An information processing method and apparatus

Abstract

The present invention discloses an information processing method and device, relating to the technical field of signal control, so as to improve the efficiency of controlling traffic lights at multiple intersections. The method includes: obtaining traffic condition information of at least one intersection in a to-be-controlled area and traffic control information of the at least one intersection; obtaining global traffic state information in the to-be-controlled area; determining adjustment information for traffic control information of a target intersection according to the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information; the target intersection being one intersection or multiple intersections of the at least one intersection; and adjusting traffic control signals of the target intersection according to the adjustment information. Embodiments of the present invention can improve the efficiency of controlling traffic lights at multiple intersections.

Description

Technical Field

[0001] The present invention relates to the technical field of signal control, and particularly to an information processing method and apparatus. Background Art

[0002] The multi-intersection traffic light control schemes in the prior art mainly include the following methods:

[0003] A scheme of independent control for multiple intersections (such as CN110114806A): that is, each intersection acts independently without explicit cooperation and communication. Assuming that each intersection independently performs its own traffic light regulation, the overall effect will also become better;

[0004] A cooperative control scheme considering the association of adjacent intersections (CN102110371A): that is, sharing their respective traffic states at the level of adjacent intersections to achieve simple cooperative control. The specific manifestation forms include hierarchical area regulation and adjacent shared information.

[0005] In the process of implementing the present invention, the inventor found the following technical problems in the prior art:

[0006] The control light models of each intersection are trained separately, with low efficiency and unable to achieve direct scheduling at the global level. Summary of the Invention

[0007] Embodiments of the present invention provide an information processing method and apparatus to improve the efficiency of multi-intersection traffic light control.

[0008] In a first aspect, embodiments of the present invention provide an information processing method, including:

[0009] Obtaining traffic condition information of at least one intersection in a to-be-controlled area and traffic control information of the at least one intersection;

[0010] Obtaining global traffic state information in the to-be-controlled area;

[0011] Determining adjustment information for traffic control information of a target intersection according to the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information; the target intersection is one intersection or multiple intersections of the at least one intersection;

[0012] Adjusting the traffic control signal of the target intersection according to the adjustment information.

[0013] Wherein, before obtaining the traffic condition information of at least one intersection in the to-be-controlled area and the traffic control information of the at least one intersection, the method further includes:

[0014] Generate a single intersection traffic control model;

[0015] The traffic control information of the at least one intersection is obtained based on the single intersection traffic control model.

[0016] Among them, the generation of the single intersection traffic control model includes:

[0017] Collect traffic condition information of at least one intersection;

[0018] Perform state conversion on the obtained traffic condition information of the at least one intersection, where the traffic condition information after state conversion has a traffic state expression in a unified format;

[0019] Divide the traffic conditions after state conversion into at least one subtask;

[0020] Use a deep learning model to jointly train the at least one subtask to obtain the single intersection traffic control model.

[0021] Among them, after obtaining the single intersection traffic control model, the method further includes:

[0022] Iteratively optimize the single intersection traffic control model to obtain an optimized single intersection traffic control model.

[0023] Among them, the iterative optimization of the single intersection traffic control model to obtain an optimized single intersection traffic control model includes:

[0024] Select a target subtask from the at least one subtask;

[0025] Construct a model copy of the single intersection traffic control model and enter the inner loop;

[0026] Run the model copy using the target subtask to obtain a running result;

[0027] Update the model weights of the single intersection traffic control model based on the running result to obtain the result of the inner loop;

[0028] Calculate the quadratic gradient update model based on the result of the inner loop to obtain an optimized single intersection traffic control model.

[0029] Among them, the determination of the adjustment information for the traffic control information of the target intersection according to the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information includes:

[0030] Process the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information by using a fully connected linear network and a convolutional network respectively to obtain a processing result;

[0031] Model the sequence relationship by using a recurrent neural network according to the processing result;

[0032] Determine the target intersection and the adjustment information of the traffic control information for the target intersection based on the hidden state mapping obtained by modeling the sequence relationship by using the recurrent neural network.

