An illegal behavior AI intelligent identification method based on big data analysis

Abstract

This invention discloses an AI-powered intelligent identification method for traffic violations based on big data analysis, comprising the following steps: acquiring historical behavior datasets, merging and performing data preprocessing to generate a sequence of violation events as input; constructing an event generation and judgment framework by calling a neural time-point process hybrid model to generate a set of mechanistic event intensities; constructing a dynamic responsibility allocation mechanism, performing weighted modulation processing to obtain modulated event intensities; performing state recursive updates to generate a violation risk state representation and performing state coupling updates; constructing a judgment-driven event generation mechanism, performing judgment-driven reweighting processing to generate judgment event generation results; performing a joint inference process to generate intelligent identification results for traffic violations; and performing feedback update processing on the event generation and judgment framework based on the intelligent identification results for traffic violations. This invention improves the stability, consistency, and interpretability of traffic violation identification.

Description

Technical Field

[0001] This invention relates to the field of intelligent recognition of road traffic violations, and in particular to an AI-based intelligent recognition method for violations based on big data analysis. Background Technology

[0002] As cities expand and the number of entities subject to regulation continues to grow, violations in areas such as road traffic, urban management, workplace safety, and public order are becoming more frequent, repetitive, and evolving. Existing violation identification technologies primarily rely on rule matching, statistical analysis, or machine learning models based on single samples to make static judgments on violations. This makes it difficult to depict the temporal patterns of violations and the risk evolution characteristics that arise as enforcement actions change.

[0003] While some existing technologies incorporate deep learning or sequence models to identify violations, they typically treat violations as independent samples, ignoring the temporal relationships between violations and the mutual influence between multiple violations. This fails to reflect how a single violation alters the risk structure of subsequent violations. Furthermore, existing methods often employ attention weights or confidence weights for result fusion, lacking a continuous evolution mechanism for responsibility allocation and risk assessment. This makes them easily equated with simple weight adjustments and unsuitable for supporting the continuous regulatory needs of complex scenarios.

[0004] Furthermore, in real-world scenarios where multiple violation generation mechanisms coexist, existing technologies typically lack a unified state scheduling and mechanism consistency constraint mechanism. Their ability to resolve conflicts in the outputs of different violation mechanisms is insufficient, resulting in limited stability and interpretability of the identification results. Especially after law enforcement intervention, existing technologies struggle to effectively utilize feedback to structurally update subsequent identification processes, failing to form a closed-loop optimized identification mechanism.

[0005] Therefore, how to provide an AI-based intelligent identification method for traffic violations based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an AI-powered intelligent identification method for traffic violations based on big data analysis. This invention constructs a sequence of violation events as an input structure based on historical behavior data sets. It combines this with a neural time-point process hybrid model to form an event generation and judgment framework. Through a dynamic responsibility allocation mechanism, a violation risk state representation, and a judgment-driven event generation mechanism, it achieves unified modeling of the temporal evolution characteristics of violation behavior and the mechanisms of multiple violation generation. Furthermore, it introduces mechanism consistency constraints and conflict resolution processing during the joint inference process, outputting intelligent identification results for traffic violations. This invention can reflect the continuous evolutionary characteristics of violation behavior, improving the stability, consistency, and interpretability of the identification results.

[0007] According to an embodiment of the present invention, an AI-based intelligent identification method for traffic violations based on big data analysis includes the following steps: Obtain a set of historical behavior data, perform data preprocessing, and obtain the input structure of the violation event sequence; The neural time-point process hybrid model is used as the basis for time event modeling. An event generation and judgment framework is constructed by combining the input structure of violation event sequence and calculating the set of event intensity of generation mechanism. In the event generation and judgment framework, a dynamic responsibility allocation mechanism is constructed. The input structure is a sequence of violation events, which is processed to generate a responsibility weight vector. This vector is then weighted and modulated with the set of mechanistic event intensities to obtain the modulated event intensity. The system receives the input structure of the sequence of violation events and the intensity of the modulated events, performs a recursive update of the state, generates a representation of the violation risk state, constructs the state coupling relationship, and performs a coupled update of the state. A decision-driven event generation mechanism is constructed, the contribution result of violation determination is calculated, and the intensity of the modulated event is subjected to decision-driven reweighting to obtain the result of the event generation. Input the sequence of violations into the structure, the responsibility weight vector, the violation risk status representation, and the judgment event generation result into the joint reasoning process, perform mechanism consistency weighted convergence and sequence consistency verification, and generate intelligent recognition results of violations. The event generation and judgment framework is updated based on the intelligent recognition results of violations.

[0008] Optionally, the generation of the violation event sequence input structure includes: Obtain the historical behavior data set from the violation monitoring system; Aggregation processing is performed on data entries belonging to the same object identifier in the historical behavior data set, and they are organized into object-level behavior data subsets. In the object-level behavioral data subset, event-level data construction processing is performed to construct each data entry with a clear event occurrence time into an event-level data unit. Each event-level data unit contains the corresponding event occurrence time, event spatial location identifier, event environmental state characteristics, and law enforcement handling information. Perform sequential consistency verification and outlier removal on the event-level data units to obtain the event-level data unit sequence; Perform missing data completion processing on the event-level data unit sequence to obtain an event-level data unit sequence with complete fields; Perform multi-field alignment processing on the complete event-level data unit sequence to form a time-aligned event-level data unit sequence; Time-aligned event-level data unit sequences are processed according to a preset time granularity, and the time interval between event occurrences and the change in event density are calculated to organize and generate the input structure of the violation event sequence.

