Vehicle operation data driven carrier vehicle portrait evaluation method and system

Abstract

The invention relates to the technical field of vehicle portrait evaluation, in particular to a vehicle operation data driven carrier vehicle portrait evaluation method and system, and the method comprises the following steps: collecting multi-modal data in a vehicle operation process, and carrying out the automatic sensing and marking of abnormal behaviors in the data through a self-supervised learning algorithm; carrying out weighted fusion, generating a comprehensive feature vector F, constructing a dynamic balance model which simultaneously considers short-term behavior fluctuation and long-term running trend of the vehicle, and simultaneously capturing short-term risk events and carrying out long-term stability analysis; and carrying out portrait generation on the operation state of the carrier vehicle, and providing corresponding intervention decision suggestions for the identified abnormal behaviors. According to the method, high reliability of abnormal behavior marking in the vehicle portrait is ensured, the long-term stability trend of the vehicle is comprehensively analyzed, the short-term risk and the long-term running state of the vehicle can be considered at the same time, and a scientific basis is provided for safety scoring and operation stability analysis of the vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle image assessment, and in particular to a method and system for transport vehicle image assessment driven by vehicle operation data. Background Art

[0002] With the rapid development of the intelligent transportation and logistics industries, the operating status and safety assessment of carrier vehicles during transportation have gradually become the focus of managers. At present, traditional vehicle assessment methods mainly rely on static data of vehicles and single-dimensional operating parameters, such as vehicle model, driver experience, vehicle maintenance records, etc. However, these methods cannot effectively capture the dynamic behavior of vehicles during actual operation, especially under complex road conditions and changeable driving conditions, and it is difficult to accurately identify potential safety hazards.

[0003] In the prior art, although some evaluation methods based on vehicle operation data have been introduced, these methods usually only analyze data of a single mode and lack comprehensive integration and processing of multimodal data. For example, a single data source such as commonly used speed cannot fully reflect the comprehensive operation status of the vehicle, resulting in large errors and omissions in abnormal behavior detection. In addition, traditional time series analysis methods mostly focus on short-term predictions and it is difficult to take into account both the short-term behavior fluctuations and long-term operation trends of the vehicle at the same time, which limits the effective evaluation of the overall safety and stability of the vehicle.

[0004] In this context, how to conduct a comprehensive and dynamic assessment of the operating status of carrier vehicles by effectively integrating multimodal data has become an urgent problem to be solved in the current technical field. Summary of the invention

[0005] The present invention provides a transport vehicle portrait evaluation method and system driven by vehicle operation data.

[0006] The vehicle operation data-driven carrier vehicle portrait assessment method includes the following steps: S1, abnormal behavior perception and labeling: The multi-modal data of the vehicle operation process is collected through the multi-sensor network deployed on the vehicle. The multi-modal data includes braking force, steering wheel angle, speed change rate, and combined with historical data and the current operation status of the vehicle, a self-supervised learning algorithm is used to automatically perceive and label abnormal behaviors in the data. The abnormal behaviors include sudden emergency braking, abnormal steering, speeding, and frequent lane changes; S2, adaptive fusion of multimodal data: for the annotated multimodal data, an adaptive data fusion algorithm is used for weighted fusion to generate a comprehensive feature vector F. Based on the credibility, relevance and correlation of different modal data with abnormal behavior, the fusion weight of each modal data is dynamically adjusted to ensure that the weight of important data is appropriately increased during the fusion process, thereby improving the model's sensitivity to abnormal behavior and evaluation accuracy; S3, Construction of dynamic balance model of short-term fluctuation and long-term trend: Based on the integrated feature vector F after fusion, a dynamic balance model is constructed that takes into account both the short-term behavior fluctuation and long-term operation trend of the vehicle. The dynamic balance model is based on the time series analysis algorithm, and the short-term abnormal behavior and long-term operation trend of the vehicle are hierarchically modeled, so as to capture both short-term risk events and long-term stability analysis. S4, portrait generation and dynamic update: Based on the dynamic balance model, a portrait of the operating status of the carrier vehicle is generated, and the portrait result includes the vehicle safety score and operation stability analysis; S5, abnormal behavior identification and intervention decision: By analyzing the generated vehicle portrait, potential abnormal behavior is identified, and combined with the long-term trend analysis in the dynamic balance model, it is determined whether the abnormal behavior has a persistent risk; for the identified abnormal behavior, corresponding intervention decision suggestions are provided, including adjusting the driving route, optimizing the driving strategy or arranging vehicle maintenance.

[0007] Optionally, the collection of multimodal data in S1 includes: Braking force data collection: Through the brake pressure sensor integrated in the brake system, the pressure applied by the brake system is monitored in real time to capture the braking force of the vehicle under different driving conditions, including emergency braking and normal braking; Steering wheel steering angle data collection: Use the angle sensor on the steering shaft to collect the steering wheel steering angle changes in real time and capture the steering wheel operation during the vehicle steering process, including steering angle, steering speed and steering wheel return status; Speed ​​change rate data collection: The wheel speed sensor integrated in the vehicle collects the vehicle's speed data in real time and calculates the rate of change of the vehicle's speed to identify the dynamic characteristics of the vehicle during acceleration, deceleration and constant speed driving.