[0033] In a second aspect, an embodiment of the present invention provides an information processing device, including:

[0034] A first acquisition module, configured to acquire the traffic condition information of at least one intersection in a to-be-controlled area and the traffic control information of the at least one intersection;

[0035] A second acquisition module, configured to acquire the global traffic state information in the to-be-controlled area;

[0036] A first determination module, configured to determine the adjustment information of the traffic control information for a target intersection according to the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information; the target intersection is one intersection or multiple intersections of the at least one intersection;

[0037] A first adjustment module, configured to adjust the traffic control signal of the target intersection according to the adjustment information.

[0038] Wherein, the device further includes:

[0039] A generation module, configured to generate a single-intersection traffic control model;

[0040] The traffic control information of the at least one intersection is obtained based on the single-intersection traffic control model.

[0041] Wherein, the generation module includes:

[0042] An acquisition sub-module, configured to acquire the traffic condition information of at least one intersection;

[0043] A conversion sub-module, configured to perform state conversion on the obtained traffic condition information of at least one intersection, wherein the traffic condition information after state conversion has a traffic state expression in a unified format;

[0044] A division sub-module, configured to divide the traffic condition after state conversion into at least one sub-task;

[0045] A training sub-module for jointly training the at least one sub-task using a deep learning model to obtain the single intersection traffic control model.

[0046] Wherein, the generation module further includes:

[0047] A processing sub-module for iteratively optimizing the single intersection traffic control model to obtain an optimized single intersection traffic control model.

[0048] Wherein, the processing sub-module includes:

[0049] A selection unit for selecting a target sub-task from the at least one sub-task;

[0050] A construction unit for constructing a model copy of the single intersection traffic control model and entering the inner loop;

[0051] A first acquisition unit for running the model copy using the target sub-task to obtain a running result;

[0052] A second acquisition unit for updating the model weights of the single intersection traffic control model based on the running result to obtain the result of the inner loop;

[0053] A third acquisition unit for calculating a quadratic gradient update model based on the result of the inner loop to obtain an optimized single intersection traffic control model.

[0054] Wherein, the first determination module includes:

[0055] A first processing sub-module for respectively processing the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information using a fully connected linear network and a convolutional network to obtain a processing result;

[0056] A second processing sub-module for modeling the sequence relationship using a recurrent neural network according to the processing result;

[0057] A first determination sub-module for determining the target intersection and the adjustment information of the traffic control information of the target intersection based on the hidden state mapping obtained by modeling the sequence relationship using the recurrent neural network.

[0058] In the embodiment of the present invention, the adjustment information of the traffic control information of the target intersection is determined by the traffic condition information of at least one intersection, the traffic control information of at least one intersection, and the obtained global traffic state information in the area to be controlled, and the traffic control signal of the target intersection is adjusted according to the adjustment information. Since the global traffic state information is considered, the efficiency of controlling traffic lights at multiple intersections can be improved by using the solution of the embodiment of the present invention. Description of the Drawings

[0059] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments of the present invention. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0060] Figure 1 is one of the flowcharts of the information processing method provided by the embodiments of the present invention;

[0061] Figure 2 is the flowchart for generating a single intersection traffic control model provided by the embodiments of the present invention;

[0062] Figure 3 is the second flowchart of the information processing method provided by the embodiments of the present invention;

[0063] Figure 4 is the system block diagram provided by the embodiments of the present invention;

[0064] Figure 5 is the schematic diagram of the single intersection traffic control model provided by the embodiments of the present invention;

[0065] Figure 6 is one of the structural diagrams of the information processing device provided by the embodiments of the present invention;

[0066] Figure 7 is the second structural diagram of the information processing device provided by the embodiments of the present invention;

[0067] Figure 8 is one of the structural diagrams of the generation module provided by the embodiments of the present invention;

[0068] Figure 9 is the second structural diagram of the generation module provided by the embodiments of the present invention;

