A tunnel passing risk prediction method and system based on multi-source data

Abstract

The application provides a tunnel passing risk prediction method and system based on multi-source data, and relates to the technical field of risk prediction. The method comprises the following steps: using a structure evaluation method based on the resilience theory to analyze the influence degree of the vehicle characteristic data on the tunnel state to obtain a characteristic influence coefficient, and dynamically adjusting the characteristic influence coefficient according to different scenes to obtain a risk weight; calculating the causal association features of the internal passing state data and the risk weight through an invariant anomaly detection method; establishing a risk prediction model, inputting real-time internal passing state data, and outputting to obtain a prediction result. According to the dynamic adjustment of the characteristic influence coefficient according to different scenes, the model can better adapt to the risk changes of the tunnel under different operation states, reduce the probability of accidents, and improve the passing efficiency and safety of the tunnel.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of risk prediction, in particular to a tunnel passing risk prediction method and system based on multi-source data. BACKGROUND

[0002] With the acceleration of urbanization, as a key node of the traffic network, the tunnel passing safety is facing severe challenges. The traditional tunnel risk prediction method has significant limitations: single data utilization, relying on traffic flow and other basic parameters, ignoring the differentiated influence of vehicle characteristics (such as large vehicle size, braking performance) on tunnel structure and passing state, leading to one-sided risk assessment; static evaluation of the influence of vehicle characteristics, using fixed weight to calculate risk, without considering the dynamic influence of scenario changes (such as heavy rain, construction) on the risk contribution of the same characteristics, for example, the risk of "vehicle speed standard deviation" in heavy rain is much higher than in sunny days, and fixed weight is difficult to adapt to actual risk fluctuations; the prediction model lacks a dynamic updating mechanism and cannot respond to real-time drift of tunnel internal state (such as sudden equipment failure, sudden change of traffic flow) and causal relationship, resulting in a decrease in long-term prediction accuracy and difficulty in meeting the safety management and control needs in complex environments.

[0003] In the prior art, risk correlation analysis relies on simple correlation, which is easy to misjudge false correlation as causal relationship, further exacerbating prediction bias. Therefore, there is an urgent need for a tunnel passing risk prediction method and system based on multi-source data to accurately quantify the scenario risk contribution of vehicle characteristics, capture real causal relationships, and improve the timeliness and accuracy of tunnel passing risk prediction. SUMMARY

[0004] The present application provides a tunnel passing risk prediction method and system based on multi-source data to solve the defects that different vehicle characteristics are easily ignored in the prior art, and the model cannot adapt to scenario changes and dynamic updates.

[0005] In one aspect, the present application provides a tunnel passing risk prediction method based on multi-source data, comprising:

[0006] Obtaining vehicle flow data of the road network around the tunnel, filtering vehicle characteristic data passing through the tunnel from the vehicle flow data, and collecting real-time tunnel internal passing state data.

[0007] Using a structure evaluation method based on resilience theory to analyze the influence degree of vehicle characteristic data on tunnel state to obtain a characteristic influence coefficient, and dynamically adjusting the characteristic influence coefficient according to different scenarios to obtain a risk weight.

[0008] Calculating the causal relationship characteristics of the internal passing state data and the risk weight by an invariant anomaly detection method.

[0009] Establish a risk prediction model, update the risk prediction model according to risk weights and causal correlation characteristics, input real-time internal traffic status data, and output the prediction results.

[0010] The present invention provides a tunnel traffic risk prediction method based on multi-source data, comprising:

[0011] Feature extraction is performed on vehicle flow data, and vehicles that will pass through the tunnel are screened out based on the time when the vehicles enter and leave the tunnel, the license plate recognition system, the license plate records of the tunnel entrances and exits, and the vehicle's driving path information.

[0012] Vehicles are classified, and their length, width, height, and weight are obtained to obtain vehicle attribute characteristics. Status characteristics are then obtained by analyzing the vehicle's braking performance, tire wear, light signal effectiveness, cargo loading stability, and fuel and battery status.

[0013] The present invention provides a tunnel traffic risk prediction method based on multi-source data, comprising:

[0014] The resilience index is defined by the ability of the tunnel structure to recover normal function after being disturbed.

[0015] Based on the different responses of the tunnel structure under vehicle passing and leaving conditions and combined with toughness indicators, damage data is generated, and the fragility curve is fitted to calculate the probability of the tunnel's damage state under different vehicle characteristics.

[0016] An exponential function is selected as the repair function for the functional recovery process of the tunnel structure after damage.

[0017] The functional state after damage is determined based on the damage state probability, and the repair function is adjusted to obtain the functional function of the tunnel structure.

[0018] The characteristic influence coefficient is obtained by combining the influence of vehicle characteristics on the toughness index.

[0019] The present invention provides a tunnel traffic risk prediction method based on multi-source data, including: the formula for damage state probability is expressed as:

[0020] ;

[0021] Where, The tunnel structure is strong Lower damage level Greater than or equal to The probability of is the cumulative distribution function of the standard normal distribution, The degree of damage caused The characteristic strength of is the shape parameter of the fragility curve.