[0008] Optionally, the automatic perception and labeling in S1 specifically includes: S11, comparative analysis of historical data and current data: comparing and analyzing the current operation data of the vehicle with the historical data, the current operation data includes the braking force B, the steering angle of the steering wheel A, and the speed change rate V, and the historical data includes the normal behavior mode and abnormal behavior record of the vehicle under the same or similar driving conditions; S12, training of self-supervised learning model: by using a large amount of labeled historical data to train the self-supervised learning model, the characteristic distribution of the vehicle under normal driving and abnormal behavior states is learned, including the characteristic patterns of sudden braking, abnormal steering, speeding and frequent lane changing behaviors; S13, abnormal behavior detection and labeling: During vehicle operation, the self-supervised learning model analyzes the current data in real time. When it detects that the current data matches the learned abnormal behavior characteristics, it automatically labels it as abnormal behavior. Abnormal behaviors include sudden braking, abnormal steering, and frequent lane changes.

[0009] Optionally, the S2 specifically includes: S21, Credibility calculation of modal data: Calculate the credibility of each modal data based on the annotated abnormal behavior data ,in Indicates specific modal data, including braking force , Steering wheel angle , speed change rate ,The credibility calculation is based on the accuracy and frequency of abnormal behaviors detected in historical data by each modality; S22, correlation calculation of modal data: by analyzing the correlation between modal data, including their correlation with abnormal behavior, to calculate the correlation weight ,in and Representing different modal data, the correlation weights were determined by calculating the Pearson correlation coefficient; S23, Dynamic Adjustment of Fusion Weights: Combining Credibility and relevance weight , dynamically adjust the fusion weight of each modality data , the fusion weight is expressed as: ,in, Represents modal data The final fusion weight of . The larger the weight, the greater the proportion of the modality data in the fusion process. S24, weighted fusion: using dynamically adjusted fusion weights Perform weighted fusion on each modal data to generate a comprehensive feature vector , used for subsequent vehicle portrait evaluation: .

[0010] Optionally, the construction of the dynamic balance model of short-term fluctuations and long-term trends in S3 specifically includes: S31, organize the comprehensive feature vector F in the form of a time series to form a time series data set ,in Indicates the time period of observation; S32, Hierarchical Modeling: Short-term behavior fluctuation modeling: The autoregressive moving average model in the time series analysis algorithm is used to model the short-term fluctuations in the time series data and capture the abnormal behavior of the vehicle in a short period of time, including sudden braking and abnormal steering, which is expressed as: ,in, is the comprehensive feature vector at the current moment, is the comprehensive feature vector of the previous moment, and They represent the autoregressive and moving average parts in the autoregressive moving average model, is white noise, is the white noise term at the previous moment; Long-term operation trend modeling: A long-term time series prediction model is used to extract the long-term trend of vehicle operation status and capture the stability and potential risks of vehicles over a long period of time, expressed as: , where LSTM is a long-term time series prediction model (long short-term memory network), Represents the long-term trend feature vector. LSTM uses the input time series data Extracting long-term trends; S34, construction of dynamic balance model: Integrate the results of short-term fluctuation model and long-term trend model to form dynamic balance model M. Dynamic balance model M simultaneously analyzes the short-term behavioral fluctuations and long-term operating trends of the vehicle, and captures short-term risk events and long-term stability characteristics in vehicle portrait assessment.

[0011] Optionally, the dynamic equilibrium model is expressed as: ,in, for The output of the dynamic equilibrium model at time is the short-term behavior fluctuation characteristic vector, is the long-term trend eigenvector, and To adjust the coefficients, ensure that the model can balance short-term and long-term factors.

[0012] Optionally, the dynamic equilibrium model is used in S4 Short-term behavior fluctuation characteristics of output and long-term operating trend characteristics Calculating the vehicle's safety score , which comprehensively considers the abnormal behavior of the vehicle in the short term (such as sudden braking, abnormal steering, etc.) and the stability in long-term operation, and is expressed as: ,in, is the adjustment coefficient of the safety score, Indicates the length of the time window; The S4 is based on the long-term trend characteristics captured in the dynamic equilibrium model Perform vehicle operation stability analysis. Vehicle operation stability analysis identifies potential operational instability by evaluating the consistency and regularity of the vehicle during long-term operation, expressed as: ,in, Represents the operational stability score, where lower values ​​indicate more stable operations.

[0013] Optionally, S4 further includes integrating the scoring results, wherein the scoring results integration includes integrating the calculated security scores and operational stability analysis results Integrate to generate a comprehensive score of vehicle operating status , the integration formula is: ,in, and They are the weight coefficients of the safety score and the operational stability score, respectively, and are used to adjust the impact of the two in the comprehensive score. Used to reverse the stability score so that higher stability (i.e. lower value) contributes more positively to the overall score; Image generation: using the comprehensive score calculated above , generating a profile of the operating status of the carrier vehicle. This profile visualizes the safety and operational stability of the vehicle, helping managers quickly assess the overall operating status of the vehicle, identify potential risk points, and provide a basis for decision-making.

