Urban traffic spatiotemporal resource optimization method and device based on space computing

Abstract

The present invention discloses a method and device for optimizing urban traffic spatiotemporal resources based on space computing. First, a space constellation is designed to obtain urban satellite remote sensing images of a specified area; and sky sensing and calculation are realized to obtain real-time traffic flow information of a specified area within a set time interval; a traffic simulation tool is used to construct a signal-controlled intersection to obtain simulated traffic flow information; and training data sets are constructed based on the simulated traffic flow information, and a lane configuration agent and a signal timing agent based on reinforcement learning are trained using a deep Q network algorithm; finally, the trained lane configuration agent is used to regularly evaluate the state of the lane configuration and determine whether the lane direction needs to be adjusted; the trained signal timing agent monitors the traffic flow parameters in real time, and adjusts the signal timing parameters to optimize the traffic flow. The present invention can achieve the optimal configuration of traffic spatiotemporal resources.

Description

Technical Field

[0001] The present invention relates to the field of intelligent computing, and in particular to a method and device for optimizing urban traffic spatiotemporal resources based on space computing. Background Art

[0002] With the rapid development of urbanization, traffic congestion is becoming increasingly serious. Traditional traffic control methods rely on ground sensors and video surveillance, which have problems such as limited coverage, difficult maintenance, and untimely data updates. The development of satellite remote sensing technology has made large-scale, real-time traffic monitoring possible, but how to effectively apply satellite data to traffic control remains a challenge.

[0003] With the advancement of big data and artificial intelligence technologies, more and more research has begun to focus on how to use these technologies to optimize urban traffic management, and research on the optimization of time and space resources in urban traffic has made significant progress. However, these methods have inherent drawbacks. In terms of time optimization, rule-based methods rely on the experience of human experts and cannot adapt to highly dynamic traffic conditions; model-based predictive control and adaptive control methods require traffic flow models based on specific patterns or rules to solve optimization problems. Usually, these models are too complex and difficult to adjust and optimize; agent-based reinforcement learning methods have more research on optimizing signal timing, but relatively less research on optimizing spatial resources (such as the control of variable lanes). In terms of spatial optimization, fixed lane configuration may lead to a waste of time and space resources when traffic demand changes, when the green light time in a specific direction is allocated but no vehicles pass, or when poor signal timing leads to extremely unbalanced queue lengths in different lanes. Moreover, optimizing spatial resources or time resources alone is not enough to improve the overall performance of intersections, and time and space resources are interdependent. For example, fixed lane configuration may waste time resources, and fixed signal timing may lead to a waste of space resources. Summary of the invention

[0004] In view of the shortcomings of the current methods for obtaining traffic flow parameters and the existing spatiotemporal resource optimization methods, the present invention proposes a traffic spatiotemporal resource optimization method and device based on space computing and hierarchical reinforcement learning, which aims to obtain real-time traffic flow parameter data through the sky sensing and calculation of space constellations, and to coordinately optimize space and time resources through hierarchical reinforcement learning to meet the spatiotemporal resource optimization under different traffic demands, thereby improving the operating efficiency of urban traffic.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] A method for optimizing urban traffic spatiotemporal resources based on space computing, the method comprising the following steps:

[0007] Step 1: Design a space constellation to obtain satellite remote sensing images of cities in a designated area;

[0008] Step 2: Implement weather sensing and calculation to obtain real-time traffic flow information of a specified area within a set time interval;

[0009] Step 3: Use a traffic simulation tool to construct a signal-controlled intersection, define a road network description file, a traffic demand file, and a signal control plan, simulate the signal-controlled intersection, and obtain simulated traffic flow information;

[0010] Step 4: construct a lane configuration agent based on reinforcement learning, construct a lane configuration training data set based on the simulated traffic flow information, and use a deep Q-network algorithm to train the lane configuration agent so that it can regularly adjust the lane configuration according to the long-term traffic flow pattern; construct a signal timing agent based on reinforcement learning, construct a signal timing training data set based on the simulated traffic flow information, and use a deep Q-network algorithm to train the signal timing agent so that it can adjust the signal timing in real time according to the short-term traffic flow pattern;

