Road grading speed limiting method and device based on deep learning and storage medium

Abstract

The invention discloses a road grading speed limiting method based on deep learning, and the method comprises the steps: obtaining expressway flow parameters, inputting the expressway flow parameters into a strategy model, and enabling the strategy model to output a grading speed limiting strategy based on the expressway flow parameters; inputting the grading speed limit strategy into an evaluation model, and outputting grading speed limit evaluation indexes by the evaluation model; and according to the evaluation index and based on a graded speed limiting strategy, adjusting a speed limiting indication sign arranged beside the expressway. According to the technical scheme provided by the invention, the traffic smoothness can be effectively improved, and the expressway graded speed limiting method based on deep learning has the advantages of high reliability, low cost and high expandability.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic safety facilities, and in particular to a road classification speed limit method, device and storage medium based on deep learning. Background Art

[0002] Highways provide good conditions for the smooth driving of vehicles with their high efficiency and speed, but the frequency of traffic accidents is also increasing. Studies have shown that 50% of the non-recurring congestion on highways is caused by traffic accidents. In order to solve the above problems, in recent years, many scholars have proposed to use variable speed limits (VSL) to publish reasonable speed limits through roadside information boards, intervene in the traffic flow in the accident bottleneck area, and effectively improve the traffic efficiency. This has become one of the current mainstream methods. Existing studies have explored the accuracy of speed limits by constructing various mathematical models, but lack comprehensive consideration of the location of graded speed limits, speed limit reduction, information board spacing, and buffer distance from the accident bottleneck area. Summary of the invention

[0003] The main purpose of the present invention is to provide a road classification speed limit method, device and storage medium based on deep learning, aiming to solve the above technical problems.

[0004] To achieve the above objectives, the present invention provides a road classification speed limit method based on deep learning.

[0005] The road classification speed limit method based on deep learning comprises the following steps: Obtaining highway traffic parameters, and inputting the highway traffic parameters into a strategy model, wherein the strategy model outputs a hierarchical speed limit strategy based on the highway traffic parameters; Inputting the graded speed limit strategy into an evaluation model, the evaluation model outputting an evaluation index of the graded speed limit; The speed limit signs set on the side of the expressway are adjusted according to the evaluation index and based on the graded speed limit strategy.

[0006] In one embodiment, the step of inputting the highway traffic parameters into a strategy model, wherein the strategy model outputs a graded speed limit strategy based on the highway traffic parameters comprises: The dimension of the highway traffic parameters is transformed to 64 dimensions through 1D convolution, then transformed to 128 dimensions for dimensionality upgrading, linearly transformed in 128 dimensions, then mapped to 64 dimensions, and finally mapped to the output variable dimension; The input data and output data are fused and spliced ​​as the input of the evaluation model.

[0007] In one embodiment, the formula for the step of transforming the dimension of the highway traffic parameter to 64 dimensions through 1D convolution, then transforming it to 128 dimensions for dimensionality upgrading, performing linear transformation in 128 dimensions, then mapping it to 64 dimensions, and finally mapping it to the output variable dimension is: ; in, represents the output result of the i-th level, Style The convolutional layer of level 1 is level results.

[0008] In one embodiment, the step of inputting the graded speed limit strategy into an evaluation model, wherein the evaluation model outputs an evaluation index of the graded speed limit comprises: Initialize hidden state and cell status , and input the hierarchical speed limit strategy data Pass to LSTM layer; According to the input data and the hidden state of the previous moment, calculate the input gate , Forget Gate and output gate The value of Compute candidate cell states: ,in, To calculate candidate cell states of , Weight matrix and bias term; Update cell status: ,in, is the updated cell state; Update hidden state: Update the hidden state based on the output gate and the updated cell state ,in, is the updated hidden state; Output the hidden state of the last time step .

