A road traffic safety facility information collection and processing system
By integrating a dynamic health assessment model with multiple data sources and a distributed emergency response mechanism, the problems of inaccurate health assessment of transportation facilities and untimely emergency response in existing technologies have been solved. This has enabled accurate assessment of facility health status and efficient emergency response, thereby improving the intelligence and reliability of traffic safety management.
Patent Information
- Application Number
- CN202611284211.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-24
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-08-24
AI Technical Summary
Existing technologies are insufficient to comprehensively assess the overall health status of road traffic safety facilities, resulting in low reliability of traffic control decisions. Centralized emergency response models are prone to untimely responses due to information processing bottlenecks in emergency situations, making it difficult to determine accident liability and assess handling efficiency, and failing to provide effective data support for optimizing traffic control strategies.
The system employs a data collection module to acquire data on structural dynamics, physical integrity, and material degradation. A dynamic health assessment model is used to calculate the health index of transportation facilities. An anomaly monitoring module is used to conduct multi-scale anomaly monitoring and generate early warning signals. An emergency linkage module broadcasts collaborative information packets between facility nodes. A cloud server performs distributed log aggregation and causal reconstruction to generate a full-process handling record report.
It enables precise and quantitative assessment of facility health status and early proactive warning, builds a decentralized distributed emergency response mechanism, improves the accuracy and timeliness of warnings, ensures the efficiency and reliability of emergency response, and provides traceability of the entire process of handling records and the basis for liability determination, thereby enhancing the refinement and intelligence of traffic safety management.
Smart Images

Figure CN122842364B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent transportation systems, specifically to a road traffic safety facility information collection and processing system. Background Technology
[0002] Road traffic safety facilities, as a crucial component of road traffic control systems, directly impact the effective transmission of traffic control commands and the safe operation of traffic flow. These facilities, such as traffic signs, guardrails, and traffic lights, form the foundation for orderly traffic management. With the increasing complexity of road networks, these facilities are numerous and widely distributed, making their operational reliability critical to the overall effectiveness of the traffic control system. These facilities are constantly subjected to multiple factors, including vehicle dynamic loads, severe weather, and accidental collisions, leading to structural damage, material deterioration, and ultimately, partial or complete functional failure. Therefore, how to efficiently, accurately, and proactively monitor and manage the health status of this vast number of traffic safety facilities is a pressing challenge.
[0003] Most systems rely solely on video or single-type sensor data for assessment, making it difficult to comprehensively evaluate the overall health of facilities. This one-sided data collection leads to inaccurate assessments of facility status, consequently affecting the reliability of traffic control decisions. When anomalies are detected, information is typically reported to a centralized management platform, which then issues instructions. This centralized emergency response model suffers from long chains and significant latency, easily leading to delayed responses due to information processing bottlenecks in emergencies, hindering the formation of rapid and effective regional collaborative responses. After an accident or malfunction occurs, the entire process record from early warning and response to the end of the handling is often fragmented and incomplete. This makes it difficult to determine responsibility for the accident, assess handling efficiency, and provide effective data support and management loops for the continuous optimization of traffic control strategies.
[0004] To address this, a road traffic safety facility information collection and processing system is proposed. Summary of the Invention
[0005] The purpose of this invention is to provide a road traffic safety facility information collection and processing system to solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides the following technical solution: A road traffic safety facility information collection and processing system, comprising: The data collection module acquires structural dynamics data, physical integrity data, and material degradation data of road traffic safety facilities; The indicator fusion module calculates the health index of traffic facilities based on structural dynamics data, physical integrity data, and material degradation data through a dynamic health assessment model for traffic safety facilities. The anomaly monitoring module utilizes a traffic anomaly identification engine and a traffic trend prediction engine to perform multi-scale anomaly monitoring of the health index of traffic facilities and generate early warning signals for traffic safety facilities. The emergency response module triggers a traffic safety facility warning signal. The traffic safety facility node generates a digitally signed collaborative information packet, which is broadcast to nearby traffic safety facility nodes via the traffic communication link. The nearby traffic safety facility nodes respond to the emergency and enhance the collaborative information packet before continuing to broadcast it.
[0007] In the traceability management module, traffic safety facilities upload local collaborative information packages. The cloud server analyzes the collaborative information packages based on distributed log aggregation and causal reconstruction methods to generate a full-process handling record report.
[0008] Preferably, the structural dynamics data, physical integrity data, and material degradation data of the road traffic safety facilities specifically include: The structural dynamics data includes triaxial tilt angle, angular velocity, impact acceleration values, and continuous micro-vibration data; the physical integrity data includes the on / off state of conductive circuits and resistance values; and the material degradation data includes the cumulative number of freeze-thaw cycles, corrosion potential, and cumulative vibration energy.
[0009] Preferably, the calculation process of the traffic facility health index is as follows: based on structural dynamics data and physical integrity data, the data is processed by an acute injury diagnosis module based on fuzzy logic; probabilistic reasoning is performed on acute injury patterns, including structural bending, structural fracture, and foundation loosening, to generate a multi-dimensional acute injury diagnosis vector; based on the acute injury diagnosis vector, the acute injury index is quantified by norm calculation; a chronic disease diagnosis module is used to process material degradation data; the degree of chronic injury is defined as a latent state, and material degradation data is used as the observation value of the latent state; a Bayesian filtering loop including prediction and update steps is used to continuously infer the posterior probability distribution of the latent state, and the expected value of the posterior probability distribution is calculated as the chronic disease index; the acute injury index and the chronic disease index are weighted and nonlinearly fused to generate the traffic facility health index.
[0010] Preferably, the traffic anomaly identification engine and the traffic trend prediction engine specifically involve: the traffic anomaly identification engine processing the sequence data of the traffic facility health index within a short time window; extracting a multi-dimensional feature vector of the short-time sequence of the traffic facility health index, including time-domain statistical features and frequency-domain features; matching the multi-dimensional feature vector with a standard waveform feature pattern library for acute abnormal events, and outputting an abnormal traffic event matching report; and the traffic trend prediction engine performing statistical analysis on historical data of the traffic facility health index over a long time series; if the statistical results show a monotonically decreasing trend, further calling a time series prediction model to make predictions and generate a predicted value for the aging of traffic safety facilities. Preferably, the process of generating the traffic safety facility warning signal is as follows: The abnormal traffic event matching report is spliced with the traffic safety facility aging prediction value, input into the preset multi-level traffic management risk decision rule library, and the output results are integrated to generate a traffic safety facility early warning signal; the multi-level traffic management risk decision rule library includes three levels of decision rules, specifically high-level alarm generation rules, medium-level alarm generation rules and low-level alarm generation rules.
[0011] Preferably, the working process of the emergency response module is as follows: The facility node triggered by the traffic safety facility warning signal acts as the originating node, generates an initial collaborative information packet, and digitally signs it using the originating node's key. It then broadcasts the initial collaborative information packet to all neighboring nodes within its communication range. These neighboring nodes, acting as receiving nodes, receive the initial collaborative information packet, verify the validity of the digital signature, and parse the packet. Based on the content of the initial information packet, they initiate an emergency response, adding the emergency response action and execution timestamp to the initial collaborative information packet to generate an enhanced collaborative information packet. The enhanced collaborative information packet is then locally backed up and relayed to neighboring nodes until the maximum propagation hops specified in the initial information packet are reached.
[0012] Specifically, the initial collaborative information package includes event identification information, traffic safety facility early warning signals, and propagation control information, specifically: Event identification information includes the originating node identifier, timestamp, and geographic coordinates; traffic safety facility early warning signals include alarm level, traffic safety facility aging prediction value, and abnormal traffic event matching report; propagation control information includes recommended emergency response and maximum propagation hops.