[0011] Step 5: The real-time traffic flow information of the designated area obtained in step 2 is input into the trained lane configuration agent and signal timing agent. The trained lane configuration agent periodically evaluates the status of the lane configuration and decides whether the lane direction needs to be adjusted. The trained signal timing agent monitors the traffic flow parameters in real time and adjusts the signal timing parameters to optimize the traffic flow, thereby realizing the optimization of urban traffic spatiotemporal resources.

[0012] Furthermore, the designing of the space constellation in step 1 specifically includes determining the number of satellites, orbital altitude, orbital inclination and satellite distribution pattern.

[0013] Furthermore, the step 2 comprises the following steps:

[0014] S2.1: Obtain high-resolution satellite remote sensing images sufficient to identify vehicles and roads;

[0015] S2.2: Perform image preprocessing operations such as radiometric calibration, orthorectification, atmospheric correction and image fusion on the acquired high-resolution satellite remote sensing images to improve image quality;

[0016] S2.3: Identify the region of interest;

[0017] S2.4: Extract traffic flow related features through image segmentation, feature matching and pattern recognition;

[0018] S2.5: According to the characteristics of vehicles in satellite remote sensing images, object-oriented classification or support vector machine classification methods are used to extract texture features, train classifiers, and realize vehicle detection;

[0019] S2.6: Extract traffic flow parameters from vehicle detection results.

[0020] Furthermore, the signal-controlled intersection has four entrances, namely, east, south, west and north, each entrance has six lanes, the first two lanes from the left at each entrance are left-turn lanes, and the third lane from the left is a variable lane, that is, the lane is set for left turn or straight going; the first lane from the right is a right-turn lane, and the second and third lanes from the right are straight going lanes.

[0021] Furthermore, the state of the lane configuration agent is the past The average speed of each possible turning traffic flow in the time step, the action is the lane turning configuration of the variable lane at the signal-controlled intersection, and the reward is the ... The change in the number of vehicles passing through the signal-controlled intersection within a time step.

[0022] Furthermore, the state of the signal timing agent is the past The maximum queue length and number of vehicles passing each possible turning traffic flow in the time step, the action is to set a possible hyperparameter for the signal timing of the signal-controlled intersection, which is the minimum and maximum possible green light duration of the phase; the reward is The change in the sum of the total waiting times of all vehicles at the signalized intersection during the time step.

[0023] Furthermore, the loss function during the training of the lane configuration agent and the signal timing agent is the difference between the Q value predicted by the current strategy and the actually observed Q value.

[0024] A system for optimizing urban traffic spatiotemporal resources based on space computing, the system comprising one or more processors for implementing a method for optimizing urban traffic spatiotemporal resources based on space computing.

[0025] An electronic device, comprising:

[0026] one or more processors;

[0027] A storage device is used to store one or more programs. When the one or more programs are executed by the electronic device, the electronic device implements a method for optimizing urban traffic spatiotemporal resources based on space computing.

[0028] A computer-readable storage medium stores a program, which, when executed by a processor, implements a method for optimizing urban traffic spatiotemporal resources based on space computing.

[0029] The beneficial effects of the present invention are as follows:

[0030] 1. The present invention calculates orbital parameters and satellite distribution, designs space constellations, realizes comprehensive monitoring of urban traffic, can obtain traffic data in real time, and perform sky sensing and calculation. It uses the computing resources on the satellite to process remote sensing images in real time, extract traffic flow information, and improve the efficiency and real-time performance of data processing;

[0031] 2. The present invention is based on hierarchical reinforcement learning, and simultaneously trains lane configuration agents and signal timing agents, so that they can be dynamically adjusted according to long-term and short-term traffic flow patterns, and can achieve optimal configuration of traffic spatiotemporal resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Designing a schematic for a space computing constellation.

