A Method for Rating Poor Driving Behavior Based on Improved Extended Confidence Rule Reasoning

Abstract

This invention provides a method, system, storage medium, and electronic device for rating poor driving behavior based on improved extended belief rule reasoning, relating to the field of road traffic safety technology. In this invention, the extended belief rule reasoning (EBRB) method is introduced, utilizing a data-driven decision model to effectively solve the combinatorial explosion problem of the rule base and improve the system's rationality and interpretability. The SC clustering algorithm is introduced, proposing a new activation rule determination method, which effectively reduces the number of rule accesses, improves system operating efficiency, and enhances the consistency among candidate activation rules. Furthermore, using the aforementioned activation rule determination method, a user poor driving behavior rating model based on SC_EBRB is constructed, improving the objectivity and accuracy of poor driving behavior evaluation.

Description

Technical Field

[0001] This invention relates to the field of road traffic safety technology, and specifically to a method, system, storage medium, and electronic device for rating poor driving behavior based on improved extended confidence rule reasoning. Background Technology

[0002] With the rapid development of my country's economy, the number of motor vehicles has increased dramatically, and the transportation industry has developed and progressed rapidly. However, while promoting social and economic development and facilitating people's daily travel, it has also brought about problems that cannot be ignored, such as traffic safety and environmental pollution. Drivers are the main players in road traffic, and their driving behavior is closely related to road safety and is an important factor affecting traffic safety.

[0003] Existing driving behavior assessment models still exhibit significant subjectivity, demanding not only high professional competence from raters but also background knowledge relevant to the research question. With the rapid development of vehicle-to-everything (V2X) and sensor technologies, intelligent connected transportation systems have emerged, gradually enabling real-time vehicle monitoring and information transmission, thus providing new avenues for driving behavior research.

[0004] The Extended Belief Rule Base (EBRB) model is a more flexible and widely applicable complex decision-making model capable of accurately representing uncertain information with probabilistic, fuzzy, and incomplete characteristics. Furthermore, extended belief rules are automatically generated from given sample input-output data pairs, effectively avoiding the combinatorial explosion problem of traditional rule bases. The model comprises two parts: knowledge representation and knowledge reasoning. Knowledge representation is implemented through an extended belief rule base expert system, while knowledge reasoning is achieved using an Evidential Reasoning (ER) algorithm.

[0005] However, traditional EBRB activation rule determination methods have some shortcomings in practical operation, which become more apparent when the rule base is large. On the one hand, traditional activation rule determination methods access and activate almost all rules in the rule base, resulting in high computational costs and long processing times, affecting practical efficiency. On the other hand, due to the large number of activation rules, inconsistencies exist in the rules activated by the same rule, thus affecting the accuracy of the inference results. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a method, system, storage medium, and electronic device for rating poor driving behavior based on improved extended confidence rule reasoning, thus solving the technical problem of unreasonable driving behavior assessment models.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A method for rating poor driving behavior based on improved extended confidence rule reasoning includes:

[0011] S1. Obtain historical vehicle data and the level of bad driving behavior of several users, and obtain evaluation indicators of bad driving behavior based on the historical vehicle data;

[0012] S2. Standardize the evaluation index values ​​and bad driving behavior levels for each user, obtain the input data of the antecedent and consequent of the rule respectively, and form corresponding standardized input data pairs; combine the standardized input data pairs corresponding to all users to form a standardized dataset.

[0013] S3. Based on each standardized input data pair, obtain the belief distribution structure of the preceding and resulting attributes of the extended confidence rule, and generate the corresponding extended confidence rule; the extended confidence rules corresponding to all standardized input data pairs are used to form an extended confidence rule library.

[0014] S4. Based on the belief distribution structure of the preceding attributes of each extended confidence rule, the extended confidence rule base is clustered into several sub-rule bases using the SC clustering algorithm.

[0015] S5. Convert the vehicle data of the user to be analyzed into a belief distribution structure, calculate the similarity of the belief distribution structure between the preceding attributes of each rule in the extended belief rule base, obtain candidate activation rules, and use the sub-rule base where the candidate activation rules are located as the candidate activation sub-rule base.

[0016] S6. Based on the alternative activation sub-rule base and its updated initial rule weights, the ER method is used to obtain the level of bad driving behavior of the user to be analyzed.

[0017] Preferably, in S3:

[0018] The input data of the antecedent of the rule in the standardized input data pair Belief distribution structure transformed into the preceding attribute

[0019]

[0020]

[0021] α i，j′ =0; j′=1,...,Ji j′≠j, j+1

[0022] Among them, A i，j For the i-th preceding attribute U i The utility value of the j-th reference level, which changes as the parameters are learned; α i，j For input data Transform to U i The matching degree at the j-th reference level, i.e., U i The degree of belief at the j-th reference level; J i For U i The number of reference levels; T is the number of preceding attributes;

[0023] Similarly, the input data D of the consequent of the rule in the standardized input data pair * Belief distribution structure transformed into outcome attributes

[0024] Let the k-th rule R k The representation is: if U is {A, α} k}, then V is {D, β} k};

[0025] In the antecedent of the rule, U = {U i} represents the preceding attribute of the rule; A = {A i} represents the reference level for the preceding attribute of the rule. For U i The utility value of the reference level; For in R k The belief level of the reference level for all preceding attributes. For R k Chinese Description U i The degree of belief, and satisfying For R k The distribution structure of beliefs about the preceding attributes E(A) k ), then U i The belief distribution structure is represented as

[0026] In the consequent of a rule, V is the result attribute of the rule; D = {D n , n = 1, ..., N} are the reference levels of the result attributes of the rule, and there are N reference levels in total for the result attributes; For R k The belief level of the outcome attribute, i.e., using the reference value D = {D n The degree of belief in the outcome attribute, and satisfying the following conditions: like This indicates that the result of the k-th rule is complete; otherwise, it is incomplete.