[0022] The application provides a tunnel passing risk prediction method based on multi-source data, comprising: dividing multiple scenes according to actual tunnel operation, and converting each scene into a quantitative parameter, wherein the quantitative parameter comprises a traffic pressure parameter, an environmental interference parameter, a space constraint parameter and a consequence amplification parameter.

[0023] According to the quantitative parameter, a scene sensitivity factor of a vehicle feature in different scenes is calculated.

[0024] According to the feature influence coefficient and the scene sensitivity factor, a risk weight of each vehicle feature in different scenes is calculated.

[0025] Real-time scene parameters are collected at a preset time interval, and are compared with the quantitative parameters to determine whether a preset similarity is met, if yes, corresponding risk weights are called, and if not, current risk weights are continuously used.

[0026] The application provides a tunnel passing risk prediction method based on multi-source data, comprising: extracting key features from internal passing data, and constructing a detection model through definition of condition invariable regularization.

[0027] According to the detection model, abnormal changes are defined according to changes in data distribution and causal relationship, and causal influences of different key features on corresponding risk weights are analyzed.

[0028] The total effect of internal passing state data on risk weights is calculated by using a ridge regression method, false effects are identified by the detection model, and causal correlation features are obtained by subtraction.

[0029] The application provides a tunnel passing risk prediction method based on multi-source data, comprising: determining an analysis target by quantifying the influence degree of internal passing state data on risk weights, and taking features related to risk weights as training data.

[0030] A regression model is constructed by using a ridge regression method, and is trained by using the training data, and a regularization term is introduced to define a loss function.

[0031] Optimized parameters are obtained by calculating regression model parameters by minimizing the loss function.

[0032] A ridge regression model is constructed by using the optimized parameters, and the total effect of internal passing state data on risk weights is calculated.

[0033] The application provides a tunnel passing risk prediction method based on multi-source data, comprising: selecting a machine learning model as a risk prediction model, taking vehicle feature data and internal passing data as initial input features, taking historical risk events as label data, and dividing the label data into a training set and a validation set according to a preset proportion.

[0034] The risk prediction model is adjusted according to the causal correlation characteristics, and the performance of the updated model is verified using the verification set data.

[0035] The change data is obtained by monitoring the changes of the internal traffic state data and the risk weight in real time, and the causal correlation characteristics are updated continuously.

[0036] According to the change data and the updated causal correlation characteristics, and combining the online learning algorithm, the risk prediction model parameters are updated.

[0037] The present application provides a tunnel traffic risk prediction method based on multi-source data, comprising: extracting the same data as the key characteristics from the real-time internal traffic state data as the key characteristic organization, and extracting the core causal characteristics from the key characteristic organization based on the updated causal correlation characteristics.

[0038] According to the corresponding risk weight called by the real-time scene parameters, the dynamic input feature set is formed by fusing the core causal characteristics.

[0039] The dynamic input feature set is input into the updated risk prediction model, and the contribution proportion of different key characteristic organizations is adjusted according to the current risk weight.

[0040] The core causal characteristics reaching the preset correlation strength are calculated preferentially, and the prediction result is output by matching with the historical risk events. The prediction result includes the risk level, the core risk factor and the risk evolution trend.

[0041] On the other hand, the present application also provides a tunnel traffic risk prediction system based on multi-source data, which adopts any one of the tunnel traffic risk prediction methods based on multi-source data as described above, and the early warning system comprises:

[0042] The data collection module is used for acquiring the vehicle flow data of the road network around the tunnel, screening the vehicle characteristic data passing through the tunnel from the vehicle flow data, and collecting the internal traffic state data of the tunnel in real time.

[0043] The weight acquisition module is used for obtaining the characteristic influence coefficient by analyzing the influence degree of the vehicle characteristic data on the tunnel state using the structure evaluation method based on the resilience theory, and dynamically adjusting the characteristic influence coefficient to obtain the risk weight according to different scenes.

[0044] The feature acquisition module is used for calculating the causal correlation characteristics of the internal traffic state data and the risk weight by the invariant anomaly detection method.

[0045] The prediction output module is used for establishing a risk prediction model, updating the risk prediction model according to the risk weight and the causal correlation characteristics, inputting the real-time internal traffic state data, and outputting the prediction result.

[0046] The tunnel passing risk prediction method and system based on multi-source data provided by the application calculate the characteristic influence coefficient based on a structure evaluation method based on the resilience theory, then dynamically adjust the risk weight by combining the scene sensitive factor, so that the risk proportion of the same characteristic flexibly changes with the scene, solving the problem of rigid static evaluation, and accurately adapting the risk characteristics of different scenes.

