Traffic risk hidden danger monitoring method based on artificial intelligence

Abstract

The invention discloses a traffic risk hidden danger monitoring method based on artificial intelligence, and relates to the technical field of intelligent traffic. The method comprises the following steps: collecting traffic operation data and road meteorological data in real time; a BP neural network model is constructed, artificial feature engineering and stacked auto-encoder automatic learning are combined, risk influence factor vectors are generated, and road section risk values are predicted through nonlinear mapping and classified; generating an initial disposal strategy by using a particle swarm algorithm, verifying and optimizing in combination with a traffic flow simulation model, and dynamically adjusting parameters such as signal lamps and reversible lanes; based on regional path guidance (sorting according to destination distances) and dynamic traffic efficiency calculation, hierarchical management and control are implemented. According to the method, through a prediction-disposal-verification closed loop mechanism, the accident rate and the congestion duration are reduced, and the efficient configuration and the safety level of traffic resources in a complex road network environment are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent transportation, and particularly relates to a traffic risk hidden danger monitoring method based on artificial intelligence. BACKGROUND

[0002] With the acceleration of urbanization and the continuous growth of traffic demand, traffic congestion, traffic accidents and sudden disaster risks have become key problems restricting the safe and efficient operation of modern transportation systems. Traditional traffic management methods mainly rely on static rules and manual experience, which are difficult to cope with complex and variable traffic scenes and sudden events. In addition, the existing technology often has a lag in risk warning, making it difficult to achieve the organic combination of pre- prevention and dynamic intervention, resulting in low resource allocation efficiency, slow accident handling response and other problems.

[0003] In recent years, the rapid development of artificial intelligence technology has provided a new idea for traffic risk management. For example, a traffic flow prediction model based on deep learning can mine potential risk patterns through multi-source data such as traffic flow, speed, and weather information, while machine learning algorithms such as BP neural network and particle swarm optimization algorithm can realize the automatic extraction of risk factors and the dynamic optimization of disposal strategies. However, the existing AI applications still have the following limitations: Single feature extraction: Most models only rely on artificially defined features (such as traffic flow fluctuation rate and visibility decay rate), ignoring the nonlinear interaction between traffic flow and weather conditions (such as the combined feature of precipitation intensity + speed standard deviation), resulting in limited prediction accuracy.

[0004] Static risk disposal strategy: Traditional control measures (such as fixed lane number and preset signal period) are difficult to dynamically adjust according to real-time risk levels, lacking a linkage mechanism for coordinated control of high-risk road sections and resource optimization of low-risk road sections.

[0005] Insufficient multi-source data fusion: The existing system has limited integration capability for traffic operation data, weather data, and road structure state (such as slope displacement and bridge deformation), making it difficult to build a global perception intelligent monitoring system.

[0006] The present application is aimed at the above technical bottlenecks and proposes a traffic risk hidden danger monitoring method based on artificial intelligence SUMMARY

[0007] In order to overcome the shortcomings and deficiencies of the existing technology, the present application adopts the following technical solutions: The traffic risk hidden danger monitoring method based on artificial intelligence has the following process: Step 1, data acquisition: real-time acquisition of traffic operation data by monitoring equipment; at the same time, real-time acquisition of road weather data in a future time period from a weather platform; Step two, risk prediction: obtain historical traffic data and historical meteorological data; input the historical traffic data and the historical meteorological data into a BP neural network for training to establish a BP neural network model, then input the traffic operation data and the road meteorological data of each road section into the BP neural network model, obtain the predicted risk data of each road section, and divide the road sections into high-risk road sections, medium-risk road sections and low-risk road sections based on the road section risk values in the predicted risk data; Step three, risk disposal strategy generation and optimization: obtaining a first risk disposal strategy according to the predicted risk data by using a particle swarm algorithm, and adjusting the first risk disposal strategy according to the simulation results of the traffic flow simulation model to obtain a second risk disposal strategy; Step four, traffic control scheme formulation and execution: obtaining traffic demand information, executing path guidance based on the second risk disposal strategy, and dynamically adjusting lane configuration and signal control parameters.

[0008] Preferably, the process of establishing the BP neural network model in step two comprises: inputting the historical traffic data and the historical meteorological data into the input layer of the BP neural network, obtaining risk influence factors according to the historical traffic data and the historical meteorological data, randomly obtaining first weight values and first bias values of the risk influence factors, and obtaining first prediction values according to the first weight values and the first bias values by using a first activation function; taking the first prediction values as inputs of a hidden layer, obtaining loss values according to the first prediction values and true risk values by using a loss function, adjusting the first weight values and the first bias values according to the loss values, obtaining second weight values and second bias values, repeating the above steps n times until the loss values are less than or equal to a loss expected threshold, then obtaining final weight values and final bias values; and obtaining predicted risk data as outputs of an output layer by using a second activation function according to the final weight values and the final bias values.

[0009] Preferably, the process of obtaining the risk influence factors adopts a hybrid mode combining manual definition and automatic learning; wherein 12 types of core risk factors including traffic flow fluctuation rate, speed variation coefficient, lane occupancy rate peak value, illegal density, accident frequency, visibility attenuation rate, road surface temperature extreme value, precipitation intensity grade, wind force grade, road surface humidity threshold, road curvature, and intersection density are extracted from original data by using manual feature engineering; 8 types of high-order feature factors including traffic flow-speed joint distribution characteristics reflecting traffic flow state high-order mode and precipitation intensity + speed standard deviation combination characteristics reflecting meteorological-traffic interaction characteristics are generated by using a stacked autoencoder to perform nonlinear transformation on the original data; and the manual features and the automatic features are combined into a 20-dimensional risk influence factor vector.

