CN114333361A - Signal lamp timing method and device - Google Patents
Signal lamp timing method and device Download PDFInfo
- Publication number
- CN114333361A CN114333361A CN202210159646.9A CN202210159646A CN114333361A CN 114333361 A CN114333361 A CN 114333361A CN 202210159646 A CN202210159646 A CN 202210159646A CN 114333361 A CN114333361 A CN 114333361A
- Authority
- CN
- China
- Prior art keywords
- traffic flow
- intersection
- flow data
- time period
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Traffic Control Systems (AREA)
Abstract
The invention discloses a signal lamp timing method and a signal lamp timing device, which respectively obtain traffic flow data of a plurality of intersections in a target road network in a historical time period corresponding to a target prediction time period, input the traffic flow data into a graph volume network model for predicting the traffic flow data, obtain the traffic flow data of each intersection in the target prediction time period output by the graph volume network model, and obtain the traffic flow data of the target intersection in the target prediction time period; constructing a signal lamp timing model based on traffic flow data of the target intersection in a target prediction time period; and determining the optimal signal lamp timing information of the target intersection in the target prediction time period by using the signal lamp timing model. The invention can automatically determine and adjust the signal lamp timing according to the traffic flow data, improves the timing efficiency and timing accuracy of the signal lamp, does not need to observe the traffic condition of the intersection through manpower to perform manual signal lamp timing, avoids the consumption of manpower resources and reduces the timing cost of the signal lamp.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a signal lamp timing method and a signal lamp timing device.
Background
With the high-speed development of national economy and the acceleration of urbanization process, urban traffic becomes a bottleneck restricting the sustainable development of urban economy and society. The urban traffic control technology is an important way to effectively solve the urban traffic problem and improve the urban traffic operation efficiency.
In urban traffic control technology, signal lamp timing is an important task. The signal lamp timing can comprise determining the red lamp timing time length, the green lamp timing time length and the like at the intersection.
At present, signal lamp timing is carried out by observing the traffic condition of the intersection through naked eyes by working personnel based on personal experience and knowledge.
However, there are many intersections in the city and there are situations such as traffic conditions changing quickly, and the way of manually performing signal lamp timing needs to consume many human resources, and the efficiency is low.
Disclosure of Invention
In view of the above problems, the present invention provides a method and an apparatus for timing a signal lamp, which overcome the above problems or at least partially solve the above problems, and the technical solution is as follows:
a signal lamp timing method, comprising:
respectively obtaining traffic flow data of a plurality of intersections in a target road network in a historical time period corresponding to a target prediction time period;
inputting the traffic flow data of each intersection in the historical time period into a trained graph volume network model for predicting the traffic flow data to obtain the traffic flow data of each intersection in a target prediction time period, which is output by the graph volume network model;
obtaining traffic flow data of a target intersection in the target prediction time period from the traffic flow data of each intersection in the target prediction time period;
constructing a signal lamp timing model based on traffic flow data of the target intersection in the target prediction time period;
determining the optimal signal lamp timing information matched with the traffic flow data of the target intersection in the target prediction time period by utilizing the signal lamp timing model; the optimal signal timing information at least comprises: an optimal green light configuration duration and/or an optimal signal light period.
Optionally, after obtaining the traffic flow data of each intersection in the historical time period, the graph convolution network model calculates the traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters, and the graph matrix and the adjacency matrix corresponding to the road network information of the target road network, and obtains and outputs the traffic flow data of each intersection in the target prediction time period.
Optionally, an activation function is set in the graph convolution network model; the calculating of the traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters, and the graph matrix and the adjacency matrix corresponding to the road network information of the target road network comprises:
and the graph convolution network model inputs traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters and a graph matrix and an adjacent matrix corresponding to the road network information of the target road network into the activation function for calculation.
Optionally, each intersection is provided with a flow sensor for monitoring traffic flow data of the intersection; the method for respectively obtaining the traffic flow data of a plurality of intersections in the target road network in the historical time period corresponding to the target prediction time period comprises the following steps:
and respectively obtaining the traffic flow data of each intersection sent by each flow sensor in the historical time period.
Optionally, the allowable travel directions in the target intersection include a first number of phase directions; the traffic flow data of the target crossing in the target prediction time period comprises: traffic flow data in each of the phase directions within the target prediction period.
Optionally, the constructing a signal lamp timing model based on the traffic flow data of the target intersection in the target prediction time period includes:
constructing the first number of single-phase congestion models, wherein each single-phase congestion model is the same;
correspondingly weighting each single-phase congestion model based on the traffic flow ratio of the traffic flow data of each phase direction in the target prediction time period at the target intersection;
and integrating the weighted single-phase congestion models into the signal lamp timing model.
Optionally, the determining, by using the signal lamp timing model, optimal signal lamp timing information matched with traffic flow data of the target intersection in the target prediction time period includes:
determining the signal lamp timing model as an objective function;
determining a variable space for searching an optimal solution for the objective function based on a preset green light display duration range and a preset signal light period range;
and performing space search in the variable space by adopting a genetic algorithm of self-attenuation variation rate, and determining the optimal signal lamp timing information when the objective function obtains the optimal solution.