[0027] θ k For R k The initial rule weights, satisfying 0 < θ k ≤1, δ={δ i Let {i = 1, ..., T} be the relative weights of each preceding attribute in the rule, satisfying 0 ≤ δ i ≤1;

[0028] All standardized input data pairs and their corresponding extended confidence rules together constitute the extended confidence rule base R, denoted as R = {R...} k , k = 1, ..., L}, where L is the number of rules.

[0029] Preferably, S4 includes:

[0030] S41. Treat each rule in the extended confidence rule base as a corresponding data point, calculate the similarity between the belief distribution structures of the preceding attributes of the rules, and use this as the distance between the corresponding data points. The formula for calculating the similarity is as follows:

[0031]

[0032]

[0033]

[0034]

[0035] Among them, dis(R) k R k′ ) represents the k-th rule R k and the k′th rule R k′ The distance between them, the magnitude of which is used to characterize the similarity between them, k≠k′; for and Individual matching degree between them; Δα i，m Indicates in R k and R k′ In the middle, use A i，m Description U i Belief level and The similarity, Δα i，n In the same way; u(A i，m ) represents A i，m The utility, u(A) i，n Similarly;

[0036] Based on the similarity scores, construct a fully connected matrix:

[0037]

[0038] Where FM is a fully connected matrix;

[0039] S42. Calculate the degree matrix and Laplacian matrix based on the fully connected matrix. The calculation formula is as follows:

[0040] DM = sum(FM)

[0041] LM = DM--FM

[0042] Where DM is the degree matrix; LM is the Laplace matrix;

[0043] S43. Standardize the Laplace matrix;

[0044] LM* = DM (-1 / 2) (LM)DM (-1 / 2)

[0045] S44. Calculate the eigenvalues ​​and eigenvectors of the standardized Laplacian matrix;

[0046] S45. Perform K-means clustering on the eigenvectors of the Laplacian matrix;

[0047] S46. Output the clustering results, including the following:

[0048] Each rule belongs to a cluster labeled = {label} k Given the clusters {k = 1, ..., L}, with M clusters, the extended confidence rule base R is divided into M sub-rule bases R. m , m=1,...,M.

[0049] Preferably, in S5:

[0050] Transform the vehicle data X′ of the user to be analyzed into a belief distribution structure, and calculate the similarity of the belief distribution structure between it and the preceding attributes of each rule in the extended belief rule base;

[0051]

[0052] Where, dis(X′,R) k ) represents the belief distribution structure corresponding to X′ and R. k The distance between the belief distribution structures of the preceding attributes, the magnitude of which is used to characterize the similarity between the two; for E(A′ i )and Individual matching degree between them; E(A′) i ) represents the belief distribution structure corresponding to X′;

[0053] Let R be the rule with the highest similarity. max and R max The sub-rule library in which it is located serves as the alternative activated sub-rule library.

[0054] Preferably, in step S6, the ER method is used to obtain the level of poor driving behavior of the user to be analyzed, including:

[0055] Standardize the evaluation index values ​​corresponding to the vehicle data X′ of the user to be analyzed, obtain the input data A′ of the antecedent of the corresponding rule, and calculate the relationship between A′ and the k0th candidate activation rule. Individual matching degree of the preceding attribute

[0056] Get Activation weights:

[0057]

[0058]

[0059] in, The initial weight of the k0th rule in the candidate activation sub-rule library after the update; L0 is the number of candidate activation rules; The relative weight of the i-th preceding attribute after standardization; For the k0th alternative activation rule The activation weights represent The degree of activation by A′, and satisfying like Then A′ is not activated. Otherwise A′ is activated

[0060] The ER method is used to aggregate the consequents of the candidate activation rules;

[0061]

[0062]

[0063] in, Input data for the successors of all alternative activation rules; Belief level for the outcome attribute of all candidate activation rules; for The belief level of the n0th outcome attribute;

[0064] Obtain the level of poor driving behavior of the user to be analyzed;

[0065]

[0066] Wherein, D′ is the classification result of the extended confidence rule reasoning on A′, which serves as the level of the user’s bad driving behavior to be analyzed.

[0067] Preferably, the updated initial rule weight of the k0th rule in the candidate activation sub-rule base. The acquisition process is as follows:

[0068] Select two rules from the sub-rule base Rm. and k≠k′,R m The CPC has L m Rules; among which:

[0069] for if Then D = {(D1, β)} 1，k ), ..., (D N ,β N，k )};

[0070] for If U1 is Then D = {(D1, β)} 1，k′ ), ..., (D N ,β N，k′ )};

[0071] in, They are respectively and Middle U i The belief distribution structure is represented as and {(D1,β 1，k ), ..., (D N ,β N，k )}、{(D1,β 1，k′ ), ..., (D N ,β N，k′ )} are respectively and The belief distribution structure of D;

[0072] Obtain each and Similarity of belief distribution structure among the preceding attributes and the similarity of belief distribution structures among outcome attributes

[0073]

[0074]

[0075] in, for It is a rule The preceding attribute U i The distribution structure of beliefs for It is a rule The preceding attribute U i The distribution structure of beliefs; D k and D k′ They are respectively and The belief distribution of outcome attribute D, i.e., {(D1, β} 1，k ), ..., (D N ,β N，k )} and {(D1, β 1，k′ ), ..., (D N ,β N，k′ )}.

[0076] For rules and In the preceding attribute U i Individual matching degree on S(E(D) k ), E(D k′ )) is a rule and The degree of individual matching on the resulting attribute D;

[0077] calculate and degree of consistency And the degree of inconsistency Incons(k);

[0078]

[0079]

[0080] The degree of inconsistency ξ of the extended confidence rule base is calculated based on the degree of consistency and inconsistency among rules. Incons ;

[0081]

[0082] Then the sub-rule base R m The initial weight of each rule can be updated as follows:

[0083]

[0084] Order rule base R m As a candidate activation sub-rule base, and let k = k0, then obtain the updated initial rule weight of the k0th rule in the candidate activation sub-rule base.

