Smart city traffic management method based on big data

Abstract

The invention discloses a smart city traffic management method based on big data, and belongs to the field of traffic management, and the management method comprises the following specific steps: I, obtaining real-time traffic data through a plurality of channels, and constructing a traffic data association library in combination with historical traffic data; iI, constructing an urban traffic network model, synchronizing the collected traffic data in real time, and predicting the traffic flow of each road in the city; long-term congestion caused by a short-term strategy can be avoided, the signal coordination capability of a large-scale road network is improved, the method adapts to multiple traffic modes, and the refinement degree of signal lamp regulation is improved; the feasibility and rationality of path planning can be improved, a high-quality path can be provided in limited time, the accuracy of dynamic path planning is improved, the overall traffic efficiency is optimized, and the urban average congestion time is shortened.

Description

Technical Field

[0001] The present invention relates to the field of traffic management, and in particular to a smart city traffic management method based on big data. Background Art

[0002] Urban traffic congestion is becoming increasingly serious, resulting in longer travel time, increased energy consumption, and worsening environmental pollution, which seriously affects the quality of life of residents and the sustainable development of cities. Traditional traffic management methods rely on preset rules and manual intervention, lack the ability to perceive and predict real-time traffic conditions, and are difficult to adapt to the increasingly complex traffic environment; therefore, it is particularly important to invent a smart city traffic management method based on big data.

[0003] After searching, Chinese patent number CN118736819A discloses a smart city traffic management method based on artificial intelligence. Although this invention prevents the occurrence of traffic accidents based on risk factors and regional information and reduces the probability of traffic accidents, it is prone to long-term congestion caused by short-term strategies, reduces the signal coordination ability of large-scale road networks, cannot adapt to various traffic modes, and reduces the degree of refinement of traffic light control; in addition, the feasibility and rationality of path planning in existing smart city traffic management methods are low, and high-quality paths cannot be provided within a limited time, reducing the accuracy of dynamic path planning; for this reason, we propose a smart city traffic management method based on big data. Summary of the invention

[0004] The purpose of the present invention is to solve the defects in the prior art and to propose a smart city traffic management method based on big data.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A smart city traffic management method based on big data, the specific steps of the management method are as follows:

[0007] Ⅰ. Obtain real-time traffic data through multiple channels and combine it with historical traffic data to build a traffic data association database;

[0008] Ⅱ. Build a city traffic network model, synchronize the collected traffic data in real time, and predict the traffic flow of each road in the city;

[0009] III. After the traffic flow prediction of each road is completed, the signal light timing strategy is dynamically adjusted according to the traffic conditions of different roads;

[0010] IV. Analyze lane usage in real time based on monitoring data, and optimize and adjust lane allocation strategies in real time for emergencies or changes in traffic flow during peak hours;

[0011] V. Based on traffic forecast results, diversion suggestions are pushed to drivers, and driving routes are dynamically adjusted in the event of sudden congestion.

[0012] As a further solution of the present invention, the specific steps of constructing the traffic data association database in step I are as follows:

[0013] S1.1: Real-time traffic data is obtained through various channels such as city sensors, cameras, GPS devices and mobile terminals. Abnormal data points in each group of traffic data are detected and removed using the Z-score method. The missing data are then filled using the Lagrange interpolation method. The time interpolation method is used to align the timestamps of traffic data from different sources. The GPS coordinates of each group of traffic data are then converted into standard latitude and longitude formats.

[0014] S1.2: Extract the text of each traffic event from the traffic database according to the predefined rules, and then use the named entity recognition technology to extract entities from each group of traffic event texts, including roads, vehicles, traffic lights and events. Then use NLP technology to identify the relationship between the extracted entities, and store each entity and its corresponding relationship in the form of triples. The constructed triple data is stored in the Neo4j library to build a complete traffic data association library;

[0015] S1.3: Estimate the probability of a vehicle moving from section A to section B based on the random walk algorithm, and predict the missing relationships between entities through the graph completion algorithm. Then, based on the continuous update of traffic data, add new traffic events or road changes to the traffic data association library, and remove the relationships between entities that are no longer applicable.