[0013] Preferably, the generation process of the full-process handling record report is as follows: The cloud server receives locally backed-up collaborative information packets from facility nodes and performs distributed log aggregation based on event identification information and propagation hop count to reconstruct a global traffic safety facility response event sequence. Based on the global traffic safety facility response event sequence and collaborative information packets, a hierarchical causal model is used for causal analysis. The hierarchical causal model is defined as a three-layer network structure containing an entity layer, a context layer, and an effectiveness layer, where the effectiveness layer contains latent variables that cannot be directly observed. The global event sequence is mapped to the observed variables of the entity layer and the context layer, respectively. Maximum likelihood estimation is used to calculate the causal path coefficients connecting all layer variables to generate a hierarchical causal relationship path graph. Based on the hierarchical causal relationship path graph, a full-process handling record report is constructed, containing traffic event information in three dimensions: time series, geography, and logic.
[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention integrates structural dynamics data, physical integrity data, and material degradation data of road traffic safety facilities, and constructs a dynamic health assessment model and an anomaly identification and trend prediction engine to achieve accurate and quantitative assessment and early proactive warning of facility health status. This method overcomes the shortcomings of existing technologies that rely on a single data source, leading to inaccurate assessments and only providing reactive warnings. It can identify potential risks in facilities in advance, transforming safety management from reactive response to proactive intervention, significantly improving the accuracy and timeliness of early warnings.
[0015] 2. This invention constructs a decentralized distributed emergency response mechanism by having facility nodes generate digitally signed collaborative information packets and broadcast them to neighboring nodes after an early warning is triggered. This mechanism eliminates dependence on a central server, shortens the transmission path of emergency information, reduces the risk of single-point failures in the system, and achieves faster and more reliable regional collaborative responses. Especially when dealing with sudden, multi-point concurrent events, it demonstrates unparalleled efficiency and resilience compared to traditional centralized models.
[0016] 3. This invention uploads collaborative information packets from each node to the cloud, uses distributed log aggregation and causal relationship-based reconstruction methods to obtain a full-process handling report, achieving traceability of the entire emergency response process from its source and ensuring its authenticity. This method not only guarantees the immutability of the handling process but also preserves all content in its original state, providing a reliable basis for determining responsibility for the handling results afterward. It also enables effective evaluation of handling effectiveness and the formulation of contingency plan improvement measures, forming a closed-loop management system of "monitoring-early warning-response-source tracing," thus improving the refinement and intelligence of traffic safety management. Attached Figure Description
[0017] Figure 1This is a schematic diagram of the structure of a road traffic safety facility information collection and processing system according to the present invention; Figure 2 This is a schematic diagram of the time prediction model of a road traffic safety facility information collection and processing system according to the present invention; Figure 3 This is a flowchart illustrating a road traffic safety facility information collection and processing system according to the present invention. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] Example 1: This embodiment monitors and deploys a central crash barrier system on a road segment. This segment experiences an average daily traffic flow exceeding 80,000 vehicles, with a high proportion of heavy-duty vehicles, and has long suffered from issues such as facility fatigue and aging, and frequent structural impacts. Taking the central crash barrier system as an example, the specific modules of this system are described in detail.
[0020] This invention proposes a road traffic safety facility information collection and processing system, referring to... Figure 1 The specific modules consist of: The data collection module acquires structural dynamics data, physical integrity data, and material degradation data of road traffic safety facilities; The indicator fusion module calculates the health index of traffic facilities based on structural dynamics data, physical integrity data, and material degradation data through a dynamic health assessment model for traffic safety facilities. The anomaly monitoring module utilizes a traffic anomaly identification engine and a traffic trend prediction engine to perform multi-scale anomaly monitoring of the health index of traffic facilities and generate early warning signals for traffic safety facilities. The emergency response module is triggered by a traffic safety facility warning signal. The traffic safety facility node generates a digitally signed collaborative information packet, which is broadcast to nearby traffic safety facility nodes via a traffic communication link. The nearby traffic safety facility nodes respond to traffic safety emergencies, enhance the collaborative information packet, and then continue to broadcast the collaborative information packet.
[0021] In the traceability management module, traffic safety facilities upload local collaborative information packages. The cloud server analyzes the collaborative information packages based on distributed log aggregation and causal reconstruction methods to generate a full-process handling record report.
[0022] Furthermore, the structural dynamics data, physical integrity data, and material degradation data of road traffic safety facilities are acquired. The specific implementation process for the aforementioned data collection module is as follows: Acquiring structural dynamics data is primarily used to reflect the dynamic response of traffic safety facilities under the influence of the external environment. It can reveal the stability and fatigue level of the facilities. Specifically, this includes obtaining tilt angle data through a three-axis gyroscope installed on the guardrail post; angular velocity sensors installed on the top of the facility to measure the rotational speed of the facility, which can indicate whether the facility has encountered a sudden external impact or structural loosening; acceleration sensors installed at key locations on guardrails, signposts, and other facilities to record the acceleration changes generated by the equipment when it encounters external forces (such as collisions), with a sampling frequency of 100Hz, which can quickly respond to instantaneous impacts; and micro-vibration sensors deployed at the base of the facility or structural connection points to capture the continuous micro-vibrations of the facility.
[0023] Obtaining physical integrity data can reflect whether a facility has failed or aged due to damage. This includes monitoring the electrical connection status of facilities such as guardrails or warning signs by installing conductive loop monitoring modules within the facility. If an electrical break occurs (e.g., due to facility damage or loose wiring), it indicates that the facility's integrity has been compromised. Resistance sensors installed at critical connection points of the facility detect changes in resistance as current flows through the facility structure, determining whether there is a risk of poor electrical contact due to corrosion, loosening, or other factors.
[0024] Acquiring material degradation data is used to assess the degree of wear, corrosion, or aging during long-term use of facilities. Specifically, this includes recording the number of freeze-thaw cycles using temperature sensors and an environmental humidity monitoring module; the system determines the facility's durability based on set freeze-thaw standards; and monitoring the corrosion potential of the facility's metal components using electrochemical corrosion sensors. If the potential is below a given threshold, it indicates that corrosion has occurred, leading to reduced load-bearing capacity or even failure. A given threshold for corrosion failure is set (this threshold is calibrated based on the standard electrochemical polarization curve of the facility's metal material and historical degradation statistics; for example, for a conventional carbon steel guardrail foundation, this threshold can be set to -450mV). When the collected potential value is less than or equal to this given threshold, it is determined that electrochemical corrosion has occurred in the facility's foundation.
[0025] Vibration sensors are used to measure the energy accumulated by vibrations on a facility, serving as an important indicator for evaluating the degree of damage under long-term vibration. Under heavy traffic conditions, the cumulative vibration energy can lead to fatigue failure of the structure. Specifically, the calculation logic for the cumulative vibration energy is as follows: within a given statistical time window, a discrete acceleration amplitude sequence from the vibration sensor is obtained; for each sampling period, the recorded acceleration amplitude is squared and then multiplied by the sensor's sampling time interval; finally, the products obtained from all sampling periods within the time window are summed, and this sum is used to approximately quantify the total structural fatigue energy borne by the facility.
[0026] This invention collects three different types of structural information: first, information on the dynamic characteristics of the facility (such as impact acceleration); second, information on physical integrity (such as corrosion potential); and third, information related to material degradation (such as accumulated vibration energy). Through the effective integration and analysis of these various types of information, a comprehensive and multi-layered description of the facility's condition can be achieved. The resulting data sources are not only rich and comprehensive, but also greatly reduce the possibility of erroneous judgments due to relying on only one type of data.
[0027] Furthermore, based on structural dynamics data, physical integrity data, and material degradation data, a traffic facility health index is calculated using a dynamic health assessment model for traffic safety facilities. The specific implementation process for the aforementioned index fusion module is as follows: Based on structural dynamics data and physical integrity data, the system assesses the condition of the facility through a fuzzy logic acute injury diagnosis module.