[0085] A poor driving behavior rating system based on improved extended confidence rule reasoning includes:

[0086] The acquisition module is used to acquire historical vehicle data and the level of bad driving behavior of several users, and to acquire evaluation indicators of bad driving behavior based on the historical vehicle data.

[0087] The module is used to standardize the evaluation index values ​​and bad driving behavior levels for each user, obtain the input data of the antecedent and consequent of the rule respectively, and form corresponding standardized input data pairs; the standardized input data pairs corresponding to all users are combined to form a standardized dataset.

[0088] The generation module is used to obtain the antecedent and consequent of the extended confidence rule for each standardized input data pair, as the belief distribution structure of the antecedent attribute and the result attribute, and generate the corresponding extended confidence rule; the extended confidence rules corresponding to all standardized input data pairs are used to form the extended confidence rule library.

[0089] The clustering module is used to cluster the extended confidence rule base into several sub-rule bases based on the belief distribution structure of the preceding attributes of each extended confidence rule, using the SC clustering algorithm.

[0090] The activation module is used to convert the vehicle data of the user to be analyzed into a belief distribution structure, a belief distribution structure between the preceding attributes, obtain candidate activation rules, and use the sub-rule library where the candidate activation rules are located as the candidate activation sub-rule library.

[0091] The rating module is used to obtain the level of bad driving behavior of the user to be analyzed by using the ER method based on the alternative activation sub-rule library and its updated initial rule weights.

[0092] A storage medium storing a computer program for rating poor driving behavior based on improved extended confidence rule reasoning, wherein the computer program causes a computer to perform the poor driving behavior rating method as described above.

[0093] An electronic device, comprising:

[0094] One or more processors;

[0095] Memory; and

[0096] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing the poor driving behavior rating method as described above.

[0097] (III) Beneficial Effects

[0098] This invention provides a method, system, storage medium, and electronic device for rating poor driving behavior based on improved extended confidence rule reasoning. Compared with existing technologies, it has the following advantages:

[0099] This invention introduces the Extended Belief Rule Inference (EBRB) method, utilizing a data-driven decision-making model. This effectively solves the combinatorial explosion problem of the rule base and improves the system's rationality and interpretability. It also introduces the SC clustering algorithm, proposing a new method for determining activation rules. This effectively reduces the number of rule accesses, improving system efficiency, and enhances the consistency among candidate activation rules. Furthermore, using the aforementioned activation rule determination method, an SC_EBRB-based user poor driving behavior rating model is constructed, improving the objectivity and accuracy of poor driving behavior evaluation. Attached Figure Description

[0100] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, 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.

[0101] Figure 1 This is a block diagram illustrating a method for rating undesirable driving behavior based on improved extended confidence rule reasoning, as provided in an embodiment of the present invention. Detailed Implementation

[0102] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0103] This application provides a method, system, storage medium, and electronic device for rating poor driving behavior based on improved extended confidence rule reasoning, thereby solving the technical problem of unreasonable driving behavior assessment models.

[0104] The technical solution in this application is to solve the above-mentioned technical problems, and the general idea is as follows:

[0105] The embodiments of the present invention take into account the importance of adopting scientific and reasonable methods to evaluate users' poor driving behavior. On the one hand, it can provide targeted suggestions for improving drivers' poor driving behavior, reducing drivers' driving risks and reducing the occurrence of traffic accidents. On the other hand, it can provide important references for driver training design and personalized car insurance customization.

[0106] The existing technology is an extended confidence rule inference model. This model extends the traditional rule base by embedding the belief distribution into the antecedent terms of the rules, effectively coordinating the relationship between input and antecedent variables, and providing a more flexible way to integrate mixed input information. Simultaneously, the system can automatically generate extended confidence rules from given sample input-output data pairs, without involving complex rule generation mechanisms, and effectively solving the combinatorial explosion problem of the rule base.

[0107] However, in practical applications, the EBRB system still has some shortcomings:

[0108] 1. Traditional extended confidence rules are generated directly from sample data, resulting in an excessively large rule base and a large number of inefficient rules. Furthermore, the activation rule determination process of the traditional EBRB method requires accessing all extended confidence rules, and the large rule base leads to high computational costs and low operating efficiency.

[0109] 2. The identified alternative activation rules are inconsistent. For the same input data, multiple rules with different results may be activated. When processing high-dimensional data or data with noise, the inconsistency will be more serious, thus affecting the classification accuracy of the system.

[0110] In this embodiment of the invention, the Extended Belief Rule Inference (EBRB) method is introduced, utilizing a data-driven decision-making model. This effectively solves the combinatorial explosion problem of the rule base and improves the rationality and interpretability of the system. The SC clustering algorithm is introduced, proposing a new method for determining activation rules. This effectively reduces the number of rule accesses, improves system operating efficiency, and enhances the consistency among candidate activation rules. Furthermore, using the aforementioned activation rule determination method, a user's poor driving behavior rating model based on SC EBRB is constructed, improving the objectivity and accuracy of poor driving behavior evaluation.

[0111] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0112] Example:

[0113] like Figure 1 As shown, this embodiment of the invention provides a method for rating poor driving behavior based on improved extended confidence rule reasoning, including:

[0114] S1. Obtain historical vehicle data and the level of bad driving behavior of several users, and obtain evaluation indicators of bad driving behavior based on the historical vehicle data;

[0115] S2. Standardize the evaluation index values ​​and bad driving behavior levels for each user, obtain the input data of the antecedent and consequent of the rule respectively, and form corresponding standardized input data pairs; combine the standardized input data pairs corresponding to all users to form a standardized dataset.

[0116] S3. Based on each standardized input data pair, obtain the belief distribution structure of the preceding and resulting attributes of the extended confidence rule, and generate the corresponding extended confidence rule; the extended confidence rules corresponding to all standardized input data pairs are used to form an extended confidence rule library.