[0009] In one embodiment, the input gate is calculated based on the input data and the hidden state at the previous moment. , Forget Gate and output gate The value of is calculated as: ; ; ; in, For input gate , The weight matrix and bias term of ; : Forget Gate , The weight matrix and bias term of ; Output gate , The weight matrix and bias term of .

[0010] In one embodiment, the calculation formula for calculating the candidate cell state is: ; in, To calculate candidate cell states of , Weight matrix and bias term.

[0011] In addition, to achieve the above-mentioned purpose, the present invention also provides a road classification speed limit method based on deep learning, and the road classification speed limit method based on deep learning includes: a memory, a processor, and a road classification speed limit program stored in the memory and executable on the processor, and when the road classification speed limit program is executed by the processor, the steps of the road classification speed limit method as described above are implemented.

[0012] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which a road classification speed limit program is stored, and when the road classification speed limit program is executed by a processor, the steps of the road classification speed limit method as described above are implemented. The beneficial effects that can be achieved by the present invention are as follows: a road graded speed limit method based on deep learning proposed in an embodiment of the present invention obtains highway flow parameters and inputs the highway flow parameters into a strategy model, and the strategy model can quickly output a graded speed limit strategy; and outputs evaluation indicators for the graded speed limit strategy through an evaluation model. After the evaluation indicators meet the requirements, the speed limit signs set on the side of the highway can be adjusted according to the graded speed limit strategy, which can effectively improve traffic flow. The highway graded speed limit method based on deep learning in this application has the advantages of high reliability, low cost and strong scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of the structure of a device in a hardware operating environment involved in an embodiment of the present invention; Figure 2 A schematic diagram showing a comparison of average formation time according to an embodiment of the present invention; Figure 3 This is a schematic diagram of average delay time comparison in an embodiment of the present invention; Figure 4This is a schematic diagram of average speed comparison in an embodiment of the present invention; Figure 5 Schematic diagram of comparison of standard deviation of average travel time according to an embodiment of the present invention; Figure 6 Schematic diagram of comparison of average speed standard deviations according to an embodiment of the present invention.

[0014] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0015] 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.

[0016] The specific embodiments of the road classification speed limiting device of the present invention are basically the same as the embodiments of the road classification speed limiting method described below, and will not be described in detail here.

[0017] Reference Figure 1 The first embodiment of the present invention provides a road classification speed limit method based on deep learning, and the road classification speed limit method based on deep learning includes: Obtaining highway traffic parameters, and inputting the highway traffic parameters into a strategy model, wherein the strategy model outputs a hierarchical speed limit strategy based on the highway traffic parameters; Inputting the graded speed limit strategy into an evaluation model, the evaluation model outputting an evaluation index of the graded speed limit; The speed limit signs set on the side of the expressway are adjusted according to the evaluation index and based on the graded speed limit strategy.

[0018] The construction of the hierarchical speed limit method model depends on two submodules. First, the policy model PolicyNet and the evaluation model SumoNet need to be built. PolicyNet mainly uses the convolution layer in the CNN model to slide a fixed-size window on the input data to perform convolution operations, extract features within the window, and then map these features to the output layer to output the hierarchical speed limit policy.

[0019] SumoNet takes the hierarchical speed limit policy output by PolicyNet as input to simulate SUMO operation. It uses the LSTM model to capture the long-term dependencies of time series data in the SUMO simulation process, outputs evaluation indicators of the hierarchical speed limit method, and verifies the effect of the hierarchical speed limit method.

[0020] For input First, a policy is generated through PolicyNet, so that this output Represents the corresponding strategy. Next, we use the fusion module DFM (Data Fusion Module), which is a data fusion module that outputs the fused data. , and finally a SumoNet module is used to calculate the final input result , the whole process is shown as follows: ; ; ; The purpose is to Minimum, find the best solution .

