Vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening

Abstract

The invention discloses a vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening, and relates to the technical field of vehicle safety. The method comprises the following steps: analyzing the generality of historical spontaneous combustion vehicles, carrying out coarse screening on full-dose vehicles, and narrowing a sample range; a plurality of risk indexes are derived through historical and current national standard data, and dimensions are split based on the distinction degree of the indexes; risk factors are constructed according to dimensions, and screening conditions and sorting schemes are set; risk vehicles are screened in the dimension, and the sorting score is converted into a centesimal risk score; and carrying out weighted average on the risk scores of the multiple dimensions to obtain a total risk score, and carrying out secondary screening on the risk vehicles based on the total risk score. According to the method, national standard data are fully utilized, multiple rounds of screening and sorting are carried out from score to total by splitting dimensions, the spontaneous combustion risk of the vehicle is comprehensively evaluated, attention and intervention of a corresponding degree are given according to the risk level of the vehicle, a vehicle factory has an opportunity to dispose in advance, actual vehicle burning is avoided as much as possible, and the safety of the electric vehicle is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle safety, and in particular relates to a vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening. Background Art

[0002] With the continuous development of new energy vehicle technology, improving safety is an important topic in new energy vehicle research. Compared with fuel-powered vehicles, new energy vehicles have more battery safety hazards, among which vehicle spontaneous combustion is the most serious safety issue. Once it occurs, it will cause great losses to all parties and is one of the main bottlenecks restricting the development of new energy vehicles.

[0003] For example, Chinese patent CN114643929A discloses a vehicle battery pack spontaneous combustion automatic alarm system and a vehicle, wherein the vehicle battery pack spontaneous combustion automatic alarm system includes: a monitoring system, an automatic alarm system, an early warning system and a positioning system. The monitoring system is arranged on the battery pack to monitor the status of the battery pack; the monitoring system communicates with the automatic alarm system, and the automatic alarm system communicates with the network fire alarm system; the monitoring system communicates with the early warning system, and the early warning system communicates with the electronic equipment of the vehicle owner; the positioning system is suitable for locating the vehicle, and the positioning system communicates with the automatic alarm system. The vehicle battery pack spontaneous combustion automatic alarm system can monitor the risk of spontaneous combustion of the battery pack when the vehicle is not in use, and promptly notify the owner when there is a risk of spontaneous combustion. When a spontaneous combustion risk occurs, it can automatically alarm and provide the vehicle location information to the fire alarm, so as to buy a lot of time for the rescue of the vehicle and prevent the fire from spreading further. For example, Chinese patent CN109649301A also involves a spontaneous combustion early warning system and method for vehicles. A spontaneous combustion warning system for a vehicle comprises: a spontaneous combustion prediction device for obtaining battery-related parameters of the vehicle and judging whether the vehicle is about to spontaneously combust based on the battery-related parameters; and a spontaneous combustion warning device for sending an alarm to objects around the vehicle when the spontaneous combustion prediction device judges that the vehicle is about to spontaneously combust.

[0004] Based on the above-mentioned existing technologies, it is obvious that if the risk of spontaneous combustion of vehicles can be assessed in advance, high-risk vehicles can be dealt with in a timely manner, and the probability of vehicle burning incidents can be reduced as much as possible, the market competitiveness of new energy vehicles can be greatly improved and user acceptance of electric vehicles can be enhanced. Summary of the invention

[0005] The purpose of the present invention is to provide a vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening. By combining the current performance and historical cumulative performance of battery signals under different working conditions, different weights are assigned to indicators of different importance in different time periods, and a more comprehensive and accurate assessment of the spontaneous combustion risk is performed. The existing problems are solved, and vehicles with medium and high spontaneous combustion risks are identified. The corresponding degree of attention and intervention are given according to the vehicle risk level, so as to avoid the actual occurrence of vehicle burns as much as possible and improve the safety of electric vehicles.

[0006] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0007] The present invention is a vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening, comprising the following steps:

[0008] Conduct commonality screening of historical spontaneous combustion vehicles to narrow the sample scope: Analyze the basic information and characteristic performance of historical spontaneous combustion vehicles, circle out vehicles that may be at risk and meet the evaluation conditions based on battery model, mileage range, data magnitude and other obvious common characteristics, and initially narrow the sample scope;

[0009] Indicator derivation and dimension splitting: Combine different horizontal indicators and vertical time windows, classify and compare all indicators, and divide indicators into separate dimensions and supplementary dimensions according to their differentiation;

[0010] Acquire data according to the divided dimensions and construct risk factors that characterize the degree of risk based on the application model;

[0011] Screen and sort vehicles by risk in each dimension to obtain risk scores in each dimension;

[0012] The total risk score is calculated by combining the risk scores of each dimension, and the comprehensively screened high-risk vehicles are output.

[0013] Furthermore, the method of deriving indicators and splitting dimensions includes:

[0014] Horizontally, the deviation of the horizontal value, difference, and change rate of each indicator is obtained based on different working conditions;

[0015] Vertically, obtain the cumulative deviation of each indicator on the time axis and its performance within the preset period;

[0016] Based on the spontaneous combustion samples and business experience input, the discrimination and reliability of all indicators are analyzed, and the indicators with discrimination higher than the preset threshold are divided into separate dimensions, and the remaining indicators are merged into a supplementary dimension as a whole.

[0017] Furthermore, the method for constructing the risk factor includes:

[0018] For monomer inconsistency, the entropy method is used to output the abnormal coefficient;

[0019] For static pressure difference, rule aggregation is applied to output abnormal factors;

[0020] For supplementary risk factors, the risk accumulation model and classification model are applied to output the risk probability.