[0069] Figure 10 is the structural diagram of the processing sub-module provided by the embodiments of the present invention;

[0070] Figure 11 is the structural diagram of the first determination module provided by the embodiments of the present invention;

[0071] Figure 12 is the structural diagram of the information processing device provided by the embodiments of the present invention. Detailed implementation manners

[0072] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0073] See Figure 1 , Figure 1 which is a flowchart of the information processing method provided by the embodiment of the present invention. As Figure 1 shown, it includes the following steps:

[0074] Step 101, obtain the traffic condition information of at least one intersection in the area to be controlled and the traffic control information of the at least one intersection.

[0075] Among them, the area to be controlled may refer to any area that includes at least one intersection. Different intersections have different traffic condition information and traffic control information. Among them, the traffic condition information includes: the number of waiting vehicles at the intersection, the average passing time of vehicles at the intersection, the average number of passing vehicles at the intersection, and the vehicle distribution density at the intersection; the traffic state information may include the real-time state of traffic lights, etc.

[0076] Among them, the traffic control information of the at least one intersection is obtained based on the single-intersection traffic control model in the embodiment of the present invention. Therefore, the embodiment of the present invention may also include generating a single-intersection traffic control model. At each intersection, the traffic control information of the intersection can be obtained by running the single-intersection traffic control model.

[0077] In practical applications, diverse traffic data of multiple intersections can be collected, and the collected data can be converted into a consistent traffic state expression through a traffic state conversion method. Based on the traffic intersection form (such as a three-way intersection, a crossroads, and other forms of intersections) and its traffic conditions (combinations of different traffic flows in different directions at different times), different single-intersection traffic regulation sub-tasks are divided. Through joint learning and iterative optimization on multiple different single-intersection traffic regulation sub-tasks, a single-intersection traffic control model with fast adaptation ability that can handle various traffic conditions is trained. Based on this single-intersection traffic control model, during the use at different intersections, using its specific traffic conditions, the single-intersection traffic control model corresponding to the intersection can be quickly optimized.

[0078] As Figure 2 shown, the steps of generating a single-intersection traffic control model include:

[0079] Step 201, collect the traffic condition information of at least one intersection.

[0080] Among them, the at least one intersection may have different intersection forms, such as crossroads, T-junctions, etc. The traffic condition information may include: the number of waiting vehicles at the intersection, the average passing time of vehicles at the intersection, the average number of passing vehicles at the intersection, the vehicle distribution density at the intersection, the real-time state of traffic lights, etc.

[0081] Step 202: Perform state conversion on the obtained traffic condition information of at least one intersection. Among them, the traffic condition information after state conversion has a traffic state expression in a unified format.

[0082] In the embodiment of the present invention, having a traffic state expression in a unified format means that for intersections with different physical forms (such as three-way intersections, crossroads, etc.), traffic states that were originally inconsistent and could not be directly shared (such as traffic flow matrices with different numbers, and changes in the positional relationships between different directions, etc.) are converted into pairwise combinations between directions that have a competitive relationship. It is achieved by competing and selecting between pairs, so that the model trained thereby has a strong generalization ability. Or, the traffic condition information can also be classified based on different intersection forms, and then data and features are directly shared between intersections of each class.

[0083] Step 203: Divide the traffic condition after state conversion into at least one subtask.

[0084] In the embodiment of the present invention, the dataset of each intersection is regarded as an independent subtask. It is also possible to fuse the data of intersections with similar physical forms and traffic flow states to obtain multiple representative intersection datasets, and use this intersection dataset as the subtask for single-intersection regulation.

[0085] Step 204: Use a deep learning model to jointly train the at least one subtask to obtain the single-intersection traffic control model.