[0047] The tunnel passing risk prediction method and system based on multi-source data provided by the application continuously optimize the model parameters based on "real-time internal state change + risk weight update + dynamic adjustment of causal correlation characteristics", so that the model can respond to risk changes in real time, solving the problem of model staticization, and realizing active risk prevention and control. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0049] Figure 1 is one of the tunnel passing risk prediction method flowcharts based on multi-source data provided by the embodiment of the application;

[0050] Figure 2 is the second tunnel passing risk prediction method flowchart based on multi-source data provided by the embodiment of the application;

[0051] Figure 3 is Figure 2 is the method flowchart for acquiring the characteristic influence coefficient;

[0052] Figure 4 is the structure diagram of the tunnel passing risk prediction system based on multi-source data provided by the embodiment of the application. DETAILED DESCRIPTION

[0053] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in conjunction with the drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the protection scope of the present application.

[0054] The technical solutions in the present application will be described clearly and completely below in conjunction with the drawings in the present application. Figures 1-4 A tunnel passing risk prediction method and system based on multi-source data are described.

[0055] As shown in Figure 1 and Figure 2 , a tunnel passing risk prediction method based on multi-source data provided by an embodiment of the present application comprises:

[0056] Obtain vehicle flow data of a road network around the tunnel, filter vehicle feature data passing through the tunnel from the vehicle flow data, and collect tunnel internal passing state data in real time. The method for obtaining vehicle flow data can comprise: contacting the local traffic management department or the highway management bureau to obtain the vehicle flow data of the tunnel and its surrounding road network. Real-time vehicle flow data is obtained by using sensors and cameras in the intelligent transportation system. Vehicle detection sensors (such as loop coils, microwave radars, video monitoring cameras, etc.) are installed at the tunnel entrance, exit and surrounding key road sections to ensure that the equipment can accurately collect vehicle flow data. The data includes vehicle passing time, vehicle speed, vehicle type, license plate number, etc.

[0057] The step of obtaining vehicle feature data comprises:

[0058] Feature extraction is performed on the vehicle flow data, and vehicles that will pass through the tunnel are filtered out based on the time of vehicle entering and leaving the tunnel, the license plate recognition system, the license plate record at the tunnel entrance and exit, and the driving path information of the vehicle.

[0059] Classify vehicles, obtain vehicle attribute features such as length, width, height and weight of the vehicle, and analyze state features such as braking performance, tire wear degree, light signal effectiveness, cargo loading stability, and fuel and battery status of the vehicle.

[0060] The influence degree of vehicle feature data on the tunnel state is analyzed by using a structure evaluation method based on the resilience theory to obtain a feature influence coefficient, and the risk weight is obtained by dynamically adjusting the feature influence coefficient according to different scenarios.

[0061] As shown in Figure 3 , the step of obtaining the feature influence coefficient comprises:

[0062] Resilience is defined by the ability of a tunnel structure to recover normal function after being disturbed. Resilience indicators can include the following:

[0063] Rapid repairability: refers to the speed at which the tunnel structure recovers its function after being damaged.

[0064] Performance loss: indicates the degree of functional loss of the tunnel structure during the damage period.

[0065] Performance recovery: refers to the degree of functional recovery of the tunnel structure during the repair process.

[0066] Structural toughness index: comprehensively reflects the overall toughness of the tunnel structure during damage and repair processes.

[0067] Repairability index: indicates the degree to which the tunnel structure can be repaired after damage.

[0068] Based on the different responses of the tunnel structure under vehicle passing and leaving conditions and combined with the toughness index to generate damage data, the fragility curve is fitted and the probability of damage state of the tunnel under different vehicle characteristics is calculated. The formula is expressed as:

[0069] ;

[0070] Where, The tunnel structure is strong Lower damage level Greater than or equal to The probability of is the cumulative distribution function of the standard normal distribution, The degree of damage caused The characteristic strength of is the shape parameter of the fragility curve.

[0071] The exponential function is selected as the repair function of the tunnel structure after damage and the formula is expressed as:

[0072] ;

[0073] Where, It's time Tunnel function status at the time is the initial functional state, is the final functional state, is the initial time, is the recovery rate parameter.

[0074] The functional state after damage is determined based on the probability of the damage state, and the repair function is adjusted to obtain the functional function of the tunnel structure, which is expressed as follows:

[0075] ;

[0076] wherein, is the tunnel function state at time , is the initial function state, is the final function state, is the initial time, is the recovery rate parameter, is the function state after damage.

[0077] And the feature influence coefficient is obtained by combining the influence amount of the vehicle feature on the resilience index, and the formula is expressed as:

[0078] ;

[0079] wherein, is the influence coefficient of the first vehicle feature on the resilience index, is the first vehicle feature.

[0080] The step of obtaining the risk weight includes:

[0081] According to the actual operation of the tunnel, a plurality of scenes are divided, and each scene is converted into a quantitative parameter, and the quantitative parameter includes a traffic pressure parameter, an environmental interference parameter, a space constraint parameter and a consequence amplification parameter. The traffic pressure parameter: the ratio of traffic flow to design flow.