[0014] Optionally, the S5 specifically includes: Abnormal behavior identification: safety scoring based on vehicle operation status profile and operational stability results , identify potential abnormal behaviors, and mark them as potential abnormal behaviors when the safety score is significantly lower than the set threshold or the operation stability results fluctuate greatly; Long-term trend analysis: Identify potential abnormal behaviors and compare them with the long-term trend feature vectors in the dynamic equilibrium model. Conduct comparative analysis to observe changes in long-term trend characteristics and determine whether abnormal behavior is just a short-term fluctuation or a persistent risk; Intervention decision recommendations: Provide corresponding intervention decision recommendations based on the identified abnormal behaviors and their persistent risks.

[0015] The vehicle operation data-driven carrier vehicle portrait evaluation system is used to implement the above-mentioned vehicle operation data-driven carrier vehicle portrait evaluation method, and includes the following modules: Multimodal data acquisition module: collects multimodal data during vehicle operation through a multi-sensor network deployed on the vehicle, including braking force, steering wheel angle, and speed change rate; Abnormal behavior perception and labeling module: Analyzes the collected multimodal data through a self-supervised learning algorithm, automatically perceives and labels abnormal behaviors, including sudden braking, abnormal steering, and speeding, and generates labeled behavior data; Adaptive data fusion module: For the annotated behavior data, an adaptive data fusion algorithm is used to dynamically adjust the fusion weight of each modal data based on the credibility, relevance and correlation of each modal data with abnormal behavior to generate a comprehensive feature vector; Dynamic balance model module: Utilizes the integrated feature vector after fusion to build a dynamic balance model that takes into account both the short-term behavior fluctuations and long-term operation trends of the vehicle, thus realizing hierarchical modeling and evaluation of the vehicle's operating status; Vehicle operation status portrait generation module: Based on the dynamic balance model, it calculates the vehicle's safety score and operational stability analysis results, and generates a vehicle operation status portrait to present the vehicle's overall operation status; Risk analysis and intervention decision module: Analyze the generated vehicle portrait, identify potential abnormal behaviors, and combine the long-term trend analysis in the dynamic balance model to determine whether the abnormal behaviors have persistent risks, and provide decision-making suggestions, including adjusting the driving route, optimizing driving strategies, or arranging vehicle maintenance.

[0016] Beneficial effects of the present invention: The present invention, through the combination of a self-supervised learning algorithm and an adaptive data fusion algorithm, can effectively integrate multimodal data from different sensors: braking force, steering wheel angle and speed change rate. By dynamically weighting these multimodal data, the accuracy of detecting abnormal vehicle behavior is significantly improved, especially when capturing risk behaviors such as sudden braking and abnormal steering that occur in a short period of time. It can be more accurate than the existing technology. This comprehensive data integration and optimization processing ensures the high reliability of abnormal behavior annotation in vehicle portraits, laying the foundation for subsequent analysis and intervention.

[0017] The present invention, through the constructed dynamic balance model, can not only capture the abnormal fluctuations of the vehicle in a short period of time, but also comprehensively analyze the long-term stability trend of the vehicle by hierarchical modeling of short-term behavioral fluctuations and long-term operating trends, so that the present invention can take into account both the short-term risks and long-term operating status of the vehicle, and provide a scientific basis for the vehicle's safety score and operational stability analysis. Compared with traditional models, the present invention can more comprehensively evaluate the operating status of the vehicle, especially in judging the continuous risk of abnormal behavior, and has obvious advantages.

[0018] The present invention, through the generated vehicle portrait and its analysis results, combined with the long-term trend data in the dynamic balance model, can accurately judge the potential continuous risk of abnormal behavior, and for the identified high-risk behavior, can provide targeted intervention decision suggestions, such as adjusting the driving route, optimizing the driving strategy or arranging vehicle maintenance. The precise intervention based on comprehensive scoring and trend analysis helps to effectively reduce the potential risks in vehicle operation and improve the overall operation safety and operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 A schematic diagram of the evaluation method flow in an embodiment of the present invention; Figure 2 A schematic diagram of constructing a dynamic balance model of short-term fluctuations and long-term trends according to an embodiment of the present invention; Figure 3 Schematic diagram of an evaluation system module according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0022] It should be noted that the references to "one embodiment", "embodiment", "exemplary embodiments", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0023] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0024] like Figure 1-Figure 2 As shown, the vehicle operation data-driven carrier vehicle portrait evaluation method includes the following steps: S1, abnormal behavior perception and labeling: The multi-modal data collected during vehicle operation is collected through a multi-sensor network deployed on the vehicle. The multi-modal data includes braking force, steering wheel angle, and speed change rate. Combined with historical data and the current operating status of the vehicle, a self-supervised learning algorithm is used to automatically perceive and label abnormal behaviors in the data. Abnormal behaviors include sudden emergency braking, abnormal steering, speeding, and frequent lane changes. S2, adaptive fusion of multimodal data: for the annotated multimodal data, an adaptive data fusion algorithm is used for weighted fusion to generate a comprehensive feature vector F. Based on the credibility, relevance and correlation of different modal data with abnormal behavior, the fusion weight of each modal data is dynamically adjusted to ensure that the weight of important data is appropriately increased during the fusion process, thereby improving the model's sensitivity to abnormal behavior and evaluation accuracy; S3, Construction of dynamic balance model of short-term fluctuation and long-term trend: Based on the integrated feature vector F after fusion, a dynamic balance model is constructed that takes into account both the short-term behavior fluctuation and long-term operation trend of the vehicle. The dynamic balance model is based on the time series analysis algorithm, and the short-term abnormal behavior and long-term operation trend of the vehicle are hierarchically modeled, thereby capturing short-term risk events and long-term stability analysis at the same time. S4, portrait generation and dynamic update: Based on the dynamic balance model, the operating status of the carrier vehicle is profiled. The portrait results include the vehicle's safety score and operational stability analysis; S5, abnormal behavior identification and intervention decision: By analyzing the generated vehicle portrait, potential abnormal behavior is identified, and combined with the long-term trend analysis in the dynamic balance model, it is determined whether the abnormal behavior has a persistent risk; for the identified abnormal behavior, corresponding intervention decision suggestions are provided, including adjusting the driving route, optimizing the driving strategy or arranging vehicle maintenance.

