Road traffic signal lamp management system based on big data

Abstract

The invention relates to the technical field of intelligent traffic management, and discloses a road traffic signal lamp management system based on big data, which comprises a data acquisition and fusion module for acquiring traffic sensor data, visual data, GPS track data and external environment data, generating unified traffic state data through space-time alignment, data filtering and multi-modal data fusion; and the flow prediction module is used for predicting the traffic flow of a single intersection and a region by using a time sequence model and a graph neural network based on the traffic state data generated by the data acquisition and fusion module. Through multi-modal data acquisition and space-time alignment, Kalman filtering, Bayesian inference and other technologies, deep fusion of different data sources is realized, the comprehensiveness and accuracy of data are significantly improved, dynamic changes of a complex traffic scene can be truly reflected, high-quality basic data are provided for subsequent flow prediction and optimization control, and a core technical support is formed.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent traffic management, and in particular to a road traffic signal light management system based on big data. Background Art

[0002] In recent years, the urbanization process has been accelerating, the complexity of urban traffic networks has increased significantly, and traffic congestion has become more serious. Traditional traffic light management systems mostly use static control methods based on fixed cycles or simple flow detection, which cannot adapt to dynamically changing traffic flows in real time, and the management method has the following problems and defects:

[0003] The data source of traditional traffic light management systems mainly relies on a single sensor. The data collected is limited in dimension and cannot fully reflect the dynamic characteristics of traffic conditions. In complex traffic scenarios, such as traffic fluctuations during peak hours and abnormal events, the system cannot provide reliable data support.

[0004] Existing traffic light optimization methods mostly perform local optimization on a single intersection, using preset rules or simple algorithms based on traffic statistics. They are unable to coordinate and control intersections within a region, resulting in a reduction in the balance of the overall traffic flow and the formation of new congestion on some sections of the road, further reducing regional traffic efficiency.

[0005] When faced with abnormal events such as traffic accidents, emergency rescue vehicles’ priority passage needs, or sudden congestion, traditional systems mainly rely on manual intervention for adjustment and are unable to respond quickly to emergencies, resulting in longer response times and easily causing further chaos in traffic order.

[0006] With the continuous development of urban traffic, the spatial and temporal distribution of traffic flow has shown obvious dynamic changes. The traditional traffic light management system lacks real-time feedback and dynamic learning mechanisms. When the traffic pattern changes, the optimization effect decreases rapidly or even fails.

[0007] When introducing artificial intelligence technology to optimize traffic lights, its application scope is usually limited to local problems. It lacks a complete technical chain from data perception, traffic prediction, optimization control to abnormal response. In addition, most systems cannot fully utilize emerging technologies such as distributed computing and edge computing, and have obvious deficiencies in real-time and scalability.

[0008] Therefore, those skilled in the art provide a road traffic signal light management system based on big data to solve the above-mentioned problems. Summary of the invention

[0009] In view of the deficiencies in the prior art, the present invention provides a road traffic signal light management system based on big data to solve the problems raised in the above background technology.

[0010] To achieve the above objectives, the present invention is implemented through the following technical solutions: A road traffic signal light management system based on big data, comprising:

[0011] Data collection and fusion module, which collects traffic sensor data, visual data, GPS trajectory data and external environment data, and generates unified traffic status data through spatiotemporal alignment, data filtering and multimodal data fusion;

[0012] The traffic prediction module predicts the traffic flow at single intersections and regions using time series models and graph neural networks based on the traffic status data generated by the data collection and fusion module;

[0013] The single intersection dynamic optimization module generates a signal light cycle and phase adjustment plan using a reinforcement learning algorithm based on the single intersection traffic data predicted by the traffic prediction module;

[0014] The multi-intersection linkage optimization module uses the distributed optimization algorithm to coordinate the traffic lights at multiple intersections in the region based on the regional traffic flow prediction data generated by the flow prediction module and the traffic light scheme generated by the single-intersection dynamic optimization module.

[0015] The anomaly detection and response module receives the visual data and traffic data transmitted by the data acquisition and fusion module, identifies traffic accidents, congestion and vehicle priority requirements through anomaly detection algorithms, and generates traffic light adjustment plans under abnormal conditions;

[0016] The feedback and iterative optimization module updates the optimization algorithm and prediction model based on the traffic light adjustment plans generated by the single-intersection dynamic optimization module and the multi-intersection linkage optimization module, as well as the adjustment plans generated by the anomaly detection and response module, combined with the real-time traffic status data from the data acquisition and fusion module.

[0017] Preferably, the data acquisition and fusion module includes:

[0018] The data acquisition unit is used to collect traffic flow, speed and vehicle density data. Specifically, the traffic flow q(t), speed v(t) and vehicle density ρ(t) are obtained through the geomagnetic coil sensor, which satisfies the following relationship:

[0019] q(t)=ρ(t)·v(t),

[0020] Among them, q(t) represents the number of vehicles passing the sensor per unit time, ρ(t) represents the vehicle density per unit length of road, and v(t) represents the average speed of vehicles on the road section;

[0021] The data processing unit is used to filter the traffic flow, speed and density data obtained by the data acquisition unit, and estimate and denoise the observed data through the Kalman filter formula. The specific formula is as follows:

[0022] in, is the state estimate after filtering, z k is the observed value at the current moment, K k is the Kalman gain, is the estimated value of k-1 at the previous moment;

[0023] The Kalman gain is calculated as:

[0024] Among them, P k-1 is the state prediction error covariance of the previous step, H is the observation matrix, R is the observation noise covariance, H T is the transpose of the observation model matrix;

[0025] The data fusion unit is used to align the processed multimodal data in time and space and achieve fusion through Bayesian inference, satisfying the following formula:

[0026]

[0027] Among them, P(S|D1, D2, ..., D n ) is the posterior probability of the current traffic state S after fusion, P(S is the prior probability of the traffic state, P(D i |S) is the observed data D i Conditional probability under traffic state S.