[0009] Optionally, the generation of the set of mechanistic event intensities includes: In the event generation and judgment framework, a neural time point process hybrid model is used as the basis for time event modeling, and a set of sub-event generation units is constructed, with each sub-event generation unit corresponding to a violation generation mechanism. The sequence of violations is input into each sub-event generation unit. Based on the time-point process modeling mechanism of the neural time-point process hybrid model, and combined with the sequence of violations, time event modeling processing is performed. In each sub-event generation unit, based on the time event modeling processing results, the historical impact aggregation and weighted accumulation of the events that have occurred in the input structure of the violation event sequence are performed to obtain the historical impact vector, and the mechanism mapping calculation is performed to obtain the mechanism event intensity. The output mechanistic event intensities are aggregated to generate a set of mechanistic event intensities.

[0010] Optionally, the generation of the modulation event intensity includes: A dynamic responsibility allocation mechanism is constructed within the event generation and judgment framework, and a corresponding responsibility status representation is established for each violation generation mechanism. The sequence of violations is input into a dynamic responsibility allocation mechanism, which inputs the time interval between events, changes in event density, spatial location identifiers of events, environmental characteristics of events, and law enforcement information into the structure. Time-based responsibility recursion is then performed to generate a responsibility weight vector for the current moment. In the time-based responsibility derivation process, the changing trend of the time interval between events is used as the basis for time modulation of responsibility growth and decay, and law enforcement information is used as the triggering condition for responsibility inhibition. The delay inhibition effect is introduced, responsibility inhibition adjustment is performed, and a responsibility weight vector is generated by the combined effect of time evolution and law enforcement inhibition. Normalize the responsibility weight vector after time responsibility recursion processing and write it into the internal state of the sub-event generation unit set; In the set of sub-event generation units, the normalized responsibility weight vector corresponding to the current moment is read, and it is combined with the set of mechanism event intensity to perform weighted modulation processing, and responsibility evolution modulation is performed to obtain the modulated event intensity.

[0011] Optionally, the generation and state coupling update of the violation risk status representation include: Within the event generation and judgment framework, a process for generating a representation of violation risk status is constructed. The sequence of violation events and the intensity of modulation events are input into the violation risk status representation generation process in chronological order of event occurrence. Based on the violation risk status representation corresponding to the previous event and the intensity of modulation events corresponding to the current event, the status is recursively updated to generate the violation risk status representation corresponding to the current event. A state coupling relationship is constructed based on the violation risk status representation corresponding to the current event. The state coupling relationship is a state coupling mapping between the violation risk status representation and the set of sub-event generation units. In the state coupling relationship, state scheduling processing is performed on each sub-event generation unit based on the violation risk state representation; The state coupling relationship and the modulation event intensity are input into the set of sub-event generation units. Based on the activation and inhibition states of each sub-event generation unit, state coupling update is performed to obtain the coupled event state.

[0012] Optionally, the generation of the determination event generation result includes: Within the event generation and judgment framework, a judgment-driven event generation mechanism is constructed; The violation risk status representation, responsibility weight vector, coupled event status and modulation event intensity are input into the judgment-driven event generation mechanism to perform judgment contribution calculation and generate violation judgment contribution results. Based on the contribution results of violation determination, the event generation path is selected to determine the event generation path status corresponding to the current event occurrence time. When the result of the violation judgment crosses the preset path switching threshold, the event generation path switching process is triggered, which changes the combination structure of the sub-event generation units participating in the current event generation in the sub-event generation unit set. Based on the event generation path state selected by the decision driver, the modulation event intensity is reweighted by the decision driver to generate the decision event generation result.

[0013] Optionally, the generation of the intelligent identification result of the violation includes: Within the event generation and decision-making framework, a joint reasoning process is constructed; During joint reasoning, the consistency of the execution mechanism for generating results of judgment events is weighted and converged based on the responsibility weight vector during the joint reasoning process. During the mechanism consistency weighted aggregation process, based on the time sequence of events in the input structure of the violation event sequence, the order consistency check is performed on the violation generation mechanism label corresponding to the judgment event generation result to determine mechanism conflicts. When a mechanism conflict is detected, a resolution process is performed based on the conflict resolution rules; After completing the mechanism consistency weighted aggregation and conflict resolution processing, the intelligent identification result of the violation behavior is generated based on the judgment event generation result after consistency verification and conflict resolution.

[0014] Optionally, the feedback update process includes: The intelligent identification results of violations are input into the feedback update processing module in the event generation and judgment framework, and the feedback constraint information corresponding to the current event occurrence time is determined based on the violation judgment label and the violation generation mechanism label. Based on feedback constraint information, the dynamic responsibility allocation mechanism is processed to update the responsibility weights. The process of updating the status of violation risk is subject to status transition update processing. Perform decision path update processing on the decision-driven event generation mechanism; After completing the feedback update processing of the dynamic responsibility allocation mechanism, the violation risk status representation update process, and the judgment-driven event generation mechanism, the updated event generation and judgment framework is generated.