[0016] As a further solution of the present invention, the specific steps of predicting the traffic flow of each road in the city in step II are as follows:

[0017] S2.1: Collect historical traffic data and external factors, match historical traffic data with corresponding external factors, preprocess each set of matched data, divide each set of processed data into training set, test set and validation set, and then create and initialize a set of traffic prediction models based on the hybrid architecture of CNN and LSTM network;

[0018] S2.2: The training set is input into the traffic prediction model. The traffic prediction model forward propagates the training set, outputs the global average pooled feature vector through multi-layer 1D convolution and pooling of the one-dimensional convolution branch of the CNN layer of the model, and then extracts the spatial correlation between different roads through multi-layer transposed convolution, and analyzes the impact of the congestion of each road on the surrounding roads;

[0019] S2.3: The traffic prediction model then propagates and calculates the training set through the input gate, forget gate, and output gate in the LSTM layer, and outputs the trend of future traffic flow. The CNN layer output and the LSTM layer output are then passed to the fully connected layer, and the two sets of output results are processed nonlinearly through the Softmax activation function. The output layer outputs the predicted road traffic flow value and road congestion.

[0020] S2.4: The loss value between the prediction result of the traffic prediction model and the actual traffic situation is calculated through the MSE function, and then the calculated loss value is back-propagated from the output layer of the traffic prediction model, and the gradient of the loss value to the model parameters is calculated, and then the parameters are updated using the Adam optimizer. After each round of training, the validation set is input into the traffic prediction model and its corresponding loss value is calculated. If the validation loss does not decrease for several consecutive times, the training is stopped, otherwise the traffic prediction model is retrained and validated;

[0021] S2.6: After training, the performance of the traffic prediction model is evaluated through the test set, and the trained traffic prediction model is deployed to the monitoring platform. The collected real-time traffic data is input into the trained traffic prediction model, and the predicted traffic flow values ​​and road congestion conditions of each road are obtained through the forward propagation of the traffic prediction model.

[0022] As a further solution of the present invention, the specific steps of dynamically adjusting the signal light timing strategy in step III are as follows:

[0023] S3.1: Collect the traffic conditions of the current road intersection, including the queue length of each direction lane, the current signal light status, the historical traffic flow trend and the road prediction results, and set the state vector of the current road according to the collected data. Then, construct an action set according to extending the current green light time, shortening the current green light time and switching the signal light phase.

[0024] S3.2: Based on the traffic light control and random road factors in the action set, calculate the transition probability of the traffic condition at each road intersection to the next random traffic condition, set the reward function based on minimizing the vehicle waiting time and maximizing the road traffic rate, and calculate the reward value of the traffic condition change at each road intersection after each traffic light control action is selected;

[0025] S3.3: Calculate the long-term benefit of taking any traffic light control action in the action set under the current traffic conditions at the road intersection, and select a series of traffic light control actions that maximize the long-term cumulative reward. Then use Q-learning to update the rules, continuously simulate and select the traffic light control strategy with the largest long-term cumulative reward, until the strategy reward value converges to the preset range, then output the series of traffic light control actions that maximize the long-term cumulative reward, and use it as the traffic light timing strategy for traffic light control.

[0026] As a further solution of the present invention, the lane allocation strategy described in step IV is optimized and adjusted in real time for emergencies or changes in traffic flow during peak hours:

[0027] S4.1: Obtain traffic data of each lane in real time, including flow, average speed, occupancy rate and lane change demand, and set the state set of each lane. Then, construct a lane optimization objective function based on minimizing the total lane congestion index, randomly initialize a population containing multiple lane allocation schemes, and calculate the fitness of each lane allocation scheme through the lane optimization objective function, and select the allocation scheme with the lowest current fitness as the optimal solution;

[0028] S4.2: Set a set of coefficient vectors, and generate a random number between 0 and 1 in each round of iteration. If the generated random number is greater than the preset probability parameter, the coefficient A and step length L for controlling the position update are calculated based on the coefficient vector, and the remaining individuals are judged according to the size of the direction to shrink around the optimal solution;

[0029] S4.3: If |A| < 1, it means that the remaining individuals are surrounding the optimal solution, and the positions of the remaining individuals are updated according to the optimal solution, that is, the lane allocation scheme corresponding to each individual is adjusted. If |A| ≥ 1, it means that the remaining individuals are far away from the optimal solution. Through random search, a random position is selected in the population space, and the positions of the remaining individuals are updated. After each iteration, the coefficient vector is updated based on the linear decreasing rule.