[0086] In the embodiment of the present invention, the deep learning model used is not limited. On the basis of forming multiple subtasks, a unified single-intersection deep learning model, that is, a single-intersection traffic control model, is constructed in the embodiment of the present invention to complete the regulation of the single-intersection traffic state. When designing this model, the network structure is designed as a combination of a multi-layer linear network and a convolutional network. First, the multi-layer linear network receives and processes the underlying traffic flow characteristics (the number of vehicle flows in different lanes in each direction, the current signal light state), and then through pairwise combination, a feature matrix is constructed. The convolutional network further extracts features and outputs the final regulation signal, that is, which direction (lane) should be selected to pass within the current time period.

[0087] To further improve the accuracy of the model, on the basis of the above process, it may further include:

[0088] Step 205: Iteratively optimize the single intersection traffic control model to obtain an optimized single intersection traffic control model.

[0089] Specifically, select a target subtask from the at least one subtask, where the target subtask can be a randomly selected subtask. Then, construct a model copy of the single intersection traffic control model and enter the inner loop, and use the target subtask to run the model copy to obtain a running result. After that, update the model weights of the single intersection traffic control model based on the running result to obtain the result of the inner loop. Finally, calculate the second-order gradient to update the model based on the result of the inner loop to obtain an optimized single intersection traffic control model.

[0090] In practical applications, the core purpose of multi-task joint training is to enable the model to master the ability of fast learning through learning multiple subtasks, so as to quickly iterate a model with excellent performance for a given intersection dataset. Specifically, after initializing the model, randomly sample subtasks from multiple subtasks, construct a model copy and enter the inner loop, normally optimize and update the model weights based on the subtasks, and after completing the subtasks, exit the inner loop, calculate the second-order gradient to update the model based on the inner loop update result, and then continuously resample subtasks and iterate until the performance no longer improves. After obtaining this optimized model, in the inference stage, it can be fine-tuned based on its own dataset at each intersection to achieve fast iteration and reach the optimal result on it, thus realizing generalization and promotion.

[0091] Step 102: Obtain the global traffic state information within the area to be controlled.

[0092] The global traffic state information refers to the traffic state information used to reflect the entire area to be controlled. For example, the average throughput rate of the traffic network, the spatial vehicle density distribution of the traffic network, the average passing speed of each section of the traffic network, etc.

[0093] Step 103: Determine the adjustment information for the traffic control information of the target intersection according to the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information; the target intersection is one or more intersections of the at least one intersection.

[0094] Specifically, in this step, the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information are processed by using a fully connected linear network and a convolutional network respectively to obtain a processing result. Then, according to the processing result, a recurrent neural network is used to model the sequence relationship. Finally, based on the hidden state mapping obtained by modeling the sequence relationship with the recurrent neural network, the target intersection and the adjustment information of the traffic control information for the target intersection are determined.

[0095] Step 104: Adjust the traffic control signal of the target intersection according to the adjustment information.

[0096] In the embodiment of the present invention, the adjustment may include, for example, changing the control duration, status, etc. of the traffic control signal.

[0097] In the embodiment of the present invention, the adjustment information of the traffic control information for the target intersection is determined by the traffic condition information of at least one intersection in the area to be controlled, the traffic control information of at least one intersection, and the obtained global traffic state information, and the traffic control signal of the target intersection is adjusted according to the adjustment information. Since the global traffic state information is considered, the efficiency of controlling traffic lights at multiple intersections can be improved by using the solution of the embodiment of the present invention.

[0098] See Figure 3 , Figure 3 which is a flowchart of the information processing method provided by the embodiment of the present invention. In the embodiment of the present invention, the entire processing flow may include: iterative optimization of the single-intersection regulation model and global traffic control and perception.

[0099] As Figure 3 shown, the iterative optimization of the single-intersection regulation model includes the following steps:

[0100] Step 301: Collect the traffic condition information of at least one intersection.

[0101] Among them, the at least one intersection may have different intersection forms, such as crossroads, T-junctions, etc. The traffic condition information may include: the number of waiting vehicles at the intersection, the average passing time of vehicles at the intersection, the average number of passing vehicles at the intersection, the vehicle distribution density at the intersection, the real-time state of the traffic light, etc.