[0015] The beneficial effects of this invention are: First, this invention constructs an event generation and judgment framework based on the input structure of violation event sequences, and introduces a neural time-point process hybrid model to perform time modeling of violation behavior. This makes violation behavior no longer regarded as independent discrete events, but as a continuous process with time dependence and evolutionary laws for analysis. In this way, it can characterize the continuous impact of a violation behavior on the subsequent violation risk structure, and significantly improve the accuracy and stability of intelligent violation behavior recognition in the time dimension.

[0016] Secondly, this invention models the temporal evolution of responsibility weights for different violation generation mechanisms through a dynamic responsibility allocation mechanism, enabling the responsibility weight vector to possess historical memory and evolutionary path characteristics. This avoids equating responsibility weights with instantaneous attention weights, thereby achieving continuous adjustment of the degree of participation of violation generation mechanisms and improving the adaptability and judgment consistency in scenarios where multiple mechanisms coexist during violation identification.

[0017] Furthermore, this invention constructs a violation risk state representation and generates state coupling relationships based on this state. The violation risk state is used as a global scheduling state shared across sub-event generation units to uniformly modulate the participation state of each sub-event generation unit. This enables violation risks to propagate collaboratively between different violation generation mechanisms, effectively improving the completeness and interpretability of risk characterization in complex violation scenarios.

[0018] Furthermore, this invention uses a judgment-driven event generation mechanism to control the selection of event generation paths by using the contribution results of violation judgments, rather than simply applying numerical weighting. This allows the judgment results to influence the event generation structure in reverse. In the joint reasoning process, mechanism consistency constraints and conflict resolution are introduced to reduce the impact of conflicts between different violation generation mechanisms on the overall recognition results and enhance the reliability of intelligent violation recognition results. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of an AI-based intelligent identification method for traffic violations based on big data analysis proposed in this invention. Figure 2 This is a schematic diagram of the event generation and determination framework in this invention; Figure 3 This is a schematic diagram of the joint reasoning process and conflict resolution mechanism in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figure 1-3 A method for intelligent identification of traffic violations based on big data analysis includes the following steps: A historical behavior data set is obtained, which includes the identifier of the object to be identified, the time of the event, the spatial location identifier of the event, the environmental status characteristics of the event, and law enforcement handling information. The historical behavior data set is sorted according to the time of the event and data preprocessing is performed according to a preset time granularity. The data preprocessing includes event-level data construction processing, abnormal data removal processing, missing data completion processing, and multi-field alignment processing to obtain the input structure of the violation event sequence. The neural time-point process hybrid model is used as the basis for time event modeling. An event generation and judgment framework is constructed by combining the input structure of violation event sequence. The event generation and judgment framework includes a set of sub-event generation units, which is a set of functional units in the event generation and judgment framework. Each sub-event generation unit receives the input structure of violation event sequence and calculates the intensity of the mechanism event corresponding to a violation generation mechanism, and outputs a set of mechanism event intensities. In the event generation and judgment framework, a dynamic responsibility allocation mechanism is constructed. The event occurrence time interval, event density change, event spatial location identifier, event environmental state characteristics, and law enforcement handling information in the input structure of the violation event sequence are input into the dynamic responsibility allocation mechanism. The mechanism processes and generates a responsibility weight vector. The responsibility weight vector is normalized so that the sum of the components of the responsibility weight vector is a preset constant. The responsibility weight vector and the mechanism event intensity set are input into the sub-event generation unit set for weighted modulation processing to obtain the modulated event intensity. In the event generation and judgment framework, a violation risk state representation generation process is constructed. The violation risk state representation generation process receives the input structure of the violation event sequence and the modulation event intensity, performs state recursive update according to the event occurrence time order, generates the violation risk state representation, and constructs the state coupling relationship based on the violation risk state representation. The state coupling relationship is a state coupling mapping of the sub-event generation unit set. The state coupling relationship and the modulation event intensity are input into the sub-event generation unit set to perform state coupling update to obtain the coupled event state. In the event generation and judgment framework, a judgment-driven event generation mechanism is constructed. The violation risk status representation, responsibility weight vector, coupled event status, and modulation event intensity are input into the judgment-driven event generation mechanism to calculate the violation judgment contribution result. The violation judgment contribution result is used to perform judgment-driven reweighting on the modulation event intensity to obtain the judgment event generation result. The process inputs the sequence of violations, the responsibility weight vector, the violation risk status representation, and the results of the judgment events into a joint reasoning process. Based on the responsibility weight vector, it performs a weighted aggregation of the results of the judgment events to ensure consistency of the mechanism. Based on the time sequence of events in the sequence of violations, it performs a sequence consistency check on the weighted aggregation results to generate intelligent identification results of violations. The intelligent identification results of violations include violation judgment labels and violation generation mechanism labels. Based on the intelligent identification results of violations, feedback update processing is performed on the event generation and judgment framework. The feedback update processing includes updating the responsibility weight of the dynamic responsibility allocation mechanism, updating the state transition process of the violation risk status representation, and updating the judgment path of the judgment-driven event generation mechanism, resulting in the updated event generation and judgment framework.