[0082] The environmental interference parameter: the ratio of rainfall to 20mm / h plus the ratio of low-illumination time to 1 hour.

[0083] The space constraint parameter: the sum of the construction road occupation rate and the lane reduction ratio.

[0084] The consequence amplification parameter: the ratio of dangerous goods vehicle frequency to 2 vehicles / hour plus historical accident severity.

[0085] According to the quantitative parameter, the scene sensitivity factor of the vehicle feature in different scenes is calculated, and the formula is expressed as:

[0086] ;

[0087] wherein, is the sensitivity factor of the first vehicle feature in the scene, is the sensitivity base value of the vehicle feature, is the association weight of the first vehicle feature and the first scene, is the first vehicle feature, is the first​ quantization parameter.

[0088] According to the characteristic influence coefficient and the scene sensitive factor, the risk weight of each vehicle characteristic in different scenes is calculated, and the formula is expressed as:

[0089]

[0090] In the formula, is the risk weight of the first vehicle characteristic in the scene , the number of vehicle characteristics is , and the th vehicle characteristic is

[0091] According to the preset time interval, the real-time scene parameters are collected and compared with the quantization parameters, and it is judged whether the preset similarity is met. If yes, the corresponding risk weight is called, otherwise the current risk weight is continued to be used.

[0092] The causal correlation characteristics of the internal traffic state data and the risk weight are calculated by the invariant anomaly detection method.

[0093] The steps of obtaining the causal correlation characteristics include:

[0094] The key features are extracted from the internal traffic data, which can include: the key features include traffic flow: the number of vehicles passing per minute.

[0095] Speed distribution: average speed, speed standard deviation, and proportion of speeding vehicles.

[0096] Vehicle distance and density: headway, lane occupancy.

[0097] Lane change frequency: number of lane changes per minute.

[0098] Abnormal driving behavior: number of emergency braking, lane deviation frequency.

[0099] Vehicle type composition: large / small vehicle proportion, presence of special vehicles.

[0100] Air quality: CO concentration, smoke concentration.

[0101] Visibility: visible distance.

[0102] Temperature and humidity and road surface state: temperature, humidity, water film thickness, road foreign matter.

[0103] Lighting and light data: average brightness, brightness uniformity, glare intensity.

[0104] Equipment operating state: lighting system failure rate, ventilation system exhaust volume, fire fighting system state. ​​​

[0105] Structural safety data: lining displacement, surrounding rock pressure.

[0106] Emergency and management data: construction road occupation information, temporary control measures.

[0107] Personnel behavior data: fatigue driving characteristics, distracted behavior.

[0108] The detection model is constructed by defining conditional invariant regularization, which is expressed as:

[0109] ;

[0110] Where, is the model prediction value, is the regularization parameter, and are data points in different environments, are all the parameters in the model, It is The risk weight of the sample, is the number of data points used for regularization.

[0111] Define abnormal changes based on changes in data distribution and causal relationships identified by the detection model, and analyze the causal impact of different key features on the corresponding risk weights, which may include:

[0112] The causal effect of traffic volume on risk weight: how much the risk weight increases for every 10% increase in traffic volume.

[0113] The causal effect of vehicle speed distribution on risk weight: the impact of average vehicle speed, standard deviation of vehicle speed and the proportion of speeding vehicles on risk weight.

[0114] The causal effect of vehicle distance and density on risk weights: The impact of headway and lane occupancy on risk weights.

[0115] Causal effect of lane changing frequency on risk weight: The impact of lane changing frequency on risk weight.

[0116] The causal effect of abnormal driving behavior on risk weight: the impact of the number of sudden braking and lane departure frequency on risk weight.

[0117] The causal effect of vehicle model composition on risk weights: the impact of the proportion of large cars / small cars and the existence of special vehicles on risk weights.

[0118] The causal effect of air quality on risk weights: the impact of CO concentration and smoke concentration on risk weights.

[0119] Causal effect of visibility on risk weights: The impact of visibility on risk weights.

[0120] Temperature and humidity and road condition causal effect on risk weight: temperature, humidity, water film thickness, and road foreign matter effect on risk weight.

[0121] Lighting and light data causal effect on risk weight: average brightness, brightness uniformity, and glare intensity effect on risk weight.

[0122] Equipment operating status causal effect on risk weight: lighting system failure rate, ventilation system exhaust volume, and fire safety system status effect on risk weight.

[0123] Structure safety data causal effect on risk weight: lining displacement and surrounding rock pressure effect on risk weight.

[0124] Emergency and management data causal effect on risk weight: construction road occupation information and temporary control measures effect on risk weight.

[0125] Personnel behavior data causal effect on risk weight: fatigue driving characteristics and distraction behavior effect on risk weight.

[0126] The total effect of internal traffic state data on risk weight is calculated using the ridge regression method, false effects are identified by detecting the model, and the causal correlation features are obtained by subtraction.