[0010] Preferably, the real risk value is generated by weighted fusion of the accident risk value and the congestion risk value; wherein the weight coefficient of the accident risk value is greater than the weight coefficient of the congestion risk value; and the real risk value ranges from 0 to 1 after normalization processing.

[0011] Preferably, in step two, the road section with a risk value greater than or equal to the first risk threshold is defined as a high-risk road section, the road section with a risk value less than or equal to the second risk threshold is defined as a low-risk road section, and the road section with a risk value between the second risk threshold and the first risk threshold is defined as a medium-risk road section.

[0012] Preferably, in step three, the risk association information of the high-risk road section is obtained, the risk association information including the road types, road location data, traffic flow data and signal control data of other road sections except the high-risk road section, the risk location data of the high-risk road section is obtained, the road section disposal distance is obtained according to the road location data and the risk location data, the risk disposal weight value of each road section is determined according to the road section disposal distance and the road type, the road section with the maximum risk disposal weight value is obtained and defined as the cooperative control road section of the high-risk road section; the associated optimization information of the low-risk road section is obtained, the associated optimization information including the road types, road location data, traffic flow data and signal control data of other road sections except the low-risk road section, the optimization location data of the low-risk road section is obtained, the road section optimization distance is obtained according to the road location data and the optimization location data, the traffic flow guiding value of each road section is determined according to the road section optimization distance and the road type, the road section with the maximum traffic flow guiding value is obtained and defined as the flow guiding road section of the low-risk road section, and the above steps are repeated to obtain the first risk disposal strategy.

[0013] Preferably, determining the traffic flow guiding value of each road section according to the road section optimization distance and the road type comprises: obtaining a distance influence coefficient according to the road section optimization distance, the distance influence coefficient being a normalized coefficient calculated based on the road section optimization distance; obtaining a road grade coefficient according to the road type, the road grade coefficient being a continuous function based on the real risk value, wherein the coefficient of the high-risk road section decreases as the risk value decreases, the coefficient of the low-risk road section increases as the risk value increases, and the coefficient of the medium-risk road section is 0; and performing weighted calculation on the distance influence coefficient and the road grade coefficient to obtain the traffic flow guiding value.

[0014] Preferably, determining the risk treatment weight value of each road section according to the road section treatment distance and the road type comprises: obtaining a distance attenuation coefficient according to the road section treatment distance, the distance attenuation coefficient being a normalized coefficient calculated based on the road section treatment distance, reflecting the influence degree of the spatial distance between the target road section and the current road section on the risk treatment effect, the closer the road section distance, the higher the priority of risk treatment, and the farther the distance, the lower the priority; obtaining a road risk coefficient according to the road type, the road risk coefficient being a compensation function based on the risk diffusion cost, reflecting the priority of the road section to treatment, wherein the coefficient of the high-risk road section is negative and the higher the risk value, the greater the absolute value, the coefficient of the low-risk road section is positive and the lower the risk value, the greater the value, and the coefficient of the medium-risk road section is 0; and performing weighted calculation on the distance attenuation coefficient and the road risk coefficient to obtain the risk treatment weight value.

[0015] Preferably, in step four, it further comprises regional path induction: including regional division, path recommendation and priority sorting, wherein the regional division is to divide the city into several traffic blocks according to the road network topology, the path recommendation is to induce vehicles to bypass to low-risk road sections through V2X communication or APP push around the high-risk road sections, and the priority sorting is to sort according to the distance between the vehicle destination and the high-risk road section to preferentially induce transit vehicles to bypass; preferably, in step four, it further comprises a traffic efficiency calculated by a formula, which takes into account the actual traffic capacity of the road section, the number of lanes, the basic traffic capacity of a single lane and an adjustment coefficient; preferably, in step four, it further comprises a strategy of dynamic tidal lane and adaptive signal control: the current flow and the predicted flow are obtained in real time through the road section monitoring device, if the current flow is close to congestion, i.e. greater than or equal to the proportion of the actual traffic capacity of the road section, the tidal lane is enabled, if the current flow is low flow, i.e. less than or equal to the proportion of the actual traffic capacity of the road section, the number of open lanes is reduced, the basic lane is reserved, and the current number of lanes is maintained in the intermediate state, and the signal light cycle is adjusted in real time according to the flow.

[0016] In summary, due to the adoption of the above technical solutions, the present application has the following advantages: 1. The present application can accurately predict the risk level of each road section by real-time collection of traffic operation data and road weather data, combined with the hybrid feature extraction mode of BP neural network, artificial definition of core risk factors and automatic learning of high-order feature factors, and can classify and control the road sections into high, medium and low risk road sections, thereby providing accurate basis for subsequent treatment strategies.

[0017] 2. After the first risk treatment strategy is generated by the particle swarm algorithm, the second risk treatment strategy is obtained by adjusting and optimizing the simulation results of the traffic flow simulation model, so that the strategy can better adapt to the actual traffic conditions and improve the treatment effect on traffic risks.

[0018] 3、The application realizes accurate screening and guiding of the controlled object through regional path induction, cooperates with dynamic tidal lane and self-adaptive signal control, can effectively reduce accident risk and relieve traffic congestion, realizes efficient allocation of traffic resources, and comprehensively improves the safety level and operation efficiency of the traffic system. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of these drawings.

