + 0.05 × ( (mean / 1.0)(16);

[0188] The value of α ranges from [0.05, 0.1].

[0189] S323. Obtain the deviation weighting coefficient:

[0190] The deviation weighting coefficient β reflects the contribution of the deviation between the data characteristics and the normal threshold to the weight adjustment, and is positively correlated with the importance of the parameter and the degree of impact of the fault.

[0191] Define the parameter impact factor I as follows: I = (Frequency of safety accidents caused by parameter failure / Total frequency of safety accidents) × 0.5 + (Performance degradation caused by parameter failure / Maximum allowable performance degradation) × 0.5;

[0192] Calculate the I value of each parameter and classify them according to their I values.

[0193] Example 1: Voltage fault, known total safety incident frequency = 100 times, voltage fault-induced safety incident frequency = 80 times, maximum permissible performance degradation = 80%, voltage fault-induced performance degradation = 80%.

[0194] calculate: =0.9,

[0195] Result: I=0.9, which satisfies I≥0.7, and is a core parameter.

[0196] Example 2: Insulation resistance fault. Given: total safety accident frequency = 100 times, safety accident frequency caused by insulation resistance fault = 26 times, maximum permissible performance degradation = 80%, performance degradation caused by insulation resistance fault = 30%, calculate: =0.3175, that is, I=0.3175,

[0197] It satisfies 0.3≤I<0.7, and is considered an auxiliary parameter.

[0198] Calculate the I value of each parameter and classify them according to the I value: I ≥ 0.7 are core parameters for voltage faults and current faults; I ≤ 0.3 are auxiliary parameters for temperature faults, insulation resistance faults, and communication faults.

[0199] The core parameters are voltage and current, and I≥0.7.

[0200] The auxiliary parameters are temperature and insulation resistance, with a value of 0.3 ≤ I < 0.7. 0.7 and 0.3 are cutoff values ​​based on statistical fitting of the parameter influence.

[0201] β=0.1+0.1×(I / 1.0), (17);

[0202] Core parameters after calculation It should be located in the range [0.17, 0.2], auxiliary parameter. It should be located in the interval [0.13, 0.17]. If , If the parameter does not fall within the preset range, the parameter execution range is constrained. Specifically, if the calculated parameter value is less than the lower limit of the corresponding range, the parameter is taken as the lower limit value of the range. If the calculated parameter value is greater than the upper limit of the corresponding range, the parameter is taken as the upper limit value of the range, thus completing the parameter correction.

[0203] S33. Obtain the final weight W of the fault and dynamically adjust it according to the fault type:

[0204] S331. Based on data features extracted from the fault type library and combined with preset fault matching logic, the weight values ​​of corresponding faults are dynamically adjusted. The core logic is that the more prominent the feature, the higher the weight. The final weight W is calculated using a weighted summation formula, taking into account both the feature frequency and feature bias to avoid the limitations of single-dimensional judgment. The formula is:

[0205] (18);

[0206] in, The initial weights after normalization are N, which is the number of times the data features match the fault characterization, such as the number of times the jump point occurs or the number of times the invalid value occurs, and Δ is the deviation value between the data features and the normal threshold, such as the jump difference value or temperature error.

[0207] This application dynamically adjusts weights, enabling the weight coefficients of various factors / indicators / faults to be dynamically modified based on real-time status and environmental changes. This upgrades the decision-making system from "static and rigid" to "adaptive and highly flexible," always adhering to the core needs of real-world scenarios. The process and effect data of dynamic weight adjustment can be accumulated into scenario-based decision-making experience, such as "optimal weight configuration under high-temperature conditions" and "weight balancing strategy when multiple faults occur concurrently." This experience can be solidified into a standard strategy library to support subsequent systems in quickly adapting to similar scenarios. At the same time, it provides quantitative basis for model iteration, promoting the upgrade of the decision-making system from "passive response" to "proactive prediction."

[0208] S332. To ensure priority identification of emergency faults, a special rule is set. The special rule is that when the strong characteristics of sudden faults in the fault matching logic are met, W is directly set to 1.0. See the specific fault matching and weight adjustment logic for details.

[0209] The specific fault matching and weight adjustment logic is as follows:

[0210] Sampling chip failure: When the difference between the sampled values ​​of voltage, temperature, current, and insulation resistance is greater than the preset threshold, or the number of invalid sampled values ​​exceeds 5, the final weight W will increase by 0.1 to 0.2 for each occurrence of the above features, with a maximum adjustment to 0.8.

[0211] Temperature sampling anomaly sub-fault: Under conditions of NTC breakage, short circuit, or temperature rise, the sampling value of a certain NTC does not change (the values ​​of 20 consecutive sampling points remain unchanged, and the sampling values ​​of other NTCs change normally), the final weight W is directly adjusted to 1.0;

[0212] Current sampling abnormal sub-fault: Under continuous constant speed driving conditions, the total current shows a jump point corresponding to no change in operating conditions. Excluding the current change caused by normal operating conditions, the final weight W is directly adjusted to 1.0.

[0213] Communication failure: Null values ​​appeared in the collected operational data. After excluding invalid values ​​and 0 values ​​caused by sampling anomalies, the final weight W increased by 0.3 to 0.5.

[0214] Differential voltage fault: The maximum difference in individual unit voltage exceeds 0.2V and lasts for more than 5 seconds, while the total voltage data does not change. The final weight W increases by 0.2 to 0.3.