[0030] S4.4: If the generated random number is less than or equal to the preset probability parameter, the distance between the positions of the remaining individuals in the population and the optimal solution position is calculated, and the movement law of the individuals of each population along the spiral trajectory approaching the optimal solution is simulated through the spiral motion formula to update the positions of the individuals in the population;

[0031] S4.5: Repeatedly select the optimal solution and iteratively update the position until the fitness value converges to the preset threshold range, then compare the fitness values ​​of individuals in each group of populations, and output the lane allocation plan with the highest fitness value, and adjust the current lane usage based on the lane allocation plan.

[0032] As a further solution of the present invention, the specific steps of dynamically adjusting the driving route in step V are as follows:

[0033] S5.1: Establish a road network graph, in which nodes represent intersections and edges represent roads. According to the node where the vehicle is currently located, the target node, the current time, and the predicted traffic conditions, establish a vehicle path planning state set, and then form a corresponding path based on the vehicle selecting the next intersection from the current intersection;

[0034] S5.2: Take the node where the vehicle is currently located in the road network as the root node, and calculate the UCB values ​​of the nodes connected to the node. Based on the upper confidence interval selection strategy, gradually select the node with the largest UCB value until an unexplored node is found. Then select the adjacent intersection of the current intersection and add it to the search tree as a new candidate path to expand the new path;

[0035] S5.4: Starting from the newly expanded node, randomly simulate a complete path to the target node and calculate the total travel time. Then trace the simulation results back to the root node and update the number of visits and average travel time of each node in the same path.

[0036] S5.5: Repeat the selection, expansion, simulation and backtracking process until the path travel time changes within a preset range during multiple rounds of iterations, then traverse each group of paths constructed in the road network graph, select the path with the minimum travel time as the optimal path for the vehicle, and send a new path planning instruction to the navigation system.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. This smart city traffic management method based on big data collects the traffic conditions of the current road intersections, including the queue length of lanes in each direction, the current signal light status, historical traffic flow trends and road prediction results, and sets the state vector of the current road according to the collected data. Then, an action set is constructed based on extending the current green light time, shortening the current green light time and switching the signal light phase. According to the signal light control and random road factors in the action set, the transfer probability of the traffic conditions of each road intersection to the next random traffic condition is calculated. The reward function is set based on minimizing the vehicle waiting time and maximizing the road traffic rate, and the traffic conditions of each road intersection after each signal light control action is selected are calculated. The reward value of the situation change is calculated, and the long-term benefit of taking any signal light control action in the action set under the current traffic conditions at the road intersection is calculated. A series of signal light control actions that maximize the long-term cumulative reward are selected, and then the Q-learning update rules are used to continuously simulate and select the signal light control strategy with the largest long-term cumulative reward until the strategy reward value converges to the preset range. After that, a series of signal light control actions that maximize the long-term cumulative reward are output and used as the signal light timing strategy for signal light control. This can avoid long-term congestion caused by short-term strategies, improve the signal coordination capability of large-scale road networks, adapt to various traffic modes, and improve the refinement of signal light control.

[0039] 2. The smart city traffic management method based on big data establishes a road network diagram and a set of vehicle path planning states, and then selects the next intersection from the current intersection based on the vehicle to form a corresponding path. The node where the vehicle is currently located in the road network diagram is used as the root node, and the UCB values ​​of the nodes connected to the node are calculated. Based on the upper confidence interval selection strategy, the node with the largest UCB value is gradually selected until an unexplored node is searched. Then, the adjacent intersection of the current intersection is selected and added to the search tree as a new candidate path to expand the new path. Starting from the newly expanded node, a complete path is randomly simulated until the target node. The total travel time is calculated, and then the simulation results are traced back to the root node, and the number of visits and average travel time of each node in the same path are updated. The selection, expansion, simulation and backtracking process are repeated until the path travel time changes in multiple rounds of iterations. The value converges to the preset range, and then the groups of paths constructed in the road network diagram are traversed, and the path with the minimum travel time is selected as the optimal path for the vehicle. At the same time, new path planning instructions are sent to the navigation system, which can improve the feasibility and rationality of path planning, provide high-quality paths within a limited time, improve the accuracy of dynamic path planning, optimize the overall travel efficiency, and reduce the average congestion time in the city. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0041] Figure 1 This is a flowchart of a smart city traffic management method based on big data proposed by the present invention. DETAILED DESCRIPTION