[0020] Figure 1 The flowchart of the traffic risk hidden danger monitoring method based on artificial intelligence of the present application is shown; Figure 2 The flowchart of step two of the present application is shown; Figure 3 The flowchart of step three of the present application is shown. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0022] In addition, the described features, structures or characteristics can be combined in any suitable way in one or more example embodiments. In the following description, many specific details are provided to give a sufficient understanding of example embodiments of the present disclosure. However, those skilled in the art will realize that the technical solutions of the present disclosure can be practiced without one or more of the specific details, or can employ other methods, components, steps, etc. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring aspects of the present disclosure.

[0023] Embodiment 1: Referring to Figure 1 The traffic risk hidden danger monitoring method based on artificial intelligence of the present embodiment is shown as follows: Step one, data acquisition.

[0024] Real-time traffic operation data is obtained by using a monitoring device, and the traffic operation data includes traffic flow data, vehicle speed data, lane occupancy rate data, illegal data and accident data; The traffic volume data includes average traffic volume per unit time and traffic volume distribution per unit time; The vehicle speed data includes interval average vehicle speed, instantaneous vehicle speed, and vehicle speed standard deviation; The lane occupancy rate data includes time occupancy rate and space occupancy rate; the violation data includes violation type, violation frequency, and violation location; The accident data includes accident type, accident casualty, accident property loss, and accident cause; the road weather data in a future time period is obtained from a meteorological platform in real time, and the road weather data includes visibility, road temperature, precipitation intensity, wind force level, road humidity, and ice and snow coverage.

[0025] Step two, risk prediction.

[0026] Referring to Figure 2 As shown in the figure, the step two process is as follows: Obtain historical traffic data and historical weather data; The historical traffic data includes historical traffic volume data, historical vehicle speed data, historical lane occupancy rate data, historical violation data, and historical accident data; the historical weather data includes historical visibility, historical road temperature, historical precipitation intensity, historical wind force level, historical road humidity, and historical ice and snow coverage.

[0027] The historical traffic data and the historical weather data are input into a BP neural network for training to establish a BP neural network model, which specifically includes: The historical traffic data and the historical weather data are input into the input layer of the BP neural network, the risk influence factors are obtained according to the historical traffic data and the historical weather data, the first weight value and the first bias value of the risk influence factors are randomly obtained, the first prediction value is obtained according to the first weight value and the first bias value by using the first activation function, the first prediction value is taken as the input of the hidden layer, the loss value is obtained according to the first prediction value and the true risk value by using the loss function, the first weight value and the first bias value are adjusted according to the loss value, the second weight value and the second bias value are obtained, the above steps are repeated n times until the loss value is less than or equal to the expected loss threshold, then the final weight value and the final bias value are obtained, and the predicted risk data is obtained by using the second activation function according to the final weight value and the final bias value, as the output of the output layer; The first weight value is a randomly initialized weight parameter (denoted as W1) in the linear transformation from the input layer to the hidden layer of the BP neural network, which is used for linear weighting of the input risk influence factor vector to generate the first prediction value.

[0028] The first bias value is a randomly initialized bias parameter (denoted as b1) in the linear transformation from the input layer to the hidden layer of the BP neural network, which cooperates with the first weight value to adjust the baseline of the linear transformation to generate the first prediction value.

[0029] The first activation function in the hidden layer is used to introduce nonlinearity to the first prediction value (linear transformation result) and map the linear output to nonlinear features, thereby enhancing the expression ability of the model and generating the hidden layer output H.

[0030] The first prediction value input layer is linearly transformed by the first weight value (W1) and the first bias value (b1).

[0031] The second weight value is the weight parameter from the input layer to the hidden layer adjusted according to the loss value during the BP neural network training process, which is used to map the hidden layer output to the second prediction value.

[0032] The second bias value is the bias parameter (denoted as b2) from the hidden layer to the output layer adjusted according to the loss value during the BP neural network training process, which cooperates with the second weight value to adjust the linear transformation reference and generate the second prediction value.

[0033] The final weight value is the weight parameter (W2) from the hidden layer to the output layer after multiple iterations (repeated n times) of training, which satisfies the condition that the loss value is less than or equal to the expected threshold, and is used to generate the second prediction value.

[0034] The final bias value is the bias parameter (b2) from the hidden layer to the output layer after multiple iterations of training, which satisfies the condition that the loss value is less than or equal to the expected threshold, and cooperates with the final weight value to generate the second prediction value.

[0035] The second activation function in the output layer is used to convert the second prediction value (linear transformation result) into the final road segment risk value (such as Softmax, linear activation), ensuring that the output range is consistent with the true risk value (normalized to [0, 1]), and generating the final road segment risk value R.

[0036] The risk influence factor acquisition process is: a hybrid mode of manual definition and automatic learning is adopted; 12 types of core risk factors (traffic flow fluctuation rate, speed variation coefficient, lane occupancy rate peak value, violation density, accident frequency, visibility attenuation rate, road surface temperature extreme value, precipitation intensity grade, wind grade, road surface humidity threshold, road curvature, intersection density) are extracted from the original data by manual feature engineering; automatic feature learning is performed by stacking autoencoders to perform nonlinear transformation on the original data, generating 8 types of high-order feature factors, which are combined with the manual features to form a 20-dimensional risk influence factor vector. High-order feature factors include high-order patterns of traffic flow state: such as joint distribution characteristics of traffic flow-speed (reflecting the correlation between congestion and speed). Meteorological-traffic interaction features: such as the combination of precipitation intensity + speed standard deviation (reflecting the driving behavior fluctuation in rainy weather).

