Road safety management system

Abstract

The invention relates to the technical field of road management, in particular to a road safety management system, which comprises a data acquisition module used for acquiring road risk data; the data preprocessing module is used for preprocessing the road risk data; the risk analysis and early warning module is used for performing risk assessment on the actual highway risk data according to the risk assessment model; the security check strategy determination module is used for determining a corresponding security check strategy according to the risk early warning information; the security management and control module is used for checking and verifying the settlement application through a checking node corresponding to the responsible person node; the educational training module is used for generating corresponding training content according to the risk early warning information and the security check strategy; and the target examination module is used for examining and evaluating the risk disposal result in the training completion data and the safety file information through an examination and evaluation model. According to the invention, targeted assessment can be carried out on the road risk data, and the efficiency and accuracy of road safety management and control are significantly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of highway management, and particularly relates to a highway safety management system. BACKGROUND

[0002] In the field of highway safety control, with the continuous increase of highway mileage and the increasing traffic flow, highway safety is facing more complex challenges, and higher requirements for the efficiency and accuracy of highway safety management systems are put forward. At present, the existing highway safety management system has many defects in risk data evaluation and personnel management. In the aspect of highway risk data evaluation, due to the complex and changeable environment along the highway, involving road conditions, weather conditions, traffic flow and other multi-source data, and these data have the characteristics of mass, heterogeneity and dynamic change. The existing system often lacks effective data integration and analysis means, and it is difficult to deeply mine and correlate the risk data of different sources and different formats, and it is impossible to realize targeted evaluation. For example, for mountainous highways, in the rainy season, the system cannot comprehensively analyze rainfall, geological conditions, past accident data and other factors, accurately evaluate the risk probability of disasters such as landslides and mudslides, and thus cannot take effective preventive measures in advance.

[0003] In the aspect of data analysis, the system algorithm stays at the threshold judgment level, lacks correlation mining of historical accident data and environmental impact factors, cannot establish a risk evolution model, and is difficult to realize the upgrade from passive early warning to active prediction. The lack of data collection and analysis directly causes the disconnection of each link of risk control. After the early warning is triggered, due to the lack of structured risk level classification standards, the safety inspection task allocation is random, and it is difficult to accurately match the hidden danger position and inspection resources; the problems found in the inspection cannot be traced digitally, and the implementation of control measures cannot be tracked in real time; the training content is not updated dynamically combined with actual risk cases, which leads to slow improvement of the risk identification ability of practitioners; the evaluation system is not linked with the risk disposal effect, and it is difficult to form effective responsibility constraints. Each link forms a management island due to data incompatibility and standard deficiency, which seriously affects the efficiency and accuracy of highway safety control.

[0004] Chinese patent publication No. CN119296325A discloses a highway safety operation intelligent management system, which includes a traffic intelligent monitoring subsystem, an information management subsystem, an intelligent dispatching subsystem, and a safety early warning and rescue subsystem. The traffic intelligent monitoring subsystem is used to monitor the highway in real time and obtain initial monitoring information of the highway. A system information management model is constructed in the information management subsystem, and the initial monitoring information is processed by the system information management model to obtain target monitoring information. The intelligent dispatching subsystem formulates a safety dispatching scheme for the highway based on the target monitoring information. The safety early warning and rescue subsystem analyzes and predicts the safe operation of the highway in combination with the target monitoring information and the safety dispatching scheme. Although this scheme can realize real-time monitoring, information processing, and safety early warning and dispatching of the highway, it lacks targeted collection and analysis of highway risk data, and does not construct a closed-loop management mechanism from risk early warning to safety inspection, control, training, and evaluation, resulting in incomplete risk disposal process, lack of supervision and guidance for the execution of the responsible person, and reduction of the efficiency and accuracy of highway safety control. SUMMARY

[0005] To this end, the present application provides a highway safety management system to overcome the problem of lack of targeted collection and analysis of highway risk data, and lack of construction of a closed-loop management mechanism from risk early warning to safety inspection, control, training, and evaluation, resulting in incomplete risk disposal process and lack of supervision and guidance for the execution of the responsible person.

[0006] To achieve the above-mentioned purpose, the present application provides a highway safety management system, which comprises: a data acquisition module for acquiring highway risk data through a data interface of a system interaction interface; a data preprocessing module for preprocessing the highway risk data to obtain actual highway risk data; a risk analysis and early warning module for risk assessment of the actual highway risk data according to a risk assessment model to obtain risk early warning information containing risk level parameters, disposal suggestions, and audit paths, and sending the risk early warning information to corresponding audit nodes through a message queue; a safety inspection strategy determination module for determining a corresponding safety inspection strategy according to the risk early warning information and sending the safety inspection strategy to corresponding responsible person nodes through a message queue; a safety control module for receiving a case closing application of the responsible person node and auditing and verifying the case closing application through the audit node corresponding to the responsible person node to obtain an audit and verification result, and outputting corresponding safety archive information when the audit and verification result meets a preset review condition; An education and training module is configured to generate corresponding training content based on the risk early warning information and the safety inspection strategy, and to push the training content to the responsible person node through the system interaction interface, and to record the training completion data of the responsible person node; A target assessment module is configured to assess the training completion data and the risk disposal result in the safety archive information through an assessment evaluation model to obtain an assessment result, and to feed back the assessment result to the safety management and control module through a message queue.

[0007] Compared with the prior art, the application has the following advantages: In the prior art, the traffic intelligent monitoring subsystem mainly performs real-time monitoring of the expressway and obtains "initial monitoring information", but does not focus on the key object "highway risk data", resulting in generalized monitoring information and lack of targeted extraction and analysis of risk factors (such as accident hazards, equipment failures, environmental risks, etc.). The application directly focuses on core information related to risks (such as road section accident rate, equipment abnormal data, meteorological disaster warning, etc.) by using the data interface of the system interaction interface to specially collect "highway risk data" through the data collection module, avoiding interference from irrelevant data. At the same time, the data preprocessing module processes the collected raw risk data to obtain "actual highway risk data", further improving the data quality and enabling more accurate identification of potential risks in the operation of the expressway, providing a reliable data basis for subsequent risk analysis, early warning and disposal, and solving the problem of inaccurate risk identification and missing key risk points caused by data generalization in the prior art.

[0008] In the prior art, although a safety early warning and rescue subsystem is included, the overall process only covers monitoring, information processing, dispatching scheme development and safety condition prediction, and does not form a complete closed loop from risk early warning to specific disposal, execution supervision and effect evaluation, such as missing links of how to implement specific inspection measures after early warning, how to verify the execution of the responsible person, and how to assess the disposal effect. The application generates risk early warning information including "risk level parameters, disposal suggestions and audit paths" through the risk analysis and early warning module, and sends it to the audit node. The safety inspection strategy determination module develops an inspection strategy based on the early warning information and sends it to the responsible person node. The safety management and control module receives the case closing application and verifies it through the audit node, and outputs the safety archive. The education and training module generates training content based on the early warning information and the inspection strategy and pushes it to the responsible person node, and records the training data. The target assessment module assesses the training data and the risk disposal result through an assessment evaluation model, and feeds back to the safety management and control module. This closed loop process covers the whole chain of risk early warning, strategy development, execution supervision, training guidance and effect evaluation, ensuring that each link has a clear responsible subject and execution standard, solving the problem of incomplete process and lack of supervision and guidance of the execution of the responsible person in the prior art, and improving the systematicness and effectiveness of risk disposal.