[0215] Overcurrent fault: The absolute value of the total current exceeds 1.2 times the rated current and the duration exceeds 2 seconds, and the final weight W is increased by 0.3 to 0.5;

[0216] Over-discharge fault, individual cell voltage below 2.5V and total current equal to discharge current, ultimately increases weight W by 0.3~0.5.

[0217] This invention sets priorities through special rules formed by fault matching and weight adjustment logic, achieving orderly handling of "prioritizing critical issues and addressing emergencies first." This transforms the entire method from "passive detection" to "proactive and orderly management," prioritizing the handling of high-risk faults and preventing major accidents. After prioritization, the system or personnel will focus on high-risk faults immediately, blocking the risk of safety accidents and core business interruptions from the source and minimizing economic losses. This method effectively optimizes resource allocation, improves the efficiency of the entire fault handling process, quickly stops losses, and is more intelligent.

[0218] S34. Associate the fault type, final weight W, and fault characteristics to construct the mapping logic of fault type, final weight, and fault characteristics.

[0219] S341. Match the fault type database collected by the host computer. If a certain feature is found, the corresponding fault type can be matched.

[0220] S342. Adjust the weight corresponding to this fault type. After all fault weights are calculated, filter by calculating the fault probability P-value. The fault probability is:

[0221] P = (Final weight of the fault W / Sum of final weights of all faults) × 100% Quantifies the probability of a fault occurring, determines the final weight W of the fault, calculates the proportion of W to the sum of final weights of all faults, and multiplies this proportion by 100% to obtain 0-100%.

[0222] Example 1: Overcurrent fault. Given that the final weight of overcurrent fault is W1=0.6, the final weight of communication fault is W2=0.3, and the final weight of temperature fault is W3=0.1, calculate the sum of the final weights of all faults: Calculate the overcurrent fault probability P1: The result is P1=60%>50%, so overcurrent faults are selected as high-priority candidates.

[0223] Example 2: Temperature fault. Given that the final weight of voltage fault W1 = 0.4, the final weight of current fault W2 = 0.35, and the final weight of temperature fault W3 = 0.25, calculate the sum of the final weights of all faults: Calculate the probability of temperature-related failure P3: The result is P3 = 25% ≤ 50%. At this point, the 1-2 types with the highest probability are selected as the medium-low priority candidate set.

[0224] When the failure probability P > 50%, the failure type is selected as a high-priority candidate set. If all failure probabilities P are ≤ 50%, the 1-2 types with the highest probabilities are selected as medium-low priority candidate sets, and the selection results are directly used as input to the secondary model.

[0225] Traditional fault diagnosis relies on the experience of technicians, which can easily lead to misjudgments and omissions when dealing with complex equipment and multi-system interconnected faults. Moreover, complex faults require a long time to investigate. In contrast, the fault type identification model constructed in this application learns from historical fault data, data features and fault matching logic, and through dynamic weight adjustment and probability calculation, it can quickly lock in the range of possible fault types. It can output fault types and locate the model in seconds, with an identification accuracy far higher than that of manual identification. It is especially suitable for multi-type, high-concurrency fault scenarios and provides a foundation for subsequent sub-fault type localization.

[0226] S4. Construct a sub-fault type identification model;

[0227] S41. Construct a fault feature library (as shown in Table 2), and sort out the sub-fault types of the sampled chip faults to facilitate accurate determination of the fault type;

[0228] Voltage sampling anomalies manifest as voltage sample value jumps exceeding the threshold, excessively high frequency of invalid / empty values, and excessive voltage deviation. Temperature sampling anomalies are caused by aging, breakage, or short circuit of the NTC sensor, specifically manifested as no temperature change, fixed extreme values, and errors exceeding the range. Current sampling anomalies are caused by current sensor sampling abnormalities, manifested as invalid current values, excessive deviation from the sampling range, and jumps without matching operating conditions. Insulation resistance sampling anomalies manifest as insulation resistance sample value jumps exceeding the threshold and excessively high frequency of invalid values.

[0229] Table 2 Fault Feature Database

[0230]

[0231] S42. Set the sub-fault type filtering rules and determine the sub-fault type.

[0232] Feature matching degree M is introduced as the core screening indicator. The formula for calculating feature matching degree M is as follows:

[0233] M = (Number of features matching the subdivision / Total number of fault features in this subdivision) × + (Cumulative deviation of subdivided features / Maximum permissible deviation of subdivided features) × Where ω1 is the feature number weight, and the total number of features = 1. =0.4, total characteristic number =2, =0.45, Total characteristic number =3 =0.55, The cumulative weight for the deviation is 1- .

[0234] When the feature matching degree M of a certain sub-fault is greater than or equal to 0.7, the sub-fault is regarded as a candidate sub-fault. If there are multiple sub-faults with a feature matching degree M greater than or equal to 0.7, the priority of the sub-faults is sorted from high to low according to the value of M.

[0235] The total number of subdivided fault features refers to the number of feature items of the fault type corresponding to the target sub-fault type in the fault feature library shown in Table 2. The number of features that meet the subdivided features refers to the number of features that meet the judgment conditions after the collected data is compared with the features of each fault type corresponding to the sub-fault type in Table 2. The cumulative deviation of subdivided features refers to the sum of the deviations of all fault features that meet the matching conditions relative to their judgment threshold.