[0102] Step 302: Perform state conversion on the obtained traffic condition information of at least one intersection.

[0103] In the embodiments of the present invention, for intersections with different physical forms (such as three-way intersections, crossroads, etc.), traffic states that were originally inconsistent and could not be directly shared (such as traffic flow matrices with different numbers, and changes in the positional relationships between different directions, etc.) are converted into pairwise combinations between directions that have competitive relationships with each other. It is achieved by competing and selecting between pairs, so that the model trained thereby has strong generalization and promotion capabilities. Alternatively, traffic condition information can also be classified based on different intersection forms, and then data and features can be directly shared between intersections of each class.

[0104] Step 303: Divide the traffic condition after state conversion into at least one subtask.

[0105] In the embodiments of the present invention, the dataset of each intersection is regarded as an independent subtask, or the data of intersections with similar physical forms and traffic flow states can be fused to obtain multiple representative intersection datasets, and the intersection dataset is used as the subtask for single-intersection regulation.

[0106] Step 304: Jointly train multiple tasks to obtain a single-intersection traffic control model.

[0107] On the basis of forming multiple subtasks, in the embodiments of the present invention, a unified single-intersection deep learning model, that is, a single-intersection traffic control model, is constructed to complete the regulation of the single-intersection traffic state.

[0108] When designing this model, the network structure is designed as a combination of a multi-layer linear network and a convolutional network. First, the multi-layer linear network receives and processes the underlying traffic flow features (the number of vehicles in each lane in each direction, the current signal light state), and then through pairwise combinations, a feature matrix is constructed. The convolutional network further extracts features and outputs the final regulation signal, that is, which direction (lane) should be selected to pass during the current time period.

[0109] Step 305: Adjust the single-intersection traffic control model obtained according to the traffic conditions of each intersection.

[0110] Based on the model obtained in step 304, during the use at different intersections, using its specific traffic conditions, the single-intersection traffic control model corresponding to the intersection is quickly optimized.

[0111] Another example Figure 3 As shown, the global traffic control and perception include the following steps:

[0112] Step 306: Collect global traffic state information.

[0113] The global traffic status information may include, for example, the average throughput rate of the traffic network, the spatial vehicle density distribution of the traffic network, the average passing speed of each section of the traffic network, etc.

[0114] Step 307: Obtain traffic condition information and traffic control information from multiple intersections.

[0115] Step 308: Determine the intersection to be adjusted and the adjustment method.

[0116] Here, based on the information obtained in step 306 and step 307, determine the intersection to be adjusted and the adjustment method for it. For example, it is determined that intersection A needs to be adjusted, and the control duration of its red light is increased.

[0117] In practical applications, steps 306 - 308 can be repeatedly executed to achieve a better traffic state.

[0118] Such as Figure 4 shown, is the system block diagram of an embodiment of the present invention. The traffic information collected at a single intersection includes: the number of waiting vehicles at the intersection, the average passing time of vehicles at the intersection, the average number of passing vehicles at the intersection, the vehicle distribution density at the intersection, the real-time state of traffic lights, etc. The global traffic information collected includes the average throughput rate of the traffic network, the spatial vehicle density distribution of the traffic network, the average passing speed of each section of the traffic network, etc. Among them, the input of a single intersection agent is the number of waiting vehicles and vehicle distribution density in different directions of the intersection, etc., and the output is the traffic signal state (red / green); the input of the global agent is the global state information of the traffic network, and the output is the intersection number that requires upper-layer intervention.

[0119] Such as Figure 5 shown, is the schematic diagram of the single intersection traffic control model in an embodiment of the present invention. Among them, the figure shows the information interaction diagram between the underlying agent and the upper-layer agent, as well as the schematic diagram of the underlying agent network model based on a multi-layer neural network and the schematic diagram of the network model of the upper-layer global traffic perception and regulation unit.