[0028] Preferably, the traffic prediction module includes:

[0029] The single intersection prediction unit is used to generate traffic status data of a single intersection based on the data acquisition and fusion module, and predict the future traffic flow of the single intersection using a long short-term memory network model. The long short-term memory network model includes the following state update formula:

[0030] Forget gate: f t =σ(W f ·[h t-1 , x t ]+b f ),

[0031] Among them, f t is the forget gate activation value, W f is the forget gate weight matrix, h t-1 is the hidden state of the previous time step, x t is the input feature of the current time step, b fis the forget gate bias term, σ is the activation function;

[0032] Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i ),

[0033]

[0034] Among them, i t is the input gate activation value, is the candidate memory state, W i and W i is the weight matrix of the input gate and the candidate memory state, b i and C are the corresponding bias terms, tanh is the hyperbolic tangent function;

[0035] Current memory status update:

[0036] Among them, C t is the memory state of the current time step, C t-1 is the memory state of the previous time step;

[0037] Output gate: o t =σ(W o ·[h t-1 , x t ]+b o ), h t =o t tanh(C t ),

[0038] Among them, t is the output gate activation value, h t is the hidden state of the current time step, W o is the output gate weight matrix, b o is the output gate bias term;

[0039] The regional traffic prediction unit is used to combine the regional traffic status data generated by the data acquisition and fusion module, and use the graph neural network to predict the traffic flow of multiple intersections in the region. The graph neural network is based on the following node feature update formula:

[0040]

[0041] Among them, H (l) is the node feature matrix of the lth layer, is the normalized adjacency matrix, W (l) is the weight matrix of the lth layer, and σ is the activation function.

[0042] Preferably, the single intersection dynamic optimization module includes a state perception unit, a reinforcement learning decision unit and a signal light execution unit:

[0043] The state sensing unit receives the single intersection traffic prediction data of the traffic prediction module and the real-time traffic status data of the data acquisition and fusion module to determine the traffic status of the current intersection, including:

[0044] Traffic flow q(t0, the number of vehicles passing through the intersection per unit time;

[0045] The average waiting time w(t) represents the average stay time of the vehicle queue, and the calculation formula is:

[0046]

[0047] Where L(t) is the length of the vehicle queue, ρ(t) is the vehicle density, v(t0 is the average vehicle speed;

[0048] The output traffic state S of the state perception unit t It is expressed as:

[0049] S t ={q(t),w(t),φ(t)},

[0050] Among them, φ(t) is the phase state of the current traffic light.

[0051] Preferably, the reinforcement learning decision unit is based on the traffic state S provided by the state perception unit. t , using deep reinforcement learning method to generate traffic light optimization decision A t , its core includes the following parts:

[0052] Action A t : A t ={ΔT, Δφ},

[0053] Among them, △T is the adjustment amount of the signal light cycle, and △φ is the adjustment amount of the signal light phase switching;

[0054] Reward R t : The reward function is designed to reduce the average waiting time and improve traffic efficiency. The specific formula is:

[0055] R t = -α·w(t)+β·q(t),

[0056] Among them, w(t) is the average waiting time, q(t) is the traffic flow, α and β represent the weight coefficients of waiting time and traffic flow;

[0057] The Q-value update formula based on reinforcement learning is:

[0058]

[0059] Among them, Q(S t ,A t ) indicates that in state S t Next, execute action A t The reward value, η is the learning rate, R t is the immediate reward value, γ is the discount factor, In the next state S t+1 The maximum expected reward under

[0060] The signal light execution unit generates an optimization decision A based on the reinforcement learning decision unit. t , adjust the cycle and phase of the signal light, including:

[0061] Modify the signal light cycle T: T t+1 =T t +ΔT,

[0062] Among them, T t The current signal light cycle, the cycle adjustment amount generated by the △T decision;

[0063] Modify the signal light phase φ: φ t+1 =φ t +Δφ,

[0064] Among them, φ t is the current signal light phase time, the phase adjustment amount generated by the △φ decision.

[0065] Preferably, the multi-intersection linkage optimization module includes:

[0066] The regional collaborative modeling unit receives the traffic flow prediction data of multiple intersections generated by the regional traffic prediction unit in the traffic prediction module, and constructs a traffic graph model of the regional road network with multiple intersections, which is defined as:

[0067] G=(V,E),

[0068] Among them, V is the intersection set, E is the road set;

[0069] The linkage optimization unit uses a distributed optimization algorithm to perform linkage optimization on the traffic lights at multiple intersections based on the traffic map G = (V, E) generated by the regional collaborative modeling unit, including:

[0070] The optimization goal is defined as minimizing the total average waiting time and total travel time at intersections in the area. The objective function is:

[0071] Among them, w i (t) is the intersection v i The average waiting time, Tij (t) is the road e ij The average travel time, λ is the trade-off coefficient;

[0072] The linkage optimization unit adjusts the signal light cycle T i Dynamic adjustment to achieve coordinated control of multiple intersections:

[0073] T i (t+1)=T i (t)+ΔT i ,

[0074] Among them, T i (t) is the intersection v i At time t, the signal light cycle, △T i is the cycle-adjusted amount;

[0075] Balanced traffic distribution is achieved by adjusting the phase of traffic lights between adjacent intersections. The adjustment formula is:

[0076] φ i (t+1)=φ i (t)+Δφ i ,

[0077] Among them, φ i (t) is the intersection v i The phase of the signal light at time t, △φ i Phase adjustment amount.

[0078] Preferably, the multi-intersection linkage optimization module includes:

[0079] The distributed computing unit realizes multi-intersection linkage optimization through distributed reinforcement learning or genetic algorithm, specifically including genetic algorithm and distributed reinforcement learning algorithm;

[0080] The genetic algorithm comprises:

[0081] Chromosome encoding: The signal light cycle T of each intersection i and phase φ i Encoded as chromosome C, it is expressed as:

[0082] C={T1,φ1,T2,φ2,…,T N ,φ N},

[0083] Where N is the number of intersections in the area;

[0084] Fitness function: The fitness function is defined as the inverse function of the optimization objective and is expressed as:

[0085]

[0086] Among them, F(C) represents the objective function value of the state set C, V is the node set, and w i (t) is the weight or loss of node i at time t, E is the edge set, T ij (t) represents the impact or cost at time t, and λ is a weight parameter used to balance the node loss and edge loss;

[0087] The distributed reinforcement learning algorithm includes:

[0088] State space: S t ={H i , W ij ∣v i ∈V,e ij ∈E},

[0089] Among them, S t represents the state set of the system at time t, H i represents the state variable of node i, W ij Represents edge e ij The weight or status of

[0090] Action Space: A t ={ΔT i , Δφ i ∣v i ∈V},

[0091] Among them, A t represents the updated set of nodes at time t, △T i Variable T representing node i at time t i The change in △φ i The variable φ represents the node i at time t i The amount of change;

[0092] Reward function: The reward function is defined as the weighted sum of the total waiting time and the travel time in the area:

[0093]

[0094] Among them, R t represents a certain benefit or cost function at time t, w i (t) represents the weight or cost of node i, T ij (t) represents the weight or cost at time t, λ is the balance coefficient;

[0095] Distributed training: Each intersection independently trains a local reinforcement learning model, and the regional control center achieves global optimization by aggregating local reward values;

[0096] The traffic light execution unit receives the traffic light adjustment plan calculated by the linkage optimization unit, and sends the adjustment plan to the traffic light equipment at each intersection in real time through the regional controller.