[0021] Furthermore, the method for obtaining the abnormal coefficient includes:

[0022] Select the monomer voltage array of the national standard message in the past n1 days, split the monomer voltage array, and obtain the monomer voltage matrix; n1 is a preset value;

[0023] Replace invalid values ​​with normal values ​​of upper and lower frames to clean data;

[0024] Taking m frames of messages as an iteration window, the voltage interval is divided based on the voltage value distribution in the window; m is a preset value;

[0025] Discretize the voltage value, count the number of samples falling in different voltage ranges, and calculate the entropy value of each battery cell in the window; standardize the entropy value to obtain the abnormal coefficient;

[0026] The calculation formula of entropy value is: E(x) = -∑p(xi)logp(xi)

[0027] Where E(x) is the sample entropy value; p(xi) is the probability density of the time occurrence in the i-th voltage interval; n is the number of voltage intervals;

[0028] The calculation formula of the anomaly coefficient is: A=|E–E ave | / σ E

[0029] Where, E is the entropy value of a battery cell; E ave is the average entropy value of multiple cells; σ E is the standard deviation of the entropy values ​​of multiple cells.

[0030] Furthermore, the method for obtaining the abnormal factor includes:

[0031] When the vehicle is stationary for more than four hours, the voltage difference when it is powered on again is a static voltage difference with high business value. The message with a static voltage difference exceeding 30 mV is of great concern to battery manufacturers. Therefore, the static voltage difference of the first frame message of the day is selected. If the highest cell voltage or the lowest cell voltage of the first frame is an invalid value, the static voltage difference of the first frame message is skipped and the static voltage difference of the next frame message is used.

[0032] When the static voltage difference of the message exceeds 0.03V, it is considered as abnormal static voltage difference of the day;

[0033] Count the percentage of abnormal days in the past 3 days, 7 days, one month, and two months;

[0034] Output anomaly factors based on rule aggregation.

[0035] Furthermore, the method for obtaining the risk probability includes:

[0036] Select indicators based on discrimination, and map the indicator value to the range of (0, 1) to obtain the risk value according to the degree to which the indicator exceeds the threshold and the amount of data;

[0037] The risk values ​​of the messages in the past n2 days are averaged to obtain the cumulative risk value of each indicator, where n2 is the preset value;

[0038] Combined with historical spontaneous combustion samples, the weight of each indicator is calculated based on the application classification model to obtain the risk probability corresponding to the indicator.

[0039] Furthermore, the method of screening and risk ranking vehicles within each dimension includes:

[0040] First, the risky vehicles in the dimension are screened out according to the threshold, and then the risk factors are sorted from small to large according to the risk degree. The serial number corresponding to the risk factor is divided by the total number of vehicles screened out in the dimension to obtain the risk score in percentage.

[0041] Furthermore, when obtaining the risk score, a baseline adjustment step of the risk score is also included, and the baseline adjustment step includes:

[0042] If the risky vehicles screened out do not exceed 5% of the total number of vehicles, the value range of the risk score of the corresponding dimension will be proportionally adjusted from [0, 100] to [60, 100].

[0043] Furthermore, the method for calculating the total risk score includes:

[0044] Evaluate the effect by dimensions, assign weights according to the hit effect of each dimension, and take a weighted average of the risk scores of each dimension to get a total risk score in percentage.

[0045] Vehicles whose total risk score exceeds the specified threshold are considered high-risk vehicles, and the vehicle's VIN, total risk score, risk score in each dimension, and index value are output;

[0046] Among them, the sum of the weights is equal to 1.

[0047] Furthermore, the weight distribution is:

[0048] The default weights of monomer inconsistency, static pressure difference, and supplementary risk factors are 0.4, 0.35, and 0.25, respectively;

[0049] When the risks corresponding to monomer inconsistency, static pressure difference, and supplementary risk factors are divided into x1, x2, and x3, the total risk is 0.4*x1+0.35*x2+0.25*x3.

[0050] The present invention has the following beneficial effects:

[0051] The present invention measures the spontaneous combustion risk of vehicles on the market through risk assessment, which is a relatively mild method, identifies vehicles with relatively high risks and focuses on them to prevent the occurrence of low-probability events; multi-dimensional sampling and multi-round screening and sorting can effectively control the proportion of risky vehicles, and at the same time, have a greater grasp of the risk judgment of top-ranked vehicles; the time window is combined with the time series characteristics to make the assessment of vehicle risks more comprehensive and detailed; the assessment dimensions of the spontaneous combustion risk of electric vehicles are enriched, the false alarm rate and missed alarm rate of risky vehicles are reduced, and through risk identification and pre-processing, efforts are made to reduce the occurrence of actual vehicle burns and enhance the confidence of car owners in the safety of electric vehicles.

[0052] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0054] Figure 1 It is a flow chart of the spontaneous combustion risk assessment method based on multi-dimensional screening and sorting of the present invention;

[0055] Figure 2 This is a funnel diagram of the data volume of a link predicted by a certain car factory A in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0057] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0058] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0059] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0060] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0061] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0062] Embodiment 1:

[0063] The present invention is a vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening, comprising the following steps:

[0064] Conduct commonality screening of historical spontaneous combustion vehicles to narrow the sample scope: Analyze the basic information and characteristic performance of historical spontaneous combustion vehicles, circle out vehicles that may be at risk and meet the evaluation conditions based on battery model, mileage range, data magnitude and other obvious common characteristics, and initially narrow the sample scope;

[0065] Indicator derivation and dimension splitting: Combine different horizontal indicators and vertical time windows, classify and compare all indicators, and divide indicators into separate dimensions and supplementary dimensions according to their differentiation. Specifically, the indicator derivation and dimension splitting methods are as follows:

[0066] The national standard fields are derived into risk-related factors one by one, and the discrimination between the confirmation factors of spontaneous combustion and non-spontaneous combustion vehicles is compared;

[0067] Risk factors are combined with time windows to consider long-term trend characteristics and recent prominent features;

[0068] Based on the importance of features and business experience, the feature factors are classified and sorted, the primary and secondary are distinguished, the dimensions are split, and internal risk assessment is performed by dimension. The final risk score is then obtained by combining the assessment results of each dimension.