[0021] The present application obtains highway traffic parameters and inputs the highway traffic parameters into a strategy model, and the strategy model can quickly output a graded speed limit strategy; and outputs evaluation indicators for the graded speed limit strategy through an evaluation model. After the evaluation indicators meet the requirements, the speed limit signs set on the side of the highway can be adjusted according to the graded speed limit strategy, which can effectively improve traffic flow. The graded speed limit method for highways based on deep learning in the present application has the advantages of high reliability, low cost, and strong scalability.

[0022] Strategy model construction and work: In order to improve the generalization ability of the model, convolutional layers are used to construct the PolicyNet module. Compared with fully connected layers, the use of convolutional layers can also greatly reduce the number of model parameters.

[0023] PolicyNet first increases the dimension of the input variable. Since the traffic operation status characterization data (flow, density, speed) is generally one-dimensional sequence data, a 1D convolution is used to transform the input variable dimension to 64 dimensions, and then to 128 dimensions for dimensionality increase. A linear transformation is performed in the 128 dimensions, and then mapped to 64 dimensions, and finally mapped to the output variable dimension. In PolicyNet, a data fusion module DFM (Data Fusion Module) is designed to fuse and splice the input data and output data using channel splicing as the input parameter of the next level model. The above process is shown in the following formula:

[0024] in represents the output result of the i-th level, Style The convolutional layer of level 1 is level results.

[0025] Evaluation model construction and work: SUMO simulation of traffic operation status is a nonlinear process, and the model gradient cannot be directly calculated for back propagation, but it changes over time and has time series properties.

[0026] Therefore, LSTM is used to construct the SumoNet module as the proxy model of SUMO. The input of this module is the hierarchical speed limit policy given by the PolicyNet module, the output is the hierarchical speed limit effect evaluation index, and the loss function is designed as the mean square error of the traffic operation status evaluation index.

[0027] The calculation process of the LSTM layer mainly involves input data, hidden state, cell state and LSTM weight parameters. The calculation steps are as follows: (1) Initialize hidden state and cell status , input the hierarchical speed limit strategy data Passed to the LSTM layer. The hidden state dimension, used to store the information learned by the model in the sequence; cell state The dimension is the same as the hidden state, which is used to store long-term memory in the sequence. is the input data of time step t, b represents batch_size, s represents the time step, and f represents the number of features.

[0028] (2) Calculate gating information: The weight parameters of the LSTM layer include the weight matrix of the input gate, forget gate, output gate, and the corresponding bias term. The input gate is calculated based on the input data and the hidden state of the previous moment. , Forget Gate and output gate The value of .

[0029] ; ; ; in, For input gate , The weight matrix and bias term of ; : Forget Gate , The weight matrix and bias term of ; Output gate , The weight matrix and bias term of .

[0030] (3) Calculate candidate cell states:

[0031] in, To calculate candidate cell states of , Weight matrix and bias term.

[0032] (4) Update cell status: ,in, is the updated cell state, (5) Update hidden state: Update the hidden state based on the output gate and the updated cell state ,in, is the updated hidden state.

[0033] (6) Output: The output of the LSTM layer is the hidden state of the last time step .

[0034] Simulation Example 1) Simulation scene construction This application is based on the open source simulation software SUMO to build an active induction scenario of graded speed limit on the main line of the highway under traffic accident conditions, and dynamically interacts with Python through the built-in Traci module of the software to verify the effectiveness of the graded speed limit.

[0035] 2) Construction of simulated road network This paper takes a section of Shanghai-Hangzhou-Ningbo Expressway as an example to build a 9.2-kilometer-long, one-way, four-lane expressway section with a default speed limit of 33.33 m / s. 36 detector points are deployed with an average spacing of 272 m. Each detector point is equipped with an E1 detector on each lane to monitor the vehicle running status.

[0036] 3) Simulation parameter settings The lane-changing model in the simulation process adopts the LC2013 model, the car-following model adopts the Krauss model, and the vehicle types are set to three categories: cars, trucks, and buses. The settings of vehicle driving behavior and simulation environment related parameters based on the traffic flow data survey on actual roads are shown in Table 1.