[0097] Preferably, the anomaly detection and response module includes:

[0098] The anomaly detection unit receives the traffic sensor data and visual data from the data acquisition and fusion module, and detects traffic anomalies by combining time series analysis and computer vision algorithms, including:

[0099] Visual anomaly detection uses the target detection algorithm to analyze the video stream collected by the camera and extract traffic anomaly targets. The detection content includes illegal parking, traffic accidents, etc. The target detection outputs the bounding box B and category label C of the abnormal target:

[0100] B = {(x1, y1, x2, y2)}, C∈{illegal parking, accident, congestion},

[0101] Among them, B represents the spatial location where the abnormal event occurs, and C represents the type of abnormality detected.

[0102] Combined with the time series traffic flow data, the anomaly detection algorithm is used to identify potential anomalies in the traffic flow, and a feature vector is constructed based on the traffic flow q(t), waiting time w(t) and road travel time T(t):

[0103] X t ={q(t),w(t),T(t)},

[0104] The calculation formula of abnormal score is:

[0105] Among them, X t is the feature vector at time t, path length (X t , T i ) represents the feature point X t In the tree i The path length in , S t is the abnormal score at time t;

[0106] The anomaly classification unit classifies the anomaly type and generates the corresponding priority according to the output of the anomaly detection unit. The anomaly type classification rules are as follows:

[0107] Vehicle priority category: including emergency vehicles, fire trucks or police cars’ priority needs;

[0108] Intersection congestion: Based on time series anomaly detection t >threshold, the current road section is determined to be congested;

[0109] Traffic accident category: Category C = accident in the visual inspection result, which is determined to be a traffic accident;

[0110] The priority P of each abnormal event is generated according to the following rules:

[0111] Vehicle priority, traffic accidents, intersection congestion, etc.

[0112] Preferably, the anomaly detection and response module includes:

[0113] The response plan generation unit generates a real-time response plan based on the abnormal type and priority of the abnormal classification unit and the control strategy of the single intersection dynamic optimization module and the multi-intersection linkage optimization module, including:

[0114] If the abnormal type is vehicle priority, the phase of the corresponding intersection signal light is dynamically adjusted to open the green channel. The signal light phase adjustment formula is:

[0115] φ i (t+1)=φ i (t)+Δφ priority ,

[0116] Among them, φ i (t) is the current signal light phase, △φ priority Priority phase adjustment, φ i (t+1 is the phase of the signal light at time t+1;

[0117] If the abnormal type is congestion at the intersection, adjust the signal light cycle to relieve traffic pressure. The signal light cycle adjustment formula is: T i (t+1)=T i (t)+ΔT congestion ,

[0118] Among them, T i (t) is the current signal light cycle, △T congestion Congestion relief cycle adjustment, T i (t+1 is the signal light cycle at time t+1;

[0119] If the abnormal type is a traffic accident, the multi-intersection linkage optimization module is combined to recalculate the regional traffic light cycle and phase to generate a global response plan;

[0120] The feedback adjustment unit monitors the execution effect of the response plan based on the real-time traffic status data of the data acquisition and fusion module, and dynamically modifies the response plan based on the feedback data. The modification logic is as follows:

[0121] If congestion persists, increase the cycle adjustment: ΔT congestion ←ΔT congestion +δ, where △Tcongestion It represents the change of the state variable related to "congestion", and δ is the incremental value;

[0122] When the accident is handled, the default traffic light cycle and phase are restored.

[0123] Preferably, the feedback and iterative optimization module monitors traffic performance in real time, including average vehicle waiting time and traffic rate, and uses historical data to update parameters of the long short-term memory network and the reinforcement learning model.

[0124] The present invention provides a road traffic signal light management system based on big data. It has the following beneficial effects:

[0125] 1. The present invention realizes the deep integration of different data sources through multimodal data collection and spatiotemporal alignment, Kalman filtering, Bayesian inference and other technologies, significantly improves the comprehensiveness and accuracy of data, can truly reflect the dynamic changes of complex traffic scenes, and provide high-quality basic data for subsequent traffic prediction and optimization control, forming the core technical support.

[0126] 2. The present invention adopts regional collaborative modeling and distributed optimization algorithms to realize the linkage control of traffic lights at multiple intersections. The genetic algorithm is combined with reinforcement learning to optimize the global traffic network, solving the problem of overall efficiency decline due to local lag in traditional single-intersection optimization methods, significantly improving the overall traffic efficiency of regional traffic, reducing the risk of systemic congestion, and demonstrating a technological breakthrough in global optimization capabilities.

[0127] 3. The present invention combines computer vision detection with timing analysis technology, which can accurately identify complex abnormal events such as traffic accidents, congestion, and priority traffic needs, and generate targeted traffic light adjustment plans based on the abnormal type. The response plan is automated and accurate, shortening the abnormal processing time and significantly improving the robustness of the system. It solves the inefficient problem of abnormal event processing relying on manual intervention in traditional systems, and is an important innovation facing the actual traffic complexity.

[0128] 4. The present invention can dynamically monitor the execution effect of the optimization plan through a real-time feedback mechanism, and iteratively update the traffic prediction model and optimization algorithm. It can adapt to long-term changes in traffic patterns, maintain continuous improvement in system performance, ensure the adaptability of the system in diversified traffic scenarios, and avoid failure problems caused by model aging and changes in traffic patterns in traditional traffic light management. It is a technical embodiment of self-learning and self-adaptive capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0129] Figure 1 It is a system framework diagram of the present invention. DETAILED DESCRIPTION

[0130] In order to make the technical personnel in the technical field understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in combination with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a partial embodiment of the present invention, not a complete embodiment. Based on the embodiment of the present invention, other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0131] The present invention is described in detail below in conjunction with the accompanying drawings:

[0132] Example:

[0133] Please refer to the attached Figure 1 The embodiment of the present invention provides a road traffic signal light management system based on big data, including:

[0134] Data collection and fusion module, which collects traffic sensor data, visual data, GPS trajectory data and external environment data, and generates unified traffic status data through spatiotemporal alignment, data filtering and multimodal data fusion;

[0135] The traffic prediction module predicts the traffic flow at single intersections and regions using time series models and graph neural networks based on the traffic status data generated by the data collection and fusion module;

[0136] The single intersection dynamic optimization module generates a signal light cycle and phase adjustment plan using a reinforcement learning algorithm based on the single intersection traffic data predicted by the traffic prediction module;

[0137] The multi-intersection linkage optimization module uses the distributed optimization algorithm to coordinate the traffic lights at multiple intersections in the region based on the regional traffic flow prediction data generated by the flow prediction module and the traffic light scheme generated by the single-intersection dynamic optimization module.