[0069] Data is obtained according to the divided dimensions, and risk factors representing the degree of risk are constructed based on the application model. Specifically, the risk factor construction method is as follows: (1) For single cell inconsistency, the abnormal coefficient and standard deviation are constructed by the entropy method to represent the disorder degree of the single cell; (2) For static pressure difference, the severity of the pressure difference abnormality is represented by the sum of the proportion of abnormal days in different time windows. The shorter the time window is, the higher the weight of the corresponding abnormal days as the denominator; (3) For multiple supplementary risk indicators, different risk values ​​are set for the degree of exceeding the threshold of each indicator, and the message risks in the same time period are accumulated to obtain the cumulative risk score of each indicator, which represents the degree and frequency of exceeding the standard of the indicator;

[0070] Screen and sort vehicles by risk in each dimension to obtain risk scores in each dimension;

[0071] The total risk score is calculated by combining the risk scores of each dimension, and the comprehensively screened high-risk vehicles are output.

[0072] As an embodiment provided by the present invention, preferably, the method for obtaining the abnormal factor includes:

[0073] When the vehicle is stationary for more than four hours, the voltage difference when it is powered on again is a static voltage difference with high business value. The message with a static voltage difference exceeding 30 millivolts is of great concern to battery manufacturers. Therefore, the static voltage difference of the first frame message of the day is selected. If the highest cell voltage or the lowest cell voltage of the first frame is an invalid value, the static voltage difference of the first frame message is skipped and the static voltage difference of the next frame message is used.

[0074] When the static voltage difference of the message exceeds 0.03V, it is considered as abnormal static voltage difference of the day;

[0075] Count the percentage of abnormal days in the past 3 days, 7 days, one month, and two months;

[0076] Output anomaly factors based on rule aggregation.

[0077] As an embodiment provided by the present invention, preferably, the method for obtaining the risk probability includes:

[0078] Select indicators based on discrimination, and map the indicator value to the range of (0, 1) to obtain the risk value according to the degree to which the indicator exceeds the threshold and the amount of data;

[0079] The risk values ​​of the messages in the past n2 days are averaged to obtain the cumulative risk value of each indicator, where n2 is the preset value;

[0080] Combined with historical spontaneous combustion samples, the weight of each indicator is calculated based on the application classification model to obtain the risk probability corresponding to the indicator.

[0081] As an embodiment provided by the present invention, preferably, the method for screening and risk ranking vehicles within each dimension includes:

[0082] First, the risky vehicles in the dimension are screened out according to the threshold, and then the risk factors are sorted from small to large according to the risk degree. The serial number corresponding to the risk factor is divided by the total number of vehicles screened out in the dimension to obtain the risk score in percentage.

[0083] As an embodiment provided by the present invention, preferably, the numerical value of the risk factor is converted into a risk score in percentage, and the steps are as follows:

[0084] (1) For vehicles with a risk factor value of 0, the risk score is directly set to 0 and they can be initially eliminated from the dimension;

[0085] (2) For vehicles with non-zero risk factor values, they are sorted in ascending order of risk. If two vehicles have the same risk factor value, they have the same serial number, and the serial number of the next vehicle is increased by 2. The serial number is divided by the total number of vehicles with non-zero risk factors to obtain the risk score in percentage.

[0086] (3) If the number of vehicles with risk factors greater than 0 is small and the factor standard is relatively strict, the risk score based on the percentage system is scaled and translated to adjust the value range.

[0087] As an embodiment provided by the present invention, preferably, the method for calculating the total risk score includes:

[0088] Evaluate the effect by dimensions, assign weights according to the hit effect of each dimension, and take a weighted average of the risk scores of each dimension to get a total risk score in percentage.

[0089] Vehicles whose total risk score exceeds the specified threshold are considered high-risk vehicles, and the vehicle's VIN, total risk score, risk score in each dimension, and main indicator values ​​are output;

[0090] The sum of the weights is equal to 1, and the risk scores of different dimensions are weighted averaged, which not only highlights the important dimensions, but also meets the situation where the medium risks in multiple dimensions lead to overall high risks. The weight is a parameter with a default value and can be adjusted according to the actual situation.

[0091] As an embodiment provided by the present invention, preferably, the weight distribution is:

[0092] The default weights of monomer inconsistency, static pressure difference, and supplementary risk factors are 0.4, 0.35, and 0.25, respectively;

[0093] When the risks corresponding to monomer inconsistency, static pressure difference, and supplementary risk factors are divided into x1, x2, and x3, the total risk is 0.4*x1+0.35*x2+0.25*x3.

[0094] Among them, we currently have access to only national standard message data. The relevant indicators of the national standard are only battery signals such as voltage, current, temperature and resistance. We cannot get pressure and gas sensor data, so we can only make full use of national standard data for risk assessment. The national standard data of a certain vehicle includes the current battery signal and the historical battery signal, which is the message data uploaded every 10 seconds since the vehicle was used. The data we used n days before the car burn belongs to the historical national standard data, and the data used on the day of the car burn (the characteristics before the car burn occurred) belongs to the current data.

[0095] For differentiation: take inventory of all possible features, analyze the proportion of vehicles with such features among spontaneous combustion vehicles, and compare them with the proportion of vehicles with such features among non-spontaneous combustion vehicles. For example, overcharging and over-discharging are behaviors that damage the battery, and most spontaneous combustion vehicles will have a certain degree of overcharging and over-discharging 7 days before the car burns. The cumulative risk value of overcharging and over-discharging of non-spontaneous combustion vehicles is generally lower than that of spontaneous combustion vehicles.