[0117] S4. Based on the belief distribution structure of the preceding attributes of each extended confidence rule, the extended confidence rule base is clustered into several sub-rule bases using the SC clustering algorithm.

[0118] S5. Convert the vehicle data of the user to be analyzed into a belief distribution structure, calculate the similarity of the belief distribution structure between the preceding attributes of each rule in the extended belief rule base, obtain candidate activation rules, and use the sub-rule base where the candidate activation rules are located as the candidate activation sub-rule base.

[0119] S6. Based on the alternative activation sub-rule base and its updated initial rule weights, the ER method is used to obtain the level of bad driving behavior of the user to be analyzed.

[0120] In this embodiment of the invention, the Extended Belief Rule Inference (EBRB) method is introduced, utilizing a data-driven decision-making model. This effectively solves the combinatorial explosion problem of the rule base and improves the rationality and interpretability of the system. The SC clustering algorithm is also introduced, proposing a new method for determining activation rules. This effectively reduces the number of rule accesses, improving system operating efficiency, and also enhances the consistency among candidate activation rules. Furthermore, using the aforementioned activation rule determination method, a user's poor driving behavior rating model based on SC_EBRB is constructed, improving the objectivity and accuracy of poor driving behavior evaluation.

[0121] The following will detail each step of the above solution:

[0122] In step S1, historical vehicle data and the level of bad driving behavior of several users are obtained, and evaluation indicators of bad driving behavior are obtained based on the historical vehicle data.

[0123] To fully utilize vehicle-to-everything (V2X) and sensor data, this step first requires combining expert knowledge to identify data items in the collected data that reflect driver behavior characteristics. These characteristic data items will then be used as evaluation indicators for the level of poor driving behavior, serving as the antecedent attributes U = {U} in the subsequent step S3 for constructing the extended confidence rule base. i Let i = 1, ..., T}. Specifically, in this embodiment of the invention, five evaluation indicators are selected as the preceding attributes. Among them, U1 is rapid acceleration, U2 is sudden braking, U3 is sharp turning, U4 is long idling, and U5 is malfunctioning driving, and the user's poor driving behavior level is used as the result attribute D.

[0124] In step S2, the evaluation index values ​​and bad driving behavior levels of each user are standardized, and the input data of the antecedent and consequent of the rule are obtained respectively, forming corresponding standardized input data pairs; the standardized input data pairs corresponding to all users are used to form a standardized dataset.

[0125] After determining the evaluation indicators, this step requires further processing of sensor data to construct an extended confidence rule base. Based on relevant theoretical knowledge, the calculation methods for the evaluation indicators and the threshold values ​​for each indicator level are determined. The user's value on each evaluation indicator is then calculated as the initial input data for the rule antecedent. In this embodiment of the invention, to more intuitively and clearly demonstrate the user's values ​​on each indicator, expert opinions are incorporated to set the indicator level thresholds, transforming the initial input data into standardized input data—that is, converting specific indicator values ​​into indicator levels.

[0126] Specifically, standardized input data can be represented as in Indicates the driver's frequency of rapid acceleration. Indicates the frequency level of emergency braking by the driver. This indicates the driver's frequency level of sharp turns. Indicates the frequency level of long idling by the driver. This indicates the frequency level of driver malfunctions. Similarly, the user's poor driving behavior level is D. * .

[0127] A corresponding to each user * and D * A standardized input data pair is formed, and all standardized input data pairs constitute a standardized input dataset of number L, denoted as X = {x}. i , i = 1, ..., L}.

[0128] In step S3, based on each standardized input data pair, the belief distribution structure of the preceding and resulting attributes of the extended confidence rule is obtained, and the corresponding extended confidence rule is generated; the extended confidence rules corresponding to all standardized input data pairs together constitute the extended confidence rule library, including:

[0129] The input data of the antecedent of the rule in the standardized input data pair Belief distribution structure transformed into the preceding attribute

[0130]

[0131]

[0132] α i，j′ =0; j′=1,...,J i j′≠j, j+1

[0133] Among them, A i，j For the i-th preceding attribute U i The utility value of the j-th reference level, which changes as the parameters are learned; α i，j For input data Transform to U i The matching degree at the j-th reference level, i.e., U i The degree of belief at the j-th reference level; J i For U i The number of reference levels; T is the number of the preceding attributes.

[0134] Similarly, the input data D of the consequent of the rule in the standardized input data pair * Belief distribution structure transformed into outcome attributes

[0135] Let the k-th rule R k The representation is: if U is {A, α} k}, then V is {D, β} k}

[0136] In the antecedent of the rule, U = {U i} represents the preceding attribute of the rule; A = {A i} represents the reference level for the preceding attribute of the rule. For U i The utility value of the reference level; For in R k The belief level of the reference level for all preceding attributes. For R k Chinese Description U i The degree of belief, and satisfying For R k The distribution structure of beliefs about the preceding attributes E(A) k ), then U i The belief distribution structure is represented as

[0137] In the consequent of a rule, V is the result attribute of the rule; D = {D n , n = 1, ..., N} are the reference levels of the result attributes of the rule, and there are N reference levels in total for the result attributes; For R k The belief level of the outcome attribute, i.e., using the reference value D = {D n The degree of belief in the outcome attribute, and satisfying the following conditions: like This indicates that the result of the k-th rule is complete; otherwise, it is incomplete.

[0138] θ k For R k The initial rule weights, satisfying 0 < θ k ≤1, δ={δ i Let {i = 1, ..., T} be the relative weights of each preceding attribute in the rule, satisfying 0 ≤ δ i ≤1;

[0139] All standardized input data pairs and their corresponding extended confidence rules together constitute the extended confidence rule base R, denoted as R = {R...} k , k = 1, ..., L}, where L is the number of rules.

[0140] In step S4, based on the belief distribution structure of the preceding attributes of each extended confidence rule, the SC clustering algorithm is used to cluster the extended confidence rule base into several sub-rule bases.