[0037] Wherein, the real risk value is the key label of BP neural network training, used to measure the severity and frequency of risk events (such as accidents, congestion) actually occurred in the historical period of the road section, and the real risk value is generated by weighted fusion of accident risk value (direct risk) and congestion risk value (indirect risk): the accident risk value reflects the severity and frequency of historical accidents of the road section, ; the congestion risk value reflects the state and duration of historical congestion of the road section, ; ; ; , , , , and are weight coefficients, , since the direct harm of the accident has higher priority; , respectively reflect the influence priority of accident severity and accident frequency on accident risk, if the road section is a high-speed tunnel (once an accident occurs, the severity is extremely high but the frequency is low), the severity can be focused on, and set , , if the road section is a school gate branch road (frequent small accidents such as scratches but light consequences), the two can be balanced, and set , ; , reflect the influence of congestion frequency and congestion duration, if the road section is a main road during morning and evening peak hours (congestion is frequent but single duration is short), the congestion frequency can be focused on, and set , ; if the road section is a dead-end road (congestion occurs less frequently but once congestion occurs, it lasts for a long time), the duration can be focused on, and set , ; , respectively represent the weights of accident risk and congestion risk, usually , if the road section is an urban expressway (the consequences of accidents are extremely serious), set , , if the road section is a suburban sightseeing road (accidents are rare and congestion has a more prominent impact on the travel experience), set , (alpha3 is still higher).

[0038] Accident severity score: quantitative evaluation of the severity of historical accidents on the road section, which comprehensively considers consequences such as casualties, property losses, and social impacts.

[0039] Accident frequency score: quantitative evaluation of the frequency of historical accidents on the road section, which is usually calculated based on the number of accidents in a statistical period (such as month or year).

[0040] Congestion frequency score: Quantitative evaluation of the frequency of historical congestion on a road segment, calculated based on the number of congestion occurrences within a statistical period (e.g., month, week).

[0041] Congestion duration score: Quantitative evaluation of the duration of historical congestion on a road segment, calculated based on the average duration, maximum duration, or total congestion duration within a statistical period.

[0042] Input the traffic operation data and road weather data of each road segment into the BP neural network model to obtain the predicted risk data of each road segment, retrieve the road segment risk value in the predicted risk data, define the road segment with a road segment risk value greater than or equal to the first risk threshold as a high-risk road segment, and define the road segment with a road segment risk value less than or equal to the second risk threshold as a low-risk road segment; define the road segment with a road segment risk value between the second risk threshold and the first risk threshold as a medium-risk road segment.

[0043] High-risk road segment: Road segment risk value ≥ first risk threshold, immediate coordinated control (e.g., adjusting signal lights, diverting vehicles) is required.

[0044] Low-risk road segment: Road segment risk value ≤ second risk threshold, optimization of traffic efficiency (e.g., opening more lanes) is possible.

[0045] Medium-risk road segment: Risk value between 0.3 and 0.8, continuous monitoring and disposal planning are required.

[0046] First risk threshold (high-risk determination): Used to identify road segments that require immediate control, set in combination with historical accident / congestion severity cases and road network carrying capacity limits. For example, statistically analyze the risk value distribution of historical high-risk road segments, and take the minimum value of 90% high-risk cases as the threshold.

[0047] Second risk threshold (low-risk determination): Used to identify road segments that can optimize efficiency, set in combination with historical low-risk road segment stable states and optimal traffic efficiency intervals. For example, statistically analyze the risk value distribution of historical low-risk road segments, and take the maximum value of 90% low-risk cases as the threshold.

[0048] Example (urban expressway scenario) First risk threshold (e.g., 0.8): If a certain urban expressway, historically requires coordinated control of road segments, the risk value is as low as 0.78. To cover extreme cases, set the first risk threshold = 0.8 (risk value ≥ 0.8 is determined as high-risk). Second risk threshold (e.g., 0.3): Historically, road segments that can open more lanes have a risk value as high as 0.32. To accurately identify low-risk, set the second risk threshold = 0.3 (risk value ≤ 0.3 is determined as low-risk).

[0049] Wherein, the road segment risk value is the output result of the BP neural network model, which is used to quantify the comprehensive traffic risk level of each road segment in the future time period. The essence is the predicted value of the road segment risk by the model, which is generated based on the input risk influence factors (such as traffic flow fluctuation rate, visibility attenuation rate, etc.) through nonlinear mapping. The production process is as follows:

[0050] Input layer: The 20-dimensional risk influence factors (12 types of artificial factors + 8 types of automatic factors) extracted in step two are input into the input layer of the BP neural network as the basis for model prediction.

[0051] Hidden layer: The risk influence factor vector is transmitted to the hidden layer through nonlinear transformation of the input layer, and the first weight value W1 and the first bias value b1 are randomly initialized to perform linear transformation, generating the first prediction value (H = W1X + b1) , where X is the input vector). The first prediction value introduces nonlinearity through the first activation function (such as Sigmoid, ReLU), generating the hidden layer output (H = f(H)) ).

[0052] Output layer: The hidden layer output H is transmitted to the output layer, and the final weight value W2 and the final bias value b2 are linearly transformed to generate the second prediction value (Y = W2H + b2) ). The second prediction value generates the road segment risk value (Y) through the second activation function (such as Softmax, linear activation), which is consistent with the true risk value (normalized to [0, 1]).

[0053] Second prediction value: The result of linear transformation of the hidden layer output H through the final weight value and the final bias value, which generates the final risk value after inputting the second activation function.

[0054] Step three, risk disposal strategy generation and optimization.