[0025] The acquisition of multimodal data in S1 includes: Braking force data collection: Through the brake pressure sensor integrated in the brake system, the pressure applied by the brake system is monitored in real time to capture the braking force of the vehicle under different driving conditions, including emergency braking and normal braking; Steering wheel steering angle data collection: Use the angle sensor on the steering shaft to collect the steering wheel steering angle changes in real time and capture the steering wheel operation during the vehicle steering process, including steering angle, steering speed and steering wheel return status; Speed ​​change rate data collection: The wheel speed sensor integrated in the vehicle collects the vehicle's speed data in real time and calculates the rate of change of the vehicle's speed to identify the dynamic characteristics of the vehicle during acceleration, deceleration and constant speed driving; ,in, represents the rate of change of speed, Indicates the vehicle at the current time speed, Indicates that the vehicle was at the previous moment speed, Indicates the time interval in seconds.

[0026] The automatic perception and annotation in S1 specifically include: S11, comparative analysis of historical data and current data: comparing and analyzing the current operation data of the vehicle with the historical data, the current operation data includes the braking force B, the steering angle of the steering wheel A, and the speed change rate V, and the historical data includes the normal behavior mode and abnormal behavior record of the vehicle under the same or similar driving conditions; S12, training of self-supervised learning model: by using a large amount of labeled historical data to train the self-supervised learning model, the characteristic distribution of the vehicle under normal driving and abnormal behavior states is learned, including the characteristic patterns of sudden braking, abnormal steering, speeding and frequent lane changing behaviors; S13, abnormal behavior detection and labeling: During vehicle operation, the self-supervised learning model analyzes the current data in real time. When it detects that the current data matches the learned abnormal behavior characteristics, it automatically labels it as abnormal behavior. Abnormal behaviors include sudden braking, abnormal steering, and frequent lane changes. Specifically: Sudden braking: When the braking force suddenly increases and exceeds the normal range, the model marks it as sudden braking behavior; Abnormal steering: When the steering wheel angle changes at an abnormally fast rate, or the angle exceeds the normal turning range, the model marks it as abnormal steering behavior; Speeding: When the speed change rate exceeds the current road conditions and historical normal values, the model marks it as speeding behavior; Frequent lane changing: When the steering wheel angle changes frequently in a short period of time and the angle of each change exceeds the normal value, the model marks it as frequent lane changing behavior.

[0027] The self-supervised learning model is trained using historical data. The model uses a self-supervised learning objective function, which is expressed as: ,in, is the sample size, is the reconstructed vector predicted by the model, is the input data vector, including braking force, steering wheel angle, and speed change rate. The model minimizes the reconstruction error to learn characteristic patterns of normal behavior.

[0028] During operation, the current Time data vector Enter the trained model and calculate the difference between it and the reconstruction of historical data: ; If the difference Exceeding the set abnormal threshold , it is marked as abnormal behavior: = ; Specific abnormal behavior annotation basis: Emergency brake: When and , marked as emergency braking behavior; Abnormal turn: When and , marked as abnormal steering behavior; Speeding: When and , marked as speeding behavior; Frequent lane changes: In a short period of time, if Exceeding normal steering angle multiple times and , marked as frequent lane changing behavior.

[0029] S2 specifically includes: S21, Credibility calculation of modal data: Calculate the credibility of each modal data based on the annotated abnormal behavior data ,in Indicates specific modal data, including braking force , Steering wheel angle , speed change rate , the credibility calculation is based on the accuracy and frequency of abnormal behaviors detected by each modal data in historical data, and the calculation formula is: ,in, is modal data The number of abnormal behaviors correctly detected in is modal data The total number of abnormal behaviors detected in; S22, correlation calculation of modal data: by analyzing the correlation between modal data, including their correlation with abnormal behavior, to calculate the correlation weight ,in and Representing different modal data, the correlation weight is determined by calculating the Pearson correlation coefficient: ; in, For modal data and The correlation between For modal data In the The specific value at each sampling point, For modal data In the The specific value at each sampling point, For modal data The mean of For modal data The mean of is the total number of sampling points; S23, Dynamic Adjustment of Fusion Weights: Combining Credibility and relevance weight , dynamically adjust the fusion weight of each modality data , the fusion weight is expressed as: ,in, Represents modal data The final fusion weight of . The larger the weight, the greater the proportion of the modality data in the fusion process. S24, weighted fusion: using dynamically adjusted fusion weights Perform weighted fusion on each modal data to generate a comprehensive feature vector , used for subsequent vehicle portrait evaluation: The comprehensive feature vector F provides a comprehensive description of the vehicle's operating status at different time periods. It integrates multimodal data from braking force, steering wheel angle, and speed change rate through an adaptive data fusion algorithm. These data reflect the key behavioral parameters of the vehicle under various driving conditions. By fusing these data, the feature vector F can comprehensively capture the overall operating status of the vehicle at different time points, including potential abnormal behaviors and routine operations, and provide a comprehensive, general description of the vehicle's operating performance based on multi-dimensional data.