[0138] The anomaly detection and response module receives the visual data and traffic data transmitted by the data acquisition and fusion module, identifies traffic accidents, congestion and vehicle priority requirements through anomaly detection algorithms, and generates traffic light adjustment plans under abnormal conditions;

[0139] The feedback and iterative optimization module updates the optimization algorithm and prediction model based on the traffic light adjustment plans generated by the single-intersection dynamic optimization module and the multi-intersection linkage optimization module, as well as the adjustment plans generated by the anomaly detection and response module, combined with the real-time traffic status data from the data acquisition and fusion module.

[0140] Through the multimodal fusion of traffic sensors, visual equipment, GPS trajectory data and external environment data, comprehensive perception of traffic conditions is achieved; through spatiotemporal alignment, data filtering and Bayesian inference technology, the accuracy and consistency of data are significantly improved, and the changes in complex traffic scenes can be dynamically and accurately reflected, providing high-quality data support for traffic flow prediction and traffic light optimization;

[0141] Combining time series models and graph neural networks, it can accurately predict single intersection and regional traffic flow, capture short-term changes in traffic flow and the correlation between intersections in the road network, and provide a scientific decision-making basis for dynamic optimization of single intersections and linkage optimization of multiple intersections, effectively avoiding the decline in optimization effect caused by traffic prediction errors in traditional traffic light management.

[0142] The reinforcement learning algorithm is used to learn the current traffic status of the intersection in real time and adaptively adjust the cycle and phase of the traffic lights to reduce vehicle waiting time and improve the traffic efficiency of the intersection. Compared with the optimization method of static traffic light control or preset rules, it can dynamically adapt to the fluctuation of traffic flow and significantly improve the traffic performance of a single intersection.

[0143] Through regional collaborative modeling and distributed optimization algorithms, global collaborative control of traffic lights at multiple intersections is achieved to solve the regional imbalance problem caused by single intersection optimization in traditional systems. Through the combination of genetic algorithms and reinforcement learning, the cycle and phase of traffic lights at intersections in the region are optimized, effectively reducing the total waiting time and systemic congestion risks within the region, and improving the overall traffic operation efficiency.

[0144] Through computer vision technology and time series analysis methods, it can accurately identify abnormal events such as traffic accidents, congestion and vehicle priority. Combined with the automated response mechanism, it can quickly generate signal light adjustment plans for abnormal situations, shorten the response time, avoid the further spread of abnormal events to the road network, and enhance the robustness and emergency response capabilities of the system.

[0145] By real-time monitoring of the execution effect of the traffic light adjustment plan and using feedback data to dynamically update the traffic prediction model and optimization algorithm, the system's adaptability and long-term stability can be improved. Its dynamic iteration mechanism can cope with long-term changes in traffic flow patterns, ensure continuous improvement of system performance, and avoid optimization failures due to model aging or traffic changes.

[0146] The data acquisition and fusion module includes:

[0147] The data acquisition unit is used to collect traffic flow, speed and vehicle density data. Specifically, the traffic flow q(t), speed v(t) and vehicle density ρ(t) are obtained through the geomagnetic coil sensor, which satisfies the following relationship:

[0148] q(t)=ρ(t)·v(t),

[0149] Among them, q(t) represents the number of vehicles passing the sensor per unit time, ρ(t) represents the vehicle density per unit length of road, and v(t0 represents the average speed of vehicles on the road section;

[0150] The data processing unit is used to filter the traffic flow, speed and density data obtained by the data acquisition unit, and estimate and denoise the observed data through the Kalman filter formula. The specific formula is as follows:

[0151] in, is the state estimate after filtering, z k is the observed value at the current moment, K k is the Kalman gain, is the estimated value of k-1 at the previous moment;

[0152] The Kalman gain is calculated as:

[0153] Among them, P k-1 is the state prediction error covariance of the previous step, H is the observation matrix, R is the observation noise covariance, H T is the transpose of the observation model matrix;

[0154] The data fusion unit is used to align the processed multimodal data in time and space and achieve fusion through Bayesian inference, satisfying the following formula:

[0155]

[0156] Among them, P(S|D1, D2, ..., D n ) is the posterior probability of the current traffic state S after fusion, P(S is the prior probability of the traffic state, P(D i |S) is the observed data D i Conditional probability under traffic state S.

[0157] Through the synergy of the data acquisition unit, data processing unit and data fusion unit, comprehensive perception and high-precision processing of traffic status data are achieved. The vehicle flow, speed and density data collected by the geomagnetic coil are combined with Kalman filtering to denoise the original observation data and generate accurate traffic status estimates. At the same time, Bayesian inference is used to align and fuse multimodal data in time and space, and integrate data from different sensors into a unified traffic status description. The multi-level data processing and fusion method effectively improves the comprehensiveness, accuracy and consistency of the data, provides high-quality basic data support for subsequent traffic prediction and optimization control, and ensures the reliability and real-time performance of the system in complex traffic scenarios.

[0158] The traffic prediction module includes:

[0159] The single intersection prediction unit is used to generate traffic status data of a single intersection based on the data acquisition and fusion module, and use the long short-term memory network model to predict the future traffic flow of the single intersection. The long short-term memory network model includes the following state update formula:

[0160] Forget gate: f t =σ(W f ·[h t-1 , x t ]+b f ),

[0161] Among them, f t is the forget gate activation value, W f is the forget gate weight matrix, h t-1 is the hidden state of the previous time step, x t is the input feature of the current time step, b f is the forget gate bias term, σ is the activation function;

[0162] Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i ),

[0163]

[0164] Among them, i t is the input gate activation value, is the candidate memory state, W i and W i is the weight matrix of the input gate and the candidate memory state, b i and C are the corresponding bias terms, tanh is the hyperbolic tangent function;

[0165] Current memory status update:

[0166] Among them, C t is the memory state of the current time step, C t-1 is the memory state of the previous time step;

[0167] Output gate: o t =σ(W o ·[h t-1 , x t ]+b o ), h t =o t tanh(C t ),

[0168] Among them, t is the output gate activation value, h t is the hidden state of the current time step, W o is the output gate weight matrix, b o is the output gate bias term;

[0169] The regional traffic prediction unit is used to combine the regional traffic status data generated by the data acquisition and fusion module, and use the graph neural network to predict the traffic flow at multiple intersections in the region. The graph neural network is based on the following node feature update formula:

[0170]

[0171] Among them, H (l) is the node feature matrix of the lth layer, is the normalized adjacency matrix, W (l0 is the weight matrix of the lth layer, and σ is the activation function.