[0037] parameter Value Car acceleration / (m·s-2) 2.6 Bus acceleration / (m·s-2) 2 Truck acceleration / (m·s-2) 1.7 Car deceleration / (m·s-2) 4.5 Bus deceleration / m·s-2) 3.8 Truck deceleration / (m·s-2) 3 Driver defects 0.5 Maximum speed of car / (m·s-1) 40 Maximum speed of bus / (m·s-1) 36.67 Maximum speed of truck / (m·s-1) 33.33 Minimum parking distance / m 1 Expected minimum headway / s 1 Detector output frequency / s 60 Simulation time / s 3600 Table 1 Simulation parameter settings 4) Construction of accident simulation scenarios The accident point is set at the transition section of the exit ramp. The vehicle.slowDown command is used to set the accident vehicle speed to 0 when the simulation runs for 10 minutes. The accident is simulated as a side collision between two vehicles, which occupies the second and third lanes. When the simulation runs for 40 minutes, the accident vehicle is moved, the accident is eliminated, and the simulation is waited for to end.

[0038] 5) Model training First, SumoNet is trained for 200 rounds. The Adam optimizer is used, and the initial learning rate is 0.001. The model input parameters are buffer length, interval distance, speed limit, road flow, and road speed. All speeds within the speed limit range are randomly selected in a traversal manner, and each combination of buffer distance, buffer length, traffic flow, and speed is traversed. The average travel time, average delay time, average speed, average travel time standard deviation, and average speed standard deviation are used as evaluation indicators. Five evaluation indicators are calculated using SUMO simulation and the loss function is calculated. SumoNet is used to fit the calculation results of SUMO simulation. The loss function sumo_loss is designed as follows:

[0039] in and They represent the actual average travel time series and the travel time series output by SumoNet, and represent the actual delay time series and the predicted delay time series respectively, and are the actual vehicle travel time standard deviation series and the predicted vehicle travel time standard deviation series, They are the actual vehicle average speed standard deviation series and the predicted average speed standard deviation series respectively.

[0040] Secondly, we train PolicyNet. The input of this part is the road flow and road speed, and the output is the buffer distance, buffer length, and graded speed limit. The training rounds are 300 times, and the optimization goal is to maximize the average speed and minimize the other evaluation indicators. The loss function Policy_loss is shown in Formula 6. The Adam optimizer is used, and the initial learning rate is 0.001.

[0041]

[0042] in represents the average travel time, Indicates the delay time. The standard deviation of vehicle travel time, Indicates average speed.

[0043] The entire running environment is Window 11 operating system, the CPU is i9-14900k, the memory is 64G, the GPU is RTX4090, the python version is 3.10, the pytorch version is 2.3.1, and the Sumo version is 1.19.0.

[0044] 6) Result analysis The input of this simulation verification scheme is road flow and road speed, and the output is a hierarchical speed limit strategy. The hierarchical speed limit strategy is put into the SUMO system for simulation calculation to obtain the simulation results, and the simulation results are compared with those without hierarchical speed limit. The input flow ranges from 30 veh / min to 115 veh / min, with an interval of 5, and the speed ranges from 15 m / s to 29 m / s, with an interval of 1.

[0045] (1) Comparison of average travel time Please refer to Figure 2 , by comparing whether the graded speed limit strategy is adopted under different road flow and road speed conditions, it can be seen that the proposed graded speed limit strategy can significantly reduce the average vehicle travel time. When the road flow exceeds 60 veh / min, the average travel time decreases significantly, especially when the road flow is 90-95 veh / min and the road speed is 27-29 m / s, the average travel time decreases by 3.72%. When the flow upstream of the accident is 65 veh / min and the speed is 28 m / s, the output optimal speed limit values ​​are 20 m / s, 18 m / s, 15 m / s, and 13 m / s, respectively, the interval distance is 200m, and the buffer length is 750m. At this time, the average travel time decreases by 6.78%.