[0127] The steps to obtain the total effect include:

[0128] By quantifying the degree of influence of internal traffic state data on risk weight, the analysis target is determined, and the features related to risk weight are used as training data. Training data can include vehicle speed standard deviation, road humidity, CO concentration, traffic volume, whether there are dangerous goods vehicles, whether construction occupies roads, etc.

[0129] A regression model is constructed by the ridge regression method, and training data is used for training. The loss function is defined by introducing a regularization term, which is expressed as:

[0130] ;

[0131] In the formula, is the loss function, is the regularization term parameter, is the regularization term of ridge regression, is the number of samples, is the target variable value of the th sample, is the intercept term, is the independent variable value of the th sample,

[0132] The regression model parameters are calculated by minimizing the loss function to obtain the optimal parameters, which is expressed as:

[0133] ;

[0134] wherein, is a feature matrix, is a target variable vector, is an identity matrix, is a regression model parameter vector, is a transpose of the feature matrix.

[0135] The ridge regression model is constructed using the optimization parameters, and the total effect of the internal traffic state data on the risk weight is calculated, which is expressed by the formula:

[0136] ;

[0137] wherein, is a total effect, is a regression coefficient of the ridge regression model, is a key feature.

[0138] The risk prediction model is established, and the risk prediction model is updated according to the risk weight and the causal association feature, the real-time internal traffic state data is input, and the prediction result is output.

[0139] The step of updating the risk prediction model includes:

[0140] The machine learning model is selected as the risk prediction model, the vehicle feature data and the internal traffic data are selected as the initial input features, the historical risk events are selected as the label data, and the training set and the validation set are divided according to the preset proportion.

[0141] The risk prediction model is adjusted according to the causal association feature, and the step includes:

[0142] According to the strength of the causal association feature, the weight of the corresponding feature in the risk prediction model is adjusted.

[0143] The interaction effect in the causal association feature is added to the risk prediction model.

[0144] The risk prediction model is retrained using the training set.

[0145] The performance of the updated model is verified using the validation set data.

[0146] The change data is obtained by monitoring the changes of the internal traffic state data and the risk weight in real time, and the causal association feature is continuously updated.

[0147] The risk prediction model parameters are updated according to the change data and the updated causal association feature, and combined with the online learning algorithm.

[0148] The step of adjusting the parameters includes:

[0149] When the newly calculated causal correlation feature changes by more than the preset threshold compared to the previous time, the current risk weight differs from the historical risk weight by more than the threshold, and the prediction error of the last 3 time windows (such as 1 window every 10 minutes) exceeds the threshold, a forced update is triggered.

[0150] From the real-time monitored internal traffic state data and risk weight, the latest multiple data before triggering the update are intercepted as incremental training samples, and the newly calculated causal correlation feature (such as the causal strength of "road surface humidity → sudden braking risk weight") is added as a new feature to the input feature set of the incremental samples.

[0151] The stable and effective core parameters in the historical training of the model (such as the basic weight of "large vehicle proportion" on risk) are retained to avoid drastic changes in parameters due to fluctuations in incremental data, and the weight of the corresponding input feature in the model is adjusted according to the updated causal correlation feature strength:

[0152] For features with increased causal strength (such as "CO concentration → dangerous goods vehicle risk weight" from 0.5 to 0.7), increase their contribution proportion in the model (such as from 1.2 times the original weight to 1.5 times).

[0153] For features with reduced causal strength (such as "vehicle flow → congestion risk weight" from 0.6 to 0.3), reduce their contribution proportion (such as from 1.0 times the original weight to 0.6 times).

[0154] Based on the incremental samples, the small batch gradient descent method is used to fine-tune the model parameters, and only a small amount of new data (such as 30 per batch) is used for updating each iteration to avoid covering historical learning results:

[0155] Focus on optimizing parameters related to "changing data" (such as coefficients related to "road surface humidity" and "sudden braking times").

[0156] Control the learning rate (such as 0.01) to ensure smooth parameter updates (such as no more than 5% of the original parameter per adjustment).

[0157] Set parameter upper and lower limits: set a reasonable range (such as 0.1-1.0) for core parameters (such as weight coefficients of causal correlation features) to prevent parameters from exceeding physical meaning due to abnormal incremental data (such as the weight of "vehicle speed standard deviation" cannot be negative).

[0158] Introduce regularization constraints: during parameter updating, by constraining the difference between "new parameters and historical parameters", ensure that the overall trend of the model is consistent with the historical performance (such as the rule that "the increase of large vehicle proportion leads to the increase of risk" is not overturned).

[0159] Preserve key causal relationships: Enforce the maintenance of core relationships that have passed causal verification (such as "presence of a hazardous materials vehicle → increased explosion risk weight"), and the parameter update amplitude must not exceed 10% of the original value to ensure the interpretability of the model logic.

[0160] The steps to obtain the prediction results include:

[0161] The data identical to the key features are extracted from the real-time internal traffic status data as the key feature organization, and the core causal features are extracted from the key feature organization based on the updated causal association features.

[0162] The corresponding risk weight is called according to the real-time scenario parameters and fused with the core causal features to form a dynamic input feature set.