[0033] Figure 2 Schematic diagram of the signalized intersection designed for SUMO.

[0034] Figure 3 Schematic diagram of hierarchical reinforcement learning.

[0035] Figure 4 This is a schematic diagram of the signal light phase.

[0036] Figure 5 Schematic diagram of using trained lane configuration agents and signal timing agents to optimize traffic spatiotemporal resources. DETAILED DESCRIPTION

[0037] The present invention will be described in detail below based on the accompanying drawings and preferred embodiments, and the purpose and effects of the present invention will become more clear. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0038] The present invention aims to optimize the spatial and temporal resources of intersections to better guide traffic flow. It can adapt to changing traffic demands by dynamically adjusting lane directions and signal timing. By using hierarchical reinforcement learning, the original reinforcement learning task is decomposed into two objectives (i.e., lane configuration and signal timing optimization), each of which can be achieved through reinforcement learning.

[0039] The urban traffic spatiotemporal resource optimization method based on space computing of this embodiment includes the following steps:

[0040] Step 1: Design the space constellation and calculate the orbital parameters and satellite distribution to obtain satellite remote sensing images of cities in the specified area.

[0041] First, the design goals of the space constellation are determined, mainly to enable regional coverage, use high-resolution remote sensing payloads to capture urban satellite remote sensing images of designated areas, use space distributed computing systems to process information, and then transmit the data in real time to the ground station for call-up.

[0042] Then determine the orbit type of the satellite. Since real-time information transmission is required, a low-orbit satellite constellation with low transmission delay is selected. At the same time, through the collaborative work of multiple low-orbit satellites, more stable and continuous services are provided. The satellites in the constellation system can also complement each other to improve the reliability and redundancy of the system. The satellite orbit height h of the present invention is 500km.

[0043] Next, the orbital inclination θ is determined. In order to improve the practicality of the constellation, a medium inclination is selected to cover most population areas, and the orbital inclination of the present invention is 53 degrees.

[0044] Finally, the constellation configuration is selected. The present invention selects the Walker constellation. The parameters of the Walker constellation are expressed by T / P / F, where T is the total number of satellites, P is the number of orbital planes, and F is the phase parameter. Since we want to obtain real-time data with an interval of 15 minutes, the orbit height of the low-orbit satellite is 500km, and the period is estimated using Kepler's third law:

[0045]

[0046] Where T is the orbital period in seconds; r is the distance from the satellite to the center of the Earth in meters; u is the Kepler constant 3.986013×10 5 km 3 / s 2 .

[0047] Since the radius of the earth R is 6371km, the distance from the satellite to the center of the earth is r=R+h=6371km+500km=6871km. Therefore, substituting it into the formula, we get the orbital period T of about 5678.8s, which is about 90min.

[0048] Since this embodiment wants to obtain real-time data with a time interval of t=15 minutes, the minimum number of satellites that need to be arranged in the orbital plane of each satellite is:

[0049] T / t=90min / 15min=6 pcs

[0050] Since the earth is a sphere, if the distance between satellites is too small, the communication between satellites may be blocked by the ground and interrupted. Therefore, it is necessary to calculate the minimum communication angle to determine the number of satellites, such as Figure 1As shown. Specifically, it can be transformed into a mathematical geometry problem: Find the value of an angle in a right triangle. It is known that the two sides are R = 6371km and r = 6871km, and these two sides are the adjacent side and the hypotenuse. Then the cosine function is defined as the ratio of the adjacent side to the hypotenuse. Let θ be the angle between the two sides, and calculate the cosine value:

[0051] cosθ=adjacent side / hypotenuse=6371 / 6871≈0.9275

[0052] Next, we use the arccosine function to find the angle θ:

[0053] θ = arccos(0.9275) ≈ 21.2°

[0054] The number of satellites that can be arranged in the satellite orbit is:

[0055] n=360 / (21.2*2)=8.5

[0056] According to the calculation results, the number of satellites that can communicate smoothly without being blocked is 9, which is the number of satellites in one orbit determined by the embodiment of the present invention.