[0046] (2) Comparison of average delay time Please refer to Figure 3 From the average delay time comparison chart, it can be seen that when the traffic volume is between 35-60 veh / min, regardless of the speed, imposing a graded speed limit will increase the vehicle delay time. This is because the traffic volume is small and the vehicle travels relatively freely. Imposing a graded speed limit will affect the normal driving of the vehicle. When the traffic volume is between 90-115 veh / min, the effect of the graded speed limit is not obvious. This is because the traffic volume is large and the vehicle driving is relatively stable. The graded speed limit will not have a significant impact on vehicle driving. When the traffic volume is between 60-90 veh / min, the vehicle driving is unstable and prone to sudden stops and brakes. The graded speed limit can effectively avoid this situation. The higher the speed, the greater the reduction in the average delay time, and the better the effect of the graded speed limit.

[0047] (3) Average speed comparison Please refer to Figure 4 , under different traffic and speed conditions, Figure 4It can be seen that the overall average speed of vehicles under the graded speed limit control is better than that of the no speed limit strategy, and the average speed is increased by 4.5%, which is a significant effect. When the flow rate is less than 60veh / min and the speed is less than 23m / s, the effect of the graded speed limit is not obvious, and even a negative effect occurs. With the increase of vehicle speed and flow rate, when the flow rate is greater than 90veh / min and the speed is greater than 25m / s, the graded speed limit has the best effect, and the average speed is increased by 8.09%. When the flow rate upstream of the accident point is 105veh / min and the speed is 28m / s, the output optimal speed limit values ​​are 27m / s, 23m / s, 17m / s, and 12m / s, the interval distance is 200m, and the buffer length is 1000m. At this time, the average travel time is reduced by 12.64%.

[0048] (4) Comparison of standard deviation of average travel time Please refer to Figure 5 By comparing the standard deviation of travel time before and after the implementation of the graded speed limit, it can be seen that the standard deviation of travel time is closely related to the flow rate. When the flow rate is less than 65 veh / min, the graded speed limit will increase the volatility of vehicle travel time. This is because the smaller the flow rate, the greater the freedom of vehicle driving. The graded speed limit will affect the free driving of the vehicle and increase the volatility. When the flow rate is greater than 65 veh / min, the graded speed limit has little effect on the volatility of vehicle travel time, and there is no obvious pattern.

[0049] (5) Comparison of average speed standard deviation Please refer to Figure 6 By comparing the speed standard deviation before and after the implementation of the graded speed limit, it can be seen that the effect of the graded speed limit is closely related to the flow rate. When the flow rate is less than 70 veh / min, the speed standard deviation decreases significantly with the increase of the flow rate. When the flow rate is greater than 70 veh / min, the effect of the graded speed limit is basically maintained within a certain range. With the increase of the flow rate, the effect of the graded speed limit is not obvious, but it can still reduce the volatility of vehicle speed.

[0050] In general, through the comparative analysis of the above five evaluation indicators, it can be seen that when the flow rate is between 65-95 veh / min and the vehicle speed is between 25-29 m / s, the graded speed limit effect is better. Since different evaluation indicators corresponding to flow rate and vehicle speed may not all be effective, the following table gives the graded speed limit strategy when the five evaluation indicators are effective at the same time: flow Speed Buffer length Separation distance Level 1 speed limit Level 2 speed limit 3-level speed limit Level speed limit 70 25 750 200 21 19 15 12 75 25 750 200 22 20 16 12 80 25 750 200 23 20 16 12 90 25 750 200 24 22 16 12 95 25 750 200 25 22 16 12 65 26 750 200 20 18 15 12 70 26 750 200 21 19 16 12 90 26 750 200 24 22 16 12 95 26 750 200 25 22 16 12 80 27 750 200 23 20 16 12 95 27 750 200 25 22 16 12 65 28 750 200 20 18 15 13 70 28 750 200 21 19 16 13 80 28 750 200 23 20 16 12 85 28 750 200 23 21 16 12 90 28 750 200 24 22 16 12 65 29 750 200 20 18 16 13 75 29 750 200 22 20 16 13 90 29 750 200 24 21 16 12 95 29 750 200 25 22 16 12 Table 3 Example of hierarchical speed limit strategy Through comparative experiments, it was found that the highway graded speed limit model based on deep learning provided by this application has the advantages of high reliability, low cost, and strong scalability, which can effectively improve traffic smoothness.