[0163] The dynamic input feature set is input into the updated risk prediction model, and the contribution ratio of different key feature organizations is adjusted according to the current risk weight.

[0164] Prioritizes the calculation of core causal features that meet the preset correlation strength, matches them with historical risk events, and outputs prediction results. The prediction results include risk level, core risk factors, and risk evolution trends.

[0165] Risk levels can be divided into four levels: "Low → Medium → High → Very High". Each level corresponds to a clear risk probability and potential consequences, making it easier to quickly determine the degree of urgency:

[0166] Low risk probability: The probability of a risk event occurring in the next 30 minutes is less than 10%.

[0167] Characteristics: stable traffic flow (vehicle speed standard deviation <5km / h), normal environment (CO concentration <50ppm, visibility >100 meters), no abnormal driving behavior (number of sudden brakes <1 time / 10 minutes).

[0168] Potential consequences: Basically no safety hazards, only slight traffic fluctuations may occur, and no special intervention is required.

[0169] Medium risk probability: The probability of a risk event occurring within the next 30 minutes is 10%~30%.

[0170] Characteristics: Slightly unstable traffic flow (vehicle speed standard deviation 5-10 km / h), slightly abnormal environmental parameters (such as road humidity 30%-50%, occasional lane change interference), and the presence of individual risky vehicles (such as an overloaded vehicle but driving stably).

[0171] Potential consequences: Local traffic slowdown or minor scratches may occur, requiring enhanced monitoring, but no emergency control is required.

[0172] High risk probability: The probability of a risk event occurring within the next 30 minutes is 30%~70%.

[0173] Features: Obvious traffic disorder (standard deviation of vehicle speed 10-20 km / h), abnormal environmental parameters (e.g. CO concentration 50-100 ppm, visibility 50-100 meters), frequent abnormal driving (brake times 3-5 times / 10 minutes, increased lane changes).

[0174] Potential consequences: Rear-end collisions, temporary equipment failure (e.g. local lighting outage), need to activate warning measures (e.g. speed limit prompts, broadcast guidance).

[0175] High risk probability: The probability of risk events occurring within the next 30 minutes is >70%.

[0176] Features: Severe loss of control of traffic flow (standard deviation of vehicle speed >20 km / h), serious over-standard environmental parameters (e.g. CO concentration >100 ppm, visibility <50 meters, road surface water film thickness >2 mm), presence of high-risk behaviors (e.g. dangerous goods vehicle illegal lane change, continuous emergency braking).

[0177] Potential consequences: Extremely likely to occur in serious accidents such as collisions, fires, and dangerous goods leaks, and immediate emergency response (e.g. closing part of the lane, guiding vehicles to evacuate) is required.

[0178] The core risk factors are the 2-3 key features selected from the "core causal characteristics" that contribute most to the risk level, and the "feature name + current state + impact logic" needs to be clear:

[0179] Core factors of basic traffic:

[0180] Example 1: "standard deviation of vehicle speed = 25 km / h (far beyond the safety threshold of 15 km / h) → traffic is extremely unstable, and the risk of rear-end collisions increases dramatically".

[0181] Example 2: "vehicle headway = 1.2 seconds (less than the safety threshold of 2 seconds) → the distance between the front and rear vehicles is too close, and there is not enough time to avoid".

[0182] Core factors of environmental state:

[0183] Example 1: "road surface humidity = 90% + water film thickness = 2.5 mm → vehicle braking distance increases by 30%, and emergency braking is prone to skidding".

[0184] Example 2: "CO concentration = 120 ppm (20% over standard) → drivers may experience dizziness and decreased reaction speed".

[0185] Core factors of vehicle behavior:

[0186] Example 1: "emergency braking times = 8 times / minute (in the last 5 minutes) → continuous braking leads to insufficient avoidance by the following vehicle, and the risk of collision accumulates".

[0187] Example 1: "Vehicle speed standard deviation = 15 km / h (above safety threshold 10 km / h) → risk of rear-end collision increases, potential for multi-vehicle pileup."

[0188] Device state class core factors:

[0189] Example 1: "Ventilation system load rate = 120% (above design value) → CO concentration continues to rise, unable to be effectively discharged."

[0190] Example 2: "Lighting brightness = 30 cd / m² (below safety threshold 50 cd / m²) → driver's vision is blurred, judgment error probability increases."

[0191] Risk evolution trend is based on "current core risk factor change rate + historical similar scenario evolution law" prediction, including dynamic change description in the next 30 minutes:

[0192] Trend direction includes upward trend, stable trend and downward trend.

[0193] Upward trend: core risk factor continues to deteriorate, risk level may upgrade (such as "current medium risk → high risk after 10 minutes"), for example: "vehicle flow increases at a rate of 50 vehicles per minute, predicts that lane occupancy will exceed 80% after 15 minutes, congestion risk upgrades."