[0022] In this embodiment, the generation of the violation event sequence input structure includes: Obtain the historical behavior data set from the violation monitoring system; Aggregation processing is performed on data entries belonging to the same object identifier in the historical behavior data set, and data entries with the same object identifier are organized into object-level behavior data subsets. In the object-level behavioral data subset, event-level data construction processing is performed based on the event occurrence time field. Each data entry with a clear event occurrence time is constructed into an event-level data unit. Each event-level data unit contains the corresponding event occurrence time, event spatial location identifier, event environmental state characteristics, and law enforcement handling information. Perform sequential consistency checks on event-level data units according to the event occurrence time, and remove abnormal data from event-level data units that do not meet the time sequence consistency requirements, so as to obtain a sequence of event-level data units with consistent time sequence. For event-level data unit sequences with consistent time order, missing data completion processing is performed. For missing fields in event environment state characteristics and law enforcement handling information, field completion is performed based on the corresponding field values ​​of adjacent event-level data units under the same object identifier, resulting in a complete event-level data unit sequence. Multi-field alignment processing is performed on the complete event-level data unit sequence. The multi-field alignment processing aligns the event spatial location identifier, event environmental state characteristics and law enforcement handling information according to the event occurrence time, forming a time-aligned event-level data unit sequence. The time-aligned event-level data unit sequence is subjected to time normalization processing according to a preset time granularity. The event occurrence time interval is calculated based on the event occurrence time of adjacent event-level data units. The event density change is calculated based on the number of event-level data units within a unit time granularity. The system then organizes and generates an input structure for the violation event sequence that includes the event occurrence time interval, event density change, event spatial location identifier, event environmental state characteristics, and law enforcement handling information.

[0023] In this embodiment, the generation of the mechanistic event intensity set includes: In the event generation and judgment framework, the neural time point process hybrid model is used as the basis for time event modeling, and a set of sub-event generation units is constructed based on the neural time point process hybrid model. Each sub-event generation unit in the set of sub-event generation units corresponds one-to-one with a violation generation mechanism. In the event generation and judgment framework, a mapping relationship between violation generation mechanism and sub-event generation unit is established. The sequence of violations is input into the structure of the sequence of violations, and each sub-event generation unit in the set of sub-event generation units is input into the structure of the sequence of violations. Each sub-event generation unit is based on the time point process modeling mechanism of the neural time point process hybrid model. It combines the event occurrence time, event occurrence time interval, event density change, event spatial location identifier, event environmental state characteristics and law enforcement handling information in the sequence of violations to perform time event modeling processing related to its corresponding violation generation mechanism. In each sub-event generation unit, based on the time event modeling processing results, historical impact aggregation processing is performed on the events that have occurred in the input structure of the violation event sequence. This historical impact aggregation process uses the current event occurrence time as a reference, calculating a time decay weight for each occurred event, and constructing an event attribute vector based on the event's spatial location identifier, environmental characteristics, and law enforcement information. The event attribute vectors of each occurred event are then weighted and accumulated according to their corresponding time decay weights to obtain a historical impact vector reflecting the degree of influence of historical events on the current moment. Each sub-event generation unit performs mechanism mapping calculations on the historical impact vector based on its corresponding violation generation mechanism to obtain the mechanism event intensity of the corresponding violation generation mechanism at the current event occurrence time. The calculation formula is as follows: ; in, This indicates the violation generation mechanism corresponding to the j-th sub-event generation unit at the current event occurrence time. The intensity of the mechanistic event is used to characterize the temporal tendency of the violation generation mechanism under the current input structure of the violation event sequence. This represents the historical influence vector obtained through historical influence aggregation. This represents the mechanism mapping weight vector embedded in the j-th sub-event generation unit. express transpose, This represents the mechanism mapping bias scalar built into the j-th sub-event generation unit. This represents a non-negative mapping function, used to ensure that the intensity of the mechanistic event is a non-negative output; The intensity of mechanistic events output by each sub-event generation unit in the sub-event generation unit set is aggregated to generate a set of mechanistic event intensities. This set of mechanistic event intensities is then output to the event generation and judgment framework for use in the subsequent dynamic responsibility allocation mechanism and violation risk status representation generation process.

[0024] In this embodiment, the generation of the modulation event intensity includes: A dynamic responsibility allocation mechanism is constructed in the event generation and judgment framework, and a corresponding responsibility status representation is established for each violation generation mechanism. The responsibility status representation is used to characterize the responsibility accumulation status of the corresponding violation generation mechanism in the historical violation event sequence. The sequence of violations is input into the dynamic responsibility allocation mechanism by inputting the time interval between events, changes in event density, spatial location identifiers of events, environmental state characteristics of events, and law enforcement information into the structure. Based on the responsibility status representation corresponding to the current event time and the responsibility status representation corresponding to the previous event time, time responsibility recursion processing is performed to generate the responsibility weight vector at the current moment. In the time-based responsibility derivation process, the changing trend of the time interval between events is used as the basis for time modulation of responsibility growth and decay, and law enforcement information is used as the triggering condition for responsibility inhibition. The delay inhibition effect is introduced, and the responsibility inhibition adjustment is performed on the violation generation mechanism corresponding to the law enforcement action in the responsibility weight vector, generating a responsibility weight vector that has been jointly affected by time evolution and law enforcement inhibition. Normalize the responsibility weight vector after time responsibility recursion processing so that the sum of each component of the responsibility weight vector is a preset constant. Then write the normalized responsibility weight vector into the internal state of the sub-event generation unit set to update the responsibility state representation in the sub-event generation unit corresponding to the violation generation mechanism. In the set of sub-event generation units, based on the updated responsibility state representation, the normalized responsibility weight vector corresponding to the current moment is read, and the normalized responsibility weight vector is combined with the set of mechanism event intensity to perform weighted modulation processing. The set of mechanism event intensity is subjected to responsibility evolution modulation to obtain the modulated event intensity, and the modulated event intensity is output to the event generation and judgment framework.