[0096] Embodiment 2:

[0097] See also Figure 1 As shown, combining the current performance and historical cumulative performance of battery signals under different working conditions, different weights are assigned to indicators of different importance in different time periods, and a more comprehensive and accurate assessment of the risk of spontaneous combustion is performed. The present invention aims to identify vehicles with medium and high spontaneous combustion risks, give corresponding levels of attention and intervention according to the vehicle risk level, avoid the occurrence of actual vehicle burns as much as possible, and improve the safety of electric vehicles. The present invention is a vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening, including the following steps:

[0098] Step 1: Initial screening of commonalities. Analyze the basic information and characteristic performance of historical spontaneous combustion vehicles, and circle out vehicles that may be at risk and meet the assessment conditions based on battery model, mileage range, data magnitude, and other obvious commonalities (commonalities: the highest temperature, maximum SOC value, and minimum SOC value that occurred seven days before the vehicle burned), and initially narrow the sample range;

[0099] Step 2: Derivation of indicators and dimension splitting. The method of derivation of indicators and dimension splitting includes:

[0100] Horizontally, the deviation of the horizontal value, difference, and change rate of each indicator is obtained based on different working conditions;

[0101] Vertically, obtain the cumulative deviation of each indicator on the time axis as well as the long-term and short-term performance;

[0102] Based on the spontaneous combustion samples and business experience input, analyze the discrimination and reliability of all indicators, divide the indicators with discrimination higher than the preset threshold into separate dimensions, and merge the remaining indicators into a supplementary dimension. Horizontally, you can see the deviation of the horizontal value, difference, and change rate of each indicator according to the working conditions; vertically, you can see the cumulative deviation of each indicator on the time axis and the long-term and short-term performance. Based on the spontaneous combustion samples and business experience input, analyze the discrimination and reliability of all indicators, and divide the dimensions according to the indicator category and importance;

[0103] Step 3: Take data from different dimensions, apply the model, and construct factors that represent the degree of risk;

[0104] For monomer inconsistency, the entropy method is used to output the abnormal coefficient;

[0105] For static pressure difference, rule aggregation is applied to output abnormal factors;

[0106] For other risk factors, the risk accumulation model and classification model are applied to output the risk probability;

[0107] Step 4: Screen and sort the vehicles in each dimension: first use the threshold to screen out risky vehicles in the dimension, then sort the risk factors from small to large in terms of risk level, and divide the sequence number by the total number of vehicles screened in the dimension to get the risk score in percentage.

[0108] Step 5: Calculate the total risk score by combining the risk scores of each dimension, and output the high-risk vehicles that have been comprehensively screened. Evaluate the effect by dimension, assign weights according to the hit effect of each dimension, and the sum of the weights is equal to 1. Take the weighted average of the risk scores of each dimension to get the total risk score in percentage. Vehicles with a total risk score exceeding the specified threshold are considered high-risk vehicles, and the vin, total risk score, risk scores of each dimension, and main indicator values ​​of the vehicle are output.

[0109] As an embodiment provided by the present invention, preferably, according to all the historical spontaneous combustion vehicles of a certain car factory, there are about 25 vehicles, all of which have two specific battery models and mileage within 1-11w kilometers. All of them have a maximum temperature of more than 40 degrees or a minimum soc value of less than 15 or a maximum soc value of more than 95 in the past seven days. Therefore, only for vehicles with battery models and mileage ranges that have experienced car burns, normal vehicles with a maximum temperature of less than 40 degrees in the past seven days and a soc value always between 16-94 are excluded, and the remaining vehicles enter the subsequent risk assessment (if samples in a certain range have never experienced a car burn and the characteristics are all within the safe range, it is assumed that there is no risk).

[0110] Embodiment three:

[0111] The present invention is a vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening, comprising the following steps:

[0112] Step 1: Analyze the basic information and common characteristics of historical spontaneous combustion vehicles, and use normal vehicles as a reference to see if all spontaneous combustion vehicles can be screened out, filter out a small number of normal vehicles, and narrow the evaluation sample;

[0113] Step 2: Based on the risk indicators derived from the national standard message data, different horizontal indicators and vertical time windows are combined, and all indicators are classified, sorted and compared. Indicators with obvious differentiation are divided into separate dimensions, and secondary indicators are merged into a supplementary dimension;

[0114] Horizontal indicators:

[0115] The voltage can be derived into overvoltage, undervoltage, large voltage difference, voltage fluctuation, single cell voltage entropy, and voltage probe failure rate under specific operating conditions (operating conditions are divided into driving conditions, charging conditions, and stationary conditions); the temperature can be derived into overtemperature, low temperature, large temperature difference, rapid temperature rise, and temperature probe failure rate under specific operating conditions; the current can be derived into large current charging and discharging, and abnormal current fluctuation; the resistance can be seen from the minimum value of the driving insulation resistance, the duration of the low resistance value, and the amplitude of the sudden drop in resistance; the SOC can correspond to overcharging and over-discharging.

[0116] Vertical indicators:

[0117] On the one hand, the cumulative risk calculation can be done by adding up the product of the degree and frequency of an indicator exceeding the threshold in the past n days. On the other hand, considering the persistence and freshness of the indicator anomaly, the weighted average of the anomaly proportions in the past three days, seven days, one month, three months, and the entire life cycle can be taken.

[0118] Dimension splitting:

[0119] At present, for a certain car manufacturer, the indicators with strong differentiation and high business attention are single-cell inconsistency and static pressure difference abnormality, which are temporarily divided into three dimensions: single-cell consistency, static pressure difference and other supplementary risk indicators.

[0120] Step 3: Construct screening conditions and risk factors for sorting by dimensions:

[0121] Cell consistency: Select the cell voltage array of the national standard message in the past 7 days, split the array, and obtain the cell voltage matrix; replace the invalid values ​​with the normal values ​​of the upper and lower frames for data cleaning; use 100 frames of messages as an iteration window, and divide the voltage interval based on the voltage value distribution in the window; discretize the voltage value, count the number of samples falling in different voltage intervals, and calculate the entropy value of each cell in the window; standardize the entropy value to obtain the abnormality coefficient;

[0122] The entropy calculation formula is: E(x) = -∑p(xi)logp(xi)

[0123] In the formula, E(x) is the sample entropy value; p(xi) is the probability density of the time occurrence in the i-th interval; i = 1, 2, 3, ..., n, n is the number of intervals.