[0030] The construction of the dynamic balance model between short-term fluctuations and long-term trends in S3 specifically includes: S31, organize the comprehensive feature vector F in the form of a time series to form a time series data set ,in Indicates the time period of observation; S32, Hierarchical Modeling: Short-term behavior fluctuation modeling: The autoregressive moving average model in the time series analysis algorithm is used to model the short-term fluctuations in the time series data and capture the abnormal behavior of the vehicle in a short period of time, including sudden braking and abnormal steering, which is expressed as: ,in, is the comprehensive feature vector at the current moment, is the comprehensive feature vector of the previous moment, and They represent the autoregressive and moving average parts in the autoregressive moving average model, is white noise, is the white noise term at the previous moment; Long-term operation trend modeling: A long-term time series prediction model is used to extract the long-term trend of vehicle operation status and capture the stability and potential risks of vehicles over a long period of time, expressed as: , where LSTM is a long-term time series prediction model (long short-term memory network), Represents the long-term trend feature vector. LSTM uses the input time series data Extracting long-term trends; S34, construction of dynamic balance model: Integrate the results of short-term fluctuation model and long-term trend model to form dynamic balance model M. Dynamic balance model M simultaneously analyzes the short-term behavioral fluctuations and long-term operating trends of the vehicle, and captures short-term risk events and long-term stability characteristics in vehicle portrait assessment.

[0031] The specific expansion of the autoregressive moving average ARMA model is as follows: ,in, Indicates at time The short-term behavior fluctuation feature vector at the moment, which captures the abnormal behavior of the vehicle in a short period of time (sudden braking, abnormal steering); Respectively, in time The comprehensive feature vectors at each moment are generated by an adaptive data fusion algorithm and include the behavior parameters of the vehicle during operation (braking force, steering wheel angle, speed change rate). is the autoregression coefficient of the ARMA model, which indicates the influence of the past comprehensive characteristic vector on the current short-term behavior fluctuation. is the moving average coefficient of the ARMA model, which indicates the influence of past white noise on current short-term behavior fluctuations. Indicates at time The white noise term at a certain moment is the part of the model that cannot be explained by past data.

[0032] The internal mechanism of the long-term time series prediction model LSTM is as follows: Forget Gate: ,in, Represents the output of the forget gate, which controls the memory unit of the previous moment How much information is retained at the current moment, represents the weight matrix of the forget gate, represents the hidden state at the previous moment, Represents the bias term of the forget gate.

[0033] Input Gate: ; ; in, Represents the output of the input gate, which controls the impact of the current input on the memory unit. represents the new candidate memory cell value, and Represent the weight matrices of the input gate and candidate memory cells respectively, and Represent the bias terms of the input gate and candidate memory unit respectively.

[0034] Update memory unit: ;in, Indicates the state of the memory unit at the current moment, combined with the memory of the previous moment and the impact of the current input.

[0035] Output Gate: ; ; in, Represents the output of the output gate, controlling the memory unit The impact on the current hidden state, Represents the hidden state at the current moment, and the output is used as the long-term trend feature vector part of is the weight matrix of the output gate, is the bias term of the output gate.

[0036] The LSTM network is a special recursive neural network designed to solve the gradient vanishing and gradient exploding problems of traditional neural networks when processing long time series data. Through its internal memory units and gating mechanisms, LSTM can effectively retain long-term memory, thereby capturing long-term dependencies in the data sequence. This is particularly important in the modeling of long-term operating trends of vehicles, because the operating status of vehicles is not only affected by recent behavior, but also by trends accumulated over a long period of time. The operating status of a vehicle may be affected by a variety of complex factors, and there are often nonlinear relationships between these factors. The LSTM network can process and learn such nonlinear relationships through its layered progressive structure, and identify important trends and patterns in time series. LSTM is suitable for processing these complex, multi-dimensional time series data.

[0037] The dynamic equilibrium model is expressed as: ,in, for The output of the dynamic equilibrium model at time is the short-term behavior fluctuation characteristic vector, is the long-term trend eigenvector, and To adjust the coefficients, ensure that the model can balance short-term and long-term factors.