[0051] The specific embodiments of the computer-readable storage medium of the present invention are basically the same as the above-mentioned embodiments of the road classification speed limit, and will not be described in detail here.

[0052] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0053] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0054] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course 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 is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0055] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A road classification speed limit method based on deep learning, characterized in that: The road classification speed limit method based on deep learning comprises the following steps: Obtaining highway traffic parameters, and inputting the highway traffic parameters into a strategy model, wherein the strategy model outputs a hierarchical speed limit strategy based on the highway traffic parameters; Inputting the graded speed limit strategy into an evaluation model, the evaluation model outputting an evaluation index of the graded speed limit; The speed limit signs set on the side of the expressway are adjusted according to the evaluation index and based on the graded speed limit strategy.

2. The road classification speed limit method based on deep learning according to claim 1 is characterized in that: The step of inputting the highway flow parameters into a strategy model, wherein the strategy model outputs a graded speed limit strategy based on the highway flow parameters comprises: The dimension of the highway traffic parameters is transformed to 64 dimensions through 1D convolution, then transformed to 128 dimensions for dimensionality upgrading, linearly transformed in 128 dimensions, then mapped to 64 dimensions, and finally mapped to the output variable dimension; The input data and output data are fused and spliced ​​as the input of the evaluation model.

3. The road classification speed limit method based on deep learning according to claim 2 is characterized in that: The formula for the steps of transforming the dimension of the highway traffic parameter to 64 dimensions through 1D convolution, then transforming it to 128 dimensions for dimensionality upgrading, performing linear transformation in 128 dimensions, then mapping it to 64 dimensions, and finally mapping it to the output variable dimension is: ； in, represents the output result of the i-th level, Style The convolutional layer of level 1 is level results.

4. The road classification speed limit method based on deep learning according to claim 3 is characterized in that: The step of inputting the graded speed limit strategy into an evaluation model, wherein the evaluation model outputs an evaluation index of the graded speed limit comprises: Initialize hidden state and cell status , and input the hierarchical speed limit strategy data Pass to LSTM layer; According to the input data and the hidden state of the previous moment, calculate the input gate , Forget Gate and output gate The value of Compute candidate cell states: ,in, To calculate candidate cell states of , Weight matrix and bias term; Update cell status: ,in, is the updated cell state; Update hidden state: Update the hidden state based on the output gate and the updated cell state ,in, is the updated hidden state; Output the hidden state of the last time step .

5. The road classification speed limit method based on deep learning according to claim 4 is characterized in that: The input gate is calculated based on the input data and the hidden state of the previous moment. , Forget Gate and output gate The value of is calculated as: ； ； ； in, For input gate , The weight matrix and bias term of ; : Forget Gate , The weight matrix and bias term of ; Output gate , The weight matrix and bias term of .

6. The road classification speed limit method based on deep learning according to claim 4 is characterized in that: The calculation formula for calculating the candidate cell state is: ； in, To calculate candidate cell states of , Weight matrix and bias term.

7. A road classification speed limit device based on deep learning, characterized in that: The road classification speed limit device based on deep learning includes: a memory, a processor, and a road classification speed limit program stored in the memory and executable on the processor. When the road classification speed limit program is executed by the processor, the steps of the road classification speed limit method as described in any one of claims 1 to 6 are implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a road classification speed limit program, and when the road classification speed limit program is executed by the processor, the steps of the road classification speed limit method according to any one of claims 1 to 6 are implemented.

Images