[0124] The anomaly coefficient based on the Z score is calculated as follows:

[0125] A=|E–Eave| / σE

[0126] Where E is the entropy of a battery cell; Eave is the average entropy value of multiple cells; σE is the standard deviation of the entropy values ​​of multiple cells.

[0127] The remaining risks are accumulated: select indicators with a certain degree of discrimination, and map the values ​​of each indicator to a risk value in the range of (0, 1) according to the degree of exceeding the threshold and the amount of data (as described in the supplementary answer in the derivation of horizontal indicators, first assign a value, then average the corresponding values ​​of the message, and obtain a risk value in the interval of (0, 1)). Average the risk values ​​of the messages in the past n days to obtain the cumulative risk value of each indicator. Combined with historical spontaneous combustion samples, the classification model is applied to calculate the weight of each indicator (label vehicles with and without spontaneous combustion 0 and 1, use the logistic regression model to train the samples (features + labels), and the model automatically runs out parameters, that is, the weight of each indicator). However, because there are too few positive samples, the risk probability determined by this method has a high false alarm rate.

[0128] Step 4: Screen the vehicles once in each dimension, and sort the screened vehicles according to the risk factors constructed in step 3 (risk factors include derived risk indicators and more abstract risk indicators after risk indicator selection and aggregation; risk values ​​have the initial single risk value, and then produce a dimensional risk value after aggregation of each dimension, which is converted into a ranked risk value, and then into a final comprehensive risk value), and convert the serial number into a ranked risk score in percentage. In addition, adjust the baseline of the risk score according to the severity of the screening. For example, if a small number of vehicles screened out by single inconsistency have high risks, the value range of the risk score of this dimension will be adjusted from [0, 100] to [60, 100] in proportion;

[0129] Step 5: Calculate the total risk score. Since the number of risky vehicles screened out by the consistency of the monomer is less than that of the risky vehicles screened out by the static pressure difference, and the number of risky vehicles screened out by the static pressure difference is far less than that of the risky vehicles screened out by the other indicators, the number of hits in each dimension is similar, the number of false positives in the first two dimensions is small, and the precision is good. The default values ​​of the provisional coefficient (the abnormal coefficient is an entropy value indicator within the dimension of monomer inconsistency. The coefficient here sets the weight for the sorted risk value output in each dimension, and the two are unrelated) are 0.4, 0.35, and 0.25, so that the high-risk vehicles finally screened out have risks in at least two dimensions, and the risky vehicles in the first two dimensions are given higher weights. The coefficient is a configurable parameter and can be adjusted manually according to the actual situation. The total risk score is calculated by weighted average of the risk scores of each dimension, and secondary screening is performed based on whether the total risk score exceeds 0.6, and the top N high-risk vehicles with the total risk score ranking are output. The business can further subdivide the high-risk vehicles into different levels and give corresponding treatment plans according to the risk level.

[0130] As an embodiment provided by the present invention, preferably, the strictness refers to the proportion of risky vehicles selected in each dimension to the total vehicles. The fewer the risky vehicles, the stricter the conditions. If about half of the risky vehicles are selected from the total vehicles, the selection conditions are too loose.

[0131] As an embodiment provided by the present invention, preferably, when obtaining the risk score, a baseline adjustment step of the risk score is also included, and the baseline adjustment step includes:

[0132] If the risky vehicles screened out do not exceed 5% of the total number of vehicles, the value range of the risk score of the corresponding dimension will be proportionally adjusted from [0, 100] to [60, 100]; if the original risk score is x1, the value range is [0, 100], and there is more than 50% confidence that all the screened vehicles are risky, and the screened vehicles have at least 60 points, then the risk score will be converted to x2, x2=60+0.4*x1.

[0133] As an embodiment provided by the present invention, preferably, a method for selecting an indicator with a certain degree of discrimination is: comparing 25+ spontaneous combustion vehicles and 2,000 randomly sampled non-spontaneous combustion vehicles, looking at the proportion of a certain indicator, such as the feature of pressure difference exceeding 0.1, there are 13 spontaneous combustion vehicles, accounting for about 50%, and 50 non-spontaneous combustion vehicles, accounting for about 2.5%, then the discrimination of this feature is (50-2.5%), which has a certain degree of discrimination and can exclude a large number of non-spontaneous combustion vehicles; taking inventory of all horizontal indicators, selecting indicators with large differences in sample concentrations between spontaneous combustion and non-spontaneous combustion.

[0134] As an embodiment provided by the present invention, preferably, if a vehicle has a static pressure difference exceeding 0.03v for 2 days in the past 3 days, exceeds 0.03v for 5 days in the past 7 days, and exceeds 0.03v for 20 days in the past month, the proportions of abnormal days are: 2 / 3, 5 / 7, 20 / 30 respectively.

[0135] As an embodiment provided by the present invention, preferably, a message with a voltage exceeding a threshold value of 390 is assigned a value of 1, otherwise it is 0, and the values ​​of all messages in the past 7 days are averaged to obtain an oversufficient value;

[0136] For messages with voltage lower than 330, the value is 1, otherwise it is 0. The value of all messages in the past 7 days is averaged to get the undervoltage score;

[0137] For messages with a pressure difference exceeding 0.05 under charging conditions, a value of (x-0.1)^2 is assigned; (x is the message pressure difference. This formula can give higher risk scores to higher pressure difference scores, with an exponential increase. At the same time, with 0.1 as the boundary, 0.05-0.1 is a general risk, and above 0.1 is a more serious risk).

[0138] The average of all charging message values ​​in the past 7 days is calculated to obtain the pressure difference score;

[0139] The voltage fluctuation is obtained by calculating the standard deviation of the voltage of all charging messages.