[0028] In this embodiment, the input variables are defined as tilt angle, angular velocity, impact acceleration, conductive circuit resistance, vibration energy, and corrosion potential; the output variables are structural bending, foundation loosening, structural fracture, and surface damage. The variables are fuzzified using a membership function.
[0029] The fuzzy variables after fuzzification are input into the fuzzy inference rule table, as shown in Tables 1 to 4. This embodiment designs 12 fuzzy rules, covering four different types of damage (bend, break, loosen, surface).
[0030] Table 1. Fuzzy Inference Rules for Structural Bending serial number IF condition THEN (bend) B1 The tilt angle is large (L). High (H) B2 The tilt angle is (M) and the angular velocity is (M). (M) B3 The tilt angle is small (S). Low (L) Table 2. Fuzzy Inference Rules for Structural Fracture serial number IF condition THEN (fracture) R1 Impact acceleration is strong (S) and angular velocity is high (H). High (H) R2 The impact acceleration is (M) and the tilt angle is (M). (M) R3 The impact acceleration is weak (W). Low (L) Table 3 Basic Loosening Fuzzy Inference Rules serial number IF condition THEN (loosening) L1 The resistance is high (H). High (H) L2 The resistance is too high (M) and the tilt angle is medium (M). (M) L3 The resistance is normal (N). Low (L) Table 4. Surface Damage Fuzzy Inference Rules serial number IF condition THEN (Surface Damage) S1 The corrosion potential is high (H) and the vibrational energy is high (H). High (H) S2 The corrosion potential is M and the vibrational energy is M. (M) S3 The corrosion potential is low (L). Low (L) In the fuzzy inference module of this system, the membership functions of each input variable are defined using trapezoidal or triangular functions. Taking the tilt angle variable as an example: a small tilt angle corresponds to the interval of 0 to 5 degrees, with a membership degree of 1 in this interval; beyond 5 degrees, the membership degree linearly decreases towards 15 degrees; a medium tilt angle corresponds to the main interval of 10 to 20 degrees; and a large tilt angle corresponds to the interval above 15 degrees, increasing towards 20 degrees with full membership. The threshold values for the remaining variables are calibrated according to the national standards for road facility safety. In particular, for the fuzzification of the corrosion potential variable, the system calibrates the membership function for carbon steel facilities: when the corrosion potential is greater than -300mV, it is determined as "low (L)" with a membership degree of 1; in the interval of -300mV to -450mV, it exhibits a Gaussian distribution, mainly corresponding to "medium (M)"; when the corrosion potential is lower than -450mV, it is determined as "high (H)" with a full membership degree of 1, indicating that severe corrosion is likely to cause damage to the surface and foundation of the facility.
[0031] When inputting data, the system employs the Mandani inference mechanism. If multiple fuzzy rules are activated simultaneously, the maximum value method is used for rule aggregation. Finally, the system uses the centroid method for defuzzification, that is, the abscissa of the geometric center of the aggregated fuzzy set graph is used as the deterministic membership output value of the corresponding damage mode (the value is between 0 and 1).
[0032] Through reasoning based on the fuzzy rule base, a fuzzy vector is generated. Each component of the four dimensions of structural bending, structural breakage, foundation loosening, and surface damage has a membership degree between 0 and 1.
[0033] .
[0034] A fuzzy diagnostic vector is generated based on fuzzy inference, encompassing four dimensions: structural bending, structural fracture, foundation loosening, and surface damage. The membership values for each dimension range from 0 to 1. The Acute Injury Index (ADI) is quantified by calculating the L2 norm of this vector. The specific calculation process is as follows: The squares of the membership values for each of the four dimensions are calculated; these squared values are summed; and the square root of this sum is taken as the ADI. .
[0035] Chronic damage to facilities is considered a latent state that cannot be directly detected by sensors. Instead, it is inferred through indirect observation of multi-dimensional data, such as material degradation. The core objective of the Bayesian update mechanism is to progressively adjust our understanding of the facility's damage state based on continuously changing observational data, and to provide a more accurate damage assessment that evolves over time.
[0036] The specific mathematical process of the Bayesian filtering loop includes: (1) Prediction step: Based on the historical state, calculate the latent damage state at the current moment. Prior probability distribution: ; (2) Update steps: Combine the material degradation observation data obtained at the current moment. Calculate the posterior probability distribution: ; in, This indicates the current hidden state of traffic safety facilities (i.e., the degree of chronic damage). It represents the implicit state of the previous moment (history); This represents the material degradation observation data acquired at the current moment (such as values obtained through sensors). This represents the time from the initial moment to the previous moment ( The collection of all historical observation data; It represents the set of all observation data, including the current time. This represents the latent damage state calculated in the prediction step at the current moment. The prior probability distribution; It is a state transition probability model; This represents the posterior probability distribution of the hidden state inferred from the previous time step; For sensor observation likelihood models; This indicates that the update step incorporates the observation data at the current moment. The posterior probability distribution of the latent states is then calculated. Finally, the expected value of this posterior probability distribution is used as the Chronic Disease Index (CDI), i.e. Its value is normalized to the interval [0, 1].
[0037] In this embodiment, the chronic disease state transition probability model adopts a first-order autoregressive physical degradation model, specifically: the latent state value at the current moment equals the latent state value at the previous moment multiplied by the degradation rate coefficient, plus the system process noise. This degradation rate coefficient is calibrated based on historical accelerated aging test data of similar facilities (e.g., 0.98 for carbon steel guardrails), and the process noise follows a normal distribution with a mean of zero and a preset variance. For the observation likelihood model, the system uses different probability distributions for three types of material degradation data: a Gaussian distribution likelihood model is used for corrosion potential data; a Poisson distribution likelihood model is used for the cumulative freeze-thaw cycle count, which is a discrete count; and a gamma distribution likelihood model is used for the cumulative vibration energy. The initial parameters of each likelihood model are obtained through maximum likelihood estimation using historical measured data. The initial prior distribution of the latent state is set to a uniform distribution between zero and one. After the system receives the first batch of sensor data, initialization and subsequent posterior probability updates are completed using a particle filtering algorithm with 500 particles.
[0038] The chronic damage index, after Bayesian updating, is no longer a single value, but a probability distribution. This distribution reflects the optimal estimate of the degree of chronic damage to the facility and the corresponding uncertainty.
[0039] To eliminate the influence of dimensional inconsistency, the system first normalizes the acute injury index by dividing it by the square root of the total number of dimensions (i.e., four dimensions).
[0040] First, calculate the comprehensive damage metric D: ; in, and The weights for acute injury and chronic disease are assigned separately (to satisfy the following conditions). In this embodiment, the data is obtained by fitting the historical equipment failure database of the road section using the logistic regression method. The weight of acute injury is set to 0.6, and the weight of chronic lesion is set to 0.4.
[0041] Subsequently, the comprehensive damage metric D is converted into a percentage-based traffic facility health index HI using an exponential decay mapping function. ; in, The sensitivity adjustment coefficient (typically set between 0.5 and 1.0) is determined based on the facility's safety requirement level. This mapping function ensures that when the facility is in good condition (D approaches 0), the health index approaches a perfect score of 100; while when severe acute injury occurs (ADI surges, causing D to increase), the health index HI decays exponentially (e.g., plummets below 25), thus sensitively reflecting the facility's endangered status.
[0042] Quantitative calibration standard for sensitivity adjustment coefficient: For core safety facilities (such as central crash barriers) on highway trunk roads and major bridges, extremely high early warning sensitivity is required. The value is set to 1.0; for conventional facilities on national and provincial highways, The value is set to 0.8; for ordinary rural roads or auxiliary road facilities, The value is set to 0.5. This allows for precise adaptation of the health index decay sensitivity under different road scenarios.