[0057] The parameters of the Walker constellation are determined as follows: the number of orbital planes P is 9, the phase parameter F is 40 degrees, and the total number of satellites T=9*9=81.

[0058] Step 2: Implement weather sensing and calculation to obtain traffic flow information of the specified area within the set time interval.

[0059] The satellite is equipped with a remote sensing payload to obtain high-resolution remote sensing images of the designated area. Based on the distributed computing capability of the constellation, it processes the panchromatic and multispectral images in real time, extracts roads, identifies vehicles, and calculates traffic parameters.

[0060] The step 2 includes the following sub-steps:

[0061] S2.1: Obtain high-resolution satellite remote sensing imagery that provides sufficient detail to identify vehicles and roads.

[0062] S2.2: Perform image preprocessing operations such as radiometric calibration, orthorectification, atmospheric correction and image fusion on the acquired high-resolution satellite remote sensing images in order to improve image quality.

[0063] S2.3: Identify the region of interest;

[0064] S2.4: Extract traffic flow related features, such as vehicle length, distance between vehicles, etc., through image segmentation, feature matching and pattern recognition methods;

[0065] S2.5: According to the characteristics of the vehicle in the image, use methods such as object-oriented classification or support vector machine classification to extract texture features, train the classifier, and realize vehicle detection.

[0066] S2.6: Extract traffic flow parameters such as vehicle density and average speed from the vehicle detection results, and estimate traffic flow parameters based on information such as the distance between vehicles and road capacity. Among them, vehicle density = number of vehicles / road length, average speed represents the distance moved by vehicles per unit time, and average speed = total moving distance / total time. Traffic flow represents the number of vehicles passing through a certain section per unit time, and traffic flow = vehicle density × average speed.

[0067] S2.7: Using the distance between vehicles and the geometric characteristics of the road, calculate the total length of the vehicle queue, queue length = ∑ (vehicle length + distance between vehicles).

[0068] After obtaining the traffic flow information, it will be transmitted to the designated ground station in real time for reference.

[0069] Step 3: Use traffic simulation tools to build a simulation scenario for an urban signal-controlled intersection.

[0070] In this embodiment, the SUMO traffic simulation tool is used to configure lanes, traffic flow, signal timing and other information to build a simulation scene of a city signal-controlled intersection. Figure 2 As shown in the figure, the scene is a signal-controlled intersection with four entrances from the east, south, west and north. Each entrance has six lanes. The first two lanes from the left of each entrance are left-turn lanes, and the third lane from the left is a variable lane, that is, the lane is set to turn left or go straight; the first lane from the right is a right-turn lane, and the second and third lanes from the right are through lanes. The intersection is equipped with an inductive traffic light with a typical four-stage cycle. The green light time of each stage can be changed in real time according to the dynamic traffic flow.

[0071] The specific parameters of the signal-controlled intersection are: lane width: 3.2 meters; lane length: 200 meters; speed limit: 50 km / h; vehicle type: passenger car; vehicle length: 5 meters; vehicle spacing: 2.5 meters; maximum speed: 50 km / h.

[0072] The SUMO traffic simulation tool is configured on a personal computer equipped with an Intel(R) Core(TM) i7-12700 2.10 GHz CPU, an Nvidia RTX 3090 GPU, and 64GB of memory.

[0073] When simulating a signalized intersection on the SUMO traffic simulation tool, you need to write the following separately:

[0074] (1) Road network description file, including information such as roads, intersections, and lanes;

[0075] (2) Traffic demand file .rou.xml, including vehicle type, departure time, route, etc.;

[0076] (3) Signal control file.add.xml;

[0077] Then, motion simulation of the signal-controlled intersection is carried out. After the simulation is completed, simulated traffic flow information is obtained.