[0055] Referring to FIG. 7, the step three process is as follows: Figure 3 According to the predicted risk data, the first risk disposal strategy is obtained by using the particle swarm algorithm: ​​According to the high-risk section, risk-related information is obtained, the risk-related information including road types, section location data, traffic flow data and signal control data of other sections except the high-risk section, risk position data of the high-risk section is obtained, section disposal distance is obtained according to the section location data and the risk position data, risk disposal weight values of each section are determined according to the section disposal distance and the road types, the section with the maximum risk disposal weight value is obtained, and it is defined as the cooperative control section of the high-risk section. The risk disposal weight value is a weighted score based on the section disposal distance and the road types, and reflects the suitability of the current section as the cooperative control section of the high-risk section. The core logic is that the closer the section distance to the high-risk section, and the more the road types need to be disposed preferentially (such as high-risk sections prefer to normal sections), the greater the risk disposal weight value, and the higher the priority. According to the low-risk section, associated optimization information is obtained, the associated optimization information including other section types, section location data, traffic flow data and signal control data except the low-risk section, optimization position data of the low-risk section is obtained, section optimization distance is obtained according to the section location data and the optimization position data, passing and dredging values of each section are determined according to the section optimization distance and the road types, the section with the maximum passing and dredging value is obtained, and it is defined as the shunt guiding section of the low-risk section, and the above steps are repeated to obtain the first risk disposal strategy. According to the section optimization distance and the road types, the passing and dredging values of each section are determined, including: a distance influence coefficient is obtained according to the section optimization distance, a road grade coefficient is obtained according to the road types, the distance influence coefficient and the road grade coefficient are weighted and calculated to obtain the passing and dredging values. The road grade coefficient is a continuous function based on the real-time risk value, and reflects the priority of the section for dredging, and is specifically: ; represents the current risk value of the section (predicted by the BP model); represents the high-risk threshold (such as 0.8); represents the low-risk threshold (such as 0.3).

[0056] The coefficient of the high-risk section ( ) decreases with the decrease of the risk value (such as from 1 to 0), reflecting the dynamic adjustment of the priority after disposal;

[0057] The coefficient of the low-risk section ( ) increases with the increase of the risk value (such as from -1 to 0), avoiding excessive shunting; The coefficient of the medium-risk section is 0, keeping neutral.

[0058] wherein the distance influence coefficient is a normalized coefficient calculated based on the link optimization distance, reflecting the influence degree of the spatial distance between the target link (such as a high-risk link or a low-risk link) and the current link on the traffic diversion or risk disposal effect, and the core logic is that the closer the link distance, the greater the mutual influence; the farther the distance, the smaller the influence. The generation process of the distance influence coefficient is as follows: (1) determine the link optimization distance, which is the spatial distance between the current link and the target link (such as a high-risk link, a low-risk link or a normal link), denoted as D1. (2) Normalization processing, mapping the actual link optimization distance D1 to the interval [0, 1] to generate the distance influence coefficient . The normalization formula is as follows: ; , wherein Dmax represents the preset maximum consideration distance, and the link beyond the distance is considered to have no significant influence; when D1 = 0 (i.e. the current link coincides with the target link), , indicating the maximum influence; when , , indicating no influence.

[0059] According to the link disposal distance and the road type, the risk disposal weight value of each link is determined, including: obtaining a distance attenuation coefficient according to the link disposal distance, and obtaining a road risk coefficient according to the road type, the road risk coefficient is a compensation function based on the risk diffusion cost, reflecting the priority of the link for disposal, and specifically: ; , , wherein the risk compensation coefficient represents the priority of the high-risk link for disposal and the low-risk link for reverse adjustment; represents the current risk value of the link; represents the high-risk threshold value; represents the low-risk threshold value; represents the medium-risk threshold value.

[0060] The coefficient of the high-risk link ( ) is negative, and the higher the risk value, the greater the absolute value (such as , ), reflecting the need for priority disposal; The coefficient of the low-risk link ( ) is positive, and the lower the risk value, the greater the value (such as , ), reflecting the need for appropriate diversion; The coefficient of the medium-risk link is 0, which remains neutral.

[0061] High-risk threshold: the upper limit criterion for the road segment risk value, when the road segment risk value ≥ high-risk threshold, the road segment is defined as a high-risk road segment. Function: identify road segments that need to take emergency control measures immediately, such as real-time diversion, signal light strong control, temporary traffic restrictions, etc. These road segments have a high probability of accidents or severe congestion and need to be prioritized.

[0062] Low-risk threshold: the lower limit criterion for the road segment risk value, when the road segment risk value ≤ low-risk threshold, the road segment is defined as a low-risk road segment. Function: identify road segments with stable traffic conditions and optimized efficiency, such as road segments with a very low probability of risk events, which can improve traffic capacity by opening more lanes or extending green light duration.

[0063] Medium-risk threshold: a transitional criterion between the high-risk threshold and the low-risk threshold, when the road segment risk value is low-risk threshold < risk value < high-risk threshold, the road segment is defined as a medium-risk road segment. Function: identify road segments that need to be continuously monitored and prevented from escalating risks, such as road segments with controllable current risks but potential fluctuations (such as peak conversion periods or weather changes that may escalate to high-risk), which need to be dynamically tracked and prepared for.