[0025] In this embodiment, the generation and state coupling update of the violation risk status representation include: In the event generation and judgment framework, a violation risk state representation generation process is constructed. The violation risk state representation generation process is a global state generation structure independent of the sub-event generation unit set, which is used to uniformly characterize the risk evolution state of the violation event sequence input structure on the overall time scale. The sequence of violation events and the intensity of modulation events are input into the violation risk status representation generation process in chronological order of event occurrence. During the generation process, the status is recursively updated based on the violation risk status representation corresponding to the previous event and the intensity of modulation events corresponding to the current event, thereby generating the violation risk status representation corresponding to the current event. A state coupling relationship is constructed based on the violation risk state representation corresponding to the current event. The state coupling relationship is a state coupling mapping between the violation risk state representation and the set of sub-event generation units, which is used to describe the unified modulation relationship of the violation risk state representation on each sub-event generation unit. In the state coupling relationship, state scheduling processing is performed on each sub-event generation unit based on the violation risk state representation to determine the active state and the suppression state of each sub-event generation unit at the current event occurrence time, and the active state and the suppression state are used as the state coupling update conditions. The state coupling relationship and the modulation event intensity are input into the sub-event generation unit set. In the sub-event generation unit set, state coupling update is performed based on the activation state and inhibition state corresponding to each sub-event generation unit to obtain the coupled event state. The coupled event state is then output to the event generation and decision framework.

[0026] In this embodiment, the generation of the event generation result includes: In the event generation and judgment framework, a judgment-driven event generation mechanism is constructed. As a generation path control structure independent of the set of sub-event generation units, the judgment-driven event generation mechanism is used to selectively control the event generation path based on the violation judgment result. The violation risk status representation, responsibility weight vector, coupled event status, and modulation event intensity are input into the judgment-driven event generation mechanism. The judgment contribution calculation is performed in the judgment-driven event generation mechanism to generate the violation judgment contribution result. The violation judgment contribution result is used to characterize the path-level influence of the current event generation result on the violation judgment. The specific calculation and processing of the determination contribution is as follows: In the determination-driven event generation mechanism, the global risk scheduling state corresponding to the current event occurrence time is determined based on the violation risk state representation. Under the constraint of the global risk scheduling state, the responsibility weight components corresponding to each violation generation mechanism in the responsibility weight vector are read to characterize the degree of responsibility participation of each sub-event generation unit in the current risk state. Combined with the coupled event state, the activation state and inhibition state of each sub-event generation unit in the current event generation path are determined, and the set of sub-event generation units participating in the current event generation is selected accordingly. The path-level convergence processing is performed on the modulation event intensity corresponding to the selected sub-event generation units. During the convergence process, the path contribution of each modulation event intensity is weighted and evaluated according to the responsibility weight vector to generate a violation determination contribution result that characterizes the degree of influence of the current event generation path on the violation determination. In the judgment-driven event generation mechanism, the event generation path is selected based on the violation judgment contribution result. The violation judgment contribution result is compared with the preset judgment interval to determine the event generation path status corresponding to the current event occurrence time. Different judgment intervals correspond to different sub-event generation units participating in the combination method. When the result of the violation judgment crosses the preset path switching threshold, the event generation path switching process is triggered in the judgment-driven event generation mechanism to change the combination structure of the sub-event generation units participating in the current event generation in the sub-event generation unit set. Based on the event generation path state selected by the decision driver, the modulation event intensity is reweighted by the decision driver to generate the decision event generation result, and the decision event generation result is output to the event generation and decision framework. The decision-driven reweighting process specifically involves: based on the event generation path state selected by the decision driver, determining the set of sub-event generation units corresponding to the current event generation path; for unselected event generation path states, performing generation suppression processing on the corresponding sub-event generation units to prevent their corresponding modulation event intensity from participating in the calculation of the current decision event generation result; for selected event generation path states, reading the modulation event intensity corresponding to each sub-event generation unit in that path state, and performing path consistency reweighting processing on the modulation event intensity according to the responsibility component in the responsibility weight vector corresponding to that event generation path state; after completing the path consistency reweighting, performing convergence processing on the modulation event intensity participating in the current event generation path to generate a decision event generation result consistent with the current event generation path state.