[0078] Step 4: Build a lane configuration agent based on reinforcement learning, build a lane configuration training dataset based on simulated traffic flow information, and use the deep Q-network algorithm to train the lane configuration agent so that it can adjust the lane configuration according to the long-term traffic flow pattern; build a signal timing agent based on reinforcement learning, build a signal timing training dataset based on simulated traffic flow information, and use the deep Q-network algorithm to train the signal timing agent so that it can adjust the signal timing in real time according to the short-term traffic flow pattern.

[0079] (1) Training of Lane Configuration Agent

[0080] The state of the lane configuration agent at time t is obs LC,t , action is a LC,t , the reward is r LC,t .

[0081] Status (obs LC,t ): Turning traffic is determined by the direction of the entrance lane and the turn. The entrance lane has four directions: east, south, west, and north. Turning is going straight or turning left. Therefore, there are eight possible turning traffic. Right-turning traffic is not considered here, as it is assumed that right turns are never controlled by traffic lights. The state of the lane configuration agent is the average speed of each possible turning traffic in the past ∆T time steps.

[0082] obs LC,t ={sp1:sp2:...:sp n}

[0083] where sp n is the average speed of the nth turning traffic flow, n=8.

[0084] Action (a LC,t ): The action of the Lane Configuration Agent is to configure the lane turning configuration of the variable lanes at the signal-controlled intersection. Each entrance lane can have multiple lane turning configurations, and the variable lane can be set to "straight" or "left turn".

[0085] a LC,t =lane_cf i ∈{lane_cf1 ,...,lane_cf n}

[0086] n=∏x dir , dir∈{west, north, east, south}

[0087] Among them, x dir is the number of lanes configured in the direction of the entrance road. dir indicates the direction of the entrance road; lane_cf i represents the steering configuration of the i-th variable lane. In this embodiment, each entrance lane has two possible lane steering configurations (straight or left turn), so there are a total of 16 (2 4 ) possible lane configurations.

[0088] Reward (r LC,t ): The reward of the lane configuration agent is based on the change of traffic flow at the intersection, that is, at time steps t and The number of vehicles passing through the signal-controlled intersection changes.

[0089] r LC,t =Numarr t' -Numarr t

[0090] Among them, Numarr t is the statistical time interval at time t Number of vehicles passing through the intersection, Numarr t' is Time statistics interval The number of vehicles passing through the intersection.

[0091] (2) Training of the signal timing agent

[0092] A signal timing agent is constructed and trained using a deep Q-network algorithm to enable it to adjust signal timing according to short-term traffic flow patterns. The state of the signal timing agent at time t is obs ST,t , action is a ST,t , the reward is r ST,t .

[0093] Status (obs ST,t ): The state of the signal timing agent is the maximum queue length and number of passing vehicles for each possible turning traffic flow in the past ∆T time steps.

[0094] obs ST,t ={len1:...:len n ;pas1:...:pas n}

[0095] In this example, n=8, where len n is the maximum queue length of the nth turning traffic flow, pas n is the number of vehicles passing through the nth turning traffic flow.

[0096] Action ST,t ): The action of the signal timing agent is to set a possible hyperparameter for the signal timing of the intersection, which is the minimum and maximum possible green light duration of the phase.

[0097] a t ST =light_cf i ∈{light_cf 1 ,...,light_cf q}

[0098] q = ∏y ph , ph∈{phase 1 , phase 2 ,..., phase k}

[0099] Among them, y ph light_cf is the possible minimum green light duration minDur and maximum green light duration maxDur for phase ph. q is the action controlled by the qth signal. In this embodiment, there are four phases (k=4), such as Figure 4 As shown. Based on the experience of traffic engineers, three intervals are set for the green light time of each phase: [15, 25], [25, 35] and [35, 45]. Therefore, q is 81 (3 4 ).