[0140] The failure rate is calculated by counting the percentage of messages with the two invalid values ​​of single-cell voltage being 2 or 6.5535. The voltage entropy is calculated using the entropy formula below.

[0141] As an embodiment provided by the present invention, preferably, a value of x-40 (x is the highest charging temperature of the message) is assigned to the message whose charging temperature exceeds the threshold value of 40, otherwise it is 0, and the values ​​of all messages charged in the past 7 days are averaged to obtain the high temperature score;

[0142] For messages with a charging temperature difference exceeding 5, a value of x-5 (x is the maximum charging temperature difference of the message) is assigned, otherwise it is 0. The values ​​of all messages charged in the past 7 days are averaged to obtain the temperature difference score;

[0143] Invalid values ​​of the temperature probe value of 255 or -40 are considered as temperature probe failures, and the proportion of temperature probe failures in the messages in the past 7 days is counted.

[0144] As an embodiment provided by the present invention, preferably, a value of 1 is assigned to a message with a charging current greater than 200A, a value of 0.8 is assigned to a message with a charging current between 190-200A, a value of 0.6 is assigned to a message with a charging current between 170-190A, and the rest are assigned a value of 0. The value assignments of all messages charged in the past 7 days are averaged to obtain a high charging current score; a charging current standard deviation exceeding a threshold in the past 7 days is regarded as an abnormal charging current fluctuation.

[0145] As an embodiment provided by the present invention, preferably, the insulation resistance under driving conditions is assigned a value of 1 if it is less than 500, otherwise 0, and the values ​​of the messages in the past 7 days are averaged to obtain a score with a relatively small resistance;

[0146] If the resistance change between consecutive messages under driving conditions exceeds 10,000, a value of 1 is assigned, otherwise it is 0. The resistance mutation score is obtained by averaging the values ​​assigned to messages in the past 7 days.

[0147] As an embodiment provided by the present invention, preferably, for soc equal to 100, a value of 1 is assigned, for soc between 97-99, a value of 0.8 is assigned, for soc between 95-96, a value of 0.6 is assigned, and the rest are assigned 0, and the average value of the assignments for all messages is sufficient;

[0148] For soc less than 5, assign a value of 1, for soc between 5-10, assign a value of 0.8, for soc between 10-15, assign a value of 0.6, and for the rest, assign a value of 0. The average value of the assignments for all messages is over-scoring.

[0149] As an embodiment provided by the present invention, preferably, in the above horizontal indicator derivation method, after setting a threshold value to confirm whether a message has a risk, the method of averaging the values ​​of the last n days to obtain a single risk score is a risk accumulation calculation method;

[0150] Persistence refers to the changing trend of the single risk score over the entire life cycle of the vehicle. One message data is taken every month to determine whether there is risk.

[0151] Freshness means that higher weight is given to message data that is closer to the current one. For example, if the risk score of a single item in the past three days is small and the risk score in the past seven days is large, the comprehensive score is 1 / 3*small risk score in the past three days + 1 / 7*large risk score in the past seven days.

[0152] As an embodiment provided by the present invention, preferably, if the risk scores of the three dimensions are x1, x2, and x3 respectively, the total risk score is 0.4*x1+0.35*x2+0.25*x3, and the weights of the first two risk scores are higher, and the latter is lower;

[0153] If a vehicle has no risk in two of the dimensions and the risk score is 0, even if the risk score in the third dimension is 100, the total risk score must be less than 50 and will not be selected as the final high risk score, because the coefficient assigned to a single risk by the total risk score formula does not exceed 0.5.

[0154] Embodiment 4:

[0155] Based on step 3 of the third embodiment, the method for data cleaning is as follows:

[0156] The first step is to obtain the cell voltage array and time in the message data as follows:

[0157]

[0158]

[0159]

[0160]

[0161] The second step is to sort the time from small to large, split the array to get the voltage value of each voltage probe, and the output voltage matrix is ​​as follows:

[0162]

[0163] Step 3: For 0 or invalid values, fill them with the value of the previous frame of the probe. If the previous frame is empty, try to fill it with the next frame. If the entire column is empty, delete the column.

[0164] Step 4: Take 100 frames of messages as an iteration window, and based on the maximum value v_max and the minimum value v_min of the voltage value in the window and the preset interval number 10, obtain 10 voltage intervals, namely [v_min, v_min+i), [v_min+i, v_min+2i), ..., [v_min+9i, v_min+10i];

[0165] If the current cell voltage has a minimum value of 3651 and a maximum value of 4120, the 10 voltage ranges are:

[0166] [3651,3697.9),[3697.9,3744.8),[3744.8,3791.7),(3791.7,3838.6],[3838.6,3885.5),[3885.5,3932.4),[3932.4,3979.3),[3979.3,4026.2),

[0167] [4026.2,4073.1),[4073.1,4120.0];

[0168] Step 5: Discretization. Replace the voltage value of each probe in 100 frames of messages with voltage intervals. There are only 10 interval values ​​in the sample. The voltage is mapped from numerical type to categorical variables 1-10, corresponding to 10 voltage intervals from small to large, as follows:

[0169]

[0170] Step 6: For each probe, count the sample proportion p in each voltage interval;

[0171] Voltage

[0172]

[0173]

[0174] Use the entropy calculation formula to sum the calculated values ​​of the 10 intervals to obtain the entropy value corresponding to the probe;

[0175] For example, the entropy value E of probe 1 is equal to:

[0176] -(0.46*log0.46+0.2*log0.2+0.113333*log0.113333+0.106667*log0.106667+0.093333*log0.093333+0.02*log0.02+0.006667*log0.006667+0+0+0)

[0177] Probe 1 Probe 2 ... Probe 91 Entropy 2.16055448984 2.606508260758449 2.1838002785 Value 1381 3 ... 1642

[0178] Step 7: Use the following formula (anomaly coefficient calculation formula) to standardize the entropy value to obtain A, which is the output result of the iteration window;

[0179] Probe 1 Probe 2 ... Probe 91 7.62676

[0181] Entropy value -0.592847 5…-0.164391

[0182] The above entropy value result can be obtained for each iteration window. If a vehicle has 3,000 messages in the past 7 days and each 100 frames is an iteration window, then 30 rows (number of iteration windows) and 91 columns (number of probes) of abnormal coefficients can be obtained.