Optionally, after the obtaining traffic flow data of the intersections in the target road network in the historical time period corresponding to the target prediction time period, the method further includes:
carrying out normalization processing on traffic flow data of each intersection in the historical time period to obtain normalized data;
inputting the traffic flow data of each intersection in the historical time period into a trained graph volume network model for predicting the traffic flow data, wherein the graph volume network model comprises the following steps:
and inputting the data after the normalization processing into the graph convolution network model.
Optionally, after obtaining the traffic flow data of each intersection output by the graph volume network model within the target prediction time period, the method further includes:
carrying out reverse normalization processing on the traffic flow data of each intersection in the target prediction time period to obtain reverse normalization processed data of each intersection;
the obtaining of the traffic flow data of the target intersection in the target prediction time period from the traffic flow data of each intersection in the target prediction time period comprises:
obtaining the data after the inverse normalization processing of the target intersection from the data after the inverse normalization processing of each intersection;
the method for constructing the signal lamp timing model based on the traffic flow data of the target intersection in the target prediction time period comprises the following steps:
and constructing a signal lamp timing model based on the data after the inverse normalization processing of the target intersection.
A signal lamp timing apparatus comprising: the device comprises a first obtaining unit, a first input unit, a second obtaining unit, a third obtaining unit, a first constructing unit and a first determining unit; wherein:
the first obtaining unit is used for respectively obtaining traffic flow data of a plurality of intersections in the target road network in a historical time period corresponding to the target prediction time period;
the first input unit is used for inputting traffic flow data of each intersection in the historical time period into a trained graph volume network model for predicting the traffic flow data;
the second obtaining unit is used for obtaining traffic flow data of each intersection output by the graph convolution network model in a target prediction time period;
the third obtaining unit is used for obtaining the traffic flow data of the target intersection in the target prediction time period from the traffic flow data of each intersection in the target prediction time period;
the first construction unit is used for constructing a signal lamp timing model based on traffic flow data of a target intersection in the target prediction time period;
the first determining unit is used for determining the optimal signal lamp timing information matched with the traffic flow data of the target intersection in the target prediction time period by using the signal lamp timing model; the optimal signal timing information at least comprises: an optimal green light configuration duration and/or an optimal signal light period.
Optionally, after obtaining the traffic flow data of each intersection in the historical time period, the graph convolution network model calculates the traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters, and the graph matrix and the adjacency matrix corresponding to the road network information of the target road network, and obtains and outputs the traffic flow data of each intersection in the target prediction time period.
Optionally, an activation function is set in the graph convolution network model; calculating the traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters and the graph matrix and the adjacency matrix corresponding to the road network information of the target road network, and setting as follows:
and the graph convolution network model inputs traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters and a graph matrix and an adjacent matrix corresponding to the road network information of the target road network into the activation function for calculation.
Optionally, each intersection is provided with a flow sensor for monitoring traffic flow data of the intersection;
the first obtaining unit is used for respectively obtaining traffic flow data of each intersection sent by each flow sensor in the historical time period.
Optionally, the allowable travel directions in the target intersection include a first number of phase directions; the traffic flow data of the target crossing in the target prediction time period comprises: traffic flow data in each of the phase directions within the target prediction period.
Optionally, the first building unit includes: the system comprises a model building unit, a weighting unit and an integration unit; wherein:
the model construction unit is used for constructing the first number of single-phase congestion models, and each single-phase congestion model is the same;
the weighting unit is used for correspondingly weighting each single-phase congestion model based on the traffic flow ratio of the traffic flow data of each phase direction in the target prediction time period at the target intersection;
the integration unit is used for integrating the weighted single-phase congestion models into the signal lamp timing model.
Optionally, the first determining unit includes: a second determining unit, a third determining unit and a fourth determining unit; wherein:
the second determining unit is used for determining the signal lamp timing model as an objective function;
the third determining unit is used for determining a variable space for searching an optimal solution for the objective function based on a preset green light display duration range and a preset signal light period range;
the fourth determining unit is configured to perform a spatial search in the variable space by using a genetic algorithm of a self-attenuation variation rate, and determine the optimal signal light timing information when the objective function obtains an optimal solution.
Optionally, the apparatus further comprises: a fourth obtaining unit;
the fourth obtaining unit is configured to, after the traffic flow data of the intersections in the target road network in the historical time period corresponding to the target prediction time period are obtained respectively, perform normalization processing on the traffic flow data of each intersection in the historical time period to obtain normalized data;
the first input unit is configured to input the normalized data into the graph convolution network model.
Optionally, the apparatus further comprises: a fifth obtaining unit;
the fifth obtaining unit is configured to, after obtaining traffic flow data of each intersection output by the graph convolution network model in a target prediction time period, perform inverse normalization processing on the traffic flow data of each intersection in the target prediction time period to obtain inverse normalization processed data of each intersection;
the third obtaining unit is configured to obtain denormalized data of the target intersection from the denormalized data of each intersection;
and the first construction unit is used for constructing a signal lamp timing model based on the data after the inverse normalization processing of the target intersection.