[0194] Stable trend: core risk factor has no significant change, risk level maintains current level, for example: "road surface humidity stabilizes at 70%, vehicle flow fluctuates <10 vehicles per minute, risk level remains medium risk for the next 20 minutes."

[0195] Downward trend: core risk factor improves, risk level may decrease (such as "current high risk → medium risk after 20 minutes"), for example: "rain subsides → road surface humidity decreases from 90% to 60%, water film thickness gradually decreases, braking conditions improve."

[0196] Key time nodes:

[0197] Example 1: "If vehicle speed standard deviation continues to rise, it is expected to reach 20 km / h in 8 minutes, and the risk level will rise from 'high' to 'extremely high'."

[0198] Example 2: "After the ventilation system is repaired, CO concentration is expected to drop to 80 ppm within 15 minutes, and risk level will decrease from 'high' to'medium'."

[0199] Influence range expansion prompt:

[0200] Example 1: "Current risk is high in only 1 lane → if vehicle flow continues to increase, it may spread to 2 and 3 lanes within 10 minutes, overall risk expands."

[0201] Example 2: "Local construction occupies the road -> the risk of the current construction section is high, and the risk of congestion of the adjacent lane will rise in 5 minutes due to the traffic diversion."

[0202] As shown in Figure 4 The embodiment provides a tunnel passing risk prediction system based on multi-source data, which can be controlled by a tunnel passing risk prediction method based on multi-source data. The risk prediction system comprises:

[0203] A data collection module is configured to obtain vehicle flow data of a road network around a tunnel, filter vehicle feature data passing through the tunnel from the vehicle flow data, and collect internal passing state data of the tunnel in real time.

[0204] A weight acquisition module is configured to analyze the influence degree of the vehicle feature data on the tunnel state by using a structure evaluation method based on the resilience theory to obtain a feature influence coefficient, and dynamically adjust the feature influence coefficient according to different scenes to obtain a risk weight.

[0205] A feature acquisition module is configured to calculate a causal correlation feature of the internal passing state data and the risk weight by using an invariant anomaly detection method.

[0206] A prediction output module is configured to establish a risk prediction model, update the risk prediction model according to the risk weight and the causal correlation feature, input real-time internal passing state data, and output a prediction result.

[0207] The present invention provides a tunnel traffic risk prediction method and system based on multi-source data. This system utilizes multi-source data fusion, including vehicle flow data from the tunnel's surrounding road network, vehicle characteristics data, and tunnel internal traffic status data. This system improves the accuracy and comprehensiveness of risk prediction, enabling a more comprehensive reflection of the complexities of tunnel traffic. Dynamically adjusts the characteristic impact coefficients based on different scenarios to obtain risk weights. Simultaneously, changes in internal traffic status data and risk weights are monitored in real time, and causal correlation features and model parameters are dynamically updated. This improves the model's dynamic adaptability and flexibility, enabling it to better respond to changing traffic and environmental conditions. By calculating the causal correlation features between internal traffic status data and risk weights using an invariant anomaly detection method, spurious effects are identified and eliminated, improving the scientific nature and reliability of risk prediction and preventing spurious correlations from interfering with prediction results. By monitoring changes in internal traffic status data and risk weights in real time and dynamically updating model parameters using an online learning algorithm, the system improves its real-time performance and accuracy, enabling it to promptly reflect changes in tunnel traffic risk. Dynamically adjusting the characteristic impact coefficients based on different scenarios allows the model to better adapt to changes in tunnel risk under different operating conditions. This dynamic adjustment mechanism enables the model to maintain high predictive performance even in the face of complex and changing traffic and environmental conditions. It provides real-time decision support for tunnel managers, enabling them to take effective management measures in a timely manner, reduce the probability of accidents, and improve tunnel efficiency and safety.