[0236] Example 1: Single sub-fault M > 0.7, sub-fault: overcurrent fault, total number of features in Table 2 = 2, therefore α1 = 0.45, α2 = 1 - 0.45 = 0.55, number of features matching subdivision: 2, cumulative deviation of subdivision features / maximum allowable deviation: 1.4 / 2.5 = 0.56 (cumulative deviation 1.4, maximum allowable deviation 2.5), calculate feature matching degree M: The result is M=0.758>0.7, so this sub-fault is selected as a candidate sub-fault.

[0237] Example 2: At least two sub-faults M > 0.7, sub-fault 1: overcurrent fault, total number of features = 2, α1 = 0.45, α2 = 0.55, number of subdivision features: 2, cumulative deviation of subdivision features / maximum allowable deviation: 1.5 / 2.5 = 0.6, calculate M1: Sub-fault 2: Voltage anomaly fault. The total number of features in Table 2 is 3, therefore α1 = 0.55, α2 = 1 - 0.55 = 0.45. The number of features meeting the subdivision criteria is 3. The cumulative deviation of the subdivision features / the maximum permissible deviation is 2.4 / 3 = 0.8. Calculate M2: M2 = 0.91 > 0.7, both sub-faults meet the conditions and are therefore considered candidate sub-faults.

[0238] This application establishes a progressive relationship between the sub-fault type identification model and the fault type identification model. After receiving the results of the fault type identification model, if the identified fault model contains sub-fault types, this model can further identify the types of sub-faults. The main fault only reflects the "phenomenon," while the sub-faults point to the "essence," accurately tracing the root cause of sub-faults, reducing fault recurrence from the source, and completely eliminating the ambiguity in fault handling. Since the maintenance processes and technical requirements of different sub-fault types are completely different, the application can achieve refined scheduling of operation and maintenance resources based on the difficulty, urgency, and technical requirements of handling different sub-faults, while improving work efficiency.

[0239] S5. Verify the fault type and sub-fault type to ensure the accuracy of fault type identification;

[0240] S51. To avoid misjudgment based on a single parameter, design correlation verification logic to analyze the characteristics of fault types and their linkage with other faults, as detailed in Table 3.

[0241] If the voltage sampling is abnormal, it is necessary to verify whether there is no jump in the total voltage data. If there is a jump, it is a sensor voltage jump caused by a voltage jump, which is not a fault characteristic. Clear all fault weights.

[0242] If no abnormality is found, the fault can be identified and the voltage sampling misjudgment caused by the total voltage fault can be eliminated.

[0243] If the current sampling is abnormal, it is necessary to verify whether the driving conditions are stable. If the driving conditions are stable, the fault is caused by the current sampling abnormality and there are no other faults.

[0244] If the driving conditions are unstable, there may be abnormal current jumps, making the fault diagnosis unreliable. Clear all fault diagnosis weights and eliminate misjudgments caused by current fluctuations due to changes in driving conditions.

[0245] An abnormal temperature sampling requires verification of whether other NTC sensor data are changing normally. If other NTC temperature sensors are changing normally, misjudgment caused by sudden changes in ambient temperature can be ruled out, confirming a temperature sampling fault, and other faults are not related. If other NTC sensors show the same change, it indicates that a drastic change in external ambient temperature is causing temperature sensor data to jump, which is not a fault characteristic; therefore, all fault weights should be cleared.

[0246] Table 3 Fault Types and Their Interactions

[0247]

[0248] S52. Establish physical location mapping rules.

[0249] The mapping rule is to associate data from different sampling channels with sampling chip numbers.

[0250] In one embodiment, the voltage sensor numbers start from A1 and end at A20. The voltage sampling numbers in the software are B0-B30, corresponding to software numbers starting from C0 and ending at C19. The remaining numbers B20-B30 are unused. These numbers can be mapped to the sensor numbers based on the software numbering. This is to locate the faulty sensor based on the identified faults, facilitating targeted repairs by after-sales personnel and saving repair time. Because the sensor layout varies across different battery packs, only the mapping rules are listed. Based on the actual physical layout of temperature, voltage, current, and insulation sensors, different channels of software data acquisition are used to achieve the mapping between physical location and data.

[0251] For example, if the third temperature data point becomes constant during the heating process, and the sampling chips are arranged from 0, then the NTC sampling chip for temperature 2 is abnormal. Based on the correspondence between the power battery channel and the cell, the correspondence between each sub-fault and the acquisition channel is sorted out. For example, different cell voltage sampling abnormalities correspond to specific voltage sampling channels, different NTC temperature sampling abnormalities correspond to specific sensor installation locations, and current sampling abnormalities correspond to current sensor installation locations. Since the mapping relationships are different for different items, only the mapping rules are listed here.

[0252] S53. Verify the candidate faults by reproducing the operating conditions to ensure the accuracy of the fault judgment and output relevant conclusions and maintenance suggestions;

[0253] The fault and its reproducibility are as follows:

[0254] Communication failure: The system needs to be restarted to reproduce the power-on and driving conditions. Observe whether the null values ​​in the data are reproduced. If they are reproduced, calculate the final weight after reproduction. If they are not reproduced, output the weight before reproduction and output a reproduction failure message.

[0255] Temperature sampling anomaly: The temperature rise condition or room temperature static condition needs to be repeated to verify whether the characteristic of no temperature change or excessive error continues. If it is reproduced, the final weight after reproduction is calculated. If it is not reproduced, the weight before reproduction is output and a reproduction failure prompt is output.

[0256] Current sampling anomaly: The continuous driving condition needs to be repeated to observe whether the current jump characteristics are reproduced. If they are reproduced, the final weight after reproduction is calculated. If they are not reproduced, the weight before reproduction is output and a reproduction failure prompt is output.