[0183] As an embodiment provided by the present invention, preferably, the method for obtaining the abnormal coefficient includes:

[0184] Select the monomer voltage array of the national standard message in the past n1 days, split the monomer voltage array, and obtain the monomer voltage matrix; n1 is a preset value;

[0185] Replace invalid values ​​with normal values ​​of upper and lower frames to clean data;

[0186] Taking m frames of messages as an iteration window, the voltage interval is divided based on the voltage value distribution in the window; m is a preset value;

[0187] Discretize the voltage value, count the number of samples falling in different voltage ranges, and calculate the entropy value of each battery cell in the window; standardize the entropy value to obtain the abnormal coefficient;

[0188] The calculation formula of entropy value is: E(x) = -∑p(xi)logp(xi)

[0189] Where E(x) is the sample entropy value; p(xi) is the probability density of the time occurrence in the i-th voltage interval; n is the number of voltage intervals;

[0190] The calculation formula of the anomaly coefficient is: A=|E–E ave | / σ E

[0191] Where, E is the entropy value of a battery cell; E ave is the average entropy value of multiple cells; σ E is the standard deviation of the entropy values ​​of multiple cells.

[0192] Embodiment five:

[0193] Based on step 4 of the third embodiment, the method for obtaining the risk factor is:

[0194] Single-body inconsistency: Screening is performed based on whether the absolute value of the anomaly coefficient exceeds the industry threshold |4| (when the absolute value of the anomaly coefficient (i.e., the standardized entropy value) exceeds 4, it is considered that the single-body inconsistency of the window is more obvious) and the maximum value of the anomaly coefficient standard deviation, and the number of iterative windows where the anomaly coefficient exceeds the threshold is output as the risk factor for sorting in this dimension (it is the risk value within this dimension and is also used for sorting in the next stage. As a sorting factor, it can be called a risk value or a risk factor).

[0195] Static pressure difference: Use the proportion of abnormal days in the past n days for screening, and sum the proportion of abnormal days in several time periods to obtain the risk factor for sorting.

[0196] Other risk indicators: Use the risk probability value output by the model (risk accumulation model + binary logistics regression model, learn the model parameters through sample training, and use the trained model to predict the probability of spontaneous combustion of other vehicles) to screen high-risk vehicles, and use the risk probability value as the ranking risk factor.

[0197] A vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening measures the spontaneous combustion risk of vehicles on the market in a relatively mild way through risk assessment, identifies vehicles with relatively high risks for special attention, and prevents the occurrence of low-probability events; sampling multi-dimensional and multi-round screening and sorting can effectively control the proportion of risky vehicles, and at the same time, have a greater grasp of the risk judgment of top-ranked vehicles; combined with the time window, it integrates the time series characteristics, and the assessment of vehicle risks is more comprehensive and detailed; it enriches the assessment dimensions of electric vehicle spontaneous combustion risks, reduces the false alarm rate and missed alarm rate of risky vehicles, and strives to reduce the occurrence of actual vehicle burns through risk identification and pre-processing, thereby enhancing car owners' confidence in the safety of electric vehicles.

[0198] Embodiment six:

[0199] Based on the above embodiments, the present application enriches the assessment dimensions of the risk of spontaneous combustion of electric vehicles, reduces the false alarm rate and missed alarm rate of risky vehicles, and strives to reduce the occurrence of actual vehicle burns through risk identification and pre-processing, thereby enhancing the confidence of car owners in the safety of electric vehicles.

[0200] We provided a prediction of 10 risky vehicles to a certain car manufacturer A, which fed back to the battery manufacturer for investigation and found that 2 of the vehicles were highly suspected to be risk packages and needed to be replaced in the market. The remaining 8 vehicles had a certain degree of undervoltage and were the focus of attention.

[0201] For example, the total number of vehicles is 22,000, and 14,000 vehicles are selected for risk assessment through preliminary screening;

[0202] 52 vehicles were selected from the dimension of monomer inconsistency, and the abnormal coefficient of entropy value of representative factors in monomer inconsistency was used as the ranking risk score, which was converted into a risk score x1 in percentage.

[0203] 61 vehicles were selected from the dimension of abnormal static pressure difference, and the sum of the days with abnormal representative factors was used as the ranking risk score, which was converted into a percentage risk score x2;

[0204] 2,000 vehicles were selected from the dimension of cumulative risk, and the probability scores obtained by the binary classification model were used as the ranking risk scores, which were converted into a risk score x3 in percentage.

[0205] The output results of the three dimensions are weighted averaged (0.4*x1+0.35*x2+0.25*x3) to get the total risk score in percentage. There are 10 vehicles with a total risk score exceeding 0.5.

[0206] Feedback is sent to the battery manufacturer. After investigation, the manufacturer confirms that two vehicles need to have their batteries replaced. The data volume funnel of a certain car manufacturer A predicts that Figure 2 As shown;

[0207] By comparing with the existing system, the present invention has the following advantages:

[0208] 1. Since there are very few spontaneous combustion vehicles in history, the positive and negative samples are extremely unbalanced, and the classification model is too absolute in predicting whether the vehicle will burn, and the accuracy is poor. This application measures the spontaneous combustion risk of vehicles in the market through risk assessment, a relatively mild method, identifies vehicles with relatively high risks and focuses on them to prevent the occurrence of low-probability events.

[0209] 2. Due to the high cost of vehicle recalls and battery pack replacements, which also have a significant impact on the owner's experience, it is necessary to reduce false positives as much as possible. This application uses multi-dimensional and multi-round screening and sorting to effectively control the proportion of risky vehicles and at the same time, have a greater degree of confidence in the risk assessment of top-ranked vehicles.