[0141] After obtaining the extended confidence rule base R, this step treats each rule in the extended confidence rule base as a corresponding data point, and uses the belief distribution structure of the preceding attributes of each rule to perform SC clustering (SC stands for spectral clustering). The specific clustering steps are as follows:

[0142] S41. Treat each rule in the extended confidence rule base as a corresponding data point, calculate the similarity between the belief distribution structures of the preceding attributes of the rules, and use this as the distance between the corresponding data points. The formula for calculating the similarity is as follows:

[0143]

[0144]

[0145]

[0146]

[0147] Among them, dis(R) k R k′ ) represents the k-th rule R k and the k′th rule R k′ The distance between them, the magnitude of which is used to characterize the similarity between them, k≠k′; for and Individual matching degree between them; Δα i，m Indicates in R k and R k′ In the middle, use A i，m Description U i Belief level and The similarity, Δα i，n In the same way; u(A i，m ) represents A i，m The utility, u(A) i，n Similarly;

[0148] Based on the similarity scores, construct a fully connected matrix:

[0149]

[0150] Where FM is a fully connected matrix;

[0151] S42. Calculate the degree matrix and Laplacian matrix based on the fully connected matrix. The calculation formula is as follows:

[0152] DM = sum(FM)

[0153] LM = DM-FM

[0154] Where DM is the degree matrix; LM is the Laplace matrix;

[0155] S43. Standardize the Laplace matrix;

[0156] LM* = DM (-1 / 2) (LM)DM (-1 / 2)

[0157] S44. Calculate the eigenvalues ​​and eigenvectors of the standardized Laplacian matrix;

[0158] S45. Perform K-means clustering on the eigenvectors of the Laplacian matrix;

[0159] S46. Output the clustering results, including the following:

[0160] Each rule belongs to a cluster labeled = {label} k Given the clusters {k = 1, ..., L}, with M clusters, the extended confidence rule base R is divided into M sub-rule bases R. m , m=1,...,M.

[0161] In step S5, the vehicle data of the user to be analyzed is transformed into a belief distribution structure. The similarity of the belief distribution structure between this structure and the preceding attributes of each rule in the extended belief rule base is calculated to obtain candidate activation rules. The sub-rule base containing the candidate activation rules is then used as the candidate activation sub-rule base. This includes:

[0162] Transform the vehicle data X′ of the user to be analyzed into a belief distribution structure, and calculate the similarity of the belief distribution structure between it and the preceding attributes of each rule in the extended belief rule base;

[0163]

[0164] Where, dis(X′,R) k ) represents the belief distribution structure corresponding to X′ and R. k The distance between the belief distribution structures of the preceding attributes, the magnitude of which is used to characterize the similarity between the two; for E(A′ i )and Individual matching degree between them; E(A′) i ) represents the belief distribution structure corresponding to X′;

[0165] Let R be the rule with the highest similarity. max and R max The sub-rule library in which it is located serves as the alternative activated sub-rule library.

[0166] In step S6, the ER method is used to obtain the level of bad driving behavior of the user to be analyzed based on the alternative activation sub-rule base and its updated initial rule weights.

[0167] Specifically, this step uses the ER method to obtain the level of poor driving behavior of the user to be analyzed, including:

[0168] Standardize the evaluation index values ​​corresponding to the vehicle data X′ of the user to be analyzed, obtain the input data A′ of the antecedent of the corresponding rule, and calculate the relationship between A′ and the k0th candidate activation rule. Individual matching degree of the preceding attribute

[0169] Get Activation weights:

[0170]

[0171]

[0172] in, The initial weight of the k0th rule in the candidate activation sub-rule library after the update; L0 is the number of candidate activation rules; The relative weight of the i-th preceding attribute after standardization; For the k0th alternative activation rule The activation weights represent The degree of activation by A′, and satisfying like Then A′ is not activated. Otherwise A′ is activated

[0173] Specifically, the updated initial rule weight of the k0th rule in the alternative activation sub-rule base. The acquisition process is as follows:

[0174] Select sub-rule base R m Two rules and k≠k′,R m The CPC has L m Rules; among which:

[0175] for if Then D = {(D1, β)} 1，k ), ..., (D N ,β N，k )};

[0176] for If U1 is Then D = {(D1, β)} 1，k′ ), ..., (D N ,β N，k′ )};

[0177] in, They are respectively and Middle U i The belief distribution structure is represented as and {(D1,β 1，k ), ..., (D N ,β N，k )}、{(D1,β 1，k′ ), ..., (D N ,β N，k′)} are respectively and The belief distribution structure of D;

[0178] Obtain each and Similarity of belief distribution structure among the preceding attributes and the similarity of belief distribution structures among outcome attributes

[0179]

[0180]

[0181] in, for It is a rule The preceding attribute U i The distribution structure of beliefs for It is a rule The preceding attribute U i The distribution structure of beliefs; D k and D k′ They are respectively and The belief distribution of outcome attribute D, i.e., {(D1, β} 1，k ), ..., (D N ,β N，k )} and {(D1, β 1，k′ ), ..., (D N ,β N，k′ )}.

[0182] For rules and In the preceding attribute U i Individual matching degree on S(E(D) k ), E(D k′ )) is a rule and The degree of individual matching on the resulting attribute D;

[0183] calculate and degree of consistency And the degree of inconsistency Incons(k);

[0184]

[0185]

[0186] The degree of inconsistency ξ of the extended confidence rule base is calculated based on the degree of consistency and inconsistency among rules.Incons ;

[0187]

[0188] Then the sub-rule base R m The initial weight of each rule can be updated as follows:

[0189]

[0190] Order rule base R m As a candidate activation sub-rule base, and let k = k0, then obtain the updated initial rule weight of the k0th rule in the candidate activation sub-rule base.