[0009] In the prior art, although there is information interaction between subsystems, efficient information transmission mechanisms (such as message queues) are not explicitly used to ensure accurate pushing and timely response of key information, resulting in information transmission delay or unclear responsibility subject. In the present application, the risk analysis and early warning module sends risk early warning information to the corresponding audit node through the message queue, the safety inspection policy determination module sends the inspection policy to the corresponding responsible person node through the message queue, and the safety management module verifies the case closing application through the audit node. As an asynchronous communication mechanism, the message queue can ensure efficient and reliable transmission of information between different nodes, avoiding information loss or delay due to network delay or system failure. At the same time, through the explicit pointing of the "corresponding audit node" and "corresponding responsible person node", the accurate positioning of the responsibility subject is realized, ensuring that the risk early warning information can be timely delivered to the audit link and the inspection policy can be accurately conveyed to the execution responsible person, avoiding the defects of low information transmission efficiency and unclear responsibility subject in the prior art, and improving the timeliness of risk disposal and the definiteness of responsibility implementation.

[0010] In the prior art, the safety early warning and rescue subsystem only focuses on emergency disposal after the event occurs, and does not involve the ability improvement and execution effect evaluation of the responsible person (such as maintenance personnel and management personnel), resulting in the possibility of repeated occurrence of similar risks. In the present application, the education and training module generates training content (such as risk disposal specification, equipment operation guide, etc.) according to the risk early warning information and the safety inspection policy, and pushes it to the responsible person node through the system interaction interface, while recording the "training completion data"; the target assessment module assesses the training completion data and the "risk disposal result" in the safety file information through the assessment evaluation model, and feeds back the assessment result to the safety management module, avoiding the problem of inadequate risk disposal due to insufficient ability or non-standard execution of the responsible person, and realizing the continuous optimization of risk disposal ability.

[0011] Further, the data preprocessing module performs outlier filtering and missing value interpolation processing on the highway risk data to obtain an intermediate data set, performs feature extraction on the intermediate data set according to a principal component analysis method to obtain a feature dimension reduction data set, and performs state estimation and synchronous correction on the feature dimension reduction data set according to an adaptive Kalman filtering algorithm to obtain actual highway risk data.

[0012] In the scheme, the data preprocessing module is used to filter abnormal values and interpolate missing values of the highway risk data, effectively eliminate invalid data, complete incomplete information, ensure data quality, and obtain high-quality intermediate data set; principal component analysis is used for feature extraction, which can reduce data dimension, reduce calculation amount and highlight key features, and obtain feature dimension reduction data set; adaptive Kalman filter algorithm is used for state estimation and synchronous correction, which can accurately capture data dynamic changes and eliminate error interference, and finally obtain accurate and reliable actual highway risk data.

[0013] Further, the risk analysis and early warning module comprises: a risk assessment model construction unit for constructing a risk assessment model according to a preset historical data set and a risk parameter correction factor; a risk assessment unit for inputting actual highway risk data into the risk assessment model for feature extraction, parameter calculation and risk level judgment analysis, and obtaining risk warning information containing risk level parameters, disposal suggestions and audit paths; a warning information sending unit for triggering the corresponding connection mechanism according to the risk warning information, and encapsulating the risk warning information into a warning information data packet conforming to the preset message queue format requirement based on the connection mechanism and the specified node identifier in the audit path, and sending the warning information data packet to the corresponding audit node through the message queue.

[0014] In the scheme, the preset historical data set refers to the past highway risk related data set collected and arranged in advance, which contains information such as various risk event occurrence and influencing factors, providing basic data support for constructing the risk assessment model; the risk parameter correction factor refers to the numerical value or variable used to adjust and correct the parameters in the risk assessment model; the risk assessment model refers to the model constructed according to the preset historical data set and the risk parameter correction factor, which is used for feature extraction, parameter calculation and risk level judgment analysis of actual highway risk data; the connection mechanism refers to the rules and ways of realizing the transmission of risk warning information according to the specified node identifier in the audit path in the message queue; the preset message queue format requirement refers to the format standard that should be followed when encapsulating the risk warning information into a data packet.

[0015] The risk assessment model is constructed by the preset historical data set and the risk parameter correction factor, which can accurately evaluate the highway risk combined with historical experience and actual situation; the model is used to obtain risk warning information containing multiple elements, which provides comprehensive reference for disposal; the risk warning information triggers the connection mechanism in the message queue, and is encapsulated and sent to the corresponding audit node according to the preset message queue format requirement, which ensures the fast, accurate and standardized transmission of information, and improves the efficiency of highway risk warning and disposal.

[0016] Further, the risk assessment model construction unit acquires a preset historical data set and pre-processes it to obtain a structured historical data set meeting modeling requirements, dynamically adjusts feature weights in the structured historical data set according to a risk parameter correction factor, obtains a model parameter set containing corrected feature values, and inputs the model parameter set and the structured historical data set into a logistic regression algorithm framework, iteratively optimizes model coefficients in the logistic regression algorithm framework by a gradient descent method, and outputs a risk assessment model when a loss function value in the logistic regression algorithm framework is lower than a preset loss threshold.

[0017] In this scheme, the logistic regression algorithm framework refers to a statistical model framework for solving binary classification problems, which predicts by establishing a logical relationship between input features and output results (usually 0 or 1), for example, when judging whether a mail is spam, the frequency of words in the mail can be used as input features, and 0 represents non-spam and 1 represents spam, the relationship between these features and the mail category is analyzed by using the logistic regression algorithm framework, the gradient descent method refers to an optimization algorithm for finding the minimum value of a function, which gradually approaches the minimum value by constantly adjusting the parameters in the opposite direction of the function gradient, and the preset loss threshold refers to a value preset for determining whether the loss function in the training process of the logistic regression algorithm framework reaches the convergence standard, for example, the preset loss threshold is set to 0.1.

[0018] By acquiring a preset historical data set and pre-processing it to obtain a structured historical data set, a standardized data basis is provided for modeling; according to the risk parameter correction factor, the feature weights are dynamically adjusted to obtain a model parameter set containing corrected feature values, so that the model is more in line with the actual situation; the model parameter set and the structured historical data set are input into the logistic regression algorithm framework, and the model coefficients are iteratively optimized by using the gradient descent method, which can continuously adjust the model parameters to improve the performance; when the loss function value is lower than the preset loss threshold, the risk assessment model is output, ensuring that the output model has high accuracy and can accurately assess the road risk.

[0019] Further, the risk assessment unit inputs the actual road risk data into the risk assessment model for feature extraction to obtain a road risk feature vector, calculates the corresponding risk probability value according to the road risk feature vector, compares the risk probability value with a preset risk level threshold interval to determine the corresponding risk level parameter, and calls a preset disposal suggestion library and an audit path template based on the risk level parameter to generate risk warning information containing the risk level parameter, disposal suggestions and audit paths.

[0020] In the scheme, the preset risk level threshold interval refers to a series of numerical ranges pre-set for dividing different levels of road risks, for example, the risk level is divided into three levels of low, medium and high, the threshold interval corresponding to the low risk may be 0-0.3, the medium risk is 0.3-0.7, and the high risk is 0.7-1. By comparing the calculated risk probability value with these intervals, the risk level is determined. The preset disposal suggestion library refers to a database pre-stored with measures and suggestions for different road risk levels, for example, it is suggested to strengthen daily patrol for low risk, arrange special inspection for medium risk, and immediately take control measures for high risk, etc. The audit path template refers to a pre-set audit process template for risk early warning information of different risk levels, which clearly defines which nodes the information needs to pass through for audit.

[0021] By inputting the actual road risk data into the risk assessment model to extract the feature vector and calculate the risk probability value, the road risk can be accurately quantified. By comparing the risk probability value with the preset risk level threshold interval, the risk level parameter can be determined, and the risk degree can be clearly divided. Based on the risk level parameter, the preset disposal suggestion library and the audit path template can be called to quickly generate risk early warning information containing comprehensive information, thereby improving the efficiency and standardization of road risk response.