[0038] Dynamic equilibrium model is used in S4 Short-term behavior fluctuation characteristics of output and long-term operating trend characteristics Calculating the vehicle's safety score , which comprehensively considers the abnormal behavior of the vehicle in the short term (such as sudden braking, abnormal steering, etc.) and the stability in long-term operation, and is expressed as: ,in, is the adjustment coefficient of the safety score, Indicates the length of the time window; S4 is based on the long-term trend characteristics captured in the dynamic equilibrium model Perform vehicle operation stability analysis. Vehicle operation stability analysis identifies potential operational instability by evaluating the consistency and regularity of the vehicle during long-term operation, expressed as: ,in, Represents the operational stability score, where lower values ​​indicate more stable operations.

[0039] S4 also includes scoring result integration, which includes integrating the calculated safety score and operational stability analysis results Integrate to generate a comprehensive score of vehicle operating status , the integration formula is: ,in, and They are the weight coefficients of the safety score and the operational stability score, respectively, and are used to adjust the impact of the two in the comprehensive score. Used to reverse the stability score so that higher stability (i.e. lower value) contributes more positively to the overall score; Image generation: using the comprehensive score calculated above , generating a profile of the operating status of the carrier vehicle. This profile visualizes the safety and operational stability of the vehicle, helping managers quickly assess the overall operating status of the vehicle, identify potential risk points, and provide a basis for decision-making.

[0040] αs (adjustment coefficient of safety score): obtained through experimental or historical data analysis and adjustment. First, αs is adjusted by performing regression analysis (such as gradient descent) on the vehicle operation data to minimize the error between the calculated safety score Ssafety and the actual observed safety situation. The purpose is to adjust the impact of short-term behavioral fluctuations and long-term operating trends on the safety score, so that the score is more in line with the actual safety situation.

[0041] βs (the weight coefficient of the safety score in the comprehensive score) is determined through expert experience or data-driven weight adjustment methods. In practical applications, the importance of the safety score in the overall assessment is determined by analyzing historical data, and an optimization algorithm is used to make adjustments to ensure that the comprehensive score Soverall reflects the correct risk level. The purpose is to determine the relative importance of the safety score in the comprehensive score and ensure that the comprehensive score accurately reflects the overall operating risk of the vehicle.

[0042] βo (the weight coefficient of the operational stability score in the comprehensive score) is similar to βs. It is determined through weight optimization of expert scoring and historical data regression analysis, and is adjusted according to the impact of vehicle operating stability on overall safety. The purpose is to adjust the impact of the operational stability score on the comprehensive score to ensure that the comprehensive score can effectively reflect the long-term stability of the vehicle.

[0043] S5 specifically includes: Abnormal behavior identification: safety scoring based on vehicle operation status profile and operational stability results , identify potential abnormal behaviors, and mark them as potential abnormal behaviors when the safety score is significantly lower than the set threshold or the operation stability results fluctuate greatly; Long-term trend analysis: Identify potential abnormal behaviors and compare them with the long-term trend feature vectors in the dynamic equilibrium model. Conduct comparative analysis to observe changes in long-term trend characteristics and determine whether abnormal behavior is just a short-term fluctuation or a persistent risk; Intervention decision recommendations: Provide corresponding intervention decision recommendations based on the identified abnormal behaviors and their persistent risks, including: Adjust driving route: When the identified abnormal behavior is related to specific road conditions, it is recommended to adjust the driving route to avoid potentially dangerous sections of the road; Optimize driving strategy: By analyzing the unstable factors in driving behavior, provide suggestions for optimizing driving strategy, such as adjusting braking force, steering angle, etc., to reduce risks; Arrange vehicle maintenance: If the abnormal behavior may be related to the vehicle's mechanical failure or system problems, it is recommended to conduct vehicle inspection and maintenance to eliminate possible failure risks in advance; Feedback and update: Compare and analyze the implemented intervention measures with the subsequent operating status of the vehicle, update the dynamic balance model and intervention strategy to continuously optimize the effectiveness of abnormal behavior identification and risk management.

[0044] like Figure 3 As shown, the vehicle operation data-driven carrier vehicle portrait evaluation system is used to implement the above-mentioned vehicle operation data-driven carrier vehicle portrait evaluation method, and includes the following modules: Multimodal data acquisition module: collects multimodal data during vehicle operation through a multi-sensor network deployed on the vehicle, including braking force, steering wheel angle, and speed change rate; Abnormal behavior perception and labeling module: Analyzes the collected multimodal data through a self-supervised learning algorithm, automatically perceives and labels abnormal behaviors, including sudden braking, abnormal steering, and speeding, and generates labeled behavior data; Adaptive data fusion module: For the annotated behavior data, an adaptive data fusion algorithm is used to dynamically adjust the fusion weight of each modal data based on the credibility, relevance and correlation of each modal data with abnormal behavior to generate a comprehensive feature vector; Dynamic balance model module: Utilizes the integrated feature vector after fusion to build a dynamic balance model that takes into account both the short-term behavior fluctuations and long-term operation trends of the vehicle, thus realizing hierarchical modeling and evaluation of the vehicle's operating status; Vehicle operation status portrait generation module: Based on the dynamic balance model, it calculates the vehicle's safety score and operational stability analysis results, and generates a vehicle operation status portrait to present the vehicle's overall operation status; Risk analysis and intervention decision module: Analyze the generated vehicle portrait, identify potential abnormal behaviors, and combine the long-term trend analysis in the dynamic balance model to determine whether the abnormal behaviors have persistent risks, and provide decision-making suggestions, including adjusting the driving route, optimizing driving strategies, or arranging vehicle maintenance.