[0043] This invention, by constructing an "acute injury index" and a "chronic lesion index" respectively, enables the system to clearly quantify and distinguish between acute injuries caused by sudden events such as external impacts and chronic aging caused by factors such as corrosion and fatigue. This ensures that the comprehensive health index can scientifically reflect the combined effects of both types of injury, resulting in a more comprehensive and reliable assessment.
[0044] Furthermore, by utilizing a traffic anomaly identification engine and a traffic trend prediction engine, multi-scale anomaly monitoring of the traffic facility health index is performed to generate early warning signals for traffic safety facilities. The specific implementation process corresponding to the aforementioned anomaly monitoring module is as follows: The traffic anomaly identification engine completes the data processing of the short time series of traffic facility health index and mines the information of traffic anomaly events contained therein; it obtains the short time series of the existing traffic facility health index and continuously obtains several traffic facility health indices within a predefined short time window.
[0045] Based on the determined time window length, features are extracted from all health index data sampled within that time period to generate a multi-dimensional feature vector. Time-domain statistical features are employed, including mean, standard deviation, maximum, minimum, skewness, and kurtosis. Frequency-domain features are also used, employing Discrete Fourier Transform to convert the collected time-domain signal sequence into a frequency-domain sequence. The specific processing steps are as follows: For a given discrete time-domain signal sampling point, the sample is multiplied pointwise by a complex exponential basis function (i.e., a complex sequence containing sine and cosine components) and summed over all sampling points to calculate the complex value corresponding to each frequency component. Subsequently, the magnitude (i.e., absolute value) of each extracted frequency complex value is calculated. This magnitude is used as the frequency amplitude of the corresponding dominant frequency, and then the dominant frequency and its corresponding energy proportion are extracted. Finally, these parameters are concatenated with time-domain statistical features such as mean and kurtosis to form a multi-dimensional feature vector.
[0046] The extracted multi-dimensional feature vectors are compared with a "standard waveform feature pattern library for acute abnormal events" built based on historical data. This library is constructed by conducting controlled simulation experiments on traffic safety facilities in a closed test site, including destructive impacts and structural shaking, and collecting standardized time-series signals for each known abnormality type. Since some accidents or equipment failures may cause specific frequencies or vibration modes, the system compares the standard waveform feature patterns of each sample in the library with the multi-dimensional feature vectors of the actually detected signals to determine whether an abnormal traffic event has occurred. During the matching comparison, the system uses a dynamic time warping (DTW) algorithm to calculate the similarity distance between the real-time multi-dimensional feature vector and each standard waveform pattern in the library. A successful match is determined only if the calculated DTW distance is less than a preset matching safety threshold. If the health index feature vector matches a waveform pattern in the standard pattern library, the system generates an abnormal traffic event matching report, indicating the specific time of the event, the possible abnormality type (such as acute injury, structural failure, etc.), and its severity level.
[0047] Reference Figure 2 The main task of the traffic trend prediction engine is to predict the long-term trend of traffic facility health status based on historical data, especially to model and predict the aging process of traffic facilities. It collects historical data on traffic facility health indices and constructs a trend analysis model. Statistical analysis is performed on the historical data to identify the changing trends of the facility health index. If the average slope of the health index fit in the long-term historical data meets the condition that it is consistently less than a preset negative threshold (e.g., -0.5% / cycle) for N consecutive statistical periods (N is 12 calendar months in this embodiment), indicating a monotonically declining trend, the system will trigger and call the time series prediction model for further analysis.
[0048] Using a time series decomposition algorithm, this embodiment employs the STL algorithm to analyze the complete historical data of the traffic facility health index into three core components with different characteristics: a smooth "trend component" reflecting the structural and long-term aging of the facilities; a "periodic component" related to seasonal climate change (such as annual freeze-thaw cycles and increased corrosion during the rainy season); and a "residual component" containing short-term, random shocks and measurement noise. When applying the STL algorithm, to adapt to the characteristics of traffic flow changes, the seasonal window length is set to 7 consecutive sampling periods (to capture the periodic impact of weekly traffic flow fluctuations on facility status). The trend window length is automatically smoothed and adaptively adjusted according to the period length. The inner loop (used to update the seasonal and trend components) iteration count is set to 2, and the outer loop (used to calculate robustness weights to reduce outlier interference) iteration count is set to 0 to ensure computational efficiency.
[0049] For each component, a prediction model best suited to its characteristics is matched. For the long-term "trend component" with gradual changes, a robust linear regression or multinomial fitting model is used for extrapolation. For the highly regular "periodic component," Fourier series or seasonal ARIMA models are used to accurately capture its future fluctuation patterns. For the complex nonlinear dynamic "residual component," gradient boosting trees are used for in-depth mining and prediction. The independent prediction results of the three components are linearly superimposed to reconstruct the predicted value of traffic safety facility aging, including the predicted value and its reliability. In the specific training and implementation process, for the prediction of the periodic component, the system adopts an autoregressive moving average model, and its autoregressive order, differencing order, and moving average order are automatically optimized using the Akaike information content criterion. For the gradient boosting tree model for the residual component prediction (specifically using the XGBoost algorithm), the data construction method of its training samples is as follows: extract historical residual data from the past 30 consecutive sampling periods as input feature vectors to predict the residual value of the next period. The core hyperparameters of this gradient boosting tree are configured as follows: maximum tree depth is set to 5, learning rate is set to 5%, and the number of iteration rounds is set to 200 rounds using five-fold cross-validation to prevent overfitting. When the prediction results of the three components are linearly superimposed, if the reconstructed health index exceeds the effective physical boundary of 0 to 100, the system will automatically truncate it to the boundary extreme value (i.e., 0 or 100) to ensure that the output index conforms to physical meaning.
[0050] This invention effectively avoids the errors and instabilities caused by a single model when fitting multiple mixed patterns by matching the most suitable prediction model to data components with different characteristics (smooth trends, regular cycles, and complex residuals). This allows for more accurate capture of each type of dynamic change, and the final integrated prediction results outperform traditional integrated models in both accuracy and robustness to noise. It can clearly determine whether future risks stem primarily from continuous structural degradation (trends) or from foreseeable seasonal stress (cycles). This deep insight provides unprecedented and clear data support for developing more scientific and targeted maintenance strategies (such as structural reinforcement or seasonal protection) and risk contingency plans.
[0051] This invention utilizes a traffic anomaly detection engine to focus on sudden changes within a short timeframe, emphasizing the detection of abrupt changes in the time domain; and a traffic trend prediction engine to focus on the long-term aging process of time series data, emphasizing the tracking of aging decay patterns in the frequency domain. The combined effect of these two engines enables a breakthrough from post-event alerts to pre-event predictions, thereby achieving proactive management. Based on comparison results from a standard waveform feature pattern library, the type of abnormal event can be preliminarily determined, rather than simply identifying an outlier, providing more background knowledge for further processing. For trend prediction, this facilitates preventative maintenance of facilities and effective asset management.
[0052] By integrating the abnormal traffic event matching report with the predicted value of traffic safety facility aging, the final prediction result is obtained, generating a comprehensive information package. The predicted value of the aging degree of traffic safety facilities is used as the aging prediction attribute vector, while each attribute of the data entry for each abnormal traffic event is used to generate an abnormal event attribute vector; at the same time, the consistency of the time dimension of the two attribute vectors is ensured. Based on the attribute vectors, combined with multi-level decision rules, a traffic safety facility early warning signal is generated.
[0053] As shown in Table 5, the multi-level traffic management risk decision-making rule base includes three levels of rules: high-level alarm rules, medium-level alarm rules, and low-level alarm rules.