[0191] The ER method is used to aggregate the consequents of the candidate activation rules;

[0192]

[0193]

[0194] in, Input data for the successors of all alternative activation rules; Belief level for the outcome attribute of all candidate activation rules; for The belief level of the n0th outcome attribute;

[0195] Obtain the level of poor driving behavior of the user to be analyzed;

[0196]

[0197] Wherein, D′ is the classification result of the extended confidence rule inference on A′, which serves as the level of the user’s poor driving behavior to be analyzed. This result can then be used to recommend driving behavior to the user or to conduct a comprehensive evaluation of subsequent driving behavior.

[0198] To verify the effectiveness and superiority of the poor driving behavior rating method based on improved extended confidence rule reasoning provided in the embodiments of the present invention, the following example is provided:

[0199] Based on expert opinions and historical data on poor driving behavior evaluation, five evaluation indicators were identified as the antecedent attributes of the rules, and thresholds for each attribute level were determined. Five datasets were randomly generated using Monte Carlo simulation, each containing 3000 pairs, including evaluation levels for the five attributes and levels of poor driving behavior. Explanations of the antecedent attributes and some data are shown in Tables 1 and 2.

[0200] Table 1 Evaluation Indicators for Unhealthy Driving Behaviors

[0201]

[0202]

[0203] Table 2 Partial User Vehicle Simulation Data

[0204] Serial Number <![CDATA[U1]]> <![CDATA[U2]]> <![CDATA[U3]]> <![CDATA[U4]]> <![CDATA[U5]]> D 1 3 5 3 5 2 4 2 3 2 3 1 2 3 3 3 3 3 3 2 3 4 1 1 3 1 5 1 5 5 5 3 1 5 4 6 5 5 3 3 2 4 … 2999 3 4 1 3 2 3 3000 1 2 2 3 2 2

[0205] The first 80% of the data in each group is used as training data to construct an extended confidence rule base, and the last 20% of the data is used as test data. The evaluation index is used as the antecedent attribute of the extended confidence rule. At the same time, combined with expert experience, the weights of each attribute are determined. The attribute weights and extended confidence rule expressions are shown in Figures 3 and 4.

[0206] Table 3 Weights of Antecedent Attributes

[0207]

[0208] Table 4 Extended Confidence Rules for Simulation Data Generation

[0209]

[0210] Based on simulation data, an extended confidence rule base is constructed, and SC clustering is performed on the rule base. Then, test data pairs are input, and alternative activation sub-rule bases are selected. The classification results are obtained through the reasoning process of the extended confidence rule base, and the evaluation level of bad driving behavior is determined.

[0211] The method proposed in this embodiment of the invention is used to evaluate the test data, specifically, the belief level of the test data at each level of the outcome attribute after model inference, as shown in Table 5:

[0212] Table 5 shows the evaluation results of some test data pairs.

[0213] Serial Number <![CDATA[D1]]> <![CDATA[D2]]> <![CDATA[D3]]> <![CDATA[D4]]> <![CDATA[D5]]> <![CDATA[D max ]]> 1 0.595854 0.244752 0.133271 0.015169 0.010954 1 2 0.443325 0.216236 0.2484 0.041821 0.050218 1 3 0.122535 0.249522 0.483341 0.105426 0.039176 3 4 0.641888 0.236331 0.09551 0.007813 0.018458 1 5 0.095249 0.262085 0.480822 0.115937 0.045906 3 6 0.66973 0.217785 0.082329 0.012169 0.017988 1 ... 2999 0.159792 0.178147 0.447839 0.122612 0.09161 3 3000 0.748303 0.100981 0.123154 0.023677 0.003884 1

[0214] Furthermore, to better compare existing methods and the improved method proposed in this embodiment of the invention, a total of 5 sets of simulation data were used to evaluate them using 3 evaluation criteria. The calculation method of the evaluation index is as follows:

[0215] (1) Assessment Accuracy (ACC), which is the accuracy of the rating of poor driving behavior. The accuracy calculation method is as follows:

[0216]

[0217]

[0218]

[0219] Among them, D i This indicates that the i-th test data pair is X.i Initial result attribute level, X represents i The result obtained after model inference is the highest level of belief in the attribute.

[0220] (2) Number of activation rules (Num): The average number of activation rules in a single experiment.

[0221] (3) Running time (Time): The total time required to complete the evaluation of test data for one experiment.

[0222] The results of the comparative experiments are shown in Table 6:

[0223] Table 6 Comparison of Experimental Results

[0224]

[0225] This invention provides a poor driving behavior rating system based on improved extended confidence rule reasoning, comprising:

[0226] The acquisition module is used to acquire historical vehicle data and the level of bad driving behavior of several users, and to acquire evaluation indicators of bad driving behavior based on the historical vehicle data.

[0227] The module is used to standardize the evaluation index values ​​and bad driving behavior levels for each user, obtain the input data of the antecedent and consequent of the rule respectively, and form corresponding standardized input data pairs; the standardized input data pairs corresponding to all users are combined to form a standardized dataset.

[0228] The generation module is used to obtain the belief distribution structure of the antecedent and result attributes of the extended confidence rule for each standardized input data pair, and generate the corresponding extended confidence rule; the extended confidence rules corresponding to all standardized input data pairs are used to form the extended confidence rule library.

[0229] The clustering module is used to cluster the extended confidence rule base into several sub-rule bases based on the belief distribution structure of the preceding attributes of each extended confidence rule, using the SC clustering algorithm.

[0230] The activation module is used to convert the vehicle data of the user to be analyzed into a belief distribution structure, calculate the similarity of the belief distribution structure between the preceding attributes of each rule in the extended belief rule base, obtain candidate activation rules, and use the sub-rule base where the candidate activation rules are located as the candidate activation sub-rule base.

[0231] The rating module is used to obtain the level of bad driving behavior of the user to be analyzed by using the ER method based on the alternative activation sub-rule library and its updated initial rule weights.

[0232] This invention provides a storage medium storing a computer program for rating poor driving behavior based on improved extended confidence rule reasoning, wherein the computer program causes a computer to perform the poor driving behavior rating method as described above.