[0027] In this embodiment, the generation of the intelligent identification result of the violation includes: In the event generation and judgment framework, a joint reasoning process is constructed. The joint reasoning process serves as a reasoning structure based on mechanism consistency constraints and is used to perform mechanism consistency verification and conflict resolution on the event generation results. The sequence of violations, the responsibility weight vector, the violation risk status representation, and the judgment event generation results are input into the joint reasoning process. During the joint reasoning process, the judgment event generation results are subjected to mechanism consistency weighted convergence based on the responsibility weight vector. The mechanism consistency weighted convergence aims to maintain the consistency of the violation generation mechanism in the time evolution process and is used to exclude generation results with mechanism conflicts within the same event occurrence time range. During the mechanism consistency weighted aggregation process, based on the time sequence of events in the input structure of violation event sequence, the sequence consistency check is performed on the violation generation mechanism labels corresponding to the event generation results. When the violation generation mechanism labels corresponding to adjacent event occurrence times are inconsistent in the time evolution relationship, it is determined that there is a mechanism conflict. When a mechanism conflict is detected, the conflict state is resolved based on the conflict resolution rules. The conflict resolution rules include: performing suppression and adjustment processing on the participation weight of the corresponding violation generation mechanism in the judgment event generation result with mechanism conflict based on the responsibility weight vector, and performing risk redistribution processing on the risk contribution of the corresponding violation generation mechanism based on the violation risk state representation, so that the judgment event generation result after resolution is consistent in time sequence and mechanism participation, so as to reduce the impact of mechanism conflict on the joint reasoning result; After completing the mechanism consistency weighted aggregation and conflict resolution processing, based on the judgment event generation results after consistency verification and conflict resolution, a violation behavior intelligent recognition result is generated. The violation behavior intelligent recognition result includes a violation judgment label and a violation generation mechanism label. The process of generating intelligent identification results for violations includes: during joint reasoning, based on the judgment event generation results after consistency verification and conflict resolution, determining the state of the effective event generation path corresponding to the current event occurrence time; under the constraint of the effective event generation path state, parsing the sub-event generation unit participation structure retained in the judgment event generation results, extracting the violation generation mechanism that has the main generation contribution in the event generation path; based on the extracted violation generation mechanism and its corresponding judgment event generation results, determining the violation judgment result corresponding to the current event occurrence time; and combining and mapping the determined violation judgment result with the corresponding violation generation mechanism to form an intelligent identification result for violations containing violation judgment labels and violation generation mechanism labels.

[0028] In this embodiment, the feedback update process includes: The intelligent identification result of the violation is input into the feedback update processing module in the event generation and judgment framework. In the feedback update processing module, the feedback constraint information corresponding to the current event occurrence time is determined based on the violation judgment label and the violation generation mechanism label. The feedback constraint information is used to characterize the deviation relationship between the current event generation and judgment result and the historical generation behavior. Based on feedback constraint information, the dynamic responsibility allocation mechanism is updated by adjusting the responsibility components corresponding to each violation generation mechanism in the responsibility weight vector. This allows the responsibility weight vector to gradually weaken the degree of responsibility participation corresponding to violation generation mechanisms that are inconsistent with the violation judgment results in subsequent event occurrences, and to strengthen the degree of responsibility participation corresponding to violation generation mechanisms that are consistent with the violation judgment results. Based on feedback constraint information, a state transition update process is performed on the update process of the violation risk status representation. By correcting the state transition direction of the violation risk status representation in the time recursion process, the violation risk status representation corresponding to the time of subsequent events can reflect the risk evolution trend after judgment and confirmation. Based on feedback constraint information, the judgment path update process is performed on the judgment-driven event generation mechanism. By adjusting the mapping relationship between the event generation path state and the judgment contribution interval, the event generation path selected by the judgment driver in the subsequent event generation process is consistent with the confirmed violation generation mechanism. After completing the feedback update processing of the dynamic responsibility allocation mechanism, the violation risk status representation update process, and the judgment-driven event generation mechanism, an updated event generation and judgment framework is generated, and the updated event generation and judgment framework is used for the subsequent generation, judgment, and joint reasoning process of violation events.

[0029] Example 1: To verify the feasibility of this invention in practice, it was applied to the intelligent identification of traffic violations in a comprehensive urban regulatory scenario. In this scenario, the regulatory targets cover road traffic participants, operators within key regulatory areas, and continuously regulated production and business entities. Related violations are characterized by high frequency, uneven temporal distribution, and significant impact from enforcement actions. Traditional methods based on rule thresholds or single-sample judgments are prone to instability in identifying repeat violations and insufficient characterization of the evolutionary trends of violations in this scenario, making it difficult to support continuous monitoring and risk warning requirements.

[0030] In practical applications, the system continuously collects and forms a historical behavior data set, which includes the identifier of the object to be identified, the time of the event, the spatial location of the event, the environmental characteristics of the event, and law enforcement handling information. Through data preprocessing, a sequence of violation events is formed as an input structure, enabling the same object's violations occurring at different times to be uniformly represented in a time-series manner. Based on this, the system calls a neural time-point process hybrid model to construct an event generation and judgment framework, modeling violations as time-dependent event processes rather than independent sample points, thereby characterizing the temporal tendency of violations.

[0031] During the operation of the event generation and judgment framework, the set of sub-event generation units calculates the intensity of mechanistic events for different violation generation mechanisms. A dynamic responsibility allocation mechanism, combining the changes in event occurrence time intervals, event density, and law enforcement handling information reflected in the violation event sequence input structure, generates a responsibility weight vector and weights and modulates the intensity of the mechanistic events, enabling the system to reflect the inhibitory effect of law enforcement handling on the risk of subsequent violations. Simultaneously, the system generates a violation risk state representation based on the modulated event intensity and applies this risk state uniformly to multiple sub-event generation units through state coupling relationships, ensuring consistency in risk assessment results across different violation generation mechanisms.