[0054] Each level of rules is based on different types of abnormal events, facility health status, and aging trends to assess the probability of the facility being in different real-world states, forming a probability distribution. Then, for each feasible decision option, the system calculates the expected total cost of each option using Bayesian decision theory. The specific calculation process is as follows: First, the posterior probabilities of the facility being in various real-world states (such as structural fracture, slight bending, etc.) output by the model are obtained; second, a preset cost matrix is established, representing the quantified loss value resulting from taking a specific decision under a certain real-world state (for example, when the actual cause is structural fracture but the decision is a missed alarm in routine monitoring, a very high safety penalty cost value, such as 1000, is set; when a non-damaged facility issues a false alarm decision to shut down, a relatively low interference cost value, such as 50, is set); then, for a specific optional decision, its quantified loss cost value under each real-world state is multiplied by the corresponding posterior probability, and the multiplications under all possible states are summed to obtain the expected total cost of that specific decision; finally, the system iterates through and compares the expected total cost calculation results of all optional decisions, selects the decision option that minimizes the expected total cost, and outputs the corresponding traffic safety facility warning signal.
[0055] This embodiment establishes a complete four-by-four-dimensional expected cost matrix. The rows of the matrix represent the four hidden states of the facility (undamaged, slightly damaged, foundation loose, structural fracture), and the columns represent four optional decisions (routine monitoring, periodic inspection, shutdown for repair, immediate shutdown). Cost values are set based on actual economic losses and the cost of safety accidents: for example, taking routine monitoring in the structural fracture state (i.e., failing to report a high-risk state) will lead to a serious accident, so its safety penalty cost is set to 1000; while taking immediate shutdown in the undamaged state (i.e., false alarm shutdown) has an interference economic cost of 50; the cost values for matching the correct decision in other states are set to zero or a minimum. Considering the difference in mathematical definitions between fuzzy membership degrees and probability distributions, the system introduces the Softmax function as a mapping bridge. Specifically, the system inputs the unnormalized confidence scores of the various damage mode membership values defuzzified by the fuzzy inference module into the Softmax function, nonlinearly mapping them to a probability distribution with a strict sum of 1, which serves as the posterior probability input for the facility in each corresponding true state. By multiplying this probability distribution with the cost matrix and summing the results, we can obtain the expected loss of each decision and select the strategy with the minimum expected loss as the final alarm level.
[0056] By introducing a decision cost matrix and expected cost calculation, early warning decision-making is transformed from relying on preset, empirical IF-THEN rules into a mathematical optimization process based on probability statistics and risk quantification. The output of each alarm level aims to minimize a calculable expected cost function that integrates safety and economic factors. This provides a solid and traceable mathematical basis for every decision, completely eliminating the logical rigidity and subjective assumptions that may arise in traditional rule bases when facing complex situations, making the entire decision-making process more scientific, objective, and transparent.
[0057] Table 5 Traffic Management Risk Decision-Making Rules Rule Number Input conditions Output Decision level H1 The anomaly event matching report includes "structural fracture" and a "severely declining" predicted value for facility aging. High-level alarm: Immediate shutdown and inspection High level H2 The abnormal event matching report contained "basal loosening," and the aging prediction value showed a health index decline of more than 15%. High-level alarm: Service outage and rapid repair required. High level M1 The anomaly event matching report contained "structural bending," and the facility aging prediction value showed a slight decrease. Medium-level alarm: Schedule regular checks. Intermediate level M2 The anomaly matching report included "surface damage," and the facility aging prediction value showed a slight decrease (less than 10%). Medium-level alarm: Regularly check and record. Intermediate level L1 The abnormal event matching report contained "increased vibration energy," but the predicted value for facility aging did not show a significant decrease. Low-level alarms: routine monitoring low level L2 No significant abnormalities were found in the abnormal event matching report, and the aging prediction value showed that the health index was stable or slightly declining. Low-level alarm: Monitoring continues low level High-level alarm (Level H): A high-level alarm is generated when a major traffic anomaly occurs (such as structural fracture or severe foundation loosening) and the facility's health condition is predicted to deteriorate rapidly. These alarms typically require immediate attention and repair or replacement.
[0058] Medium-level alarm (Level M): A medium-level alarm is generated when traffic facilities suffer some damage (such as structural bending or surface damage) and the facility's health status prediction shows a slight decline. This type of alarm reminds relevant personnel to conduct regular inspections and maintain monitoring of the facility.
[0059] Low-level alarm (Level L): When there is a minor abnormality or no obvious abnormality in the traffic facility, only a low-level alarm is generated. It is usually used to remind the facility to be monitored regularly.
[0060] This invention obtains corresponding fault hazard information by performing fuzzy comprehensive evaluation on abnormal matching results and aging prediction values. After automatically analyzing and judging the above two indicators using a pre-set multi-level risk decision rule base, it can generate information prompts with corresponding levels, realizing a process from simple to complex. This avoids the time lag and uncertainty caused by human factors that previously relied on manual judgment. In the event of an emergency, it ensures that relevant decision-makers can make a rapid and consistent response. In addition, the alarm levels are divided into high, medium and low levels, enabling relevant personnel to grasp the importance of the event at the first time and mobilize relevant emergency rescue forces in a targeted manner, adhering to the principle of rescuing important parts before rescuing secondary parts, which greatly improves the effectiveness and rationality of emergency rescue work.
[0061] Furthermore, triggered by a traffic safety facility warning signal, the traffic safety facility node generates a digitally signed collaborative information packet, which is broadcast to neighboring traffic safety facility nodes via a traffic communication link. The neighboring traffic safety facility nodes then initiate a traffic safety emergency response, enhance the collaborative information packet, and continue broadcasting it. Corresponding to the aforementioned emergency linkage module, the specific implementation process is as follows: When a traffic safety facility warning signal is triggered, the facility node that triggered the event (i.e., the originating node) generates an initial coordination information packet, which contains the following three key elements: Event identification information: Originating node identifier, using the facility's unique identifier; timestamp, recording the exact time the event was triggered; geographic coordinates, used to represent the geographic location where the event occurred.
[0062] Traffic safety facility early warning signals include: alarm level, indicating the severity of the current event; traffic safety facility aging prediction value; and abnormal traffic event matching report, which includes detailed information about the event, such as type and scope of impact.
[0063] Propagation control information: Recommended emergency response; maximum propagation hops and a unique global message ID (generated by a hash algorithm based on the originating node's MAC address and the event timestamp). Set the maximum propagation hops (e.g., 5 hops). Once the maximum hops are exceeded, information propagation will stop.
[0064] The initial collaborative information includes the core elements of the event itself (what it is, where it is, when it happened, and how serious it is), as well as control information such as "recommended emergency response" and "maximum propagation hops." This enables receiving nodes to easily obtain the event description and quickly take appropriate action; and by limiting the propagation distance during the information propagation process, it effectively suppresses information storms, maintaining order in information dissemination.
[0065] In terms of security, the digital signature of the initial collaborative information packet adopts the Elliptic Curve Digital Signature Algorithm (ECDSA algorithm, key length 256 bits). The private key is hard-coded into the built-in security chip when the facility leaves the factory and cannot be read. The corresponding public key certificate is pre-issued by the traffic management department and stored in all facility nodes along the route, thereby ensuring the efficiency of the verification process.
[0066] Furthermore, in the execution mapping and energy consumption degradation logic of emergency response: after neighboring nodes parse the information packets, they map the standard action pool according to the alarm level (e.g., high-level alarms are forcibly mapped to lane closure instructions, and medium-level alarms are mapped to speed limit reminders). When a node's own battery level is lower than 20% of the facility's current full charge capacity limit, the system determines that the battery is critically low and triggers a degradation strategy. In degradation mode, the node will abandon energy-intensive mechanical actions or high-brightness screen displays, retaining only the operation of extremely low-power radio frequency chips to focus on relaying collaborative information packets to maintain the communication lifeline.