[0257] Differential pressure fault: After standing for 2 hours, remeasure the individual unit voltage to verify whether the differential pressure is still excessive. If it is reproduced, calculate the final weight after reproduction. If it is not reproduced, output according to the weight before reproduction and output a reproduction failure prompt.

[0258] For overcurrent faults, over-discharge faults, and sampling chip faults, the corresponding triggering conditions need to be reproduced to verify whether the fault characteristics persist. If they are reproduced, the final weight after reproduction is calculated. If they are not reproduced, the weight before reproduction is output, and a reproduction failure message is output.

[0259] After reproduction, the final weight of the fault is recalculated. The formula for calculating the recurrence probability P_rep and the recurrence weight is as follows:

[0260] ,(19)

[0261] in, The initial weights are calculated and normalized based on historical fault data. The frequency weighting coefficient corresponds to the number of times the fault feature occurs. The characteristic deviation weighting coefficient, This represents the number of times the fault characteristics detected during the reproduction phase occurred. This represents the cumulative deviation of features detected during the reproduction phase. The calculation process during the reproduction phase involves re-collecting operating data under the reproduced conditions, extracting fault features, and counting the number of times the features occur. Calculate the cumulative deviation of the features Substitute into the above formula to calculate After obtaining the corresponding faults for all candidate faults Then, it is normalized, and the recurrence probability P_rep is calculated using the following formula:

[0262] (20);

[0263] If P_rep is still greater than 50%, it is considered a high-priority fault and must be processed within 24 hours. For candidate faults that still meet the output conditions after reproduction, calculate the comprehensive priority score S, and output the unique core fault. The priority score S is:

[0264] S = P_rep × H × γ, (21);

[0265] H represents the quantification value of the severity of the fault. It is based on the fault risk assessment standard for the power battery industry, and is divided into 1-5 levels according to the severity of the fault. For example, for an overcurrent fault, H is 5, because this fault can directly cause thermal runaway and equipment fire, belonging to the highest severity level. Given: P_rep (probability of fault reproduction) = 80%, γ (reproducibility reliability coefficient) = 0.95, priority score calculation: S = 80% × 5 × 0.95 = 3.8. For another example, for a communication fault, H is 2, because this fault only causes data acquisition delay and does not directly affect equipment safety, belonging to the low-to-medium severity level. Given: P_rep = 60%, γ = 0.8, priority score calculation: S = 60% × 2 × 0.8 = 0.96.

[0266] Where P_rep is the probability of the fault after reproduction, H is the quantification value of the fault severity, and γ is the reliability coefficient of reproduction (γ=1.0 for 1 reproduction, γ=1.2 for 2 or more reproductions).

[0267] Judgment rules: Select the fault with the highest S as the unique output fault. If S are equal, select the fault with more recurrences. The final output must include the fault type or fault subtype, fault probability, judgment criteria, priority, unique fault judgment score, fault physical location, and maintenance recommendations.

[0268] Example 1: Overcurrent fault, P_rep>50%, known initial weight of overcurrent fault W0=0.25, Pearson correlation coefficient =0.9, Deviation weighting coefficient =0.185, other candidate faults communication fault Wrep=0.20. Calculate the final weight of the overcurrent fault: Wrep=0.25+0.095×1+0.185×0.30=0.4005, calculate the sum of the final weights of all candidate faults: ∑Wrep=0.4005+0.20=0.6005, calculate the probability of the overcurrent fault P_rep=0.4005 / 0.6005×100=66.7. Since P_rep>50%, the overcurrent fault meets the criteria and needs to be processed within 24 hours.

[0269] Example 2: Temperature fault, P_rep≤50%, known initial weight of temperature fault W0=0.15, mean Pearson correlation coefficient =0.6, the frequency weight is calculated. Deviation weighting coefficient =0.14, other candidate faults: voltage fault Wrep=0.30, current fault Wrep=0.20, calculate the final weight of temperature fault: Wrep=0.15+0.08×0+0.14×0.05=0.157, calculate the sum of the final weights of all candidate faults: ∑Wrep=0.157+0.30+0.20=0.657, calculate the probability of temperature fault: P_rep=0.157 / 0.657×100=23.9. Since P_rep≤50%, temperature fault does not meet the high priority condition and does not require emergency handling within 24 hours.

[0270] In some embodiments, the maintenance recommendations include:

[0271] Voltage sampling abnormality: The physical location is the corresponding voltage sampling channel and chip pin. It is recommended to replace the sampling chip or related sampling module.

[0272] Temperature sampling abnormality: If the NTC sensor is abnormal, the physical location is the corresponding NTC installation location. It is recommended to replace the faulty NTC sensor.

[0273] Current sampling abnormality: The physical location is the current sensor or its wiring; the current sensor needs to be replaced.

[0274] Insulation resistance sampling abnormality: The physical location is the insulation resistance sampling sensor, which needs to be calibrated or replaced.

[0275] Communication failure: The physical location is the connection harness or interface between the battery pack and the BMS. It is recommended to check whether the harness is loose or the interface is oxidized or corroded, clean the interface or tighten the harness again.

[0276] Differential voltage fault: The physical location is a single cell or module with abnormal voltage. It is recommended to perform cell balancing or replace the cells.

[0277] Overcurrent fault: The physical location is the charging / discharging circuit or current sensor. It is necessary to check whether the circuit is short-circuited, whether the sensor is abnormal, and to eliminate potential safety hazards. Over-discharge fault: The physical location is a single cell with too low voltage. It is necessary to check the health status of the cell and replace it if necessary.