[0233] This invention provides an electronic device, comprising:

[0234] One or more processors;

[0235] Memory; and

[0236] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing the poor driving behavior rating method as described above.

[0237] It is understood that the bad driving behavior rating system, storage medium and electronic device based on improved extended confidence rule reasoning provided in the embodiments of the present invention correspond to the bad driving behavior rating method based on improved extended confidence rule reasoning provided in the embodiments of the present invention. The explanation, examples and beneficial effects of the relevant contents can be referred to the corresponding parts of the bad driving behavior rating system method, and will not be repeated here.

[0238] In summary, compared with existing technologies, it has the following beneficial effects:

[0239] In this embodiment of the invention, the Extended Belief Rule Inference (EBRB) method is introduced, utilizing a data-driven decision model. This effectively solves the combinatorial explosion problem of the rule base and improves the rationality and interpretability of the system. The SC clustering algorithm is also introduced, proposing a new method for determining activation rules. This effectively reduces the number of rule accesses, improves system operating efficiency, and enhances the consistency among candidate activation rules. Furthermore, using the aforementioned activation rule determination method, a user's poor driving behavior rating model based on SC_EBRB is constructed, improving the objectivity and accuracy of poor driving behavior evaluation.

[0240] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0241] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for rating poor driving behavior based on improved extended confidence rule reasoning, characterized in that, include: S1. Obtain historical vehicle data and the level of bad driving behavior of several users, and obtain evaluation indicators of bad driving behavior based on the historical vehicle data; S2. Standardize the evaluation index values ​​and bad driving behavior levels for each user, obtain the input data of the antecedent and consequent of the rule respectively, and form corresponding standardized input data pairs; combine the standardized input data pairs corresponding to all users to form a standardized dataset. The standardization of each user's evaluation index value and the level of poor driving behavior involves obtaining the input data for the antecedent and consequent of the rule, including: Based on preset level thresholds, the user's evaluation index values ​​and the level of bad driving behavior are converted into corresponding standard levels, and used as input data for the antecedent and consequent of the rule. S3. Based on each standardized input data pair, obtain the belief distribution structure of the preceding and resulting attributes of the extended confidence rule, and generate the corresponding extended confidence rule; the extended confidence rules corresponding to all standardized input data pairs are used to form an extended confidence rule library. S4. Based on the belief distribution structure of the preceding attributes of each extended confidence rule, the extended confidence rule base is clustered into several sub-rule bases using the SC clustering algorithm. S5. Convert the vehicle data of the user to be analyzed into a belief distribution structure, calculate the similarity of the belief distribution structure between the preceding attributes of each rule in the extended belief rule base, obtain candidate activation rules, and use the sub-rule base where the candidate activation rules are located as the candidate activation sub-rule base. S6. Based on the alternative activation sub-rule base and its updated initial rule weights, the ER method is used to obtain the level of bad driving behavior of the user to be analyzed.

2. The method for rating poor driving behavior as described in claim 1, characterized in that, In S3: The input data of the antecedent of the rule in the standardized input data pair Belief distribution structure transformed into the preceding attribute a i,j′ = 0; j' = 1,..., J i j' ≠ j, j + 1 Among them, A i,j For the i-th preceding attribute U i The utility value of the j-th reference level, which changes as the parameters are learned; α i,j For input data Transform to U i The matching degree at the j-th reference level, i.e., U i The degree of belief at the j-th reference level; J i For U i The number of reference levels; T is the number of preceding attributes; Similarly, the input data D of the consequent of the rule in the standardized input data pair * Belief distribution structure transformed into outcome attributes The kth rule R k is represented as: if U is {A, a k}, then V is {D, b k}. In the antecedent of the rule, U = {U i } represents the preceding attribute of the rule; A = {A i } represents the reference level for the preceding attribute of the rule. For U i The utility value of the reference level; For in R k The belief level of the reference level for all preceding attributes. For R k Chinese Description U i The degree of belief, and satisfying For R k The distribution structure of beliefs about the preceding attributes E(A) k ), then U i The belief distribution structure is represented as In the consequent of a rule, V is the result attribute of the rule; D = {D n ,n=1,…,N} represents the reference levels of the result attributes of the rule, and there are N reference levels in total for the result attributes; For R k The belief level of the outcome attribute, i.e., using the reference value D = {D n The degree of belief in the outcome attribute, and satisfying the following conditions: like This indicates that the result of the k-th rule is complete; otherwise, it is incomplete. θ k is the initial rule weight for R k , and satisfies 0 < θ k ≤ 1, δ = {δ i , i = 1, …, T} is the relative weight of each antecedent attribute in the rule, and satisfies 0 ≤ δ i ≤ 1; All standardized input data pairs and their corresponding extended confidence rules together constitute the extended confidence rule base R, denoted as R = {R...} k Let k = 1, ..., L, where L is the number of rules.

3. The method for rating poor driving behavior as described in claim 2, characterized in that, S4 includes: S41. Treat each rule in the extended confidence rule base as a corresponding data point, calculate the similarity between the belief distribution structures of the preceding attributes of the rules, and use this as the distance between the corresponding data points. The formula for calculating the similarity is as follows: Among them, dis(R) k ,R k′ ) represents the k-th rule R k and the k′th rule R k′ The distance between them, the magnitude of which is used to characterize the similarity between them, k≠k′; for and Individual matching degree between them; Δα i,m Indicates in R k and R k′ In the middle, use A i,m Description U i Belief level and The similarity, Δα i,n In the same way; u(A i,m ) represents A i,m The utility, u(A) i,n Similarly; Based on the similarity scores, construct a fully connected matrix: Where FM is a fully connected matrix; S42. Calculate the degree matrix and Laplacian matrix based on the fully connected matrix. The calculation formula is as follows: DM = sum(FM) LM = DM-FM Where DM is the degree matrix; LM is the Laplace matrix; S43. Standardize the Laplace matrix; LM*=DM (-1 / 2) (LM)DM (-1 / 2) S44. Calculate the eigenvalues ​​and eigenvectors of the standardized Laplacian matrix; S45. Perform K-means clustering on the eigenvectors of the Laplacian matrix; S46. Output the clustering results, including the following: Each rule belongs to a cluster label = {label} k Given the clusters {k=1,...,L}, with M clusters, the extended confidence rule base R is divided into M sub-rule bases Ri. m ,m=1,....,M.