[0067] This invention, through this proactive "cost-saving" approach, ensures the sustainable operation and continuity of the information chain of the entire distributed emergency system. It effectively overcomes the problem of critical nodes going offline prematurely due to a single energy-intensive operation. Therefore, even under extreme weather conditions or during prolonged emergency situations, it can maintain the online rate of the core functional modules in the system, thereby enhancing the robustness and overall survivability of the decentralized emergency network in real-world, complex environments.
[0068] Upon receiving the initial coordination packet, each neighboring node (referred to as the receiving node) first extracts the global message ID from the packet and queries its local memory for the processed message cache list within the last 5 minutes. If the global message ID already exists in the list, the packet is silently discarded to effectively suppress network broadcast information storms; otherwise, it is stored in the local cache list, and then the validity of the data is verified through digital signature to ensure that the information has not been tampered with. By verifying the validity of the signature, the receiving node confirms that the data comes from a legitimate originating node. The receiving node receives and parses the initial coordination packet to obtain the event identifier, traffic safety facility warning signals, and propagation control information. Then, the receiving node uses this information to initiate emergency responses, such as dispatching emergency vehicles, activating backup traffic lights, and adjusting traffic flow on road sections.
[0069] After completing the emergency response, the receiving node updates the information packet and generates an enhanced collaborative information packet. The enhanced collaborative information packet includes the emergency response actions, the specific response measures executed by the receiving node, such as activating emergency signals, dispatching traffic management personnel, and carrying out traffic control tasks; and an execution timestamp, recording the execution time of the response measures to ensure traceability for subsequent analysis and historical data.
[0070] After completing the emergency response and generating the enhanced coordination information packet, the receiving node locally backs up the enhanced coordination information packet to prevent data loss or network interruption. The receiving node then continues to broadcast the enhanced coordination information packet to its neighboring nodes in the same way as the initial coordination information packet. This broadcasting process will continue until the preset maximum propagation hop count is reached. Before that, the information packet will be relayed to other nodes to ensure that the entire system shares response information as widely as possible.
[0071] This invention enables direct broadcast communication between facility nodes, eliminating the need for a central server to relay information, significantly reducing response time and mitigating the risk of single-point failure at the central node. This distributed mechanism demonstrates exceptional resilience and efficiency in the face of regional, multi-point concurrent disasters. Digital signatures ensure the authenticity and integrity of emergency information, preventing malicious attacks or information tampering. Neighboring nodes enhance their responses and continue broadcasting, creating a rolling flow of situational information, allowing subsequent nodes to obtain more comprehensive information and thus make better coordinated decisions.
[0072] Furthermore, traffic safety facilities upload local collaborative information packages. The cloud server analyzes these packages based on distributed log aggregation and causal reconstruction methods, generating a full-process handling record report. Corresponding to the aforementioned traceability management module, the specific implementation process is as follows: When an anomaly occurs at a transportation infrastructure node, the node generates an initial coordination information packet and an enhanced coordination information packet according to a preset communication protocol, and uploads them to a cloud server via wireless communication. This upload process is based on Low Power Wide Area Network (LPWAN), 5G, or other communication technologies suitable for large-scale IoT applications.
[0073] After receiving the collaborative information packets uploaded by each facility node, the cloud server performs distributed log aggregation on these packets. The aggregation process, based on event identifiers and propagation hops, ensures accurate reconstruction of the entire event flow according to the event's chronological order, propagation scope, and response sequence. During this process, the cloud server sorts the content of all received collaborative information packets by timestamp, geographical location, and propagation path, merging them to generate a global traffic safety facility response event sequence.
[0074] After the event sequence aggregation is completed, the cloud server performs causal analysis based on the global traffic safety facility response event sequence using a hierarchical causal model. This model is defined as a three-layer network structure, specifically including: The entity layer contains directly observable variables; the context layer represents contextual variables that affect transportation facilities; and the effectiveness layer defines potential variables that are not directly observable and indirectly affect the response and safety of transportation facilities.
[0075] The cloud server maps each event in the global event sequence to observed variables in the entity and context layers. The system then constructs this hierarchical network using a structural equation model. This model defines the structural and measurement relationships between the set of observed variables and the set of latent variables. On the cloud server, maximum likelihood estimation is used to optimize the causal path coefficients connecting variables at each level. The specific optimization process is as follows: First, based on the data samples of the global event sequence, the observed covariance matrix of the samples is calculated. Simultaneously, based on the current path parameterization assumptions of the model, the implicit theoretical covariance matrix of the model is derived. Then, a maximum likelihood loss function is constructed. The calculation logic of this loss function is to find the trace of the product of the inverse of the sample covariance matrix and the model's implicit covariance matrix, and combine the difference in their determinants to quantify the deviation between the model and the data. The system continuously adjusts the causal path coefficient values between variables through an iterative algorithm, making the theoretical covariance matrix as close as possible to the sample observed covariance matrix, until the likelihood loss function reaches its minimum value. The parameters solved at this point are the optimal causal path coefficient values. Based on the significance test of this parameter, a quantitative hierarchical causal relationship path diagram is generated.
[0076] Because the number of hops in the collaborative information packet generated by a single, sporadic traffic accident is limited, the log sample size extracted from a single event is often small (e.g., less than 30 entries). Directly using maximum likelihood estimation to solve the structural equation model may lead to statistical bias and failure. To address this, the source management module incorporates a causal estimation strategy with adaptive sample size: when the aggregated global event log entries are less than 30, the system automatically switches to partial least squares path modeling algorithm to solve the causal structure, ensuring the estimation stability under small sample conditions. In the premise of applying this algorithm, the system explicitly defines both entity-level observed variables (such as response delay and emergency action type) and context-level observed variables as formative indicators. This means that the linear combination of these observed variables jointly determines the latent variables (overall emergency response comprehensive effectiveness) in the effectiveness layer that cannot be directly observed, rather than being generated by reflection of latent variables. This ensures the operational logic consistency and convergence stability of the partial least squares path model under small sample conditions.
[0077] Meanwhile, the system will aggregate and accumulate information streams of similar historical events that occur in different time periods. Once the total amount of data in the aggregated sample pool exceeds the statistical safety threshold of one hundred, the system will automatically switch back to the maximum likelihood estimation method, which has the property of asymptotic optimization, in order to obtain more accurate causal path parameter estimates.
[0078] To ensure the structural equation model is mathematically identifiable, the network constructed in this system satisfies the overidentification condition. Specifically, the entity layer defines five observed variables (including the health index of the originating node, inter-node response delay, emergency action type, propagation hops, and geographical distance), the context layer defines three observed variables (alarm level, aging prediction value, and environmental characteristics), and the effectiveness layer extracts one latent variable that cannot be directly observed (i.e., the overall emergency response effectiveness). The total number of free parameters to be optimized within the model is less than the total number of elements in the covariance matrix of the observed data. During the solution process, the system employs a quasi-Newton iterative algorithm (such as the L-BFGS-B algorithm) to perform maximum likelihood estimation optimization. The iterative convergence criterion is set as follows: the relative change in the likelihood loss function between two consecutive iterations is less than 10. -6 The maximum number of iterations is limited to 500. After solving for the path coefficients, a bootstrap resampling method is used to test the significance of the causal path coefficients, and only significant causal paths with a p-value less than 5% are retained.
[0079] Based on the generated causal path diagram, the cloud server further constructs a full-process handling record report. This report comprehensively presents the event response from three dimensions, including: demonstrating the evolution of various causal relationships during the event response process, such as changes in facility health status, the spread of the accident, and the effectiveness of emergency measures, helping decision-makers understand the logical flow and causal relationships of event handling.