[0022] Further, the early warning information sending unit analyzes the risk early warning information to obtain connection request data containing a target node identifier, obtains a connection mechanism matched with the target node identifier in a preset message queue connection mechanism configuration table according to the connection request data, establishes a message transmission channel with the corresponding audit node through the connection mechanism, serializes and encapsulates the risk early warning information according to the preset message queue format requirement through the message transmission channel, generates an early warning information data packet, and pushes the early warning information data packet to the corresponding audit node through the routing rule of the message queue.

[0023] In the scheme, the preset message queue connection mechanism configuration table refers to a table pre-set to record the mapping relationship between different target node identifiers and corresponding connection mechanisms. The message transmission channel refers to a channel established between the early warning information sending unit and the corresponding audit node for data transmission according to the matched connection mechanism.

[0024] By analyzing the risk early warning information to obtain the target node identifier and matching the appropriate connection mechanism in the preset message queue connection mechanism configuration table, the message transmission channel with the corresponding audit node can be quickly and accurately established. The risk early warning information is serialized and encapsulated according to the preset message queue protocol format to generate an early warning information data packet, which can ensure the uniformity and standardization of the information format. Then, the data packet is pushed to the corresponding audit node through the routing rule of the message queue, realizing efficient, accurate and orderly transmission of the risk early warning information, and improving the efficiency and reliability of information transmission.

[0025] Further, the security check policy determination module comprises: a policy determination unit configured to parse the risk early warning information according to a preset parsing rule to obtain structured risk data, and match a security check policy corresponding to the structured risk data in a preset security check policy generation rule set; a policy sending unit configured to encapsulate the security check policy as a message object, and send the message object to a corresponding person in charge node through a routing rule of a message queue.

[0026] In the scheme, the preset parsing rule refers to a rule system for analyzing and processing risk early warning information, and the preset security check policy generation rule set refers to a rule set that contains a mapping relationship between various types of structured risk data and corresponding security check policies.

[0027] The risk early warning information is parsed according to the preset parsing rule, so that complex information can be converted into structured risk data, improving the standardization and availability of information. The corresponding security check policy is matched in the preset security check policy generation rule set, so that appropriate measures can be quickly and accurately formulated for different risk situations. The security check policy is encapsulated as a message object and sent to the corresponding person in charge node through the message queue routing rule, ensuring that the policy is timely and accurate, and improving the efficiency and pertinence of highway safety inspection.

[0028] Further, the policy determination unit parses the risk early warning information by using a semantic analysis algorithm in the preset parsing rule to obtain structured risk data containing a risk type identifier, a risk occurrence position coordinate and a risk level, and uses the structured risk data as a query condition to search a rule index table of the preset security check policy generation rule set to obtain a policy parameter set matching the risk type identifier, the risk occurrence position coordinate and the risk level, and combines the policy parameter set with a pre-stored security policy template to generate a structured security check policy containing a security check path planning scheme, a person in charge distribution identifier and a disposal time limit value.

[0029] In the scheme, the semantic analysis algorithm refers to an algorithm for analyzing the semantics of text and understanding the meanings of each word and sentence in the risk early warning information, and identifying key information. The rule index table refers to a table in the preset security check policy generation rule set for quickly searching the policy parameter set. The policy parameter set refers to a set of parameters related to a specific risk situation, containing specific information required for generating a security check policy, such as the inspection method, inspection frequency, etc. for a certain risk type at a specific location. The pre-stored security policy template refers to a general framework of the security check policy, which specifies the basic elements and structure that the security check policy should contain, such as the position and format of the security check path planning scheme, the person in charge distribution identifier and the disposal time limit value.

[0030] The risk early warning information is parsed by a semantic parsing algorithm, key information can be accurately extracted, structured risk data is obtained, and the accuracy of information processing is improved; the strategy parameter set matched with the structured risk data is quickly searched by using a rule index table, the search time is saved, and the efficiency is improved; the strategy parameter set is combined with the pre-stored safety policy template, a structured safety inspection policy containing comprehensive information is quickly generated, the safety inspection path, the person in charge and the disposal time limit are clear, the safety inspection work is more targeted and operable, and the orderly development of the highway safety inspection work is effectively ensured.

[0031] Further, the safety management module comprises: The case closing application auditing unit is configured to obtain a case closing application through an interface corresponding to the person in charge node, determine a corresponding auditing process engine through an auditing node corresponding to the person in charge node, and perform auditing and verification on the case closing application based on a preset verification rule in the auditing process engine to obtain an auditing and verification result containing a verification state and a specific verification item result. The safety archive generation unit is configured to, when the verification state is a pass state and a preset review condition is met, call a time stamp service to obtain a current system time as an archive time mark, and sign the case closing application in the pass state and meeting the preset review condition and the time stamp through a digital signature algorithm to generate safety archive information containing original application data, the time stamp and the digital signature.

[0032] In the scheme, the auditing process engine refers to a software component for driving and managing the auditing process according to preset rules and logic to control the flow sequence and operation steps of the case closing application between different auditing nodes, the preset verification rule refers to a specific standard and condition for auditing the case closing application, the preset review condition refers to an additional condition for further determining whether the safety archive can be generated on the basis of the pass state, and the digital signature algorithm refers to a mathematical algorithm for generating and verifying a digital signature.

[0033] The auditing process is determined through the auditing process engine, the auditing sequence and operation of the case closing application can be standardized, and the auditing efficiency is improved; the case closing application is audited and verified based on the preset verification rule, the completeness and compliance of the application content can be ensured; when the verification state is a pass state and the preset review condition is met, the archive time mark is obtained by calling the time stamp service, the archive generation time can be accurately recorded; the case closing application and the time stamp are signed through the digital signature algorithm, the safety archive information containing the original application data, the time stamp and the digital signature is generated, and the authenticity and integrity of the archive information are ensured.

[0034] Further, in the case closing application auditing unit, the process of auditing and verifying the case closing application based on the preset verification rule in the auditing process engine comprises: The multi-modal information conversion unit based on the audit process engine converts and analyzes the closed application, obtains a verification object set, and establishes an attribute mapping relationship between the preset verification rules and the verification object set; The hierarchical verification mechanism and the attribute mapping relationship of the audit process engine are used to perform rule item matching on the verification object set, to obtain a rule item matching result, and cross-verification is performed on the associated weights between the preset verification rules based on the rule item matching result, to generate intermediate verification data containing the matching states of each verification item and associated conflict information; The intermediate verification data is input into the result synthesis module of the audit process engine for conflict resolution, to obtain an audit verification result containing the verification state and the specific verification item result.

[0035] In the scheme, the verification object set refers to a set of objects obtained by the multi-modal information conversion unit based on the audit process engine after converting and analyzing the closed application, the attribute mapping relationship refers to the corresponding relationship between the preset verification rules and the verification object set established in advance, and the hierarchical verification mechanism refers to the verification method of verifying the verification object set in a certain hierarchical order in the audit process engine.