[0042] Example 1

[0043] Reference Figure 1 , a smart city traffic management method based on big data, the specific steps of the management method are as follows:

[0044] Real-time traffic data is obtained through multiple channels, and combined with historical traffic data to build a traffic data association library.

[0045] Specifically, real-time traffic data is obtained through various channels such as urban sensors, cameras, GPS devices and mobile terminals. The Z-score method is used to detect abnormal data points in each group of traffic data and remove them. Then, the Lagrange interpolation method is used to fill in the missing data, and the time interpolation method is used to align the timestamps of traffic data from different sources. Then, the GPS coordinates of each group of traffic data are converted into standard longitude and latitude formats. According to predefined rules, each traffic event text is extracted from the traffic database. Then, the named entity recognition technology is used to extract entities from each group of traffic event text, including roads, vehicles, traffic lights and events. Then, the NLP technology is used to identify the relationship between each extracted entity, and each entity and its corresponding relationship are stored in the form of triples. The constructed triple data are stored in the Neo4j library to construct a complete traffic data association library. The probability of a vehicle going from section A to section B is estimated based on the random walk algorithm, and the missing relationship between entities is predicted by the graph completion algorithm. Then, according to the continuous updating of traffic data, new traffic events or road changes are added to the traffic data association library, and the no longer applicable relationships between entities are removed.

[0046] Build an urban traffic network model, synchronize the collected traffic data in real time, and predict the traffic flow of each road in the city.

[0047] Specifically, historical traffic data and external factors are collected, and the historical traffic data are matched with the corresponding external factors. Then, the matched groups of data are preprocessed, and then the processed groups of data are divided into training set, test set and verification set. Then, according to the hybrid architecture of CNN and LSTM network, a set of traffic prediction models are created and initialized, and the training set is input into the traffic prediction model. The traffic prediction model forward propagates the training set, and outputs the global average pooling feature vector through the multi-layer 1D convolution and pooling of the one-dimensional convolution branch of the CNN layer of the model. Then, the spatial correlation between different roads is extracted through multi-layer transposed convolution, and the congestion of each road is analyzed. The traffic prediction model propagates and calculates the training set through the input gate, forget gate and output gate in the LSTM layer, and outputs the trend of future traffic flow changes. Then, the CNN layer output and the LSTM layer output are passed to the fully connected layer, and through the Sof The tmax activation function performs nonlinear processing on the two sets of output results, outputs the predicted road traffic flow values ​​and road congestion through the output layer, calculates the loss value between the predicted results of the traffic prediction model and the actual traffic conditions through the MSE function, and then backpropagates the calculated loss value from the output layer of the traffic prediction model, and calculates the gradient of the loss value to the model parameters, and then uses the Adam optimizer to update the parameters. After each round of training, the verification set is input into the traffic prediction model, and its corresponding loss value is calculated. If the verification loss does not decrease for many consecutive times, the training is stopped, otherwise the traffic prediction model is retrained and verified. After the training, the performance of the traffic prediction model is evaluated through the test set, and the trained traffic prediction model is deployed to the monitoring platform, the collected real-time traffic data is input into the trained traffic prediction model, and the predicted traffic flow values ​​and road congestion conditions of each road are obtained through the forward propagation of the traffic prediction model.

[0048] After the traffic flow prediction for each road is completed, the signal light timing strategy is dynamically adjusted according to the traffic conditions of different roads.