[0080] The generated full-process handling record report will be provided to traffic management personnel, emergency response teams, and decision-makers through a cloud platform to facilitate subsequent analysis, tracing, and improvement. Simultaneously, the report will be archived according to prescribed formats and requirements for future review, analysis, or legal compliance needs.
[0081] Considering the limited computing power of the end nodes of road traffic safety facilities, this system's algorithm adopts a time-domain decoupling strategy in its engineering deployment. For structural dynamics data requiring extremely high sampling rates (such as collision acceleration at 100 Hz), threshold comparisons are performed by the underlying hardware interrupts of the edge nodes to trigger acute damage assessments. Meanwhile, the resource-intensive Bayesian filtering loop (i.e., the update calculation for chronic material degradation) is detached from the main processing loop and configured for low-frequency asynchronous calculations in five-minute time windows. This high- and low-frequency decoupling architecture ensures that even on embedded chips with limited computing power, the particle filter algorithm can run smoothly without causing system thread blocking.
[0082] This invention aggregates distributed logs to obtain a complete and comprehensive global event overview. Based on a hierarchical causal model, it can discover and extract more information about the essence of the event, thus forming a triplet report format including time, space, and logical dimensions, which is provided to the transportation department as an important basis for subsequent work. The transportation management department can summarize lessons learned after each event, achieving a data-driven intelligent decision-making upgrade.
[0083] This invention utilizes multi-source heterogeneous sensing data from the front end, combined with intermediate state assessment and early warning functions, to activate a terminal collaborative broadcast mechanism in emergency situations. Relying on a cloud platform, it completes data analysis and tracing throughout the entire process, forming a four-loop closed management model of detection-early warning-coordination-tracing. This model effectively solves problems such as disconnect between front and back ends and information silos in existing systems: it integrates multi-source heterogeneous data to ensure accurate and reliable monitoring and early warning; it adopts a decentralized architecture to deploy emergency linkage broadcasts, improving emergency response speed; and it uses distributed log aggregation and causal reconstruction methods to restore the global sequence of events, improving the credibility and depth of post-event tracing.
[0084] Example 2: Reference Figure 3 This embodiment uses a gantry-type traffic sign on a highway as the monitoring and management object. This embodiment will specifically describe the complete workflow of the road traffic safety facility information collection and processing system described in this invention during a structural damage event caused by a vehicle collision.
[0085] In this embodiment, multiple sensors are pre-integrated on the target facility, specifically collecting the following three types of data. The first type is structural dynamics data, monitored by a built-in triaxial tilt sensor, angular velocity meter, and impact accelerometer to track changes in the facility's attitude and stress. The second type is physical integrity data, monitored by the continuity and resistance values of conductive circuits embedded in key structural nodes (such as the connection between columns and beams, foundation bolts, etc.). The third type is material degradation data; the system records long-term climate data for the region, calculating a cumulative freeze-thaw cycle count of 850; embedded corrosion sensors monitor the corrosion potential of the foundation; and cumulative vibration energy is calculated based on continuous micro-vibration data.
[0086] After receiving the aforementioned multi-source heterogeneous data, the indicator fusion module activates the traffic safety facility dynamic health assessment model for analysis and calculation. Based on the structural dynamics and physical integrity data collected in this incident, the system calls the fuzzy logic-based acute injury diagnosis module for processing. This model fuzzifies the input data and infers the most likely acute injury mode based on a preset rule base. For example, the combination of high tilt angle and high impact value strongly points to "structural bending," while the break in the conductive circuit clearly points to "structural fracture." The model integrates these possibilities to generate a multi-dimensional acute injury diagnosis vector containing the confidence levels of various injury modes. By calculating the norm of this vector, it is quantified into a single acute injury index characterizing the severity of sudden damage.
[0087] The system employs a chronic lesion diagnosis module to process long-term collected material degradation data. The "degree of chronic damage" of the facility is defined as a latent state that cannot be directly observed. The system uses data such as cumulative freeze-thaw cycles, corrosion potential, and cumulative vibration energy as a series of observations for this latent state. Through a Bayesian filtering loop that includes prediction and update steps, the model continuously infers and corrects the posterior probability distribution of this latent state. Prior to this event, based on existing data such as 850 freeze-thaw cycles and a corrosion potential of -450mV, the model's calculations indicated that the facility exhibited a moderate degree of chronic aging. The system takes the expected value of this posterior probability distribution and quantifies it as a chronic lesion index.
[0088] The index fusion module combines the previously calculated acute injury index and chronic lesion index using a weighted sum of squares nonlinear fusion algorithm to generate a comprehensive transportation facility health index. Due to the extremely high acute injury index, the fused health index instantly drops from the usual 95 (out of 100) to 25, indicating that the facility is in an extremely dangerous state.
[0089] The anomaly monitoring module monitors changes in the health index of traffic facilities. The traffic anomaly recognition engine within the module captures a sequence of data where the health index drops sharply from 95 to 25 within a very short time window (within 1 second in this embodiment). The traffic anomaly recognition engine extracts a multi-dimensional feature vector from this short-time sequence, including statistical features such as peak value, drop rate, and slope in the time domain, and energy distribution features in the frequency domain. The engine quickly matches this feature vector with a pre-stored library of standard waveform feature patterns for acute abnormal events. This feature pattern library contains waveform patterns for various events such as "strong wind impact," "minor scraping," and "severe impact." The feature vector of this event has a 98% similarity to the "severe impact" pattern, therefore the engine generates an abnormal traffic event matching report, stating "severe structural impact event."
[0090] The traffic trend prediction engine within the anomaly monitoring module continuously performs statistical analysis on long-term historical data of the target facility's health index over the past few years. The analysis shows that prior to the collision, the facility's health index exhibited a slow but clear monotonically declining trend. Upon identifying this trend, the engine automatically invokes its built-in time-series prediction model (in this example, the ARIMA model) to predict future aging trends. The prediction results generate a traffic safety facility aging forecast, indicating that, barring unforeseen events, the facility is expected to enter a risky state requiring major repairs within 18 months due to natural material degradation.
[0091] The anomaly monitoring module integrates the matching report of "severe structural impact event" with the predicted value of "18-month aging warning", compares it with the multi-level traffic management risk decision rule base, and generates traffic safety facility warning signals.
[0092] After the anomaly monitoring module determines the nature of the event, a traffic safety facility warning signal is generated, triggering the emergency response module. The target facility, as the originating node, first generates an initial collaborative information packet. This packet contains event identification information (originating node identifier: G01-K100-500-Sign, event timestamp, geographic coordinates), a traffic safety facility warning signal (alarm level: Level 1, aging prediction value: 18 months, anomaly event report: severe structural impact), and propagation control information (recommended emergency response: immediately close the right lane and notify the maintenance department, maximum propagation hops: 5).
[0093] The originating node digitally signs the packet using its own key and broadcasts it to all nearby traffic safety facility nodes within its communication range via the traffic communication link.
[0094] In this embodiment, a variable message sign located 500 meters upstream of the target facility acts as the receiving node and successfully receives the information packet. First, the validity of the digital signature is verified to confirm the reliable source of the information and that it has not been tampered with. After parsing the collaborative information packet, an emergency response is immediately executed, displaying on the screen, "Accident 500 meters ahead, right lane closed, please drive carefully." Simultaneously, this action and the current timestamp are added to the information packet, forming an enhanced collaborative information packet. Finally, the variable message sign backs up this enhanced collaborative information packet locally and continues to relay it to its neighboring nodes. This process continues until the propagation hop count reaches a preset maximum value of 5.