[0120] After the model training is completed at a single intersection according to its own intersection conditions, global multi-intersection collaborative regulation is carried out. First, each single intersection agent reports its own traffic conditions and signal light scheme (A) to the top controller. At the same time, the top controller also collects and reflects the global state information (G), such as the average passing time of vehicles in the network, the average number of passing vehicles in the network, the vehicle distribution density in the network, etc.

[0121] The top - level controller receives the above two types of information, processes and extracts features using a fully - connected linear network and a convolutional network respectively, then uses a recurrent neural network to model the sequence relationship, thereby taking into account the historical information in the network. Finally, based on the obtained hidden state, it maps to the intersection positions that need to be intervened, and adjusts their control signals to optimize the global traffic state.

[0122] An embodiment of the present invention also provides an information processing device. Refer to Figure 6 , Figure 6 which is the structural diagram of the information processing device provided by the embodiment of the present invention. Since the principle of the information processing device for solving problems is similar to that of the information processing method in the embodiment of the present invention, the implementation of this information processing device can refer to the implementation of the method, and the repeated parts will not be elaborated.

[0123] As Figure 6 shown, the information processing device 600 includes:

[0124] A first acquisition module 601, configured to acquire traffic condition information of at least one intersection in the area to be controlled and traffic control information of the at least one intersection; a second acquisition module 602, configured to acquire global traffic state information in the area to be controlled; a first determination module 603, configured to determine adjustment information for the traffic control information of the target intersection according to the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information, where the target intersection is one intersection or multiple intersections among the at least one intersection; a first adjustment module 604, configured to adjust the traffic control signal of the target intersection according to the adjustment information.

[0125] Optionally, as Figure 7 shown, the device may further include: a generation module 605, configured to generate a single - intersection traffic control model, and the traffic control information of the at least one intersection is obtained based on the single - intersection traffic control model.

[0126] Optionally, as Figure 8 shown, the generation module 605 includes:

[0127] An acquisition sub - module 6051, configured to acquire traffic condition information of at least one intersection; a conversion sub - module 6052, configured to perform state conversion on the obtained traffic condition information of the at least one intersection, where the traffic condition information after state conversion has a traffic state expression with a unified format; a division sub - module 6053, configured to divide the traffic condition after state conversion into at least one sub - task; a training sub - module 6054, configured to jointly train the at least one sub - task using a deep - learning model to obtain the single - intersection traffic control model.

[0128] Optionally, asFigure 9 As shown, the generating module 605 may further include: a processing sub-module 6055, configured to iteratively optimize the single-intersection traffic control model to obtain an optimized single-intersection traffic control model.

[0129] Optionally, as Figure 10 shown, the processing sub-module 6055 includes:

[0130] A selection unit 60551, configured to select a target sub-task from the at least one sub-task; a construction unit 60552, configured to construct a model copy of the single-intersection traffic control model and enter the inner loop; a first acquisition unit 60553, configured to run the model copy by using the target sub-task to obtain a running result; a second acquisition unit 60554, configured to update the model weights of the single-intersection traffic control model based on the running result to obtain the result of the inner loop; a third acquisition unit 60555, configured to calculate a second-order gradient to update the model based on the result of the inner loop to obtain an optimized single-intersection traffic control model.

[0131] Optionally, as Figure 11 shown, the first determination module 603 includes:

[0132] A first processing sub-module 6031, configured to process the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information by using a fully connected linear network and a convolutional network respectively to obtain a processing result; a second processing sub-module 6032, configured to model the sequence relationship by using a recurrent neural network according to the processing result; a first determination sub-module 6033, configured to determine the target intersection and the adjustment information of the traffic control information of the target intersection based on the hidden state mapping obtained by modeling the sequence relationship by the recurrent neural network.

[0133] The device provided by the embodiment of the present invention can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be elaborated here in this embodiment.