[0210] 3. This application comprehensively considers the possible abnormal conditions of various indicators, and integrates the time series characteristics with the time window, so that the assessment of vehicle risks is more comprehensive and detailed. In addition, the dimensional evaluation allows the feature logic within the dimension to be independently tuned, reducing the coupling between modules.

[0211] 4. The subsequent application of risk scores is relatively flexible. For example, further vehicle risk profiling can be made based on the trend of daily risk scores, the number of days with high risk scores, the proportion of vehicles in different categories, etc. The business application scenarios are relatively flexible and diverse.

[0212] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0213] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening, characterized in that: The following steps are involved: Conduct a preliminary screening of common characteristics of vehicles with historical spontaneous combustion to narrow the sample range; Indicator derivation and dimension splitting: Combine different horizontal indicators and vertical time windows, classify and compare all indicators, and divide indicators into separate dimensions and supplementary dimensions according to their differentiation; Acquire data according to the divided dimensions, and then construct risk factors that characterize the degree of risk based on the application model; Screen and sort vehicles by risk in each dimension to obtain risk scores in each dimension; The total risk score is calculated by combining the risk scores of each dimension, and the comprehensively screened high-risk vehicles are output.

2. The vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening according to claim 1 is characterized in that: The method of deriving indicators and splitting dimensions includes: Horizontally, the deviation of the horizontal value, difference, and change rate of each indicator is obtained based on different working conditions; Vertically, obtain the cumulative deviation of each indicator on the time axis and its performance within the preset period; Based on the spontaneous combustion samples and business experience input, the discrimination and reliability of all indicators are analyzed, and the indicators with discrimination higher than the preset threshold are divided into separate dimensions, and the remaining indicators are merged into a supplementary dimension as a whole.

3. The vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening according to claim 1 is characterized in that: The method for constructing the risk factor includes: For monomer inconsistency, the entropy method is used to output the abnormal coefficient; For static pressure difference, rule aggregation is applied to output abnormal factors; For supplementary risk factors, the risk accumulation model and classification model are applied to output the risk probability.

4. The vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening according to claim 3 is characterized in that: The method for obtaining the abnormal coefficient includes: Select the monomer voltage array of the national standard message in the past n1 days, split the monomer voltage array, and obtain the monomer voltage matrix; n1 is a preset value; Replace invalid values ​​with normal values ​​of upper and lower frames to clean data; Taking m frames of messages as an iteration window, the voltage interval is divided based on the voltage value distribution in the window; m is a preset value; Discretize the voltage value, count the number of samples falling in different voltage ranges, and calculate the entropy value of each battery cell in the window; standardize the entropy value to obtain the abnormal coefficient; The calculation formula of entropy value is: E(x) = -∑p(xi)logp(xi); Where, E(x) is the sample entropy value; p(xi) is the probability density of the time occurrence in the i-th voltage interval; i = 1, 2, 3, ..., n, n is the number of voltage intervals; The calculation formula of the anomaly coefficient is: A=|E–E ave | / σ E ; Where, E is the entropy value of a battery cell; E ave is the average entropy value of multiple cells; σ E is the standard deviation of the entropy values ​​of multiple cells.

5. The vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening according to claim 3 is characterized in that: The method for obtaining the abnormal factor includes: Select the static pressure difference of the first frame message of each day; When the highest cell voltage or the lowest cell voltage of the first frame message is an invalid value, the static pressure difference of the first frame message is skipped, and the static pressure difference of the next frame message is obtained; Count the percentage of abnormal days in the past 3 days, 7 days, one month, and two months; Output abnormal factors according to rule aggregation; Among them: when the static voltage difference of the message exceeds 0.03V, it is considered that the static voltage difference of the day is abnormal.

6. The vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening according to claim 3 is characterized in that: The method for obtaining the risk probability includes: Select indicators based on discrimination, and map the indicator value to the range of (0, 1) to obtain the risk value according to the degree to which the indicator exceeds the threshold and the amount of data; The risk values ​​of the messages in the past n2 days are averaged to obtain the cumulative risk value of each indicator, where n2 is the preset value; Combined with historical spontaneous combustion samples, the weight of each indicator is calculated based on the application classification model to obtain the risk probability corresponding to the indicator.

7. The vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening according to claim 1 is characterized in that: Methods for screening and risk ranking vehicles within each of the dimensions include: First, the risky vehicles in the dimension are screened out according to the threshold, and then the risk factors are sorted from small to large according to the risk degree. The sorting number corresponding to the risk factor is divided by the total number of vehicles screened out in the dimension to obtain the risk score in percentage.

8. The vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening according to claim 7 is characterized in that: When obtaining the risk score, a baseline adjustment step of the risk score is also included, and the baseline adjustment step includes: If the risky vehicles screened out do not exceed 5% of the total number of vehicles, the value range of the risk score of the corresponding dimension will be proportionally adjusted from [0, 100] to [60, 100].

9. The vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening according to claim 1 is characterized in that: The method of calculating the total risk score includes: Evaluate the effect by dimensions, assign weights according to the hit effect of each dimension, and take a weighted average of the risk scores of each dimension to get a total risk score in percentage. Vehicles whose total risk score exceeds the specified threshold are considered high-risk vehicles, and the vehicle's VIN, total risk score, risk score in each dimension, and index value are output; Among them, the sum of the weights is equal to 1.

10. The vehicle spontaneous combustion risk assessment method based on multi-dimensional sorting and screening according to claim 9 is characterized in that: The weight distribution is: The default weights of monomer inconsistency, static pressure difference, and supplementary risk factors are 0.4, 0.35, and 0.25, respectively; When the risks corresponding to monomer inconsistency, static pressure difference, and supplementary risk factors are divided into x1, x2, and x3, the total risk is 0.4*x1+0.35*x2+0.25*x3.

Images