[0172] By combining the single-intersection prediction unit and the regional flow prediction unit, the accuracy and reliability of traffic flow prediction are significantly improved. The single-intersection prediction unit uses the long short-term memory network model to fully capture the time dependence and dynamic change law of vehicle flow, and accurately predict future flow trends. The regional flow prediction unit is based on the graph neural network to build a correlation model between multiple intersections. By updating the node features and adjacency relationships, it predicts the flow distribution of complex traffic networks in the region. The prediction mechanism of single intersection and regional linkage realizes refined traffic status prediction, providing scientific decision-making basis for single intersection dynamic optimization and multi-intersection linkage optimization, effectively avoiding optimization failure caused by prediction errors, and improving the system's adaptability and real-time performance to complex traffic scenarios.

[0173] The single intersection dynamic optimization module includes a state perception unit, a reinforcement learning decision unit, and a signal light execution unit:

[0174] The state perception unit receives the single intersection traffic prediction data from the traffic prediction module and the real-time traffic status data from the data acquisition and fusion module to determine the traffic status of the current intersection, including:

[0175] Traffic flow q(t), the number of vehicles passing through the intersection per unit time;

[0176] The average waiting time w(t) represents the average stay time of the vehicle queue, and the calculation formula is:

[0177]

[0178] Where L(t) is the length of the vehicle queue, ρ(t) is the vehicle density, and v(t) is the average vehicle speed;

[0179] The output traffic state S of the state perception unit t It is expressed as:

[0180] S t ={q(t),w(t),φ(t)},

[0181] Among them, φ(t) is the phase state of the current signal light;

[0182] The reinforcement learning decision-making unit is based on the traffic state S provided by the state perception unit. t , using deep reinforcement learning method to generate traffic light optimization decision A t , its core includes the following parts:

[0183] Action A t : A t ={ΔT, Δφ},

[0184] Among them, △T is the adjustment amount of the signal light cycle, and △φ is the adjustment amount of the signal light phase switching;

[0185] Reward R t : The reward function is designed to reduce the average waiting time and improve traffic efficiency. The specific formula is:

[0186] R t = -α·w(t)+β·q(t),

[0187] Among them, w(t) is the average waiting time, q(t) is the traffic flow, α and β represent the weight coefficients of waiting time and traffic flow;

[0188] The Q-value update formula based on reinforcement learning is:

[0189]

[0190] Among them, Q(S t ,A t ) indicates that in state S t Next, execute action A t The reward value, η is the learning rate, R t is the immediate reward value, γ is the discount factor, In the next state S t+1 The maximum expected reward under

[0191] The traffic light execution unit generates the optimized decision A based on the reinforcement learning decision unit. t , adjust the cycle and phase of the signal light, including:

[0192] Modify the signal light cycle T: T t+1 =T t +ΔT,

[0193] Among them, T t The current signal light cycle, the cycle adjustment amount generated by the △T decision;

[0194] Modify the signal light phase φ: φ t+1 =φ t +Δφ,

[0195] Among them, φ t is the current signal light phase time, the phase adjustment amount generated by the △φ decision.

[0196] Through the synergy of the state perception unit, the reinforcement learning decision unit and the traffic light execution unit, the intelligent dynamic optimization of the traffic lights at a single intersection is realized. The state perception unit integrates the flow prediction data and the real-time traffic status, accurately calculates the key indicators such as the vehicle flow and the average waiting time, and provides accurate state input for optimization. The reinforcement learning decision unit uses the deep reinforcement learning method to adaptively generate the adjustment plan of the traffic light cycle and phase according to the traffic status, weighs the vehicle waiting time and traffic efficiency through the reward function, and improves the decision-making performance through Q value iteration. The traffic light execution unit applies the optimization plan to the traffic light equipment in real time to realize the dynamic adjustment of the traffic light cycle and phase. Through the adaptive optimization mechanism, the average waiting time of vehicles is significantly reduced, the traffic efficiency of a single intersection is improved, and it can quickly respond to the fluctuation of traffic flow, solving the problem that traditional static traffic light control cannot adapt to complex traffic scenes.

[0197] The multi-intersection linkage optimization module includes:

[0198] The regional collaborative modeling unit receives the traffic flow prediction data of multiple intersections generated by the regional traffic prediction unit in the traffic prediction module, and constructs a traffic graph model of the regional road network with multiple intersections, which is defined as:

[0199] G=(V,E),

[0200] Among them, V is the intersection set, E is the road set;

[0201] The linkage optimization unit uses a distributed optimization algorithm to perform linkage optimization on the traffic lights at multiple intersections based on the traffic map G = (V, E) generated by the regional collaborative modeling unit, including:

[0202] The optimization goal is defined as minimizing the total average waiting time and total travel time at intersections in the area. The objective function is:

[0203] Among them, w i (t) is the intersection v i The average waiting time, T ij (t) is the road e ij The average travel time, λ is the trade-off coefficient;

[0204] The linkage optimization unit adjusts the signal light cycle T i Dynamic adjustment to achieve coordinated control of multiple intersections:

[0205] T i (t+1)=T i (t)+ΔT i ,

[0206] Among them, T i (t) is the intersection v i At time t, the signal light cycle, △T i is the cycle-adjusted amount;

[0207] Balanced traffic distribution is achieved by adjusting the phase of traffic lights between adjacent intersections. The adjustment formula is:

[0208] φ i (t+1)=φ i (t)+Δφ i ,

[0209] Among them, φ i (t) is the intersection v i The phase of the signal light at time t, △φ i Phase adjustment amount;

[0210] The multi-intersection linkage optimization module includes:

[0211] The distributed computing unit realizes multi-intersection linkage optimization through distributed reinforcement learning or genetic algorithm, specifically including genetic algorithm and distributed reinforcement learning algorithm;

[0212] Genetic algorithms include:

[0213] Chromosome encoding: The signal light cycle T of each intersection i and phase φ i Encoded as chromosome C, it is expressed as:

[0214] C={T1,φ1,T2,φ2,…,T N ,φ N},

[0215] Where N is the number of intersections in the area;

[0216] Fitness function: The fitness function is defined as the inverse function of the optimization objective and is expressed as:

[0217]

[0218] Among them, F(C) represents the objective function value of the state set C, V is the node set, and w i (t) is the weight or loss of node i at time t, E is the edge set, Tij (t) represents the impact or cost at time t, and λ is a weight parameter used to balance the node loss and edge loss;

[0219] Distributed reinforcement learning algorithms include:

[0220] State space: S t ={H i , W ij ∣v i ∈V,e ij ∈E},

[0221] Among them, S t represents the state set of the system at time t, H i represents the state variable of node i, W ij Represents edge e ij The weight or status of

[0222] Action Space: A t ={ΔT i , Δφ i ∣v i ∈V},

[0223] Among them, A t represents the updated set of nodes at time t, △T i Variable T representing node i at time t i The change in △φ i The variable φ represents the node i at time t i The amount of change;

[0224] Reward function: The reward function is defined as the weighted sum of the total waiting time and the travel time in the area:

[0225]

[0226] Among them, R t represents a certain benefit or cost function at time t, w i (t) represents the weight or cost of node i, T ij (t) represents the weight or cost at time t, λ is the balance coefficient;

[0227] Distributed training: Each intersection independently trains a local reinforcement learning model, and the regional control center achieves global optimization by aggregating local reward values;

[0228] The traffic light execution unit receives the traffic light adjustment plan calculated by the linkage optimization unit, and sends the adjustment plan to the traffic light equipment at each intersection in real time through the regional controller.

[0229] Through the collaboration of regional collaborative modeling units, linkage optimization units and distributed computing units, global optimization control of traffic lights at multiple intersections is achieved. The regional collaborative modeling unit constructs a traffic map model based on the connection relationship between intersections and roads, accurately describing the dynamic interaction characteristics of each intersection and road in the region. The linkage optimization unit uses a distributed optimization algorithm to minimize the total average waiting time and travel time, dynamically adjusts the traffic light cycle and phase, and improves the overall operation efficiency of the road network. The distributed computing unit uses genetic algorithms and reinforcement learning models to balance the weight loss of nodes and edges and optimize regional traffic flow. The traffic light execution unit applies the optimization results to actual equipment in real time to ensure the efficient execution of the optimization plan. This module avoids the local lag problem caused by traditional single intersection optimization through linkage coordination among multiple intersections, greatly improves the overall traffic efficiency of regional traffic, and reduces the risk of systemic congestion. It is a technical breakthrough in solving complex road network traffic control problems.

[0230] The anomaly detection and response module includes:

[0231] The anomaly detection unit receives the traffic sensor data and visual data from the data acquisition and fusion module, and detects traffic anomalies by combining time series analysis and computer vision algorithms, including:

[0232] Visual anomaly detection uses the target detection algorithm to analyze the video stream collected by the camera and extract traffic anomaly targets. The detection content includes illegal parking, traffic accidents, etc. The target detection outputs the bounding box B and category label C of the abnormal target:

[0233] B = {(x1, y1, x2, y2)}, C∈{illegal parking, accident, congestion},

[0234] Among them, B represents the spatial location where the abnormal event occurs, and C represents the type of abnormality detected.

[0235] Combined with the time series traffic flow data, the anomaly detection algorithm is used to identify potential anomalies in the traffic flow, and a feature vector is constructed based on the traffic flow q(t), waiting time w(t) and road travel time T(t):

[0236] X t ={q(t),w(t),T(t)},

[0237] The calculation formula of abnormal score is:

[0238] Among them, X t is the feature vector at time t, path length (X t , T i ) represents the feature point X t In the tree i The path length in , S tis the abnormal score at time t;

[0239] The anomaly classification unit classifies the anomaly type and generates the corresponding priority according to the output of the anomaly detection unit. The anomaly type classification rules are as follows:

[0240] Vehicle priority category: including emergency vehicles, fire trucks or police cars’ priority needs;

[0241] Intersection congestion: Based on time series anomaly detection t >threshold, the current road section is determined to be congested;

[0242] Traffic accident category: Category C = accident in the visual inspection result, which is determined to be a traffic accident;

[0243] The priority P of each abnormal event is generated according to the following rules:

[0244] Vehicle priority, traffic accidents, intersection congestion;

[0245] The anomaly detection and response module includes:

[0246] The response plan generation unit generates a real-time response plan based on the abnormal type and priority of the abnormal classification unit and the control strategy of the single intersection dynamic optimization module and the multi-intersection linkage optimization module, including:

[0247] If the abnormal type is vehicle priority, the phase of the corresponding intersection signal light is dynamically adjusted to open the green channel. The signal light phase adjustment formula is:

[0248] φ i (t+1)=φ i (t)+Δφ priority ,

[0249] Among them, φ i (t) is the current signal light phase, △φ priority Priority phase adjustment, φ i (t+1 is the phase of the signal light at time t+1;

[0250] If the abnormal type is congestion at the intersection, adjust the signal light cycle to relieve traffic pressure. The signal light cycle adjustment formula is: T i (t+1)=T i (t)+ΔT congestion ,

[0251] Among them, T i (t) is the current signal light cycle, △T congestion Congestion relief cycle adjustment, T i (t+1 is the signal light cycle at time t+1;

[0252] If the abnormal type is a traffic accident, the multi-intersection linkage optimization module is combined to recalculate the regional traffic light cycle and phase to generate a global response plan;

[0253] The feedback adjustment unit monitors the execution effect of the response plan based on the real-time traffic status data of the data acquisition and fusion module, and dynamically modifies the response plan based on the feedback data. The modification logic is as follows:

[0254] If congestion persists, increase the cycle adjustment: ΔT congestion ←ΔT congestion +δ, where △T congestion It represents the change of the state variable related to "congestion", and δ is the incremental value;

[0255] When the accident is handled, the default traffic light cycle and phase are restored.

[0256] Through the synergy of the anomaly detection unit, the anomaly classification unit, the response plan generation unit and the feedback adjustment unit, accurate detection, real-time classification, rapid response and dynamic correction of traffic abnormal events are achieved. The anomaly detection unit combines the target detection algorithm and the time series analysis method to identify abnormal events such as illegal parking, traffic accidents, and intersection congestion in real time, and generate the spatial position and category label of the abnormal target, as well as the abnormal score. The anomaly classification unit reasonably divides the event type and generates the priority according to the detection results, providing a scientific basis for the response strategy. The response plan generation unit dynamically adjusts the cycle and phase of the traffic light according to the event type and priority, and achieves rapid and efficient response by opening the green channel, optimizing the congestion relief strategy or re-planning the regional traffic lights. The feedback adjustment unit monitors the response effect in real time, dynamically adjusts the strategy parameters according to the traffic status, and ensures the efficient handling of abnormal events. Through the automated and precise anomaly detection and response mechanism, the response time is significantly shortened, the inefficiency of traditional reliance on manual intervention is avoided, the robustness and adaptability of the system are enhanced, and the stable operation of the traffic network under emergencies is ensured.