[0036] The multi-modal information conversion unit obtains the verification object set and establishes the attribute mapping relationship, which can accurately locate the corresponding relationship between the verification rules and the key content of the closed application, and improve the pertinence of verification; the hierarchical verification mechanism and the attribute mapping relationship of the audit process engine are used for rule item matching, which can comprehensively and systematically check the closed application; cross-verification is performed on the associated weights between the preset verification rules based on the rule item matching result, which can find conflicts between rules; the intermediate verification data is input into the result synthesis module for conflict resolution, and finally the audit verification result containing the verification state and the specific verification item result is obtained, which ensures the accuracy and integrity of the audit and improves the quality and efficiency of the closed application audit. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 FIG. 1 is a structural schematic diagram of a highway safety management system according to an embodiment of the present application; Figure 2 FIG. 2 is a structural schematic diagram of a risk analysis and early warning module according to an embodiment of the present application; Figure 3 FIG. 3 is a structural schematic diagram of a safety inspection strategy determination module according to an embodiment of the present application; Figure 4 FIG. 4 is a structural schematic diagram of a safety management and control module according to an embodiment of the present application. DETAILED DESCRIPTION

[0038] The following will be further described in detail through specific embodiments: Please refer to Figure 1As shown, it is a structural schematic diagram of a highway safety management system according to an embodiment of the present application, comprising: a data collection module, configured to collect highway risk data through a data interface of a system interactive interface; a data preprocessing module, configured to preprocess the highway risk data to obtain actual highway risk data; a risk analysis and early warning module, configured to perform risk assessment on the actual highway risk data according to a risk assessment model to obtain risk early warning information containing a risk level parameter, a disposal suggestion and an audit path, and send the risk early warning information to a corresponding audit node through a message queue; a safety inspection strategy determination module, configured to determine a corresponding safety inspection strategy according to the risk early warning information, and send the safety inspection strategy to a corresponding responsible person node through the message queue; a safety management module, configured to receive a case closing application of the responsible person node, and perform audit verification on the case closing application through the audit node corresponding to the responsible person node to obtain an audit verification result, and output corresponding safety archive information when the audit verification result meets a preset review condition; an education and training module, configured to generate corresponding training content according to the risk early warning information and the safety inspection strategy, push the training content to the responsible person node through the system interactive interface, and record training completion data of the responsible person node; a target examination module, configured to perform examination and evaluation on the training completion data and a risk disposal result in the safety archive information through an examination and evaluation model to obtain an examination result, and feed back the examination result to the safety management module through the message queue.

[0039] Specifically, when the data collection module collects highway risk data through the data interface of the system interaction interface, it needs to clearly define the communication protocol and data format of the data interface of the interaction interface, such as using the RESTful interface of the HTTP protocol, and transmitting data in the JSON format. In the data collection module, a request component matching the interface is constructed, and the request header is set to include authentication information, such as an API key, to ensure legal access. The required collected highway risk data fields, such as road section number, risk type, occurrence time, etc., are defined in the request body. After the data collection module initiates the request, it receives the response data returned by the data interface. Through the data parsing component, the response content is parsed according to the JSON format, and the key risk data is extracted. For the parsed data, validity check is performed, such as checking whether the data is complete and whether the value is within a reasonable range. If the data is valid, it is stored in the local database or cache for subsequent analysis and processing; if the data is invalid, error information is recorded and a retry mechanism or notification to the management personnel is triggered. At the same time, the data collection module needs to have the ability to handle exceptions, and when there is a network failure or interface exception, it automatically retries or switches to a backup interface to ensure the continuity and stability of data collection, thereby completing the process of collecting highway risk data through the data interface of the system interaction interface.

[0040] In this embodiment, the highway risk data includes risk data, hidden danger data and accident data. The risk data mainly reflects potential dangerous factors on the highway and their possible consequences, and usually includes: risk type: such as adverse weather (heavy rain, heavy fog, ice and snow), geological disasters (landslide, mudslide), road conditions (pits, slippery), excessive traffic flow, etc.; risk location: the specific road section where the risk occurs can be accurately identified by highway milepost or latitude and longitude coordinates, for example, "G60 Shanghai-Kunming Expressway K1205+300 to K1206+100 section"; risk level: classified according to the severity and probability of the risk, such as low risk, medium risk, high risk, to provide priority reference for subsequent response measures; risk influence range: assess the areas that may be affected by the risk, such as affecting single lane, double lane or adjacent road section; risk duration: predict the length of time the risk exists, such as adverse weather is expected to last for 2 hours, road maintenance is expected to last for 3 days. The hidden danger data focuses on the defects in highway facilities or management that may cause accidents, including: hidden danger category: including road facility hidden danger (such as blurred signs and markings, damaged guardrail, insufficient lighting), bridge and tunnel hidden danger (such as structural cracks, water seepage, poor ventilation), traffic management hidden danger (such as signal light failure, unreasonable speed limit setting) and the like; hidden danger location: also positioned by highway milepost or latitude and longitude coordinates, such as "there is a transverse crack in the web plate of the 3rd span T beam of a certain bridge, located at K85+200"; hidden danger description: detailed description of the specific performance and characteristics of the hidden danger, such as "the inclination angle of the guardrail post is 15°, and 3 connecting bolts are loose"; hidden danger discovery time: record the time when the hidden danger is first discovered, to facilitate tracking the processing progress, hidden danger processing status: divided into unprocessed, processing, processed, and record the processing measures and processing results, such as "the damaged guardrail post has been replaced, and the stability of the guardrail has been restored". The accident data records the actual traffic accident information on the highway, which is crucial for analyzing the causes of the accident and developing preventive measures, including: accident basic information: including accident time (accurate to minutes), location (highway milepost or latitude and longitude coordinates), accident type (such as collision, rollover, rear-end collision, fire, etc.); accident involved parties: record the types of vehicles involved in the accident (such as cars, trucks, buses), license plate numbers, driver information (name, gender, age, driver's license number) and whether there are casualties; accident cause: preliminary analysis of the direct and indirect causes of the accident, such as speeding, fatigue driving, illegal lane changing, slippery road, etc.; accident loss situation: statistics of the number of casualties (number of deaths and injuries), vehicle damage (minor, severe, scrap) and road facility damage (such as guardrail damage length, road damage area). Accident handling: record the on-site handling measures (such as traffic control, rescue operations), accident liability determination results and subsequent compensation.

[0041] Specifically, the data preprocessing module performs outlier filtering and missing value interpolation processing on the highway risk data to obtain an intermediate data set, performs feature extraction on the intermediate data set according to a principal component analysis method to obtain a feature dimension reduction data set, and performs state estimation and synchronous correction on the feature dimension reduction data set according to an adaptive Kalman filtering algorithm to obtain actual highway risk data.

[0042] In this embodiment, the outlier filtering and missing value interpolation processing on the highway risk data includes, for numerical indicators of the highway risk data (such as vehicle speed, accident rate, road surface humidity, etc.), first calculating the median and interquartile range (IQR) of each indicator, defining the outlier threshold as the median ± 1.5*IQR, and marking the data outside the range as outliers. For time series data, a sliding window (such as 5 time points) is used, and the outliers in the window are replaced with the window median to avoid mistakenly deleting effective mutations (such as sudden data spikes caused by sudden accidents). Analyze the data missing pattern, and for numerical data, use spatiotemporal weighted interpolation. In the time dimension, use linear interpolation of adjacent time points to fill in the missing data; in the spatial dimension (different road segments), combine the same period data of adjacent road segments, and use weighted average according to the road segment distance (the weight is inversely proportional to the distance), to fill in the missing data by considering the spatiotemporal correlation and preserving the spatiotemporal distribution characteristics of the data.

[0043] According to the principal component analysis method, the feature extraction on the intermediate data set includes calculating the covariance matrix of the standardized data, solving the eigenvalues and eigenvectors, arranging the eigenvalues in descending order, selecting the first k principal components with cumulative variance contribution rate ≥ 85% as new features, realizing dimension reduction, and preserving key risk information.

[0044] The state estimation and synchronous correction on the feature dimension reduction data set includes establishing a state space model, a state equation describing the time evolution of the risk state (such as a linear model: state = state transition matrix x previous state + process noise), and an observation equation describing the linear relationship between the dimension reduction features and the state (observation = observation matrix x state + observation noise). The prediction step estimates the current state using the state transition matrix; the update step combines the dimension reduction feature observation value to calculate the Kalman gain to correct the prediction. According to the statistical characteristics (mean, variance) of the residual (the difference between the predicted observation and the actual observation), the process noise and observation noise parameters are dynamically adjusted to adapt to the uncertainty of the highway risk (such as noise increase caused by sudden weather changes), and finally the corrected actual highway risk data is output.