[0100] Reward (r ST,t ): The reward of the signal timing agent is based on the change in the total waiting time at the intersection, that is, at time steps t and The change in the sum of waiting times of vehicles at intersections.

[0101] r ST,t =W tot,t’ -W tot,t

[0102] Among them, W tot,t is in the time interval The total waiting time of all vehicles in the tot,t’ For the time interval The total waiting time of all vehicles in the

[0103] (3) Loss function

[0104] For an agent, given a state s t , action a and time step t, the deep Q network can be expressed as a state-action function Q(s t ,a ;θ), use the minimum loss function to update the parameters θ in the network, the loss function is:

[0105] L = Q(s t ,a ;θ) - (r t + γmax a′ Q(s a ,t+1, a′; θ ′ ))

[0106] This function is the loss function in the deep Q learning algorithm, which is used to train the behavior strategy of the agent in the reinforcement learning task.

[0107] L: loss function, used to measure the difference between the current strategy and the optimal strategy.

[0108] Q(s t ,a ;θ): Under the current strategy, given state s t And the state-action value function (also called Q-value) of action a. This value is calculated by the neural network parameters θ.

[0109] r t : The immediate reward obtained after performing action a at time step t.

[0110] γ: A discount factor that measures the importance of future rewards relative to current rewards. The value of γ is usually between 0 and 1, and the closer the value is to 1, the more important the future rewards are.

[0111] max a′ Q(s a ,t+1, a′; θ ′ ): The state S at the next time step t+1 t+1 The maximum Q value among all possible actions a' under . This value is calculated by the parameters θ ′ of the target network, which is a delayed updated version of the parameters θ of the main network.

[0112] θ and θ ′ : Parameters of the main network and the target network, respectively. The target network’s parameters are updated less frequently than the main network, which helps stabilize the training process.

[0113] The loss function L calculates the difference between the Q value predicted by the current policy and the actual observed Q value (that is, the reward plus the maximum Q value of the next state). By minimizing this loss function, the network parameters θ can be updated so that the agent learns a better policy.

[0114] (4) Training process, such as Figure 3 shown.

[0115] (4.1) Initialization: Randomly initialize two deep Q networks, where deep Q learning is used as the lane configuration agent and signal timing agent respectively.

[0116] (4.2) Environment reset: At the beginning of each training cycle, the simulation environment is reset.

[0117] (4.3) Lane configuration agent: In each The time step obtains the current state, which reflects the long-term traffic flow pattern. Based on the current state, the lane configuration agent configures the lane direction.

[0118] (4.4) Signal Timing Agent: Based on the actions of the Lane Configuration Agent, the Signal Timing Agent The time step obtains the current state, which reflects the short-term traffic flow pattern. Based on the current state, the signal timing agent adjusts the signal timing parameters.

[0119] (4.5) Reward acquisition and storage: The lane configuration agent The reward is obtained at the end of the time step and stored in the replay buffer 1 along with the state and action. The reward is obtained at the end of the time step and stored in the replay buffer 2 along with the state and action.

[0120] (4.6) Deep Q network update: Mini-batch samples are randomly drawn from the buffer periodically to update the deep Q network parameters of the lane configuration agent and the signal timing agent.

[0121] (7) Loop iteration: Repeat steps (3) to (6) until the maximum training cycle or maximum time is reached.

[0122] Step 5: Input the real-time traffic flow information of the designated area obtained in step 2 into the trained lane configuration agent and signal timing agent. The trained lane configuration agent will periodically evaluate the status of the lane configuration and decide whether the lane direction needs to be adjusted. The trained signal timing agent will monitor the traffic flow parameters in real time and adjust the signal timing parameters to optimize the traffic flow, thereby achieving the optimization of urban traffic spatiotemporal resources. Figure 5 shown.