[0064] The distance decay coefficient and the road risk coefficient are weighted to obtain the risk disposal weight value; the road types include high-risk road segments, low-risk road segments, and normal road segments; a traffic flow simulation model is established according to the risk disposal strategy and the simulation results are obtained, the first risk disposal strategy is adjusted according to the simulation results, and the second risk disposal strategy is obtained: wherein the distance decay coefficient is a normalized coefficient calculated based on the road segment disposal distance, reflecting the influence degree of the spatial distance between the target road segment (such as high-risk road segment, normal road segment, or low-risk road segment) and the current road segment on the risk disposal effect, the core logic is that the closer the road segment distance, the higher the priority of risk disposal, and the farther the distance, the lower the priority.

[0065] The generation process of the distance decay coefficient is as follows: (1) determine the road segment disposal distance, which is the spatial distance between the current road segment and the target road segment (such as high-risk road segment, normal road segment, or low-risk road segment), denoted as D2, (2) normalize the actual road segment disposal distance D2 to map it to the [0, 1] interval to generate the distance decay coefficient . The normalization formula is as follows: ; , which represents the preset maximum consideration distance, and road segments beyond this distance are considered to have no significant impact. When , (i.e. the current road segment coincides with the target road segment), , indicating the highest disposal priority; when , , indicating no disposal priority.

[0066] The traffic guidance value is a weighted score based on the link optimization distance and road type, reflecting the suitability of the current link as a low-risk link diversion guide link. The core logic is that the closer the link distance to the low-risk link, and the more the road type needs to be preferentially guided (such as high-risk links prefer to normal links), the greater the traffic guidance value, and the higher the priority.

[0067] The generation process of the traffic guidance value includes: calculating the traffic guidance value by weighting the distance influence coefficient and the road grade coefficient The traffic guidance value S is generated by weighted summation: is the weight coefficient;

[0068] are used to balance the distance influence (near links are easy to divert) and the road grade (high-grade roads have strong diversion capacity). If the diversion target is a low-risk link around a school (students need to be quickly diverted), the near link is preferred, and is set to If the diversion target is a holiday scenic road (main roads need to be used to divert large flow), the road grade is emphasized, and is set to .

[0069] The generation process of the risk disposal weight value includes: weighting the distance decay coefficient and the road risk coefficient to generate the risk disposal weight value W: is the weight coefficient, used to balance the distance decay (near links respond quickly) and the road risk (high-risk links need to be preferentially controlled), if the high-risk link is a tunnel (space is closed, needs to be disposed near), the distance decay is emphasized, and is set to If the high-risk link is a cross-river bridge (structure is special, risk coefficient is high), the road risk is emphasized, and is set to .

[0070] ​​​​​​​​​According to the traffic operation data, a traffic flow simulation model is established, first traffic flow increment data is determined according to the predicted risk data, and the traffic flow simulation model is updated according to the first traffic flow increment data; second traffic flow adjustment data is obtained through the first risk disposal strategy, and the second traffic flow adjustment data is input into the traffic flow simulation model to obtain traffic simulation data, the traffic simulation data including simulation flow, simulation speed and simulation density; the simulation flow speed of the traffic simulation data is called, the flow speed equal to or greater than the first risk threshold is defined as the first congestion risk, the flow speed equal to or less than the second risk threshold of the traffic flow simulation model is defined as the second smooth passing remainder, the first risk disposal strategy corresponding to the first congestion risk or the second smooth passing remainder is defined as the abnormal disposal strategy, and the abnormal disposal strategy is adjusted according to the first congestion risk and the second smooth passing remainder to obtain the second risk disposal strategy.

[0071] The first traffic flow increment data: the change amount of the traffic flow of each road section in the future time period determined based on the predicted risk data (risk value of each road section), reflecting the expected influence of risk prediction on traffic flow distribution.

[0072] The first traffic flow increment data generation process includes: according to the risk value (high, medium and low) of each road section predicted in step two, combining the corresponding relationship between risk and flow in historical traffic data (such as high-risk road sections usually accompanied by sudden flow reduction or congestion), a mapping model of risk value and flow change is established. For high-risk road sections (risk value ≥ first risk threshold), the flow reduction amount caused by risk disposal (such as vehicle detouring and signal control) is predicted, which is denoted as (negative value). For coordinated control road sections (road sections associated with high-risk road sections), the flow increase amount caused by accepting the shunting vehicles of high-risk road sections is predicted, which is denoted as (positive value). For low-risk road sections (risk value ≤ second risk threshold), the flow increase amount caused by optimizing traffic (such as opening more lanes) is predicted, which is denoted as (positive value). For shunting guide road sections (road sections associated with low-risk road sections), the flow reduction amount caused by guiding vehicles to leave is predicted, which is denoted as (negative value). The , , , are summarized by road section to form the first traffic flow increment data of each road section, which is used to update the initial flow condition of the traffic flow simulation model.

[0073] The second traffic flow adjustment data: the actual adjustment amount of the traffic flow of each road section after the implementation of the first risk disposal strategy (selection of coordinated control road sections and shunting guide road sections), reflecting the specific influence of the strategy on traffic flow distribution.

[0074] The second traffic flow adjustment data generation process includes the following steps: inputting the first risk treatment strategy (such as strong control of signal lights on high-risk road sections and activation of tidal lanes on low-risk road sections) into a traffic flow simulation model to simulate vehicle path selection behavior (based on shortest path, user equilibrium, etc. algorithms). For high-risk road sections, due to signal control or temporary speed limit, the actual flow reduction is . For cooperatively controlled road sections, due to the diversion of high-risk road sections, the actual flow increase is . For low-risk road sections, due to the opening of more lanes or the extension of green light duration, the actual flow increase is . For diversion guide road sections, due to the diversion of vehicles, the actual flow reduction is . The , , , are aggregated by road section to form the second traffic flow adjustment data , which is used to verify the effectiveness of the strategy (such as whether it causes congestion or smooth passage).