[0208] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0209] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A tunnel traffic risk prediction method based on multi-source data, characterized in that: include: Acquire vehicle flow data of the road network surrounding the tunnel, filter characteristic data of vehicles passing through the tunnel from the vehicle flow data, and collect traffic status data inside the tunnel in real time; Using a structural assessment method based on toughness theory to analyze the degree of influence of the vehicle characteristic data on the tunnel state to obtain a characteristic influence coefficient, and dynamically adjusting the characteristic influence coefficient according to different scenarios to obtain a risk weight; The step of obtaining the characteristic influence coefficient includes: Resilience is defined by the ability of the tunnel structure to recover normal function after being disturbed; Based on the different responses of the tunnel structure under vehicle passing and leaving conditions and combined with the toughness index, damage data is generated, and the fragility curve is fitted to calculate the damage state probability of the tunnel under different vehicle characteristics. The formula is expressed as follows: ; Where, The tunnel structure is strong Lower damage level Greater than or equal to The probability of is the cumulative distribution function of the standard normal distribution, The degree of damage caused The characteristic strength of is the shape parameter of the fragility curve; An exponential function is selected as the repair function for the functional recovery process of the tunnel structure after damage; Determining a functional state after damage based on the damage state probability, and adjusting the repair function to obtain a functional function of the tunnel structure; and obtaining the characteristic influence coefficient in combination with the influence of the vehicle characteristic on the toughness index; The steps of obtaining the risk weight include: Divide multiple scenarios based on actual tunnel operations and convert each scenario into quantitative parameters, including traffic pressure parameters, environmental interference parameters, spatial constraint parameters, and consequence amplification parameters; Calculating scene sensitivity factors of different scenes to the vehicle feature according to the quantified parameters; Calculate the risk weight of each vehicle feature under different scenarios based on the feature impact coefficient and the scenario sensitivity factor; Collect real-time scenario parameters at preset time intervals and compare them with the quantitative parameters to determine whether they meet the preset similarity. If so, call the corresponding risk weight; otherwise, continue to use the current risk weight; Calculating the causal correlation characteristics between the internal traffic status data and the risk weight using an invariant anomaly detection method; Establishing a risk prediction model, updating the risk prediction model according to the risk weight and the causal correlation characteristics, inputting real-time internal traffic status data, and outputting a prediction result; The step of obtaining the causal association feature includes: Extracting key features from the internal traffic data and constructing a detection model by defining conditional invariant regularization; Identify changes in data distribution and causal relationships based on the detection model to define abnormal changes, and analyze the causal impact of different key features on the corresponding risk weights; The total effect of the internal traffic status data on the risk weight is calculated using the ridge regression method, and false effects are identified through the detection model, and the causal association characteristics are obtained by subtraction.

2. The tunnel traffic risk prediction method based on multi-source data according to claim 1 is characterized in that: The step of obtaining the vehicle characteristic data includes: Extracting features from the vehicle flow data, and screening out vehicles that will pass through the tunnel based on the time when the vehicles enter and exit the tunnel, the license plate recognition system, and the license plate records at the tunnel entrances and exits, as well as the vehicle's travel path information; Vehicles are classified, and their length, width, height, and weight are obtained to obtain vehicle attribute characteristics. Status characteristics are then obtained by analyzing the vehicle's braking performance, tire wear, light signal effectiveness, cargo loading stability, and fuel and battery status.

3. The tunnel traffic risk prediction method based on multi-source data according to claim 1 is characterized in that: The steps of obtaining the total effect include: Determining an analysis objective by quantifying the degree of influence of the internal traffic status data on the risk weight, and using features related to the risk weight as training data; A regression model is constructed by a ridge regression method, and the training data is used for training, and a regularization term is introduced to define a loss function; Calculating the parameters of the regression model by minimizing the loss function to obtain optimized parameters; A ridge regression model is constructed using the optimized parameters to calculate the total effect of the internal traffic status data on the risk weight.

4. The tunnel traffic risk prediction method based on multi-source data according to claim 3 is characterized in that: The step of updating the risk prediction model includes: Selecting a machine learning model as a risk prediction model, using the vehicle feature data and the internal traffic data as initial input features, using historical risk events as label data, and dividing the data into a training set and a validation set according to a preset ratio; Adjusting the risk prediction model according to the causal association characteristics, and verifying the updated model performance using the validation set data; Real-time monitoring of changes in the internal traffic status data and the risk weights to obtain change data, and continuously updating the causal association characteristics; The risk prediction model parameters are updated based on the change data and the updated causal association features, and in combination with an online learning algorithm.

5. The tunnel traffic risk prediction method based on multi-source data according to claim 4 is characterized in that: The steps of obtaining the prediction result include: Extracting data identical to the key features from the real-time internal traffic status data as a key feature organization, and extracting core causal features from the key feature organization based on the updated causal association features; The corresponding risk weight is called according to the real-time scenario parameters and integrated with the core causal features to form a dynamic input feature set; Inputting the dynamic input feature set into the updated risk prediction model, and adjusting the contribution ratio of different key feature organizations according to the current risk weight; Prioritize calculating the core causal features that reach a preset correlation strength, and match them with the historical risk events to output the prediction results; the prediction results include risk levels, core risk factors, and risk evolution trends.

6. A tunnel traffic risk prediction system based on multi-source data, which adopts the tunnel traffic risk prediction method based on multi-source data according to any one of claims 1 to 5, characterized in that: The prediction system includes: A data collection module is used to obtain vehicle flow data of the road network around the tunnel, filter characteristic data of vehicles passing through the tunnel from the vehicle flow data, and collect traffic status data inside the tunnel in real time; a weight acquisition module, configured to analyze the degree of influence of the vehicle characteristic data on the tunnel state using a structural assessment method based on toughness theory to obtain a characteristic influence coefficient, and dynamically adjust the characteristic influence coefficient according to different scenarios to obtain a risk weight; a feature acquisition module, configured to calculate a causal correlation feature between the internal traffic status data and the risk weight using an invariant anomaly detection method; The prediction output module is used to establish a risk prediction model, update the risk prediction model according to the risk weight and the causal correlation characteristics, input real-time internal traffic status data, and output a prediction result.

Images