[0049] Specifically, the traffic conditions at the current road intersection are collected, including the queue length of lanes in each direction, the current signal light status, the historical traffic flow trend and the road prediction results, and the state vector of the current road is set according to the collected data sets. Then, an action set is constructed according to extending the current green light time, shortening the current green light time and switching the signal light phase. According to the signal light control and random road factors in the action set, the transition probability of the traffic conditions at each road intersection to the next random traffic condition is calculated. The reward function is set based on minimizing the vehicle waiting time and maximizing the road traffic rate, and the reward value of the traffic condition change at each road intersection after each signal light control action is selected is calculated. The long-term benefit of taking any signal light control action in the action set under the current traffic conditions at the road intersection is calculated, and a series of signal light control actions that maximize the long-term cumulative reward are selected. Then, the Q-learning update rule is used to continuously simulate and select the signal light control strategy with the largest long-term cumulative reward until the strategy reward value converges to a preset range. Then, a series of signal light control actions that maximize the long-term cumulative reward are output and used as the signal light timing strategy for signal light control.

[0050] Example 2

[0051] Reference Figure 1 , a smart city traffic management method based on big data, the specific steps of the management method are as follows:

[0052] Lane usage is analyzed in real time based on monitoring data, and lane allocation strategies are optimized and adjusted in real time to respond to emergencies or changes in traffic flow during peak hours.

[0053] Specifically, the traffic data of each lane is obtained in real time, including flow, average speed, occupancy rate and lane change demand, and the state set of each lane is set. Then, the lane optimization objective function is constructed based on minimizing the total lane congestion index, and a population containing multiple lane allocation schemes is randomly initialized. The fitness of each lane allocation scheme is calculated through the lane optimization objective function, and the allocation scheme with the lowest current fitness is selected as the optimal solution. A set of coefficient vectors is set, and at the same time, a random number between 0 and 1 is generated in each round of iteration. If the generated random number is greater than the preset probability parameter, the coefficient A and step size L for controlling the position update are calculated based on the coefficient vector. According to the size of the direction, it is determined whether the remaining individuals are shrinking around the optimal solution. If |A|<1, it means that the remaining individuals are surrounding the optimal solution, and the remaining individuals are positioned according to the optimal solution. Position update, that is, adjust the lane allocation scheme corresponding to each individual. If |A|≥1, it means that the remaining population individuals are far away from the optimal solution. Through random search, select random positions in the population space, update the positions of the remaining population individuals, and after each iteration, update the coefficient vector based on the linear decreasing rule. If the generated random number is less than or equal to the preset probability parameter, calculate the distance between the positions of the remaining population individuals in the population and the optimal solution position, and simulate the movement law of various population individuals along the spiral trajectory close to the optimal solution through the spiral motion formula, update the population individual position, repeatedly select the optimal solution and iteratively update the position until the fitness value converges to the preset threshold range, then compare the fitness values ​​of the population individuals in each group, output the lane allocation scheme with the highest fitness value, and adjust the current lane usage based on the lane allocation scheme.

[0054] Combined with traffic forecast results, diversion suggestions are pushed to drivers, and driving routes are dynamically adjusted in the event of sudden congestion.

[0055] Specifically, a road network graph is established, and the nodes in the road network graph represent intersections, and the edges represent roads. According to the node where the current vehicle is located, the target node, the current time, and the predicted traffic conditions, a vehicle path planning state set is established. Then, based on the vehicle selecting the next intersection from the current intersection, a corresponding path is formed. The node where the vehicle is currently located in the road network graph is taken as the root node, and the UCB values ​​of the nodes connected to the node are calculated. Based on the upper confidence interval selection strategy, the node with the largest UCB value is gradually selected until an incompletely explored node is found. Then, the adjacent intersection of the current intersection is selected and the It adds the search tree as a new candidate path to expand the new path. Starting from the newly expanded node, it randomly simulates a complete path to the target node and calculates the total travel time. The simulation result is then traced back to the root node, and the number of visits and average travel time of each node in the same path are updated. The selection, expansion, simulation and backtracking process are repeated until the path travel time changes in multiple rounds of iterations and converges to a preset range. After that, each group of paths constructed in the road network diagram is traversed, and the path with the minimum travel time is selected as the optimal path for the vehicle. At the same time, a new path planning instruction is sent to the navigation system.