[0045] For example, the intermediate data set has three indexes of vehicle speed, accident rate and road surface humidity, and the covariance matrix is calculated after standardization. The eigenvalues and eigenvectors are solved, such as eigenvalues of 5, 2 and 1, arranged in descending order. The cumulative variance contribution rate is calculated, and the cumulative contribution rate of the first two eigenvalues reaches 85% (such as (5+2) / (5+2+1)=87.5%). The eigenvectors corresponding to the first two eigenvalues are selected as new features, and the original 3D data is reduced to 2D, and the key risk information is retained. The state space model is established, the state equation is: current vehicle speed risk state = 0.8 x previous vehicle speed risk state + process noise; the observation equation is: reduced dimension feature observation value = 1.2 x vehicle speed risk state + observation noise. The state transition matrix 0.8 is used to estimate the current state in the prediction step. In the update step, the observation value is combined to calculate the Kalman gain to correct the prediction. If the residual mean and variance increase due to weather changes, the process and observation noise parameters are dynamically adjusted to adapt the algorithm to uncertainty, and the corrected vehicle speed risk data is output.

[0046] Please refer to Figure 2 As shown in FIG. 1, which is a structural schematic diagram of a risk analysis and early warning module according to an embodiment of the present application, comprising: A risk assessment model construction unit is configured to construct a risk assessment model according to a preset historical data set and a risk parameter correction factor; A risk assessment unit is configured to input actual highway risk data into the risk assessment model for feature extraction, parameter calculation and risk level judgment analysis, and obtain risk warning information containing risk level parameters, disposal suggestions and audit paths; A warning information sending unit is configured to trigger a corresponding connection mechanism according to the risk warning information, encapsulate the risk warning information into a warning information data packet conforming to the preset message queue format requirements based on the connection mechanism and the specified node identifier in the audit path, and send the warning information data packet to the corresponding audit node through the message queue.

[0047] Specifically, the risk assessment model construction unit acquires a preset historical data set and pre-processes it to obtain a structured historical data set meeting the modeling requirements, dynamically adjusts the feature weights in the structured historical data set according to a risk parameter correction factor, obtains a model parameter set containing corrected feature values, inputs the model parameter set and the structured historical data set into a logistic regression algorithm framework, iteratively optimizes the model coefficients in the logistic regression algorithm framework through the gradient descent method, and outputs a risk assessment model when the loss function value in the logistic regression algorithm framework is lower than a preset loss threshold.

[0048] Specifically, the risk assessment unit inputs the actual road risk data into the risk assessment model for feature extraction to obtain a road risk feature vector, calculates a corresponding risk probability value according to the road risk feature vector, compares the risk probability value with a preset risk level threshold interval, determines a corresponding risk level parameter, and calls a preset treatment suggestion library and an audit path template based on the risk level parameter to generate risk warning information containing the risk level parameter, the treatment suggestion, and the audit path.

[0049] Specifically, the warning information sending unit parses the risk warning information to obtain connection request data containing a target node identifier, obtains a connection mechanism matched with the target node identifier in a preset message queue connection mechanism configuration table according to the connection request data, establishes a message transmission channel with a corresponding audit node through the connection mechanism, serializes and encapsulates the risk warning information according to a preset message queue format requirement through the message transmission channel to generate a warning information data packet, and pushes the warning information data packet to the corresponding audit node through a routing rule of the message queue.

[0050] In this embodiment, in the risk assessment model construction unit, risk-related data is widely collected from past highway operations, covering risk event records under different road sections, time periods, weather, traffic flow, etc. such as accident occurrence time, location, type, loss degree, etc. The collected data is cleaned to remove errors, duplicates and invalid data, and standardized to unify the format. Then according to the modeling requirements, the data fields closely related to risk assessment are selected and integrated into a structured data set that can fully reflect the characteristics and laws of highway risks and meet the modeling requirements, as the preset historical data set. The risk parameter correction factor is an external input parameter (for example, from business rules or expert experience) used to dynamically adjust the relative importance of features in the data set. The implementation process is as follows: load the risk parameter correction factor from the configuration or external source. This risk parameter correction factor is usually a vector or mapping table, where each element corresponds to a feature in the data set (such as "weather conditions" or "traffic flow") and contains a weight adjustment coefficient (such as 0.8 to reduce weight, 1.2 to increase weight). Traverse the structured historical data set (preprocessed into table form, containing multiple feature columns and label columns). For each sample (i.e. data row), apply the correction factor to the feature value: multiply each feature value by the corresponding correction factor coefficient. For example, if the correction factor for the "accident frequency" feature is 1.5, multiply all sample "accident frequency" values by 1.5 to amplify the influence of this feature; if the factor is 0.5, reduce the influence. This is equivalent to weighting the features before model training, rather than directly modifying the model weights. After adjustment, the feature values of the data set are updated to form a "model parameter set containing corrected feature values". This model parameter set is essentially a corrected structured data set that retains the original structure but has feature values dynamically adjusted to reflect the latest changes in risk parameters. This process is dynamic because the correction factor can be updated in real time (such as monthly refresh) to ensure that the data set weights remain consistent with the current risk environment. The adjustment is completed before the data is input into the model, and changes in feature weights directly affect the importance of features in subsequent logistic regression training, but do not require modification of the algorithm itself.

[0051] In the logistic regression algorithm framework, the model parameter set (i.e., the dataset of modified feature values) and the structured historical dataset (containing labels such as whether a risk event occurred) are inputted, and the model coefficients are optimized by gradient descent method. The implementation process is as follows: the model parameter set (feature data, denoted as X) and the structured historical dataset (label data, denoted as y) are integrated into training data. X contains modified feature values, and y contains corresponding true labels (such as 0 or 1 representing whether a risk occurs or not). The logistic regression framework is instantiated, including defining the model structure (such as using the sigmoid function to map the linear combination to the probability), the loss function (such as the log loss function), and the optimization parameters (such as the learning rate). The model coefficients (weight vector) are initialized to default values (such as all zeros or small random numbers). In each iteration, the framework calculates the predicted probability (based on X and the sigmoid function) using the current model coefficients. The predicted probability is compared with the true label y, and the loss function value (such as the log loss function) is calculated, which measures the prediction error. The gradient (i.e., the partial derivative) of the loss function with respect to the model coefficients is calculated, which indicates the direction and size of the contribution of each coefficient to the error. The learning rate (preset step size) is used to scale the gradient, and the model coefficients are updated: new coefficients = old coefficients - learning rate x gradient. For example, if the gradient indicates that a certain coefficient leads to high error, the value of that coefficient is reduced. After each iteration, it is checked whether the loss function value is lower than the preset loss threshold (in this embodiment, the preset loss threshold is 0.1). If it is lower than the threshold, the iteration stops; otherwise, the next iteration is continued. At the same time, the maximum number of iterations can be set to prevent infinite loops. When the loss function value is lower than the preset loss threshold, the optimization ends, and the final model coefficients are fixed, forming a risk assessment model. The preset loss threshold can be determined by considering multiple factors: first, refer to the past experience of similar risk assessment tasks to understand the general reasonable loss range; then, according to the business requirements for assessment accuracy, if the requirement is high, set the threshold lower; combined with data quality, when the data noise is large, appropriately increase the threshold; also, through experiments, test the model performance under different thresholds, compare the evaluation indicators, and select the loss value that can make the model have better comprehensive performance in precision, recall rate, etc. as the preset loss threshold.