[0045] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0046] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A vehicle operation data-driven transport vehicle portrait assessment method, characterized in that: The following steps are involved: S1, abnormal behavior perception and labeling: The multi-modal data of the vehicle operation process is collected through the multi-sensor network deployed on the vehicle. The multi-modal data includes braking force, steering wheel angle, speed change rate, and combined with historical data and the current operation status of the vehicle, a self-supervised learning algorithm is used to automatically perceive and label abnormal behaviors in the data. The abnormal behaviors include sudden emergency braking, abnormal steering, speeding, and frequent lane changes; S2, adaptive fusion of multimodal data: for the annotated multimodal data, an adaptive data fusion algorithm is used for weighted fusion to generate a comprehensive feature vector F. Based on the credibility, relevance and correlation of different modal data with abnormal behavior, the fusion weight of each modal data is dynamically adjusted; S3, construction of dynamic balance model of short-term fluctuation and long-term trend: based on the integrated feature vector F after fusion, a dynamic balance model that considers both short-term behavior fluctuation and long-term operation trend of vehicles is constructed; The dynamic balance model is based on a time series analysis algorithm, which hierarchically models the short-term abnormal behavior and long-term operating trends of vehicles, thereby capturing both short-term risk events and long-term stability analysis; S4, portrait generation and dynamic update: Based on the dynamic balance model, a portrait of the operating status of the carrier vehicle is generated, and the portrait result includes the vehicle safety score and operation stability analysis; S5, abnormal behavior identification and intervention decision: By analyzing the generated vehicle portrait, potential abnormal behavior is identified, and combined with the long-term trend analysis in the dynamic balance model, it is determined whether the abnormal behavior has a persistent risk; for the identified abnormal behavior, corresponding intervention decision suggestions are provided, including adjusting the driving route, optimizing the driving strategy or arranging vehicle maintenance.

2. The vehicle operation data-driven carrier vehicle portrait assessment method according to claim 1, characterized in that: The acquisition of multimodal data in S1 includes: Braking force data collection: Through the brake pressure sensor integrated in the brake system, the pressure applied by the brake system is monitored in real time to capture the braking force of the vehicle under different driving conditions, including emergency braking and normal braking; Steering wheel steering angle data collection: Use the angle sensor on the steering shaft to collect the steering wheel steering angle changes in real time and capture the steering wheel operation during the vehicle steering process, including steering angle, steering speed and steering wheel return status; Speed ​​change rate data collection: The wheel speed sensor integrated in the vehicle collects the vehicle's speed data in real time and calculates the rate of change of the vehicle's speed to identify the dynamic characteristics of the vehicle during acceleration, deceleration and constant speed driving.

3. The vehicle operation data-driven carrier vehicle portrait assessment method according to claim 1, characterized in that: The automatic perception and labeling in S1 specifically include: S11, comparative analysis of historical data and current data: comparing and analyzing the current operation data of the vehicle with the historical data, the current operation data includes the braking force B, the steering angle of the steering wheel A, and the speed change rate V, and the historical data includes the normal behavior mode and abnormal behavior record of the vehicle under the same or similar driving conditions; S12, training of self-supervised learning model: by using the labeled historical data to train the self-supervised learning model, the characteristic distribution of the vehicle under normal driving and abnormal behavior states is learned, including the characteristic patterns of sudden braking, abnormal steering, speeding and frequent lane changing behaviors; S13, abnormal behavior detection and labeling: During vehicle operation, the self-supervised learning model analyzes the current data in real time. When it detects that the current data matches the learned abnormal behavior characteristics, it automatically labels it as abnormal behavior. Abnormal behaviors include sudden braking, abnormal steering, and frequent lane changes.