[0278] Over-discharge fault: The physical location is the BMS and voltage sensor. It is necessary to check whether the BMS over-discharge protection parameter settings are reasonable or whether the voltage sensor error is too large. Recalibrate the voltage sensor or modify the BMS over-discharge protection parameters.

[0279] This method verifies the model's output by reproducing the fault types or sub-fault types input into the two models mentioned above. This not only verifies the model's usefulness and accuracy but also achieves precise location by combining the correlation with operating data. The correlation verification requires analyzing the relationship between detailed fault characteristics and other operating parameters to avoid interference from external factors. Simultaneously, it outputs corresponding maintenance suggestions based on the fault type, allowing staff to directly refer to these suggestions, significantly improving work efficiency and maintenance accuracy.

[0280] Example: Fault diagnosis of sampling chip in an energy storage system

[0281] A1. Multi-condition data acquisition: Connect to the energy storage system BMS via host computer software to collect operating data under three conditions: initial power-on condition, room temperature resting condition for 2.5 hours condition, and charging / discharging temperature rise condition. The collected data includes individual cell voltage, total voltage, total current, insulation resistance, and temperature. At the same time, the acquisition channel identifier of each data is recorded.

[0282] A2. Data Preprocessing and Feature Extraction: The raw data was preprocessed using the moving average method. After noise removal, the following features were extracted: Three voltage jumps occurred in the individual cell voltage, with jump differences of 0.15V, 0.18V, and 0.20V respectively. The voltage jump threshold was calculated as 0.1V, based on 1.5 times the maximum fluctuation value of 0.067V under normal operating conditions. Four invalid voltage values ​​occurred, with a cumulative voltage deviation of 0.5V. The maximum allowable deviation was obtained based on battery voltage stability testing and was 0.8V. The total voltage data was normal, with no jumps or anomalies; other channel voltage, temperature, current, and insulation resistance data showed no obvious abnormalities.

[0283] A3. Initial screening of fault types in the fault type library using a fault type identification model:

[0284] A31. Based on the historical fault big data of the energy storage system, a sample set S (m=100) is constructed. After 800 iterations of the gradient descent algorithm, Loss=0.0008≤0.001, and the convergence is obtained to k1=0.004 and k2=0.05.

[0285] The characteristic sensitivity correlation coefficient r of the sampling chip fault i r i The mean is 0.9, and α = 0.095 is calculated using the formula α = 0.05 + 0.05 × 0.9. The voltage parameter influence factor I = 0.8, and β = 0.18 is calculated using the formula β = 0.1 + 0.1 × 0.8. The core parameters are within the calculated range. In the historical fault statistics of this energy storage system, the sampling chip faults have an F = 1.8 times / 1000 hours and H = 3, and the subdivided fault voltage sampling anomalies have an F = 1.2 times / 1000 hours and H = 3. The initial weights W0 = 0.004 × 1.2 + 0.05 × 3 = 0.0048 + 0.15 = 0.1548 are fitted. After normalization, W0 = 0.15, falling within the range of 0.1 to 0.3, which meets the requirements.

[0286] Based on the initial weight W0=0.15 for the sampling chip fault, the number of occurrences matching the fault characteristic is calculated as N=3 (number of jumps) + 4 (number of invalid values) = 7. The frequency weight coefficient α=0.095, the deviation weight coefficient β=0.18, and the deviation value Δ=(0.15+0.18+0.20) / 3+0.5=0.177+0.5=0.677. According to the formula W=W0+α×N+β×Δ=0.15+0.095×7+0.18×0.677≈0.937. After adjustment, the weights of other faults are as follows: communication instability fault 0.20, temperature sampling fault 0.25, current sampling fault 0.20, excessive voltage difference fault 0.10, overcurrent fault 0.30, and over-discharge fault 0.30.

[0287] A32. The sum of the final weights of all faults = 0.937 + 0.20 + 0.25 + 0.20 + 0.10 + 0.30 + 0.30 = 2.287. The probability of sampling chip failure P = (0.937 / 2.287) × 100% ≈ 40.97%, and the probabilities of other faults are all < 30%. Since the probabilities of all faults are ≤ 50%, the sampling chip failure is selected as a low-to-medium priority candidate fault category, thus completing the category range locking.

[0288] A4. Filter the sub-fault types of the sampled chip faults using the sub-fault type identification model;

[0289] For sampling chip faults, the existing fault feature library is activated. The total number of features for voltage sampling anomalies is 3, including jump difference exceeding the threshold, invalid value frequency, and deviation. Among the features extracted, the number of features matching this sub-fault is 3, the cumulative deviation of the sub-feature is 0.677V, and the maximum allowable deviation is 0.8V. With a total number of features of 3, ω1=0.55, ω2=1-0.55=0.45, the feature matching degree M is calculated as (3 / 3)×0.55+(0.677 / 0.8)×0.45≈0.55+0.381≈0.931≥0.7, which is selected as a candidate sub-fault type. There are no other sub-fault types that meet the conditions.

[0290] A5. First, perform correlation verification. Check the total voltage data according to the preset logic to confirm that there is no jump or abnormality in the total voltage. Eliminate the voltage sampling misjudgment caused by the total voltage fault. The verification is successful. Then, perform physical location mapping. According to the hardware layout mapping rules of the energy storage system, the single cell voltage sampling channel corresponding to channel 3 is cell number 3. Therefore, the physical location of the fault is the voltage sampling channel of cell number 3.