Claims

1. A smart city traffic management method based on big data, characterized in that: The specific steps of this management method are as follows: Ⅰ. Obtain real-time traffic data through multiple channels and combine it with historical traffic data to build a traffic data association database; Ⅱ. Build a city traffic network model, synchronize the collected traffic data in real time, and predict the traffic flow of each road in the city; Ⅲ. After the traffic flow prediction of each road is completed, the signal light timing strategy is dynamically adjusted according to the traffic conditions of different roads; IV. Analyze lane usage in real time based on monitoring data, and optimize and adjust lane allocation strategies in real time for emergencies or changes in traffic flow during peak hours; V. Based on traffic forecast results, diversion suggestions are pushed to drivers, and driving routes are dynamically adjusted in the event of sudden congestion.

2. According to the big data-based smart city traffic management method of claim 1, it is characterized in that: The specific steps for constructing the traffic data association database described in step Ⅰ are as follows: S1.1: Real-time traffic data is obtained through various channels such as city sensors, cameras, GPS devices and mobile terminals. Abnormal data points in each group of traffic data are detected and removed using the Z-score method. The missing data are then filled using the Lagrange interpolation method. The time interpolation method is used to align the timestamps of traffic data from different sources. The GPS coordinates of each group of traffic data are then converted into standard latitude and longitude formats. S1.2: Extract the text of each traffic event from the traffic database according to the predefined rules, and then use the named entity recognition technology to extract entities from each group of traffic event texts, including roads, vehicles, traffic lights and events. Then use NLP technology to identify the relationship between the extracted entities, and store each entity and its corresponding relationship in the form of triples. The constructed triple data is stored in the Neo4j library to build a complete traffic data association library; S1.3: Estimate the probability of a vehicle moving from section A to section B based on the random walk algorithm, and predict the missing relationships between entities through the graph completion algorithm. Then, based on the continuous update of traffic data, add new traffic events or road changes to the traffic data association library, and remove the relationships between entities that are no longer applicable.

3. According to claim 2, a smart city traffic management method based on big data is characterized in that: The specific steps for predicting the traffic flow of each road in the city described in step II are as follows: S2.1: Collect historical traffic data and external factors, match historical traffic data with corresponding external factors, preprocess each set of matched data, divide each set of processed data into training set, test set and validation set, and then create and initialize a set of traffic prediction models based on the hybrid architecture of CNN and LSTM network; S2.2: The training set is input into the traffic prediction model. The traffic prediction model forward propagates the training set, outputs the global average pooled feature vector through multi-layer 1D convolution and pooling of the one-dimensional convolution branch of the CNN layer of the model, and then extracts the spatial correlation between different roads through multi-layer transposed convolution, and analyzes the impact of the congestion of each road on the surrounding roads; S2.3: The traffic prediction model then propagates and calculates the training set through the input gate, forget gate, and output gate in the LSTM layer, and outputs the trend of future traffic flow. The CNN layer output and the LSTM layer output are then passed to the fully connected layer, and the two sets of output results are processed nonlinearly through the Softmax activation function. The output layer outputs the predicted road traffic flow value and road congestion. S2.4: The loss value between the prediction result of the traffic prediction model and the actual traffic situation is calculated through the MSE function, and then the calculated loss value is back-propagated from the output layer of the traffic prediction model, and the gradient of the loss value to the model parameters is calculated, and then the parameters are updated using the Adam optimizer. After each round of training, the validation set is input into the traffic prediction model and its corresponding loss value is calculated. If the validation loss does not decrease for several consecutive times, the training is stopped, otherwise the traffic prediction model is retrained and validated; S2.6: After training, the performance of the traffic prediction model is evaluated through the test set, and the trained traffic prediction model is deployed to the monitoring platform. The collected real-time traffic data is input into the trained traffic prediction model, and the predicted traffic flow values ​​and road congestion conditions of each road are obtained through the forward propagation of the traffic prediction model.