[0257] The feedback and iterative optimization module monitors traffic performance in real time, including average vehicle waiting time and traffic rate, and uses historical data to update the parameters of the long short-term memory network and reinforcement learning model.

[0258] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A road traffic signal light management system based on big data, characterized in that: include: Data collection and fusion module, which collects traffic sensor data, visual data, GPS trajectory data and external environment data, and generates unified traffic status data through spatiotemporal alignment, data filtering and multimodal data fusion; The traffic prediction module predicts the traffic flow at single intersections and regions using time series models and graph neural networks based on the traffic status data generated by the data collection and fusion module; The single intersection dynamic optimization module generates a signal light cycle and phase adjustment plan using a reinforcement learning algorithm based on the single intersection traffic data predicted by the traffic prediction module; The multi-intersection linkage optimization module uses the distributed optimization algorithm to coordinate the traffic lights at multiple intersections in the region based on the regional traffic flow prediction data generated by the flow prediction module and the traffic light scheme generated by the single-intersection dynamic optimization module. The anomaly detection and response module receives the visual data and traffic data transmitted by the data acquisition and fusion module, identifies traffic accidents, congestion and vehicle priority requirements through anomaly detection algorithms, and generates traffic light adjustment plans under abnormal conditions; The feedback and iterative optimization module updates the optimization algorithm and prediction model based on the traffic light adjustment plans generated by the single-intersection dynamic optimization module and the multi-intersection linkage optimization module, as well as the adjustment plans generated by the anomaly detection and response module, combined with the real-time traffic status data from the data acquisition and fusion module.

2. A road traffic signal light management system based on big data according to claim 1, characterized in that: The data acquisition and fusion module includes: The data acquisition unit is used to collect traffic flow, speed and vehicle density data. Specifically, the traffic flow q(t), speed v(t) and vehicle density ρ(t) are obtained through the geomagnetic coil sensor, which satisfies the following relationship: q(t)=ρ(t)·v(t), Among them, q(t) represents the number of vehicles passing the sensor per unit time, ρ(t) represents the vehicle density per unit length of road, and v(t) represents the average speed of vehicles on the road section; The data processing unit is used to filter the traffic flow, speed and density data obtained by the data acquisition unit, and estimate and denoise the observed data through the Kalman filter formula. The specific formula is as follows: in, is the state estimate after filtering, z k is the observed value at the current moment, K k is the Kalman gain, is the estimated value of the previous moment k-1; The Kalman gain is calculated as: Among them, P k-1 is the state prediction error covariance of the previous step, H is the observation matrix, R is the observation noise covariance, H T is the transpose of the observation model matrix; The data fusion unit is used to align the processed multimodal data in time and space and achieve fusion through Bayesian inference, satisfying the following formula: Among them, P(S|D1, D2, ..., D n ) is the posterior probability of the current traffic state S after fusion, P(S is the prior probability of the traffic state, P(D i |S) is the observed data D i Conditional probability under traffic state S.

3. The road traffic signal light management system based on big data according to claim 1 is characterized in that: The traffic prediction module includes: The single intersection prediction unit is used to generate traffic status data of a single intersection based on the data acquisition and fusion module, and predict the future traffic flow of the single intersection using a long short-term memory network model. The long short-term memory network model includes the following state update formula: Forget gate: f t =σ(W f ·[h t-1 , x t ]+b f ), Among them, f t is the forget gate activation value, W f is the forget gate weight matrix, h t-1 is the hidden state of the previous time step, x t is the input feature of the current time step, b f is the forget gate bias term, σ is the activation function; Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i ), Among them, i t is the input gate activation value, is the candidate memory state, W i and W i is the weight matrix of the input gate and the candidate memory state, b i and C are the corresponding bias terms, tanh is the hyperbolic tangent function; Current memory status update: Among them, C t is the memory state of the current time step, C t-1 is the memory state of the previous time step; Output gate: o t =σ(W o ·[h t-1 , x t ]+b o ), h t =o t tanh(C t ), Among them, t is the output gate activation value, h t is the hidden state of the current time step, W o is the output gate weight matrix, b o is the output gate bias term; The regional traffic prediction unit is used to combine the regional traffic status data generated by the data acquisition and fusion module, and use the graph neural network to predict the traffic flow of multiple intersections in the region. The graph neural network is based on the following node feature update formula: Among them, H (l) is the node feature matrix of the lth layer, is the normalized adjacency matrix, W (l) is the weight matrix of the lth layer, and σ is the activation function.

4. The road traffic signal light management system based on big data according to claim 1 is characterized in that: The single intersection dynamic optimization module includes a state perception unit, a reinforcement learning decision unit and a signal light execution unit: The state sensing unit receives the single intersection traffic prediction data of the traffic prediction module and the real-time traffic status data of the data acquisition and fusion module to determine the traffic status of the current intersection, including: Traffic flow q(t), the number of vehicles passing through the intersection per unit time; The average waiting time w(t) represents the average stay time of the vehicle queue, and the calculation formula is: Where L(t) is the length of the vehicle queue, ρ(t) is the vehicle density, and v(t) is the average vehicle speed; The output traffic state S of the state perception unit t It is expressed as: S t ={q(t),w(t),φ(t)}, Among them, φ(t) is the phase state of the current traffic light.

5. A road traffic signal light management system based on big data according to claim 4, characterized in that: The reinforcement learning decision unit is based on the traffic state S provided by the state perception unit. t , using deep reinforcement learning method to generate traffic light optimization decision A t , its core includes the following parts: actionA t :A t ={ΔT, Δφ}, Among them, △T is the adjustment amount of the signal light cycle, and △φ is the adjustment amount of the signal light phase switching; Reward R t : The reward function is designed to reduce the average waiting time and improve traffic efficiency. The specific formula is: R t =-α·w(t)+β·q(t), Among them, w(t) is the average waiting time, q(t) is the traffic flow, α and β represent the weight coefficients of waiting time and traffic flow; The Q-value update formula based on reinforcement learning is: Among them, Q(S t ,A t ) indicates that in state S t Next, execute action A t The reward value, η is the learning rate, R t is the immediate reward value, γ is the discount factor, In the next state S t+1 The maximum expected reward under The signal light execution unit generates an optimization decision A based on the reinforcement learning decision unit. t , adjust the cycle and phase of the signal light, including: Modify the signal light cycle T: T t+1 =T t +ΔT, Among them, T t The current signal light cycle, the cycle adjustment amount generated by the △T decision; Modify the signal light phase φ: φ t+1 =φ t +Δφ, Among them, φ t is the current signal light phase time, the phase adjustment amount generated by the △φ decision.