[0134] As Figure 12 shown, the information processing device of the embodiment of the present invention includes: a processor 1200, configured to read a program in a memory 1220 and execute the following processes:

[0135] Obtain the traffic condition information of at least one intersection in the area to be controlled and the traffic control information of the at least one intersection;

[0136] Obtain the global traffic state information in the area to be controlled;

[0137] Determine adjustment information for the traffic control information of the target intersection according to the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information; the target intersection is one intersection or multiple intersections among the at least one intersection;

[0138] Adjust the traffic control signal of the target intersection according to the adjustment information.

[0139] The transceiver 1210 is configured to receive and send data under the control of the processor 1200.

[0140] Among them, in Figure 12 The bus architecture may include any number of interconnected buses and bridges, specifically various circuits of one or more processors represented by the processor 1200 and the memory represented by the memory 1220 are linked together. The bus architecture can also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art, so they will not be further described herein. The bus interface provides an interface. The transceiver 1210 may be multiple components, that is, including a transmitter and a receiver, and provides a unit for communicating with various other devices on the transmission medium. The processor 1200 is responsible for managing the bus architecture and general processing, and the memory 1220 can store the data used by the processor 1200 when executing operations.

[0141] The processor 1200 is responsible for managing the bus architecture and general processing, and the memory 1220 can store the data used by the processor 1200 when executing operations.

[0142] The processor 1200 is further configured to read the program and execute the following steps:

[0143] Generate a single-intersection traffic control model;

[0144] The traffic control information of the at least one intersection is obtained based on the single-intersection traffic control model.

[0145] The processor 1200 is further configured to read the program and execute the following steps:

[0146] Collect traffic condition information of at least one intersection;

[0147] Perform state conversion on the obtained traffic condition information of at least one intersection, wherein the traffic condition information after state conversion has a traffic state expression in a unified format;

[0148] Divide the traffic condition after state conversion into at least one subtask;

[0149] Using a deep learning model, jointly train the at least one subtask to obtain the single intersection traffic control model.

[0150] The processor 1200 is further configured to read the program and execute the following steps:

[0151] Iteratively optimize the single intersection traffic control model to obtain an optimized single intersection traffic control model.

[0152] The processor 1200 is further configured to read the program and execute the following steps:

[0153] Select a target subtask from the at least one subtask;

[0154] Construct a model copy of the single intersection traffic control model and enter the inner loop;

[0155] Run the model copy using the target subtask to obtain a running result;

[0156] Update the model weights of the single intersection traffic control model based on the running result to obtain the result of the inner loop;

[0157] Calculate the second-order gradient to update the model based on the result of the inner loop to obtain an optimized single intersection traffic control model.

[0158] The processor 1200 is further configured to read the program and execute the following steps:

[0159] Process the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information using a fully connected linear network and a convolutional network respectively to obtain a processing result;

[0160] Model the sequence relationship using a recurrent neural network according to the processing result;

[0161] Determine the target intersection and the adjustment information of the traffic control information for the target intersection based on the hidden state mapping obtained by modeling the sequence relationship using the recurrent neural network.

[0162] The device provided by the embodiments of the present invention can execute the above method embodiments, and its implementation principle and technical effects are similar, which will not be elaborated here in this embodiment.

[0163] An embodiment of the present invention further provides a readable storage medium, on which a program is stored. When the program is executed by a processor, it implements each process of the above information processing method embodiment and can achieve the same technical effect. To avoid repetition, it will not be elaborated here. Among them, the readable storage medium includes, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.

[0164] It should be noted that in this article, the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article, or device. Without more limitations, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article, or device including that element.

[0165] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment method can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases, the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc) and includes several instructions to enable a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0166] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the purpose of the present invention and the scope protected by the claims, and all of them belong to the protection scope of the present invention.