4. The vehicle operation data-driven carrier vehicle portrait assessment method according to claim 1, characterized in that: The S2 specifically includes: S21, Credibility calculation of modal data: Calculate the credibility of each modal data based on the annotated abnormal behavior data ,in Indicates specific modal data, including braking force , Steering wheel angle , speed change rate ,The credibility calculation is based on the accuracy and frequency of abnormal behaviors detected in historical data by each modality; S22, correlation calculation of modal data: by analyzing the correlation between modal data, including their correlation with abnormal behavior, to calculate the correlation weight ,in and Representing different modal data, the correlation weights were determined by calculating the Pearson correlation coefficient; S23, Dynamic Adjustment of Fusion Weights: Combining Credibility and relevance weight , dynamically adjust the fusion weight of each modality data , the fusion weight is expressed as: ,in, Represents modal data The final fusion weight of . The larger the weight, the greater the proportion of the modality data in the fusion process. S24, weighted fusion: using dynamically adjusted fusion weights Perform weighted fusion on each modal data to generate a comprehensive feature vector , used for subsequent vehicle portrait evaluation: 。 5. The vehicle operation data-driven carrier vehicle portrait assessment method according to claim 1, characterized in that: The construction of the dynamic balance model of short-term fluctuations and long-term trends in S3 specifically includes: S31, organize the comprehensive feature vector F in the form of a time series to form a time series data set ,in Indicates the time period of observation; S32, Hierarchical Modeling: Short-term behavior fluctuation modeling: The autoregressive moving average model in the time series analysis algorithm is used to model the short-term fluctuations in the time series data and capture the abnormal behavior of the vehicle in a short period of time, including sudden braking and abnormal steering, which is expressed as: ,in, is the comprehensive feature vector at the current moment, is the comprehensive feature vector of the previous moment, and They represent the autoregressive and moving average parts in the autoregressive moving average model, is white noise, is the white noise term at the previous moment; Long-term operation trend modeling: A long-term time series prediction model is used to extract the long-term trend of vehicle operation status and capture the stability and potential risks of vehicles over a long period of time, expressed as: , where LSTM is a long-term time series prediction model, Represents the long-term trend feature vector. LSTM uses the input time series data Extracting long-term trends; S34, construction of dynamic balance model: Integrate the results of short-term fluctuation model and long-term trend model to form dynamic balance model M. Dynamic balance model M simultaneously analyzes the short-term behavioral fluctuations and long-term operating trends of the vehicle, and captures short-term risk events and long-term stability characteristics in vehicle portrait assessment.

6. The vehicle operation data-driven carrier vehicle portrait assessment method according to claim 5, characterized in that: The dynamic equilibrium model is expressed as: ,in, for The output of the dynamic equilibrium model at time is the short-term behavior fluctuation characteristic vector, is the long-term trend eigenvector, and To adjust the coefficients, ensure that the model can balance short-term and long-term factors.

7. The vehicle operation data-driven carrier vehicle portrait assessment method according to claim 6, characterized in that: The dynamic equilibrium model used in S4 Short-term behavior fluctuation characteristics of output and long-term operating trend characteristics Calculating the vehicle's safety score , expressed as: ,in, is the adjustment coefficient of the safety score, Indicates the length of the time window; The S4 is based on the long-term trend characteristics captured in the dynamic equilibrium model Perform vehicle operation stability analysis. Vehicle operation stability analysis identifies potential operational instability by evaluating the consistency and regularity of the vehicle during long-term operation, expressed as: ,in, Represents the operational stability score, where lower values ​​indicate more stable operations.

8. The vehicle operation data-driven carrier vehicle portrait assessment method according to claim 7, characterized in that: S4 also includes scoring result integration, wherein the scoring result integration includes integrating the calculated safety score and operational stability analysis results Integrate to generate a comprehensive score of vehicle operating status , the integration formula is: ,in, and are the weight coefficients of safety score and operational stability score, Used to reverse the stability score; Image generation: using the comprehensive score calculated above , generate a profile of the operating status of the carrier vehicle.

9. The vehicle operation data-driven carrier vehicle portrait assessment method according to claim 6, characterized in that: The S5 specifically includes: Abnormal behavior identification: safety scoring based on vehicle operation status profile and operational stability results , identify potential abnormal behaviors, and mark them as potential abnormal behaviors when the safety score is lower than the set threshold or the operation stability results fluctuate greatly; Long-term trend analysis: Identify potential abnormal behaviors and compare them with the long-term trend feature vectors in the dynamic equilibrium model. Conduct comparative analysis to observe changes in long-term trend characteristics and determine whether abnormal behavior is just a short-term fluctuation or a persistent risk; Intervention decision recommendations: Provide corresponding intervention decision recommendations based on the identified abnormal behaviors and their persistent risks.

10. A vehicle operation data-driven carrier vehicle portrait evaluation system, used to implement a vehicle operation data-driven carrier vehicle portrait evaluation method as claimed in any one of claims 1 to 9, characterized in that: Includes the following modules: Multimodal data acquisition module: collects multimodal data during vehicle operation through a multi-sensor network deployed on the vehicle, including braking force, steering wheel angle, and speed change rate; Abnormal behavior perception and labeling module: Analyzes the collected multimodal data through a self-supervised learning algorithm, automatically perceives and labels abnormal behaviors, including sudden braking, abnormal steering, and speeding, and generates labeled behavior data; Adaptive data fusion module: For the annotated behavior data, an adaptive data fusion algorithm is used to dynamically adjust the fusion weight of each modal data based on the credibility, relevance and correlation of each modal data with abnormal behavior to generate a comprehensive feature vector; Dynamic balance model module: Utilizes the integrated feature vector after fusion to build a dynamic balance model that takes into account both the short-term behavior fluctuations and long-term operation trends of the vehicle, thus realizing hierarchical modeling and evaluation of the vehicle's operating status; Vehicle operation status portrait generation module: Based on the dynamic balance model, it calculates the vehicle's safety score and operational stability analysis results, and generates a vehicle operation status portrait to present the vehicle's overall operation status; Risk analysis and intervention decision module: Analyze the generated vehicle portrait, identify potential abnormal behaviors, and combine the long-term trend analysis in the dynamic balance model to determine whether the abnormal behaviors have persistent risks, and provide decision-making suggestions, including adjusting the driving route, optimizing driving strategies, or arranging vehicle maintenance.

Images