6. The road traffic signal light management system based on big data according to claim 1 is characterized in that: The multi-intersection linkage optimization module includes: The regional collaborative modeling unit receives the traffic flow prediction data of multiple intersections generated by the regional traffic prediction unit in the traffic prediction module, and constructs a traffic graph model of the regional road network with multiple intersections, which is defined as: G=(V,E), Among them, V is the intersection set, E is the road set; The linkage optimization unit uses a distributed optimization algorithm to perform linkage optimization on the traffic lights at multiple intersections based on the traffic map G = (V, E) generated by the regional collaborative modeling unit, including: The optimization goal is defined as minimizing the total average waiting time and total travel time at intersections in the area. The objective function is: Among them, w i (t) is the intersection v i The average waiting time, T ij (t) is the road e ij The average travel time, λ is the trade-off coefficient; The linkage optimization unit adjusts the signal light cycle T i Dynamic adjustment to achieve coordinated control of multiple intersections: T i (t+1)=T i (t)+ΔT i , Among them, T i (t) is the intersection v i At time t, the signal light cycle, △T i is the cycle-adjusted amount; Balanced traffic distribution is achieved by adjusting the phase of traffic lights between adjacent intersections. The adjustment formula is: f i (t+1)=φ i (t)+Δφ i , Among them, φ i (t) is the intersection v i The phase of the signal light at time t, △φ i Phase adjustment amount.

7. The road traffic signal light management system based on big data according to claim 1 is characterized in that: The multi-intersection linkage optimization module includes: The distributed computing unit realizes multi-intersection linkage optimization through distributed reinforcement learning or genetic algorithm, specifically including genetic algorithm and distributed reinforcement learning algorithm; The genetic algorithm comprises: Chromosome encoding: The signal light cycle T of each intersection i and phase φ i Encoded as chromosome C, it is expressed as: C={T1,φ1,T2,φ2,…,T N ,f N }, Where N is the number of intersections in the area; Fitness function: The fitness function is defined as the inverse function of the optimization objective and is expressed as: Among them, F(C) represents the objective function value of the state set C, V is the node set, and w i (t) is the weight or loss of node i at time t, E is the edge set, T ij (t) represents the impact or cost at time t, and λ is a weight parameter used to balance the node loss and edge loss; The distributed reinforcement learning algorithm includes: State space: S t ={H i , W ij ∣v i ∈V,e ij ∈E}, Among them, S t represents the state set of the system at time t, H i represents the state variable of node i, W ij Represents edge e ij The weight or status of Action Space: A t ={ΔT i , Δφ i ∣v i ∈V}, Among them, A t represents the updated set of nodes at time t, △T i Variable T representing node i at time t i The change in △φ i The variable φ represents the node i at time t i The amount of change; Reward function: The reward function is defined as the weighted sum of the total waiting time and the travel time in the area: Among them, R t represents a certain benefit or cost function at time t, w i (t) represents the weight or cost of node i, T ij (t) represents the weight or cost at time t, λ is the balance coefficient; Distributed training: Each intersection independently trains a local reinforcement learning model, and the regional control center achieves global optimization by aggregating local reward values; The traffic light execution unit receives the traffic light adjustment plan calculated by the linkage optimization unit, and sends the adjustment plan to the traffic light equipment at each intersection in real time through the regional controller.

8. The road traffic signal light management system based on big data according to claim 1 is characterized in that: The anomaly detection and response module includes: The anomaly detection unit receives the traffic sensor data and visual data from the data acquisition and fusion module, and detects traffic anomalies by combining time series analysis and computer vision algorithms, including: Visual anomaly detection uses the target detection algorithm to analyze the video stream collected by the camera and extract traffic anomaly targets. The detection content includes illegal parking, traffic accidents, etc. The target detection outputs the bounding box B and category label C of the abnormal target: B = {(x1, y1, x2, y2)}, C∈{illegal parking, accident, congestion}, Among them, B represents the spatial location where the abnormal event occurs, and C represents the type of abnormality detected. Combined with the time series traffic flow data, the anomaly detection algorithm is used to identify potential anomalies in the traffic flow, and a feature vector is constructed based on the traffic flow q(t), waiting time w(t) and road travel time T(t): X t ={q(t),w(t),T(t)}, The calculation formula of abnormal score is: Among them, X t is the feature vector at time t, path length (X t , T i ) represents the feature point X t In the tree i The path length in , S t is the abnormal score at time t; The anomaly classification unit classifies the anomaly type and generates the corresponding priority according to the output of the anomaly detection unit. The anomaly type classification rules are as follows: Vehicle priority category: including emergency vehicles, fire trucks or police cars’ priority needs; Intersection congestion: Based on time series anomaly detection t >threshold, the current road section is determined to be congested; Traffic accident category: Category C = accident in the visual inspection result, which is determined to be a traffic accident; The priority P of each abnormal event is generated according to the following rules: Vehicle priority, traffic accidents, intersection congestion, etc.

9. The road traffic signal light management system based on big data according to claim 1, characterized in that: The anomaly detection and response module includes: The response plan generation unit generates a real-time response plan based on the abnormal type and priority of the abnormal classification unit and the control strategy of the single intersection dynamic optimization module and the multi-intersection linkage optimization module, including: If the abnormal type is vehicle priority, the phase of the corresponding intersection signal light is dynamically adjusted to open the green channel. The signal light phase adjustment formula is: f i (t+1)=φ i (t)+Δφ priority , Among them, φ i (t) is the current signal light phase, △φ priority Priority phase adjustment, φ i (t+1 is the phase of the signal light at time t+1; If the abnormal type is congestion at the intersection, adjust the signal light cycle to relieve traffic pressure. The signal light cycle adjustment formula is: T i (t+1)=T i (t)+ΔT congestion , Among them, T i (t) is the current signal light cycle, △T congestion Congestion relief cycle adjustment, T i (t+1 is the signal light cycle at time t+1; If the abnormal type is a traffic accident, the multi-intersection linkage optimization module is combined to recalculate the regional traffic light cycle and phase to generate a global response plan; The feedback adjustment unit monitors the execution effect of the response plan based on the real-time traffic status data of the data acquisition and fusion module, and dynamically modifies the response plan based on the feedback data. The modification logic is as follows: If congestion persists, increase the cycle adjustment: ΔT congestion ←ΔT congestion +δ, Among them, △T congestion Indicates the change in state variables related to "congestion", δ is the incremental value; When the accident is handled, the default traffic light cycle and phase are restored.

10. The road traffic signal light management system based on big data according to claim 1, characterized in that: The feedback and iterative optimization module monitors traffic performance in real time, including average vehicle waiting time and traffic rate, and uses historical data to update parameters of the long short-term memory network and reinforcement learning model.

Images