[0032] In the subsequent judgment process, the judgment-driven event generation mechanism calculates the violation judgment contribution result based on the violation risk status representation, responsibility weight vector, and coupled event status. Based on this, it performs judgment-driven reweighting on the modulation event intensity to form the judgment event generation result. During joint inference, the system performs mechanism consistency weighted convergence and sequence consistency verification on the judgment event generation result. When different violation generation mechanisms conflict in time sequence, conflict resolution reduces the involvement of conflicting mechanisms, ultimately outputting the intelligent identification result of the violation behavior. Based on this result, the event generation and judgment framework is updated, enabling the subsequent identification process to continuously adapt to changes in actual violation behavior.

[0033] During the experimental verification process, sample data of violations generated from continuous operation were selected for comparative testing. The method of this invention was compared with traditional methods based on rule thresholds and single time series models. Multiple performance indicators were statistically analyzed, and the following experimental results were obtained.

[0034] Table 1. Comparative Experiment Results of Intelligent Recognition Methods for Traffic Violations

[0035] As can be seen from the data in Table 1, traditional rule-based judgment methods exhibit low overall accuracy and high-risk violation recall rates. This is primarily because these methods fail to utilize the temporal correlations between violations and cannot reflect the impact of enforcement actions on subsequent behavior. While single-time-series model methods improve time-dependent modeling capabilities to some extent, their limited improvement in stability for identifying repeat violations and consistency across consecutive time periods is due to a lack of differentiation of multiple violation generation mechanisms and a unified responsibility evolution mechanism.

[0036] In comparison, the method of this invention improves the accuracy of violation identification to 91.4%, and exhibits more stable performance in terms of the stability of repeated violation identification and the consistency of identification over continuous time periods. This result demonstrates that through the synergistic effect of the event generation and judgment framework, the dynamic responsibility allocation mechanism, and the representation of violation risk status, the system can more accurately characterize the risk structure of violation behavior evolving over time. Furthermore, regarding the misjudgment rate after enforcement action, the misjudgment rate of the method of this invention decreases to 9.4%, indicating that the time evolution mechanism of the responsibility weight vector effectively reduces the problem of misjudgment in the short term after enforcement action. Without significantly increasing computational complexity, this invention achieves unified modeling of the time evolution characteristics of violation behavior, the synergistic effect of multiple violation generation mechanisms, and the impact of enforcement action feedback. This reasonably improves the stability, consistency, and practicality of the intelligent violation identification results, verifying the feasibility and effectiveness of this invention in practical application scenarios.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent identification of traffic violations based on big data analysis using AI, characterized in that, Includes the following steps: Obtain a set of historical behavior data, perform data preprocessing, and obtain the input structure of the violation event sequence; The neural time-point process hybrid model is used as the basis for time event modeling. An event generation and judgment framework is constructed by combining the input structure of violation event sequence and calculating the set of event intensity of generation mechanism. In the event generation and judgment framework, a dynamic responsibility allocation mechanism is constructed. The input structure is a sequence of violation events, which is processed to generate a responsibility weight vector. This vector is then weighted and modulated with the set of mechanistic event intensities to obtain the modulated event intensity. The system receives the input structure of the sequence of violation events and the intensity of the modulated events, performs a recursive update of the state, generates a representation of the violation risk state, constructs the state coupling relationship, and performs a coupled update of the state. A decision-driven event generation mechanism is constructed, the contribution result of violation determination is calculated, and the intensity of the modulated event is subjected to decision-driven reweighting to obtain the result of the event generation. Input the sequence of violations into the structure, the responsibility weight vector, the violation risk status representation, and the judgment event generation result into the joint reasoning process, perform mechanism consistency weighted convergence and sequence consistency verification, and generate intelligent recognition results of violations. The event generation and judgment framework is updated based on the intelligent recognition results of violations.

2. The AI-powered intelligent identification method for traffic violations based on big data analysis according to claim 1, characterized in that, The generation of the input structure for the violation event sequence includes: Obtain the historical behavior data set from the violation monitoring system; Aggregation processing is performed on data entries belonging to the same object identifier in the historical behavior data set, and they are organized into object-level behavior data subsets. In the object-level behavioral data subset, event-level data construction processing is performed to construct each data entry with a clear event occurrence time into an event-level data unit. Each event-level data unit contains the corresponding event occurrence time, event spatial location identifier, event environmental state characteristics, and law enforcement handling information. Perform sequential consistency verification and outlier removal on the event-level data units to obtain the event-level data unit sequence; Perform missing data completion processing on the event-level data unit sequence to obtain an event-level data unit sequence with complete fields; Perform multi-field alignment processing on the complete event-level data unit sequence to form a time-aligned event-level data unit sequence; Time-aligned event-level data unit sequences are processed according to a preset time granularity, and the time interval between event occurrences and the change in event density are calculated to organize and generate the input structure of the violation event sequence.