[0052] In the risk assessment unit, risk level parameters (such as "high", "medium", "low") are used to trigger subsequent operations, and the implementation process is as follows: The preset treatment suggestion library is a database or file (such as key-value pairs or tables) that stores mapping relationships, where the key is the risk level parameter, and the value is the corresponding treatment suggestion text (such as "high" risk corresponds to "immediate road closure for repair"). Based on the risk level parameter, the system performs a query: using the parameter as the key, retrieve the matching treatment suggestion from the library. For example, if the risk level is "medium", return "increase patrol frequency". Similarly, the preset audit path template library stores the mapping of risk levels and audit processes (such as "high" risk corresponds to "multi-level audit path"). The system uses the same risk level parameter to query this library to obtain the audit path description (such as "first reviewed by the road segment supervisor, then reported to the regional center"). Combine the risk level parameter, the retrieved treatment suggestion, and the audit path into a structured information object. This object usually uses a standard format (such as JSON or XML) and contains three fields: risk level parameter (direct input value), treatment suggestion (query result), and audit path (query result). For example, the output is { "risk_level": "high", "action_advice": "immediate road closure for repair", "review_path": "supervisor review -> center review"}. This object is the complete risk warning information.

[0053] The determination method of the preset risk level threshold interval is: analyze historical risk data, and count the distribution of risk probability values under different risk levels. Combine highway safety standards, industry norms, and actual management needs, and organize expert discussions. Experts divide different risk levels based on data analysis results and their own experience, such as low, medium, and high risk, and determine the risk probability value range corresponding to each level. After multiple demonstrations and adjustments, ensure that the threshold interval can accurately distinguish risk levels and provide a scientific basis for risk assessment.

[0054] The determination method of the preset treatment suggestion library is: collect successful treatment cases of past highway risk events, and summarize effective treatment measures under different risk levels and types. Organize experts in traffic management, safety, and other fields to systematically sort and classify treatment measures based on professional knowledge and practical experience. For various risk situations, develop detailed and operable treatment suggestions and store them in the database according to certain rules to form the preset treatment suggestion library, so that appropriate treatment suggestions can be quickly called according to the risk level.

[0055] In the early warning information sending unit, the connection request data contains the target node identifier (such as "node ID_001") for finding the connection mechanism. The implementation process is as follows: extract the connection request data (usually a field or attribute) from the risk early warning information, and the core is the target node identifier (for example, the string "audit node A"). The preset message queue connection mechanism configuration table is a structured storage (such as a database table, a configuration file, or a memory mapping) containing two columns: target node identifier and connection mechanism details. The connection mechanism details include protocol type (such as AMQP or MQTT), server address, port number, authentication information, etc. The system uses the target node identifier as the query key to perform a lookup in the configuration table. For example, traverse the "target node identifier" column of the table to find a row that exactly matches the input identifier, and then extract the "connection mechanism" field value of that row. If the identifier is "audit node A", the corresponding mechanism is returned (such as "protocol: AMQP, address: 192.168.1.1, port: 5672").

[0056] After establishing the message transmission channel, the risk early warning information is serialized into a protocol-compatible format. The implementation process is as follows: the risk early warning information is a structured object (such as a dictionary or a custom class instance) containing risk level parameters, disposal suggestions, and audit path fields. According to the preset message queue protocol format (such as the binary format of AMQP or Kafka), the system converts the information into a byte stream: converts to an intermediate format: first serializes the object into a general data format such as a JSON string. This ensures that the data is readable and structured. Based on the intermediate format, add protocol-specific metadata such as message headers (containing message type, version, timestamp) and message tails (such as checksums). For example, in the AMQP protocol, encapsulation includes adding attribute headers (such as priority and persistence flags), and converting the entire content into a binary byte sequence. After serialization, the output is an early warning information data packet that strictly follows the protocol format (such as a fixed-length header + a variable-length body). For example, the data packet may exist in the form of a byte array, with the header part identifying the message type and the body part containing the serialized JSON.

[0057] The determination method of the preset message queue connection mechanism configuration table is as follows: sort out the information of all audit nodes in the system, including node identifier, function, location, etc. Analyze the message transmission requirements of different audit nodes and early warning information sending units, such as transmission protocol, security mechanism, etc. According to these information, configure the corresponding connection mechanism for each audit node, such as connection method, parameter setting, etc. Organize the audit node identifier and the corresponding connection mechanism into a table to form the preset message queue connection mechanism configuration table, ensuring that the message transmission channel can be accurately established.

[0058] The determination manner of the preset message queue format requirement is: considering the technical framework and protocol used by the message queue, ensuring that the format can be correctly identified and transmitted by the system. From the data integrity, the fields that must be included in the risk warning information are determined, such as risk level, disposal suggestion, audit path, etc., the data type and length of each field are specified. For the convenience of the audit node processing, the general format such as JSON or XML is adopted, and the field arrangement order is determined. After testing and optimization, the preset message queue format requirement meeting the system requirements and convenient for operation is formed.

[0059] Referring to Figure 3 As shown in FIG. 1, which is a structural schematic diagram of the security check policy determination module of the embodiment of the present application, comprising: The policy determination unit is used for parsing the risk warning information through a preset parsing rule to obtain structured risk data, and matching a security check policy corresponding to the structured risk data in a preset security check policy generation rule set; The policy sending unit is used for encapsulating the security check policy as a message object, and sending the message object to the corresponding responsible person node through the routing rule of the message queue.

[0060] Specifically, the policy determination unit parses the risk warning information through the semantic analysis algorithm in the preset parsing rule to obtain the structured risk data containing the risk type identifier, the risk occurrence position coordinate and the risk level, and takes the structured risk data as a query condition to retrieve the policy parameter set matched with the risk type identifier, the risk occurrence position coordinate and the risk level in the rule index table of the preset security check policy generation rule set, and combines the policy parameter set with the pre-stored security policy template to generate the structured security check policy containing the security check path planning scheme, the responsible person allocation identifier and the disposal time limit value.

[0061] In this embodiment, after the policy sending unit obtains the security check policy generated by the policy determination unit, it first maps the structured data items such as policy identifier, check rule, and execution condition in the security check policy to the data fields of the message object according to the preset message format specification. The message object is defined in a format conforming to the message middleware transmission protocol, and contains a fixed data header structure and policy data payload. The data header contains metadata such as message type identifier, priority identifier, and timestamp, and the data payload stores the specific content of the security check policy according to the field mapping relationship. The encapsulated message object is delivered to the routing processing module of the message middleware. Based on the policy type identifier, risk level identifier, and other key metadata in the data header of the message object, the routing processing module performs matching query in the preconfigured routing rule set. Each routing rule contains a condition expression and a target address mapping relationship. The condition expression is used to logically judge the metadata of the message object, determine the target routing rule matched therewith, and extract the responsible person node communication address information corresponding to the routing rule. The communication address can be in the form of message queue name, topic name, or service endpoint URL. The routing processing module takes the matched responsible person node communication address as the target address, calls the transmission interface of the message queue, and sends the message object to the message queue or service endpoint identified by the corresponding communication address. In the transmission process, the message middleware serializes, transmits over the network, and ensures the reliability of the message object according to its own communication protocol specification, so as to ensure that the message object accurately reaches the message receiving end of the target responsible person node for subsequent policy analysis and execution operation.

[0062] The determination method of the preset analysis rule is as follows: the preset analysis rule contains a semantic analysis algorithm. The algorithm extracts key information features such as risk type, occurrence location, and risk level from a large number of risk warning information samples through learning and analysis, and forms a set of standardized analysis logic. Based on this, the input risk warning information is analyzed according to this logic, so as to obtain structured risk data containing risk type identifier, risk occurrence location coordinates, and risk level.