[0291] A6. Operating Condition Reproduction Verification: Reproducing the charging / discharging condition and the initial power-on condition, the voltage of channel 3 cell showed two more jumps in the collected data, with differences of 0.16V and 0.19V, both exceeding the 0.1V threshold. Two invalid values ​​appeared, accounting for 33% of the total. After reproduction, the weight Wrep = 0.15 + 0.095 × (7 + 4) + 0.18 × (0.677 + 0.35) ≈ 0.15 + 1.045 + 0.185 ≈ 1.38, which was adjusted to 1.0 due to the strong triggering characteristic. The sum of other fault weights = 0.20 + 0.25 + 0.20 + 0.10 + 0.30 + 0.30 = 1.35, and the total weight sum = 1.0 + 1.35 = 2.35. The probability of recurrence is P_rep = (1.0 / 2.35) × 100% ≈ 42.55%, the number of recurrences is 2, and γ = 1.2.

[0292] A7. Determine the unique fault and output the result: Calculate the overall priority score S = 42.55% × 3 × 1.2 ≈ 1.532. Other candidate faults, such as overcurrent fault P_rep = 12.77%, S = 12.77% × 5 × 1.0 ≈ 0.638. Select the sampling chip fault as the unique output fault. The output result is as follows: The major fault category is sampling chip fault, the precise sub-fault type is voltage sampling abnormality, the fault probability is 42.55%, the judgment basis is that under power-on and charging / discharging conditions, the voltage jump of channel 3 individual cell 5 times, all exceeding the 0.1V threshold, 6 invalid values, the total voltage data is normal, and the correlation verification is passed. The maintenance priority is medium-low priority, it is recommended to handle it within 72 hours, the unique fault determination score is 1.532, the physical location of the fault is the voltage sampling channel of individual cell 3, and the maintenance suggestion is to check the voltage sampling line connection of individual cell 3 and replace the sampling chip or related sampling module;

[0293] Upon inspection of the individual voltage sampling channel circuit, it was found that a loose connection was causing the sampling abnormality. After tightening the circuit and replacing the sampling chip, the voltage jump and invalid value characteristics in the acquired data disappeared, and the fault was eliminated.

[0294] This method constructs a fault type identification model to obtain the weights of each fault type under a dynamic algorithm, avoiding the subjectivity of manual settings and ensuring the consistency and accuracy of diagnostic results when used by different scenarios and personnel. The method's versatility is significantly improved. It identifies fault types through the fault type identification model and accurately identifies the types of sub-faults by constructing a sub-fault type identification model. Finally, it forms a dual guarantee through fault probability screening and working condition reproduction verification. Combined with the comprehensive priority score, it determines the unique fault, solving the problems of redundant output and ambiguous decision-making in traditional methods. It integrates the dynamic linkage of working conditions, features, and weights, and adjusts the weights based on dynamic features under multiple working conditions. The sensitivity of identifying gradual faults such as NTC aging is significantly improved, and the false judgment rate is reduced to below 5%. It has strong cross-scenario adaptability. The weight coefficients can be recalculated through the algorithm to adapt to different models of new energy vehicles, energy storage systems, and other scenarios, solving the problem that traditional methods can only identify broad categories of faults and blind maintenance and troubleshooting.

[0295] This embodiment also provides a computer device applicable to a power battery fault analysis method based on weighted fusion, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the power battery fault analysis method based on weighted fusion proposed in the above embodiment.

[0296] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a power battery fault analysis method based on weighted fusion as proposed in the above embodiments.

[0297] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0298] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0299] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0300] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0301] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0302] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A power battery fault analysis method based on weighted fusion, characterized in that: Includes the following steps: S1. Collect operating data of the power battery under various preset operating conditions; S2. Preprocess the running data and extract key data features; S3. Construct a fault type identification model and output the fault type. If the fault type includes sub-fault types, proceed to step S4; otherwise, proceed to step S5. S4. Construct a sub-fault type identification model and output the sub-fault types; S5. Verify the fault type or fault subtype, and output the verification conclusion and maintenance recommendations.

2. The power battery fault analysis method based on weighted fusion according to claim 1, characterized in that: The operational data includes data collected by the BMS on individual cell voltage, individual cell temperature, total voltage, total current, insulation resistance, and differential pressure, as well as data acquisition channel identifiers.

3. The power battery fault analysis method based on weighted fusion according to claim 1, characterized in that: The preset operating conditions include initial power-on operating condition, continuous driving operating condition, temperature rise operating condition, and room temperature static operating condition; Preferably, the initial power-on condition is the start-up phase within 0 to 30 seconds after the vehicle is powered on, during which initial data is collected and abnormal sensor initial response conditions are captured. Preferred, the uninterrupted driving condition is that the vehicle travels at a constant speed of 30-60 km / h for 30-60 minutes. The speed and time parameters are obtained through stability verification of the constant speed condition. Real-time current and voltage data are collected to identify current jumps under the constant speed condition. Preferably, the temperature rise condition is to raise the battery pack temperature from the current room temperature to 40°C to 50°C through charging / discharging or heating in an ambient chamber. This temperature range is obtained through NTC sensor response characteristic test. Temperature data is collected during the temperature rise process to verify the NTC sensor's response capability to temperature changes. Preferably, the room temperature static condition involves placing the battery pack in a room temperature environment of 20 to 25°C for more than 2 hours. After the internal temperature stabilizes, data is collected to eliminate interference from other operating conditions and determine the temperature error caused by NTC aging.