Claims

1. An information processing method, characterized in that, comprising: obtaining traffic condition information of at least one intersection in a to-be-controlled area and traffic control information of the at least one intersection; obtaining global traffic state information in the to-be-controlled area; determining adjustment information for traffic control information of a target intersection according to the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information; the target intersection is one intersection or multiple intersections of the at least one intersection; adjusting traffic control signals of the target intersection according to the adjustment information; wherein, before obtaining traffic condition information of at least one intersection in the to-be-controlled area and traffic control information of the at least one intersection, the method further comprises: generating a single-intersection traffic control model; the traffic control information of the at least one intersection is obtained based on the single-intersection traffic control model; the generating the single-intersection traffic control model includes: collecting traffic condition information of at least one intersection; performing state conversion on the obtained traffic condition information of at least one intersection, wherein the traffic condition information after state conversion has a traffic state expression with a unified format; dividing the traffic condition after state conversion into at least one subtask; using a deep learning model to perform joint training on the at least one subtask to obtain the single-intersection traffic control model; after obtaining the single-intersection traffic control model, the method further comprises: performing iterative optimization on the single-intersection traffic control model to obtain an optimized single-intersection traffic control model; the performing iterative optimization on the single-intersection traffic control model to obtain an optimized single-intersection traffic control model includes: selecting a target subtask from the at least one subtask; constructing a model copy of the single-intersection traffic control model and entering an inner loop; running the model copy with the target subtask to obtain a running result; updating model weights of the single-intersection traffic control model based on the running result to obtain a result of the inner loop; calculating a quadratic gradient update model based on the result of the inner loop to obtain an optimized single-intersection traffic control model; the determining adjustment information for traffic control information of a target intersection according to the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information includes: respectively using a fully-connected linear network and a convolutional network to process the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic state information to obtain a processing result; modeling sequence relationships according to the processing result using a recurrent neural network; determining the target intersection and adjustment information for traffic control information of the target intersection based on a hidden state mapping obtained by modeling sequence relationships using the recurrent neural network.

2. An information processing device, characterized in that, comprising: a first obtaining module, configured to obtain traffic condition information of at least one intersection in a to-be-controlled area and traffic control information of the at least one intersection; A second acquisition module, configured to acquire global traffic status information within the area to be controlled; A first determination module, configured to determine adjustment information for traffic control information of a target intersection according to the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic status information; the target intersection is one intersection or multiple intersections among the at least one intersection; A first adjustment module, configured to adjust the traffic control signal of the target intersection according to the adjustment information; Wherein, the device further includes: A generation module, configured to generate a single-intersection traffic control model; The traffic control information of the at least one intersection is obtained based on the single-intersection traffic control model; The generation module includes: An acquisition sub-module, configured to acquire traffic condition information of at least one intersection; A conversion sub-module, configured to perform state conversion on the acquired traffic condition information of at least one intersection, wherein the traffic condition information after state conversion has a traffic state expression with a unified format; A division sub-module, configured to divide the traffic condition after state conversion into at least one sub-task; A training sub-module, configured to jointly train the at least one sub-task by using a deep learning model to obtain the single-intersection traffic control model; The generation module further includes: A processing sub-module, configured to perform iterative optimization on the single-intersection traffic control model to obtain an optimized single-intersection traffic control model; The processing sub-module includes: A selection unit, configured to select a target sub-task from the at least one sub-task; A construction unit, configured to construct a model copy of the single-intersection traffic control model and enter an inner loop; A first acquisition unit, configured to run the model copy by using the target sub-task to obtain a running result; A second acquisition unit, configured to update the model weight of the single-intersection traffic control model based on the running result to obtain a result of the inner loop; A third acquisition unit, configured to calculate a quadratic gradient update model based on the result of the inner loop to obtain an optimized single-intersection traffic control model; The first determination module includes: A first processing sub-module, configured to respectively process the traffic condition information of the at least one intersection, the traffic control information of the at least one intersection, and the global traffic status information by using a fully connected linear network and a convolutional network to obtain a processing result; A second processing sub-module, configured to model a sequence relationship by using a recurrent neural network according to the processing result; A first determination sub-module, configured to determine the target intersection and the adjustment information for the traffic control information of the target intersection based on a hidden state mapping obtained by modeling the sequence relationship by the recurrent neural network.

Images