4. The method for smart city traffic management based on big data according to claim 3 is characterized in that: The specific steps of dynamically adjusting the signal light timing strategy described in step III are as follows: S3.1: Collect the traffic conditions of the current road intersection, including the queue length of each direction lane, the current signal light status, the historical traffic flow trend and the road prediction results, and set the state vector of the current road according to the collected data. Then, construct an action set according to extending the current green light time, shortening the current green light time and switching the signal light phase. S3.2: Based on the traffic light control and random road factors in the action set, calculate the transition probability of the traffic condition at each road intersection to the next random traffic condition, set the reward function based on minimizing the vehicle waiting time and maximizing the road traffic rate, and calculate the reward value of the traffic condition change at each road intersection after each traffic light control action is selected; S3.3: Calculate the long-term benefit of taking any traffic light control action in the action set under the current traffic conditions at the road intersection, and select a series of traffic light control actions that maximize the long-term cumulative reward. Then use Q-learning to update the rules, continuously simulate and select the traffic light control strategy with the largest long-term cumulative reward, until the strategy reward value converges to the preset range, then output the series of traffic light control actions that maximize the long-term cumulative reward, and use it as the traffic light timing strategy for traffic light control.

5. The method for smart city traffic management based on big data according to claim 1 is characterized in that: The lane allocation strategy described in step IV is optimized and adjusted in real time for emergencies or changes in traffic flow during peak hours: S4.1: Obtain traffic data of each lane in real time, including flow, average speed, occupancy rate and lane change demand, and set the state set of each lane. Then, construct a lane optimization objective function based on minimizing the total lane congestion index, randomly initialize a population containing multiple lane allocation schemes, and calculate the fitness of each lane allocation scheme through the lane optimization objective function, and select the allocation scheme with the lowest current fitness as the optimal solution; S4.2: Set a set of coefficient vectors, and generate a random number between 0 and 1 in each round of iteration. If the generated random number is greater than the preset probability parameter, the coefficient A and step length L for controlling the position update are calculated based on the coefficient vector, and the remaining individuals are judged according to the size of the direction to shrink around the optimal solution; S4.3: If |A| < 1, it means that the remaining individuals are surrounding the optimal solution, and the positions of the remaining individuals are updated according to the optimal solution, that is, the lane allocation scheme corresponding to each individual is adjusted. If |A| ≥ 1, it means that the remaining individuals are far away from the optimal solution. Through random search, a random position is selected in the population space, and the positions of the remaining individuals are updated. After each iteration, the coefficient vector is updated based on the linear decreasing rule. S4.4: If the generated random number is less than or equal to the preset probability parameter, the distance between the positions of the remaining individuals in the population and the optimal solution position is calculated, and the movement law of the individuals of each population along the spiral trajectory approaching the optimal solution is simulated through the spiral motion formula to update the positions of the individuals in the population; S4.5: Repeatedly select the optimal solution and iteratively update the position until the fitness value converges to the preset threshold range, then compare the fitness values ​​of individuals in each group of populations, and output the lane allocation plan with the highest fitness value, and adjust the current lane usage based on the lane allocation plan.

6. The smart city traffic management method based on big data according to claim 1 is characterized in that: The specific steps of dynamically adjusting the driving route described in step V are as follows: S5.1: Establish a road network graph, in which nodes represent intersections and edges represent roads. According to the node where the vehicle is currently located, the target node, the current time, and the predicted traffic conditions, establish a vehicle path planning state set, and then form a corresponding path based on the vehicle selecting the next intersection from the current intersection; S5.2: Take the node where the vehicle is currently located in the road network as the root node, and calculate the UCB values ​​of the nodes connected to the node. Based on the upper confidence interval selection strategy, gradually select the node with the largest UCB value until an unexplored node is found. Then select the adjacent intersection of the current intersection and add it to the search tree as a new candidate path to expand the new path; S5.4: Starting from the newly expanded node, randomly simulate a complete path to the target node and calculate the total travel time. Then trace the simulation results back to the root node and update the number of visits and average travel time of each node in the same path. S5.5: Repeat the selection, expansion, simulation and backtracking process until the path travel time changes within a preset range during multiple rounds of iterations, then traverse each group of paths constructed in the road network graph, select the path with the minimum travel time as the optimal path for the vehicle, and send a new path planning instruction to the navigation system.

Images