4. The power battery fault analysis method based on weighted fusion according to claim 1, characterized in that: Key data characteristics include jump points and jump differences, fixed extreme values, unchanged values, null and invalid values, and deviations; Preferably, the jump point and jump difference refer to the numerical change of adjacent sampling points in the data sequence. The change difference is calculated by subtracting the sampling value of the previous time from the current sampling value. If the difference is greater than 1.5 to 2 times the normal fluctuation range, it can be determined as a data jump point. Preferably, the fixed extreme value refers to the extreme value that remains unchanged over a long period of time in temperature sampling, where the value remains unchanged for 10 consecutive sampling points and is within the extreme value range of the sensor's measurement range. The 10 sampling points are the judgment criteria for verifying the fitting based on data stability. Preferably, the "unchanging value" refers to the NTC sampling value that does not change with temperature under temperature rise conditions, or the data that remains unchanged throughout the driving process; Preferably, null and invalid values ​​refer to abnormal data where the data field is missing, has no value, or has three consecutive invalid sampling points. The three sampling points are obtained by fitting based on the accuracy verification of invalid value judgment. Preferably, the deviation refers to the degree of deviation of each parameter from the normal threshold.

5. The power battery fault analysis method based on weighted fusion according to claim 1, characterized in that: S3 includes: S31. Construct a fault type library based on the types of power battery faults. The fault type library includes fault types and sub-fault characteristics. S32. Determine the initial weight W0, frequency weight coefficient α, and deviation weight coefficient β of the fault through a dynamic algorithm; S33. Obtain the final weight W of the fault and dynamically adjust the final weight according to the fault type; S34. Construct the mapping logic of fault type, final weight, and fault characteristics, and output the fault type.

6. The power battery fault analysis method based on weighted fusion according to claim 5, characterized in that: S32 includes: S321. Obtain the initial weight W0: ，（1）； Where k1 and k2 are the optimal coefficients, F represents the historical frequency of the fault, determined based on historical fault data, and H represents the quantitative value of the fault severity, which is divided into 5 levels according to the degree of impact of the fault on equipment safety and operation; Based on historical fault big data of similar power batteries, a sample set Y is constructed. Y={（ , , ）}（4）； in, Let be the frequency of the fault occurrence history of the i-th sample. This is the quantification value of the fault severity for the i-th sample. Let the target initial weights be those for the i-th sample. The optimal coefficients k1 and k2 are solved using the gradient descent algorithm. The objective function is to minimize the sum of squared errors between the theoretically calculated weights and the target initial weights. The objective function is: （5）； in, minloss is the objective function of gradient descent, which is the final goal of the iteration. The initial iteration values ​​of k1 and k2 are 0.003 and 0.04, respectively. Calculate the theoretical initial weight of the i-th sample for each sample. Given the sum of squared errors (Loss) from a single iteration, calculate the gradient: （6）； （7）； The coefficients k1 and k2 are updated using an adaptive step size. The update formula is as follows: （8）； （9）； k1^new and k2^new are the next update values ​​of k during the iteration process. The above update steps of k1^new and k2^new are repeated until Loss ≤ 0.001 or the number of iterations reaches 1000, and finally the optimal value is output. , Substituting into formula (1), we obtain the initial weight W0; S322, Weighting coefficient for the number of acquisitions: The frequency weighting coefficient for each fault type is denoted as α, where α ∈ [0.05, 0.1]. α=0.05+0.05× ；（16）； in, The Pearson correlation coefficient represents the number of times a feature occurs (N) and the actual probability of a fault occurring (P_actual). ∈[-1,1]; S323. Obtain the deviation weighting coefficient: The bias weighting coefficient is denoted as β: β=0.1+0.1×（I / 1.0），（17）； Where I is the parameter impact factor, I = (frequency of safety accidents caused by parameter failure / total frequency of safety accidents) × 0.5 + (performance degradation caused by parameter failure / maximum allowable performance degradation) × 0.5, and the value range of I is [0.3, 0.7].

7. The power battery fault analysis method based on weighted fusion according to claim 5, characterized in that: S33 includes: S331. Obtain the final weight W: （18）； in, Δ represents the initial weights after normalization, N is the number of times the data features match the fault characteristics, and Δ is the deviation between the data features and the normal threshold. S332. Dynamically adjust the final weight according to the fault type. The methods for dynamically adjusting the final weight include: Sampling chip failure: When the difference between the voltage, temperature, current, and insulation resistance sampling values ​​is greater than the preset threshold, or the number of invalid sampling values ​​exceeds 5, the final weight W will increase by 0.1-0.2 for each occurrence of the above features, with a maximum adjustment to 0.

8. Temperature sampling abnormality sub-fault: If the NTC is broken, short-circuited or under temperature rise conditions, the values ​​of 20 consecutive sampling points of the NTC to be tested remain unchanged, and the sampling values ​​of other NTCs change normally, the final weight W will be directly adjusted to 1.

0. Current sampling abnormal sub-fault: Under continuous constant speed driving conditions, the total current shows a jump point corresponding to no change in operating conditions. Excluding the current change caused by normal operating conditions, the final weight W is directly adjusted to 1.

0. Communication failure: Null values ​​appeared in the collected operational data. After excluding invalid values ​​and 0 values ​​caused by sampling anomalies, the final weight W was increased by 0.3-0.