[0063] The determination method of the preset security check policy generation rule set is as follows: according to past risk cases and safety management experience, the required safety check policy parameters under different risk type, occurrence location, and risk level combinations are sorted out. A rule index table is established, the risk type identifier, risk occurrence location coordinates, and risk level are taken as index items, the matched policy parameter set is stored correspondingly, and the rule set is formed.

[0064] The determination mode of the pre-stored security policy template is: according to the general process and requirements of security check, a general security check policy framework is designed in advance, which covers the format and structure of key contents such as security check path planning scheme, responsibility person distribution identifier and disposal time limit value, and forms a pre-stored security policy template.

[0065] Referring to Figure 4 As shown in the figure, it is a structural schematic diagram of the safety control module of the embodiment of the application, comprising: The case closing application auditing unit is configured to obtain the case closing application through the interface corresponding to the responsibility person node, determine the corresponding auditing process engine through the auditing node corresponding to the responsibility person node, and perform auditing and verification on the case closing application based on the preset verification rules in the auditing process engine to obtain the auditing and verification result containing the verification state and the specific verification item result. The security archive generation unit is configured to, when the verification state is a pass state and the preset review condition is met, call a time stamp service to obtain the current system time as an archive time mark, and sign the case closing application in the pass state and meeting the preset review condition and the time stamp through a digital signature algorithm to generate security archive information containing original application data, a time stamp and a digital signature.

[0066] Specifically, in the case closing application auditing unit, the process of auditing and verifying the case closing application based on the preset verification rules in the auditing process engine includes: The multi-modal information conversion unit based on the auditing process engine converts and analyzes the case closing application to obtain a verification object set, and establishes an attribute mapping relationship between the preset verification rules and the verification object set; The rule item matching result is obtained by performing rule item matching on the verification object set through the hierarchical verification mechanism and the attribute mapping relationship of the auditing process engine, and the cross verification of the associated weights between the preset verification rules is performed based on the rule item matching result to generate intermediate verification data containing the matching state of each verification item and the associated conflict information; The intermediate verification data is input into the result synthesis module of the auditing process engine for conflict resolution to obtain the auditing and verification result containing the verification state and the specific verification item result.

[0067] In this embodiment, the preset verification rules are determined according to the specification requirements and business logic of the closed application. First, the compliance standards of each key information item in the closed application are sorted out, such as data format, integrity, logical consistency, etc. Then, combined with the audit process, specific verification rules are set for different information items. For example, for the time field in the application, it is specified that its format must comply with a specific date format and cannot be earlier than the business start time. Through the multi-modal information conversion unit, the attribute mapping of the verification object set is established, and the cross-verification of hierarchical verification mechanism and correlation weight is used to ensure that the rules are comprehensive and effective. The preset review conditions are determined from the audit passing status of the closed application and the achievement of the business key indicators. For example, the audit verification result needs to be in the passing state to ensure that the application basis is compliant; at the same time, the key indicators related to the business, such as the risk disposal completion rate reaching 100%, the related document materials being submitted completely, etc. Only when the audit passing and these key business indicators are met at the same time, the preset review conditions are met, and the safe archive generation link can be entered.

[0068] The process of converting and analyzing the closed application based on the multi-modal information conversion unit includes: the multi-modal information conversion unit first analyzes the original data of the closed application (such as text, structured fields, etc.), and converts it into a unified format of verification object set through natural language processing, data standardization, etc. For example, the key information in unstructured text is extracted into structured fields, or the format of dates, amounts, etc. is standardized to ensure the consistency of subsequent analysis.

[0069] The process of establishing the attribute mapping relationship between the preset verification rules and the verification object set includes: the conditions in the verification rules (such as “the amount must be greater than 0”) are associated with the attributes in the verification object set (such as the “application amount” field). Through the rule parsing engine, the key attributes in the rule (such as field name, logical operator) are extracted and matched with the attribute name in the object set to form a rule-attribute mapping table, ensuring that the rule can accurately act on the target data.

[0070] The process of rule item matching through hierarchical verification mechanism and attribute mapping relationship includes: the hierarchical verification mechanism processes the verification object set layer by layer in a preset order (such as basic verification → business logic verification). According to the attribute mapping relationship, the rule conditions are compared with the object attribute values (such as “amount > 0” matching “application amount = 100”), and the rule item matching result (pass / fail) is generated. If the rule involves multiple attributes, the corresponding fields are located through the mapping relationship and verified jointly.

[0071] The process of cross-checking the association weight based on the rule item matching result includes: analyzing the dependency relationship between rules (such as "Rule A is the premise for the execution of Rule B"), combining the rule item matching result, and calculating the association weight. For example, if Rule A fails, Rule B cannot be executed, which is marked as an association conflict. Through the weight matrix or graph model, the logical contradiction or redundancy between rules is identified, and conflict information is generated.

[0072] The process of conflict resolution by inputting the intermediate verification data into the result synthesis module includes: the result synthesis module integrates the intermediate verification data (matching state, conflict information), and resolves the conflict through priority strategy (such as strict mode priority error reporting) or weight sorting. For example, high-weight rule conflict terminates the process and reports an error, and low-weight conflict records a warning. Finally, a unified verification result is generated, including the final state (pass / fail) and detailed verification item description.

[0073] In this embodiment, the implementation process of the education and training module can be divided into four links: data docking, content generation, targeted pushing, and data recording. First, the module is docked with the risk analysis and early warning module and the safety inspection strategy determination module through internal interfaces, and real-time risk early warning information (including risk level, disposal suggestion, etc.) and safety inspection strategies (such as inspection focus, operation specification) are obtained. Second, based on the obtained information, training content is generated: through a preset template matching risk type, such as "road surface collapse disposal process" training for high-risk road sections; combined with safety inspection strategies to supplement practical operation points, such as maintenance equipment operation regulations, and automatically integrated into online courses, picture-text manuals or short videos, etc. forms. Next, the module identifies the responsibility person node (such as the road maintenance responsible person, safety officer) through the system interaction interface, and pushes the generated training content to its account, supporting message reminder function. Finally, the module tracks the learning behavior of the responsible person in real time, records the training completion data, including course viewing progress, test scores, completion time, etc., and synchronously stores them to the system database, providing basic data support for the target evaluation module.

[0074] The evaluation and assessment model needs to first determine the core indicators of the two types of data for the evaluation and assessment of the training completion data and the risk disposal results in the safety archives. Training completion data focuses on training participation rate, time length standard rate, test scores, etc., reflecting the responsibility person's mastery of safety knowledge; risk disposal results focus on response timeliness, disposal compliance, hidden danger elimination rate and recurrence, reflecting actual operation ability.

[0075] The evaluation model sets weights for two types of indicators, such as the risk disposal effect weight higher than the training test score, to highlight practical ability. Then the indicators are quantified, converting participation rate, pass rate, etc. into percentage scores, and disposing results according to "excellent-good-poor" corresponding scores. Then the comprehensive score is calculated through the weighted formula, and combined with the dynamic adjustment mechanism, if a certain person's training score is high but the same type of risk disposal is improper, the comprehensive score will be adjusted downward to ensure that the evaluation fits the actual ability. Finally, according to the preset score interval (such as 90 points or above for excellent, 60 points or below for unqualified), the comprehensive score is converted into evaluation level to form the final evaluation result.