[0123] In order to facilitate the display of simulation and evaluate the effectiveness of the urban traffic spatiotemporal resource optimization method and device based on astronomical calculation and reinforcement learning under different traffic pressures, this embodiment performs a 24-hour simulation, covering a full day in real life. The simulation includes multiple time periods, such as morning and evening rush hours, during which the traffic volume increases. In these time periods, several stages are established according to different traffic needs. In addition, different ratios of vehicles turning left and going straight are set. and They are set to 1 hour and 15 minutes respectively, as shown in Table 1.

[0124] Table 1 Traffic flow configuration of simulation example

[0125]

[0126] The performance of the present invention in optimizing traffic spatiotemporal resources is evaluated by comparing it with the following three other models:

[0127] Baseline model: the original signalized intersection without any reinforcement learning optimization. The minimum green time and maximum green time (i.e. [minDur, maxDur]) of each phase are preset to a medium level (i.e. [25, 35]), and no variable lanes are set.

[0128] Model based on lane configuration agent: The lane configuration agent is used to adjust the traffic flow direction of the variable lanes, but the signal timing is not optimized. The minimum green light time and maximum green light time (i.e. [minDur, maxDur]) of each phase are preset to a medium level (i.e. [25, 35]).

[0129] Signal timing agent-based model: The signal timing agent is used to adjust signal hyperparameters, but no variable lane configuration is performed.

[0130] The performance indicators include average vehicle speed and average delay time, which is the average speed of all vehicles passing through the signal-controlled intersection and the average delay time when stopping at the intersection.

[0131] During the training process, the above two performance indicators of the present invention tend to converge from the beginning of training and stabilize after about 100 training cycles, which proves the effectiveness of the training strategy for optimizing the allocation of traffic spatiotemporal resources. The average driving speed and average delay time obtained by the above three models and the method of the present invention are shown in Table 2.

[0132] Table 2 Average driving speed and average delay time obtained by four different models

[0133]

[0134] The base model performed the worst due to its fixed lane assignments and fixed green time, which were unable to direct traffic according to the changing turn ratios in the simulation scenario. The performance of the lane configuration agent-based model showed that optimizing signal timing hyperparameters can significantly improve the performance of the intersection. Moreover, implementing variable lane configurations in addition to the signal timing agent further improved traffic efficiency.

[0135] Corresponding to the aforementioned embodiment of the urban traffic spatiotemporal resource optimization method based on space computing, the present invention also provides an embodiment of the urban traffic spatiotemporal resource optimization device based on space computing.

[0136] The urban traffic spatiotemporal resource optimization device based on space computing provided in an embodiment of the present invention includes one or more processors for implementing the urban traffic spatiotemporal resource optimization method based on space computing in the above-mentioned embodiment.

[0137] The embodiment of the urban traffic spatiotemporal resource optimization device of space computing of the present invention can be applied to any device with data processing capability, and the any device with data processing capability can be a device or apparatus such as a computer. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capability in which it is located to read the corresponding computer program instructions in the non-volatile memory into the memory for execution. From the hardware level, in addition to the processor, memory, network interface, and non-volatile memory, any device with data processing capability in which the device in the embodiment is located can also include other hardware according to the actual function of the device with data processing capability, which will not be described in detail.

[0138] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0139] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.

[0140] An embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the urban traffic spatiotemporal resource optimization method based on space computing in the above embodiment is implemented.

[0141] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device, such as a plug-in hard disk, a smart memory card (SmartMedia card, SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of any device with data processing capability and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0142] Those skilled in the art can understand that the above are only preferred examples of the invention and are not intended to limit the invention. Although the invention is described in detail with reference to the above examples, those skilled in the art can still modify the technical solutions recorded in the above examples or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, etc. made within the spirit and principle of the invention shall be included in the protection scope of the invention.