5. Differential voltage fault: The maximum difference in individual unit voltage exceeds 0.2V and lasts for more than 5 seconds, while the total voltage data does not change. The final weight W increases by 0.2-0.

3. Overcurrent fault: The absolute value of the total current exceeds 1.2 times the rated current and the duration exceeds 2 seconds, and the final weight W is increased by 0.3-0.5; Over-discharge fault, individual cell voltage below 2.5V and total current equal to discharge current, ultimately increase weight W by 0.3-0.

5.

8. The power battery fault analysis method based on weighted fusion according to claim 6, characterized in that: S34 includes: S341. Match fault types using data collected from the host computer; S342. Adjust the weights corresponding to the fault types. After all fault weights are calculated, filter the faults by calculating the fault probability P-value. The fault probability is: P = (Final weight of the fault W / Sum of final weights of all faults) × 100% Quantifies the probability of a fault occurring. When the failure probability P > 50%, the failure type is output as a candidate failure type. If all failure probabilities P ≤ 50%, the 1-2 failure types with the highest probabilities are selected as candidate failure types for output.

9. The power battery fault analysis method based on weighted fusion according to claim 1, characterized in that: S4 includes: S41. Construct a fault feature library, which includes fault types, sub-fault types, and features of fault types. S42. Determine the sub-fault type according to the sub-fault type filtering rules; Sub-fault type filtering rules include: Set the feature matching degree M: M = (Number of features matching the subdivision / Total number of fault features in this subdivision) × + (Cumulative deviation of subdivided features / Maximum permissible deviation of subdivided features) × ,in, The weights are based on the number of features, with a total number of features = 1. =0.4, total characteristic number =2, =0.45, Total characteristic number =3 =0.55, The cumulative weight for the deviation is 1- ; When the feature matching degree M of a certain sub-fault is greater than or equal to 0.7, the sub-fault is output as a candidate sub-fault. If there are multiple sub-faults with a feature matching degree M greater than or equal to 0.7, the sub-faults are output in order of priority according to the value of M.

10. The power battery fault analysis method based on weighted fusion according to claim 1, characterized in that: S5 includes: S51. Verify the correlation between fault types and fault characteristics. S52. Establish physical location mapping rules; S53. Verify the candidate fault types by reproducing the operating conditions, and output the verification conclusions and maintenance recommendations; Preferably, the method for verifying the operating condition reproduction of candidate fault types includes: Communication failure: The system needs to be restarted to reproduce the power-on and driving conditions. Observe whether the null values ​​in the data are reproduced. If they are reproduced, calculate the final weight after reproduction. If they are not reproduced, output the weight before reproduction and output a reproduction failure message. Temperature sampling anomaly: The temperature rise condition or room temperature static condition needs to be repeated to verify whether the characteristic of no temperature change or excessive error continues. If it is reproduced, the final weight after reproduction is calculated. If it is not reproduced, the weight before reproduction is output and a reproduction failure prompt is output. Current sampling anomaly: The continuous driving condition needs to be repeated to observe whether the current jump characteristics are reproduced. If they are reproduced, the final weight after reproduction is calculated. If they are not reproduced, the weight before reproduction is output and a reproduction failure prompt is output. Differential pressure fault: After standing for 2 hours, remeasure the individual unit voltage to verify whether the differential pressure is still excessive. If it is reproduced, calculate the final weight after reproduction. If it is not reproduced, output according to the weight before reproduction and output a reproduction failure prompt. For overcurrent faults, over-discharge faults, and sampling chip faults, the corresponding triggering conditions need to be reproduced to verify whether the fault characteristics persist. If they are reproduced, the final weight after reproduction is calculated. If they are not reproduced, the weight before reproduction is output, and a reproduction failure message is output. After reproduction, the final weight and probability of the fault are recalculated. For fault types that still meet the output conditions after reproduction, the comprehensive priority score S is calculated, and the fault type with the highest S is taken as the unique output fault type. If S is equal, the fault with more reproduction times is selected as the output fault type. The output results include fault type or fault subtype, fault probability, judgment basis, priority, unique fault judgment score, fault physical location and maintenance suggestions. Preferably, the maintenance recommendations include: Voltage sampling abnormality: The physical location is the corresponding voltage sampling channel and chip pin. It is recommended to replace the sampling chip or related sampling module. Temperature sampling abnormality: If the NTC sensor is abnormal, the physical location is the corresponding NTC installation location. It is recommended to replace the faulty NTC sensor. Current sampling abnormality: The physical location is the current sensor or its wiring; the current sensor needs to be replaced. Insulation resistance sampling abnormality: The physical location is the insulation resistance sampling sensor, which needs to be calibrated or replaced. Communication failure: The physical location is the connection harness or interface between the battery pack and the BMS. It is recommended to check whether the harness is loose or the interface is oxidized or corroded, clean the interface or tighten the harness again. Differential voltage fault: The physical location is a single cell or module with abnormal voltage. It is recommended to perform cell balancing or replace the cells. Overcurrent fault: The physical location is the charging / discharging circuit or current sensor. It is necessary to check whether the circuit is short-circuited, whether the sensor is abnormal, and to eliminate potential safety hazards. Over-discharge fault: The physical location is a single cell with too low voltage. It is necessary to check the health status of the cell and replace it if necessary. Over-discharge fault: The physical location is the BMS and voltage sensor. It is necessary to check whether the BMS over-discharge protection parameter settings are reasonable or whether the voltage sensor error is too large. Recalibrate the voltage sensor or modify the BMS over-discharge protection parameters.

Images