[0076] The construction of the evaluation model can be based on machine learning and data analysis framework, using Python's Scikit-learn as the basic tool chain, combined with TensorFlow or PyTorch to build deep neural network model. The input layer of the model receives training completion data (such as training duration, evaluation score, participation) and risk disposal results (such as rectification completion rate, review pass rate, risk reduction degree), extracts key indicators such as training participation, knowledge mastery, risk rectification rate, etc. through feature engineering. The model architecture uses a hybrid design: structured data (such as evaluation score, rectification period) uses random forest or gradient boosting tree for classification and regression analysis; unstructured data (such as training feedback text, risk description) applies natural language processing technology, extracts semantic features through BERT pre-training model. A multi-input multi-output neural network architecture is used to fuse and weight the two types of features, outputting evaluation scores and improvement suggestions. The evaluation process introduces a dynamic weight mechanism that automatically adjusts the weight proportion of training completion and risk disposal effect according to the risk level (such as setting the rectification result weight of high-risk problems to 70%). The evaluation model is cross-validated through historical data, and Keras Tuner is used for hyperparameter optimization. Finally, it is packaged into an API service through Flask or FastAPI, integrated into the target evaluation module to realize the evaluation of the person in charge.

[0077] The above is only an embodiment of the present application, and the common knowledge of specific structures and characteristics in the scheme is not described in detail, and the ordinary skilled person in the art knows all the ordinary technical knowledge in the technical field of the present application before the application date or the priority date, can know all the prior art in the field, and has the ability to apply conventional experimental means before that date, and the ordinary skilled person in the art can improve and implement the present scheme under the guidance of the present application, and some typical known structures or known methods should not be an obstacle for the ordinary skilled person in the art to implement the present application. It should be pointed out that for those skilled in the art, without departing from the structure of the present application, a number of modifications and improvements can be made, which should also be considered as the protection scope of the present application, which will not affect the effect and practicality of the patent. The protection scope of the present application should be subject to the content of its claims, and the specific implementation mode and the like in the specification can be used to explain the content of the claims.

Claims

1. A highway safety management system, characterized in that: include: The data acquisition module is used to collect highway risk data through the data interface of the system's interactive interface; The data preprocessing module is used to preprocess highway risk data to obtain actual highway risk data; The risk analysis and early warning module is used to conduct risk assessment on actual highway risk data based on the risk assessment model, obtain risk warning information including risk level parameters, handling suggestions and review paths, and send the risk warning information to the corresponding review nodes through a message queue. The security inspection strategy determination module is used to determine the corresponding security inspection strategy based on the risk warning information, and send the security inspection strategy to the corresponding responsible node through a message queue. The security management module is used to receive the case closure application from the responsible node, and to review and verify the case closure application through the review node corresponding to the responsible node, and obtain the review and verification result. When the review and verification result meets the preset review conditions, the corresponding security file information is output. The education and training module is used to generate corresponding training content based on risk warning information and safety inspection strategies, push the training content to the responsible person node through the system interface, and record the training completion data of the responsible person node. The target assessment module is used to assess the risk handling results in training completion data and safety file information through an assessment and evaluation model, obtain assessment results, and feed the assessment results back to the safety management module through a message queue.

2. The highway safety management system according to claim 1, characterized in that: The data preprocessing module performs outlier filtering and missing value interpolation on the highway risk data to obtain an intermediate dataset. It then extracts features from the intermediate dataset using principal component analysis to obtain a feature-reduced dataset. Finally, it performs state estimation and synchronization correction on the feature-reduced dataset using an adaptive Kalman filter algorithm to obtain the actual highway risk data.

3. A highway safety management system according to claim 1, characterized in that: The risk analysis and early warning module includes: The risk assessment model building unit is used to build a risk assessment model based on a preset historical dataset and risk parameter correction factors. The risk assessment unit is used to input actual highway risk data into the risk assessment model for feature extraction, parameter calculation, and risk level judgment and analysis, and to obtain risk warning information including risk level parameters, handling suggestions, and review paths. The early warning information sending unit is used to trigger the corresponding connection mechanism based on the risk early warning information, and based on the connection mechanism, according to the node identifier specified in the audit path, encapsulate the risk early warning information into an early warning information data packet that conforms to the preset message queue format requirements, and send the early warning information data packet to the corresponding audit node through the message queue.

4. A highway safety management system according to claim 3, characterized in that: The risk assessment model building unit acquires a preset historical dataset and preprocesses it to obtain a structured historical dataset that meets the modeling requirements. It then dynamically adjusts the feature weights in the structured historical dataset according to the risk parameter correction factor to obtain a set of model parameters containing the corrected feature values. The set of model parameters and the structured historical dataset are then input into the logistic regression algorithm framework. The model coefficients in the logistic regression algorithm framework are iteratively optimized using the gradient descent method. When the loss function value in the logistic regression algorithm framework is lower than the preset loss threshold, the risk assessment model is output.

5. A highway safety management system according to claim 4, characterized in that: The risk assessment unit inputs actual highway risk data into the risk assessment model for feature extraction to obtain a highway risk feature vector. Based on the highway risk feature vector, it calculates the corresponding risk probability value, compares the risk probability value with a preset risk level threshold range, determines the corresponding risk level parameter, and calls a preset disposal suggestion library and review path template based on the risk level parameter to generate risk warning information containing risk level parameters, disposal suggestions, and review paths.

6. A highway safety management system according to claim 5, characterized in that: The warning information sending unit parses the risk warning information to obtain connection request data containing the target node identifier. Based on the connection request data, it obtains a connection mechanism matching the target node identifier from the preset message queue connection mechanism configuration table. It establishes a message transmission channel with the corresponding review node through the connection mechanism, and serializes and encapsulates the risk warning information according to the preset message queue format requirements through the message transmission channel to generate a warning information data packet. Finally, it pushes the warning information data packet to the corresponding review node according to the routing rules of the message queue.

7. A highway safety management system according to claim 1, characterized in that: The security inspection strategy determination module includes: The strategy determination unit is used to parse the risk warning information through preset parsing rules to obtain structured risk data, and to match the security inspection strategy corresponding to the structured risk data in the preset security inspection strategy generation rule set. The policy sending unit is used to encapsulate the security inspection policy into a message object and send the message object to the corresponding responsible node through the routing rules of the message queue.

8. A highway safety management system according to claim 7, characterized in that: The strategy determination unit parses the risk warning information using a semantic parsing algorithm in the preset parsing rules to obtain structured risk data containing risk type identifiers, risk occurrence location coordinates, and risk levels. Using the structured risk data as query conditions, it retrieves a set of strategy parameters matching the risk type identifiers, risk occurrence location coordinates, and risk levels from the rule index table of the preset security inspection strategy generation rule set. The strategy parameter set is then combined with a pre-stored security strategy template to generate a structured security inspection strategy containing a security inspection path planning scheme, a responsible person assignment identifier, and a handling time limit.

9. A highway safety management system according to claim 1, characterized in that: The security management module includes: The case closure application review unit is used to obtain the case closure application through the interface corresponding to the responsible person node, and the review node corresponding to the responsible person node determines the corresponding review process engine, and reviews and verifies the case closure application based on the preset verification rules in the review process engine, and obtains the review verification result including the verification status and the specific verification item results. The security file generation unit is used to call the timestamp service to obtain the current system time as the file timestamp when the verification status is passed and the preset review conditions are met. It then uses a digital signature algorithm to sign the case closure application and timestamp that have passed the verification status and meet the preset review conditions, thereby generating security file information containing the original application data, timestamp, and digital signature.

10. A highway safety management system according to claim 9, characterized in that: In the case closure application review unit, the process of reviewing and verifying the case closure application based on the preset verification rules in the review process engine includes: The multimodal information conversion unit based on the review process engine performs conversion analysis on the case closure application to obtain the set of verification objects and establish the attribute mapping relationship between the preset verification rules and the set of verification objects. The hierarchical verification mechanism and attribute mapping relationship of the audit process engine are used to match the set of verification objects with rules to obtain the rule matching results. Based on the rule matching results, the association weights between the preset verification rules are cross-validated to generate intermediate verification data containing the matching status of each verification item and association conflict information. The intermediate verification data is input into the result synthesis module of the audit process engine for conflict resolution, resulting in an audit verification result that includes the verification status and the results of specific verification items.

Images