4. The method for rating poor driving behavior as described in claim 3, characterized in that, In S5: Transform the vehicle data X′ of the user to be analyzed into a belief distribution structure, and calculate the similarity of the belief distribution structure between it and the preceding attributes of each rule in the extended belief rule base; Where, dis(X′,R) k ) represents the belief distribution structure corresponding to X′ and R k The distance between the belief distribution structures of the preceding attributes, the magnitude of which is used to characterize the similarity between the two; for E(A′ i )and Individual matching degree between them; E(A′) i ) represents the belief distribution structure corresponding to X′; Let R be the rule with the highest similarity. max and R max The sub-rule library in which it is located serves as the alternative activated sub-rule library.

5. The method for rating poor driving behavior as described in claim 3, characterized in that, The ER method is used in S6 to obtain the level of poor driving behavior of the user to be analyzed, including: Standardize the evaluation index values ​​corresponding to the vehicle data X′ of the user to be analyzed, obtain the input data A′ of the antecedent of the corresponding rule, and calculate the relationship between A′ and the k0th candidate activation rule. Individual matching degree of the preceding attribute Get Activation weights: in, The initial weight of the k0th rule in the candidate activation sub-rule library after the update; L0 is the number of candidate activation rules; The relative weight of the i-th preceding attribute after standardization; For the k0th alternative activation rule The activation weights represent The degree of activation by A′, and satisfying like Then A′ is not activated. Otherwise A′ is activated The ER method is used to aggregate the consequents of the candidate activation rules; in, Input data for the successors of all alternative activation rules; The belief level of the outcome attribute for all candidate activation rules; for The belief level of the n0th outcome attribute; Obtain the level of poor driving behavior of the user to be analyzed; Wherein, D′ is the classification result of the extended confidence rule reasoning on A′, which serves as the level of the user’s bad driving behavior to be analyzed.

6. The method for rating poor driving behavior as described in claim 5, characterized in that, The updated initial rule weight of the k0th rule in the alternative activation sub-rule base The acquisition process is as follows: Select sub-rule base R m Two rules and k≠k′,R m The CPC has L m Rules; among which: for if Then D = {(D1,β)} 1,k ),…,(D N ,β N,k )}; for If u1 is but in, They are respectively and Middle U i The belief distribution structure is represented as and {(D1,β 1,k ),…,(D N ,β N,k )}、{(D1,β 1,k′ ),…,(D N ,β N,k′ )} are respectively and The belief distribution structure of D; Obtain each and Similarity of belief distribution structure among the preceding attributes and the similarity of belief distribution structures among outcome attributes in, for It is a rule The preceding attribute U i The distribution structure of beliefs for It is a rule The preceding attribute U i The distribution structure of beliefs; D k and D k′ They are respectively and The belief distribution of outcome attribute D, i.e., {(D1,β} 1,k ),...,(D N ,β N,k )} and {(D1,β 1,k′ ),...,(D N ,β N,k′ )}. For rules and In the preceding attribute U i Individual matching degree on S(E(D) k ),E(D k′ )) is a rule and The degree of individual matching on the resulting attribute D; calculate and degree of consistency And the degree of inconsistency Incons(k); The degree of inconsistency ξ of the extended confidence rule base is calculated based on the degree of consistency and inconsistency among rules. Incons ; Then the sub-rule base R m The initial weight of each rule can be updated as follows: Order rule base R m As a candidate activation sub-rule base, and let k = k0, then obtain the updated initial rule weight of the k0th rule in the candidate activation sub-rule base.

7. A rating system for poor driving behavior based on improved extended confidence rule reasoning, characterized in that, include: The acquisition module is used to acquire historical vehicle data and the level of bad driving behavior of several users, and to acquire evaluation indicators of bad driving behavior based on the historical vehicle data. The module is used to standardize the evaluation index values ​​and bad driving behavior levels for each user, obtain the input data of the antecedent and consequent of the rule respectively, and form corresponding standardized input data pairs; the standardized input data pairs corresponding to all users are combined to form a standardized dataset. The standardization of each user's evaluation index value and the level of poor driving behavior involves obtaining the input data for the antecedent and consequent of the rule, including: Based on preset level thresholds, the user's evaluation index values ​​and the level of bad driving behavior are converted into corresponding standard levels, and used as input data for the antecedent and consequent of the rule. The generation module is used to obtain the belief distribution structure of the antecedent and result attributes of the extended confidence rule for each standardized input data pair, and generate the corresponding extended confidence rule; the extended confidence rules corresponding to all standardized input data pairs are used to form the extended confidence rule library. The clustering module is used to cluster the extended confidence rule base into several sub-rule bases based on the belief distribution structure of the preceding attributes of each extended confidence rule, using the SC clustering algorithm. The activation module is used to convert the vehicle data of the user to be analyzed into a belief distribution structure, calculate the similarity of the belief distribution structure between the preceding attributes of each rule in the extended belief rule base, obtain candidate activation rules, and use the sub-rule base where the candidate activation rules are located as the candidate activation sub-rule base. The rating module is used to obtain the level of bad driving behavior of the user to be analyzed by using the ER method based on the alternative activation sub-rule library and its updated initial rule weights.

8. A storage medium, characterized in that, It stores a computer program for rating poor driving behavior based on improved extended confidence rule reasoning, wherein the computer program causes the computer to perform the poor driving behavior rating method as described in any one of claims 1 to 6.

9. An electronic device, characterized in that, include: One or more processors; Memory; as well as One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing the poor driving behavior rating method as described in any one of claims 1 to 6.

Images