3. The AI-powered intelligent identification method for traffic violations based on big data analysis according to claim 1, characterized in that, The generation of the set of mechanistic event intensity includes: In the event generation and judgment framework, a neural time point process hybrid model is used as the basis for time event modeling, and a set of sub-event generation units is constructed, with each sub-event generation unit corresponding to a violation generation mechanism. The sequence of violations is input into each sub-event generation unit. Based on the time-point process modeling mechanism of the neural time-point process hybrid model, and combined with the sequence of violations, time event modeling processing is performed. In each sub-event generation unit, based on the time event modeling processing results, the historical impact aggregation and weighted accumulation of the events that have occurred in the input structure of the violation event sequence are performed to obtain the historical impact vector, and the mechanism mapping calculation is performed to obtain the mechanism event intensity. The output mechanistic event intensities are aggregated to generate a set of mechanistic event intensities.

4. The AI-powered intelligent identification method for traffic violations based on big data analysis according to claim 1, characterized in that, The generation of the modulation event intensity includes: A dynamic responsibility allocation mechanism is constructed within the event generation and judgment framework, and a corresponding responsibility status representation is established for each violation generation mechanism. The sequence of violations is input into a dynamic responsibility allocation mechanism, which inputs the time interval between events, changes in event density, spatial location identifiers of events, environmental characteristics of events, and law enforcement information into the structure. Time-based responsibility recursion is then performed to generate a responsibility weight vector for the current moment. In the time-based responsibility derivation process, the changing trend of the time interval between events is used as the basis for time modulation of responsibility growth and decay, and law enforcement information is used as the triggering condition for responsibility inhibition. The delay inhibition effect is introduced, responsibility inhibition adjustment is performed, and a responsibility weight vector is generated by the combined effect of time evolution and law enforcement inhibition. Normalize the responsibility weight vector after time responsibility recursion processing and write it into the internal state of the sub-event generation unit set; In the set of sub-event generation units, the normalized responsibility weight vector corresponding to the current moment is read, and it is combined with the set of mechanism event intensity to perform weighted modulation processing, and responsibility evolution modulation is performed to obtain the modulated event intensity.

5. The AI-powered intelligent identification method for traffic violations based on big data analysis according to claim 1, characterized in that, The generation and coupled update of the violation risk status representation include: Within the event generation and judgment framework, a process for generating a representation of violation risk status is constructed. The sequence of violation events and the intensity of modulation events are input into the violation risk status representation generation process in chronological order of event occurrence. Based on the violation risk status representation corresponding to the previous event and the intensity of modulation events corresponding to the current event, the status is recursively updated to generate the violation risk status representation corresponding to the current event. A state coupling relationship is constructed based on the violation risk status representation corresponding to the current event. The state coupling relationship is a state coupling mapping between the violation risk status representation and the set of sub-event generation units. In the state coupling relationship, state scheduling processing is performed on each sub-event generation unit based on the violation risk state representation; The state coupling relationship and the modulation event intensity are input into the set of sub-event generation units. Based on the activation and inhibition states of each sub-event generation unit, state coupling update is performed to obtain the coupled event state.

6. The AI-powered intelligent identification method for traffic violations based on big data analysis according to claim 1, characterized in that, The generation of the determination event result includes: Within the event generation and judgment framework, a judgment-driven event generation mechanism is constructed; The violation risk status representation, responsibility weight vector, coupled event status and modulation event intensity are input into the judgment-driven event generation mechanism to perform judgment contribution calculation and generate violation judgment contribution results. Based on the contribution results of violation determination, the event generation path is selected to determine the event generation path status corresponding to the current event occurrence time. When the result of the violation judgment crosses the preset path switching threshold, the event generation path switching process is triggered, which changes the combination structure of the sub-event generation units participating in the current event generation in the sub-event generation unit set. Based on the event generation path state selected by the decision driver, the modulation event intensity is reweighted by the decision driver to generate the decision event generation result.

7. The AI-powered intelligent identification method for traffic violations based on big data analysis according to claim 1, characterized in that, The generation of the intelligent identification results of the violation includes: Within the event generation and decision-making framework, a joint reasoning process is constructed; During joint reasoning, the consistency of the execution mechanism for generating results of judgment events is weighted and converged based on the responsibility weight vector during the joint reasoning process. During the mechanism consistency weighted aggregation process, based on the time sequence of events in the input structure of the violation event sequence, the order consistency check is performed on the violation generation mechanism label corresponding to the judgment event generation result to determine mechanism conflicts. When a mechanism conflict is detected, a resolution process is performed based on the conflict resolution rules; After completing the mechanism consistency weighted aggregation and conflict resolution processing, the intelligent identification result of the violation behavior is generated based on the judgment event generation result after consistency verification and conflict resolution.

8. The AI-powered intelligent identification method for traffic violations based on big data analysis according to claim 1, characterized in that, The feedback update process includes: The intelligent identification results of violations are input into the feedback update processing module in the event generation and judgment framework, and the feedback constraint information corresponding to the current event occurrence time is determined based on the violation judgment label and the violation generation mechanism label. Based on feedback constraint information, the dynamic responsibility allocation mechanism is processed to update the responsibility weights. The process of updating the status of violation risk is subject to status transition update processing. Perform decision path update processing on the decision-driven event generation mechanism; After completing the feedback update processing of the dynamic responsibility allocation mechanism, the violation risk status representation update process, and the judgment-driven event generation mechanism, the updated event generation and judgment framework is generated.

Images