Claims

1. A method for optimizing urban traffic spatiotemporal resources based on space computing, characterized in that: The method comprises the following steps: Step 1: Design a space constellation to obtain satellite remote sensing images of cities in a designated area; Step 2: Implement weather sensing and calculation to obtain real-time traffic flow information of a specified area within a set time interval; Step 3: Use a traffic simulation tool to construct a signal-controlled intersection, define a road network description file, a traffic demand file, and a signal control plan, simulate the signal-controlled intersection, and obtain simulated traffic flow information; Step 4: construct a lane configuration agent based on reinforcement learning, construct a lane configuration training data set based on the simulated traffic flow information, and use a deep Q-network algorithm to train the lane configuration agent so that it can regularly adjust the lane configuration according to the long-term traffic flow pattern; construct a signal timing agent based on reinforcement learning, construct a signal timing training data set based on the simulated traffic flow information, and use a deep Q-network algorithm to train the signal timing agent so that it can adjust the signal timing in real time according to the short-term traffic flow pattern; The state of the lane configuration agent is the past The average speed of each possible turning traffic flow in the time step, the action is the lane turning configuration of the variable lane at the signal-controlled intersection, and the reward is the ... a change in the number of vehicles passing through the signal-controlled intersection within a time step; The state of the signal timing agent is the past The maximum queue length and number of vehicles passing each possible turning traffic flow in the time step, the action is to set a possible hyperparameter for the signal timing of the signal-controlled intersection, which is the minimum and maximum possible green light duration of the phase; the reward is The change in the sum of the total waiting time of all vehicles at a signalized intersection within a time step; Step 5: input the real-time traffic flow information of the designated area obtained in step 2 into the trained lane configuration agent and signal timing agent, and the trained lane configuration agent periodically evaluates the state of the lane configuration and determines whether the lane direction needs to be adjusted; The trained signal timing agent monitors traffic flow parameters in real time and adjusts signal timing parameters to optimize traffic flow, thereby achieving urban traffic spatiotemporal resource optimization.

2. The urban traffic spatiotemporal resource optimization method based on space computing according to claim 1 is characterized in that: Designing the space constellation in step 1 specifically includes determining the number of satellites, orbital altitude, orbital inclination and satellite distribution pattern.

3. The urban traffic spatiotemporal resource optimization method based on space computing according to claim 1 is characterized in that: The step 2 comprises the following steps: S2.1: Obtain high-resolution satellite remote sensing images sufficient to identify vehicles and roads; S2.2: Perform image preprocessing operations such as radiometric calibration, orthorectification, atmospheric correction and image fusion on the acquired high-resolution satellite remote sensing images to improve image quality; S2.3: Identify the region of interest; S2.4: Extract traffic flow related features through image segmentation, feature matching and pattern recognition; S2.5: According to the characteristics of vehicles in satellite remote sensing images, object-oriented classification or support vector machine classification methods are used to extract texture features, train classifiers, and realize vehicle detection; S2.6: Extract traffic flow parameters from vehicle detection results.

4. The urban traffic spatiotemporal resource optimization method based on space computing according to claim 1 is characterized in that: The signal-controlled intersection has four entrances, namely, east, south, west and north, each entrance has six lanes, the first two lanes from the left at each entrance are left-turn lanes, the third lane from the left is a variable lane, that is, the lane is set to turn left or go straight; the first lane from the right is a right-turn lane, and the second and third lanes from the right are through lanes.

5. The urban traffic spatiotemporal resource optimization method based on space computing according to claim 1 is characterized in that: The loss function during the training of the lane configuration agent and the signal timing agent is the difference between the Q value predicted by the current strategy and the actually observed Q value.

6. A system for optimizing urban traffic spatiotemporal resources based on space computing, characterized in that: The system includes one or more processors for implementing the urban traffic spatiotemporal resource optimization method based on space computing as described in any one of claims 1 to 5.

7. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs. When the one or more programs are executed by the electronic device, the electronic device implements the urban traffic spatiotemporal resource optimization method based on space computing as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: A program is stored thereon, and when the program is executed by a processor, the urban traffic spatiotemporal resource optimization method based on space computing as described in any one of claims 1 to 5 is implemented.

Images