[0095] During the event handling phase and subsequent event termination phase, the originating node and all neighboring nodes involved in the event response process upload the locally stored collaborative information packets (including the initial packet and enhancement packets at each level) to the cloud server. After this, the traceability management module is activated. Based on the event identification information in the collaborative information packets (with the same originating node, timestamp, and geographical location) and the number of propagation hops added between each level of collaborative information packets, the distributed log data in each region is aggregated to accurately reconstruct the global traffic safety facility response event sequence of this event.
[0096] Based on the detailed content of the global traffic safety facility response event sequence and collaborative information package, a hierarchical causal model is applied for causal reasoning analysis. The system maps specific actions in the global event sequence to observed variables at the entity level and environmental information to observed variables at the context level. The causal path coefficients connecting all levels of variables are calculated using the maximum likelihood estimation method. Based on the calculated hierarchical causal relationship path diagram, the system constructs a full-process handling record report.
[0097] This report not only includes information flow presented from a time perspective (the chronological order of events), impact from a spatial perspective (the process of information dissemination and the map of the covered areas), but also content presentation at the logical level (visually revealing the causal chain from the initial collision event to each information interaction and each safety device response), providing comprehensive and profound data support for post-event assessment, liability determination, and process optimization.
[0098] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A road traffic safety facility information collection and processing system, characterized in that, include: The data collection module acquires structural dynamics data, physical integrity data, and material degradation data of road traffic safety facilities; The index fusion module, based on structural dynamics data, physical integrity data, and material degradation data, employs a dynamic health assessment model for traffic safety facilities. This model includes an acute injury diagnosis module and a chronic lesion diagnosis module to calculate the health index of traffic facilities. Specifically, the acute injury diagnosis module uses tilt angle, angular velocity, impact acceleration, conductive circuit resistance, vibration energy, and corrosion potential as input variables. These input variables are fuzzified using a membership function. The fuzzified variables are then input into a fuzzy inference rule table. Using the Mandani inference mechanism, when multiple fuzzy rules are simultaneously activated, the maximum value method is used for rule aggregation, and the centroid method is used for defuzzification. This yields a fuzzy diagnostic vector containing four dimensions: structural bending, structural fracture, foundation loosening, and surface damage. The L2 norm of the fuzzy diagnostic vector is then calculated to obtain the acute injury... The chronic disease diagnosis module defines the degree of chronic damage as a latent state, uses material degradation data as the observation value of the latent state, and employs a Bayesian filtering loop that includes prediction and update steps to continuously infer the posterior probability distribution of the latent state. The state transition probability model adopts a first-order autoregressive physical degradation model, and the observation likelihood model uses a Gaussian distribution likelihood model for corrosion potential data, a Poisson distribution likelihood model for cumulative freeze-thaw cycles, and a gamma distribution likelihood model for cumulative vibration energy. The expected value of the posterior probability distribution is calculated as the chronic disease index. After normalizing the acute damage index, it is nonlinearly fused with the chronic disease index to obtain a comprehensive damage measure. Then, the comprehensive damage measure is converted into a percentage-based traffic facility health index through an exponential decay mapping function. The anomaly monitoring module utilizes a traffic anomaly identification engine and a traffic trend prediction engine to perform multi-scale anomaly monitoring of the health index of traffic facilities and generate early warning signals for traffic safety facilities. The emergency response module triggers a traffic safety facility warning signal. Traffic safety facility nodes generate a digitally signed collaborative information packet and broadcast it to neighboring traffic safety facility nodes via the traffic communication link. Neighboring traffic safety facility nodes respond to emergencies, enhance the collaborative information packet, and continue broadcasting it. Specifically: the facility node triggered by the traffic safety facility warning signal acts as the originating node, generates an initial collaborative information packet, and digitally signs it using its key; it then broadcasts the initial collaborative information packet to all neighboring nodes within its communication range; these neighboring nodes act as receiving nodes, receiving the initial collaborative information packet, verifying the validity of the digital signature, and parsing the initial collaborative information packet; based on the content of the initial information packet, they respond to emergencies, adding the emergency response action and execution timestamp to the initial collaborative information packet to generate an enhanced collaborative information packet; they then back up the enhanced collaborative information packet locally and continue relaying it to neighboring nodes until the maximum propagation hops specified in the initial information packet are reached. The source tracing management module involves traffic safety facilities uploading local collaborative information packages. The cloud server, based on distributed log aggregation and causal reconstruction methods, analyzes these packages and generates a full-process handling record report. Specifically, the cloud server receives collaborative information packages uploaded by each facility node, performs distributed log aggregation based on event identifiers and propagation hops, sorts the content of all received collaborative information packages by timestamp, geographical location, and propagation path, and merges them to generate a global traffic safety facility response event sequence. Based on the global traffic safety facility response event sequence and collaborative information packages, a hierarchical causal model is used for causal analysis. This hierarchical causal model is a three-layer network structure comprising an entity layer, a context layer, and an effectiveness layer, where the effectiveness layer includes unobservable data. The system identifies latent variables; maps global event sequences to observed variables at the entity and context layers respectively, and uses maximum likelihood estimation to calculate causal path coefficients connecting all levels of variables; when the aggregated global event log entries are less than thirty, it switches to partial least squares path modeling algorithm to solve for the causal structure, and defines both entity layer and context layer observed variables as formative indicators; it aggregates and adds information streams of similar historical events occurring in different time periods across events, and switches back to maximum likelihood estimation when the total amount of data in the aggregated sample pool exceeds one hundred; it generates a hierarchical causal path graph, and constructs a full-process handling record report of traffic event information containing three dimensions: time series, geography, and logic, based on the hierarchical causal path graph.
2. The road traffic safety facility information collection and processing system according to claim 1, characterized in that, The structural dynamics data, physical integrity data, and material degradation data of the road traffic safety facilities are specifically as follows: The structural dynamics data includes triaxial tilt angle, angular velocity, impact acceleration value, and continuous micro-vibration data; the physical integrity data includes the on / off state of the conductive circuit and resistance value data; the material degradation data includes the cumulative number of freeze-thaw cycles, corrosion potential, and cumulative vibration energy.
3. The road traffic safety facility information collection and processing system according to claim 1, characterized in that, The traffic anomaly identification engine and traffic trend prediction engine are specifically as follows: The traffic anomaly identification engine processes the sequence data of the traffic facility health index within a short time window; and extracts a multi-dimensional feature vector of the short-time sequence of the traffic facility health index, including time-domain statistical features and frequency-domain features. The multidimensional feature vector is matched with the standard waveform feature pattern library of acute abnormal events to output an abnormal traffic event matching report; the traffic trend prediction engine performs statistical analysis on the historical data of traffic facility health index over a long time series; if the statistical results show a monotonically decreasing trend, the time series prediction model is further called to make predictions and generate a predicted value for the aging of traffic safety facilities.
4. The road traffic safety facility information collection and processing system according to claim 1, characterized in that, The specific process for generating the traffic safety facility warning signal is as follows: The abnormal traffic event matching report is combined with the traffic safety facility aging prediction value, input into the preset multi-level traffic management risk decision rule library, and the output results are integrated to generate a traffic safety facility early warning signal; The multi-level traffic management risk decision-making rule base includes three levels of decision-making rules: high-level alarm generation rules, medium-level alarm generation rules, and low-level alarm generation rules.
5. The road traffic safety facility information collection and processing system according to claim 1, characterized in that, The initial collaborative information package includes event identification information, traffic safety facility early warning signals, and propagation control information, specifically: Event identification information includes the originating node identifier, timestamp, and geographic coordinates; traffic safety facility early warning signals include alarm level, traffic safety facility aging prediction value, and abnormal traffic event matching report; propagation control information includes recommended emergency response and maximum propagation hops.
Citation Information
Patent Citations
Highway guardrail health monitoring and early warning system based on big data
CN121768192A
Guardrail full-life-cycle health management system based on digital twinning
CN121860615A