[0075] Step four, traffic control scheme development and implementation.

[0076] Obtain the traffic demand information of the road section, including the travel area number, traffic demand quantity and traffic time period; According to the regional path induction (sorted by destination distance), the control object is selected, and the specific steps are as follows: Regional division: according to the road network topology, the city is divided into several traffic cells (such as 500m×500m grid); Path recommendation: for the areas around high-risk road sections, V2X communication or APP push is used to induce vehicles to bypass low-risk road sections; Priority sorting: according to the distance between the vehicle destination and the high-risk road section (not the absolute traffic distance), the transit vehicles are preferentially induced to bypass.

[0077] According to the second risk treatment strategy, the road section control information is obtained, including the first road section data, the second road section data and the signal control data; the signal control data of the road section control information is called to calculate the single-lane traffic efficiency, double-lane traffic efficiency and multi-lane traffic efficiency of the road section;

[0078] The traffic efficiency calculation formula is as follows: ; wherein, represents the actual traffic capacity of the road section (vehicles / hour); represents the number of lanes; represents the basic traffic capacity of a single lane (such as 1800 vehicles / hour / lane, which is preset according to road grade); The adjustment coefficient (reflecting the influence of factors such as vehicle ratio, slope, turning ratio, etc., which can be calibrated through historical data) is represented.

[0079] Dynamic reversible lanes and adaptive signal control are introduced, and the strategy is as follows: The current traffic is obtained in real time through the road monitoring device And the predicted traffic ; If (close to congestion), enable reversible lanes (such as temporarily changing the reverse lane to the same direction); if (low traffic), reduce the number of open lanes, and keep the basic lane (such as 1 lane); the intermediate state keeps the current number of lanes.

[0080] The signal light cycle is adjusted in real time according to the traffic (such as using inductive green wave control).

[0081] The beneficial effects of the embodiment are: by collecting real-time traffic and meteorological multi-source data, combining BP neural network hybrid feature extraction (manual definition + automatic learning), accurately predicting road risk and classifying control; using particle swarm algorithm to generate dynamic disposal strategy, optimizing the strategy adaptability through simulation; cooperating with regional path induction, dynamic reversible lane and adaptive signal control, reducing accident risk and relieving congestion, realizing efficient allocation of traffic resources and safety level improvement.

[0082] The formulas of the present application are all dimensionless, and the preset parameters in the formulas are set by the skilled person in the art according to the actual situation.

[0083] The weight coefficient of the present application is used to measure the influence degree of different factors or variables on a certain result or decision. The definition of weight coefficient is to assign a numerical value to each factor when comparing and evaluating multiple factors, to reflect its importance or priority. These weight coefficients can be determined according to specific circumstances and needs, and are usually formulated and confirmed by professionals or relevant parties. By reasonably setting the weight coefficient, the program or system can make more accurate decisions or predictions.

[0084] The above is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited to this, any skilled person in the art can make equivalent replacement or change according to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

[0085] The preferred embodiments of the application disclosed above are only to facilitate the elucidation of the application. The preferred embodiments do not describe all the details of the application and limit the application to the specific embodiments. Obviously, many modifications and variations can be made in light of the teachings above. The description is chosen and described in order to provide the best illustration of the application and its practical application to those skilled in the art and to enable those skilled in the art to utilize the application in its best mode. The application is only limited by the claims and their full scope and equivalents.

Claims

1. A traffic risk hidden danger monitoring method based on artificial intelligence, characterized in that, The method flow is as follows: Step one, data collection: real-time acquisition of traffic operation data by monitoring equipment; at the same time, real-time acquisition of road weather data in a future time period from a weather platform; Step two, risk prediction: acquisition of historical traffic data and historical weather data; input of the historical traffic data and the historical weather data into a BP neural network for training to establish a BP neural network model, input of traffic operation data and road weather data of each road section into the BP neural network model, acquisition of predicted risk data of each road section, and division of road sections into high-risk road sections, medium-risk road sections and low-risk road sections based on road section risk values in the predicted risk data; Step three, risk disposal strategy generation and optimization: acquisition of a first risk disposal strategy by using a particle swarm algorithm according to the predicted risk data, and adjustment of the first risk disposal strategy according to simulation results of a traffic flow simulation model to obtain a second risk disposal strategy; Step four, traffic control scheme formulation and execution: acquisition of traffic demand information, path guidance based on the second risk disposal strategy, and dynamic adjustment of lane configuration and signal control parameters. 2.The AI-based traffic risk hazard monitoring method of claim 1, wherein, The establishment process of the BP neural network model in the step two includes: input of the historical traffic data and the historical weather data into an input layer of the BP neural network, acquisition of risk influence factors according to the historical traffic data and the historical weather data, random acquisition of first weight values and first bias values of the risk influence factors, and acquisition of first predicted values according to the first weight values and the first bias values by using a first activation function; taking the first predicted values as inputs of a hidden layer, acquisition of a loss value according to the first predicted values and true risk values by using a loss function, adjustment of the first weight values and the first bias values according to the loss value, acquisition of second weight values and second bias values, repetition of the above steps n times until the loss value is less than or equal to a loss expected threshold, and then acquisition of final weight values and final bias values; and acquisition of predicted risk data as outputs of an output layer by using a second activation function according to the final weight values and the final bias values. 3.The AI-based traffic risk hazard monitoring method of claim 2, wherein, The acquisition process of the risk influence factors adopts a hybrid mode combining manual definition and automatic learning; wherein, 12 types of core risk factors including traffic flow fluctuation rate, speed variation coefficient, lane occupancy rate peak value, illegal density, accident frequency, visibility attenuation rate, road surface temperature extreme value, precipitation intensity grade, wind force grade, road surface humidity threshold, road curvature, and intersection density are extracted from original data by using manual feature engineering; 8 types of high-order feature factors including traffic flow-speed joint distribution characteristics reflecting high-order modes of traffic flow states and precipitation intensity + speed standard deviation combination characteristics reflecting meteorological-traffic interaction characteristics are generated by using a stacked autoencoder to perform nonlinear transformation on the original data; and the manual features and the automatic features are combined into a 20-dimensional risk influence factor vector. 4.The AI-based traffic risk hazard monitoring method of claim 2, wherein, The true risk value is generated by weighted fusion of an accident risk value and a congestion risk value; wherein, a weight coefficient of the accident risk value is greater than a weight coefficient of the congestion risk value; and the true risk value ranges from 0 to 1 after normalization processing. 5.The AI-based traffic risk hazard monitoring method of claim 1, wherein, In the second step, a road segment with a risk value greater than or equal to the first risk threshold is defined as a high-risk road segment, a road segment with a risk value less than or equal to the second risk threshold is defined as a low-risk road segment, and a road segment with a risk value between the second risk threshold and the first risk threshold is defined as a medium-risk road segment. 6.The AI-based traffic risk hazard monitoring method of claim 1, wherein, In the third step, risk-related information of the high-risk road segment is obtained, the risk-related information including road types, road location data, traffic flow data, and signal control data of other road segments except the high-risk road segment, risk location data of the high-risk road segment is obtained, a road segment disposal distance is obtained according to the road location data and the risk location data, a risk disposal weight value of each road segment is determined according to the road segment disposal distance and the road type, a road segment with the maximum risk disposal weight value is obtained, and the road segment is defined as a cooperative control road segment of the high-risk road segment. According to the low-risk road segment, associated optimization information is obtained, the associated optimization information including road types, road location data, traffic flow data, and signal control data of other road segments except the low-risk road segment, optimization location data of the low-risk road segment is obtained, a road segment optimization distance is obtained according to the road location data and the optimization location data, a traffic flow guiding value of each road segment is determined according to the road segment optimization distance and the road type, a road segment with the maximum traffic flow guiding value is obtained, and the road segment is defined as a shunt guiding road segment of the low-risk road segment, and the above steps are repeated to obtain a first risk disposal strategy. 7.The AI-based traffic risk hazard monitoring method of claim 6, wherein, The determination of the traffic flow guiding value of each road segment according to the road segment optimization distance and the road type includes: obtaining a distance influence coefficient according to the road segment optimization distance, the distance influence coefficient being a normalization coefficient calculated based on the road segment optimization distance; obtaining a road level coefficient according to the road type, the road level coefficient being a continuous function based on a real-time risk value, wherein the coefficient of the high-risk road segment decreases as the risk value decreases, the coefficient of the low-risk road segment increases as the risk value increases, and the coefficient of the medium-risk road segment is 0; and performing weighted calculation on the distance influence coefficient and the road level coefficient to obtain the traffic flow guiding value. 8.The AI-based traffic risk hazard monitoring method of claim 6, wherein, The determination of the risk disposal weight value of each road segment according to the road segment disposal distance and the road type includes: obtaining a distance decay coefficient according to the road segment disposal distance, the distance decay coefficient being a normalization coefficient calculated based on the road segment disposal distance and reflecting an influence degree of a spatial distance between a target road segment and a current road segment on a risk disposal effect, wherein the closer the road segment distance, the higher the priority of the risk disposal, and the farther the distance, the lower the priority; obtaining a road risk coefficient according to the road type, the road risk coefficient being a compensation function based on a risk diffusion cost and reflecting a priority of a road segment for disposal, wherein the coefficient of the high-risk road segment is a negative value and the higher the risk value, the larger the absolute value, the coefficient of the low-risk road segment is a positive value and the lower the risk value, the larger the value, and the coefficient of the medium-risk road segment is 0; and performing weighted calculation on the distance decay coefficient and the road risk coefficient to obtain the risk disposal weight value. 9.The AI-based traffic risk hazard monitoring method of claim 1, wherein, The step four further comprises regional path induction: including regional division, path recommendation and priority sorting, wherein the regional division is to divide the city into several traffic zones according to the road network topology, the path recommendation is to induce vehicles to bypass to low-risk sections through V2X communication or APP push around the high-risk sections, and the priority sorting is to sort the distance between the vehicle destination and the high-risk section to preferentially induce transit vehicles to bypass; The step four further comprises traffic efficiency, which is calculated by a formula, and the formula considers the actual traffic capacity of the section, the number of lanes, the basic traffic capacity of a single lane and the adjustment coefficient; The step four further comprises the strategy of dynamic tidal lane and adaptive signal control: the current flow and the predicted flow are obtained by real-time monitoring through the section monitoring device, if the current flow is close to congestion, that is, greater than or equal to the proportional relationship of the actual traffic capacity of the section, the tidal lane is enabled, if the current flow is low flow, that is, less than or equal to the actual traffic capacity of the section, the number of open lanes is reduced, the basic lane is reserved, the current number of lanes is maintained in the intermediate state, and the signal light cycle is adjusted in real time according to the flow.

Images