The signal lamp timing method and the signal lamp timing device respectively obtain traffic flow data of a plurality of intersections in a target road network in a historical time period corresponding to a target prediction time period; inputting traffic flow data of each intersection in a historical time period into a trained graph volume network model for predicting the traffic flow data to obtain the traffic flow data of each intersection in a target prediction time period, which is output by the graph volume network model; acquiring traffic flow data of the target intersection in the target prediction time period from the traffic flow data of each intersection in the target prediction time period; constructing a signal lamp timing model based on traffic flow data of the target intersection in a target prediction time period; determining optimal signal lamp timing information matched with traffic flow data of a target intersection in a target prediction time period by using a signal lamp timing model; the optimal signal timing information includes at least: an optimal green light configuration duration and/or an optimal signal light period. The method can predict the traffic flow data of the target intersection, construct the signal lamp timing model based on the traffic flow data of the target intersection, determine the corresponding optimal signal lamp timing information by using the signal lamp timing model, automatically determine and adjust the signal lamp timing according to the traffic flow data, improve the timing efficiency and timing accuracy of the signal lamp, avoid the consumption of human resources and reduce the timing cost of the signal lamp without observing the traffic condition of the intersection through manpower to perform manual signal lamp timing.
The foregoing description is only an overview of the technical solutions of the present invention, and the following detailed description of the present invention is provided to enable the technical means of the present invention to be more clearly understood, and to enable the above and other objects, features, and advantages of the present invention to be more clearly understood.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, it is obvious that the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a flowchart illustrating a first signal timing method according to an embodiment of the present invention;
fig. 2 is a schematic diagram illustrating a directional connection at each intersection in a road network according to an embodiment of the present invention;
FIG. 3 is a traffic phase diagram of an intersection according to an embodiment of the present invention;
FIG. 4 shows an overall flow chart of a genetic algorithm provided by an embodiment of the present invention;
fig. 5 shows a schematic structural diagram of a signal lamp timing device provided by an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
As shown in fig. 1, the present embodiment proposes a first signal lamp timing method. The method may comprise the steps of:
s101, traffic flow data of a plurality of intersections in a target road network in a historical time period corresponding to a target prediction time period are respectively obtained;
the target road network may include a target intersection that needs to perform traffic prediction and intersections with high correlation with the target intersection, such as adjacent intersections and peripheral intersections. As shown in fig. 2, a circle may represent an intersection, the number in the circle may be the mark of the intersection, and the line with an arrow between the circles represents the allowed driving direction.
Optionally, the intersection may be a T-shaped intersection, a cross-shaped intersection, or a right-angle intersection, which is not limited in the present invention.
The traffic flow data may be a flow including only motor vehicles, a flow including only non-motor vehicles, or a flow including only pedestrians. Of course, traffic flow data may also include the flow of vehicles, non-vehicles, and/or pedestrians at the same time.
It can be understood that, the present invention may determine the traffic object to be predicted in advance, and then obtain the corresponding traffic data of each intersection in the above history period based on the traffic object. For example, if the traffic object to be predicted is a motor vehicle, the present invention can obtain the traffic data of the motor vehicles at each intersection in the above history period.
In particular, an intersection may include one or more traffic phases. For example, as shown in fig. 3, the traffic phase for a motor vehicle at a certain intersection may include phase a, phase B, phase C and phase D. The arrows marked in each phase indicate the traffic flow direction which allows traffic to pass when the corresponding signal lamps display green lamps, for example, when the signal lamps corresponding to the phase A display green lamps, the oncoming vehicles on the phase A can turn left.
Specifically, the traffic flow data of an intersection in a certain time period may include the traffic flows of the intersections in respective phase time periods in turn in the time period. One phase time interval may be a corresponding traffic phase passage time interval, and the time duration of the phase time interval may include the green light display time duration and the yellow light display time duration of a signal light in the traffic phase. For example, as shown in fig. 3, if the duration of the phase time period corresponding to each of the phase a, the phase B, the phase C, and the phase D is 30 seconds, the vehicle flow at the intersection in 5 minutes is: the vehicle flow rate of the phase a in the 1 st to 30 th seconds is a1, the vehicle flow rate of the phase B in the 31 st to 60 th seconds is a2, the vehicle flow rate of the phase C in the 61 st to 90 th seconds is a3, the vehicle flow rate of the phase D in the 91 st to 120 th seconds is a4, the vehicle flow rate of the phase a in the 121 th to 150 th seconds is a5, the vehicle flow rate of the phase B in the 151 th to 180 th seconds is a6 … …, the vehicle flow rate of the phase a in the 241 th to 270 th seconds is a9, and the vehicle flow rate of the phase B in the 271 th to 300 th seconds is a10, so that the traffic flow data of the vehicle at the 5 minute intersection can be { a1, a2, a3, a4, a5, a6, … …, a9, a 10.
It can be understood that in an actual traffic environment, traffic flow data of many intersections have a certain correlation, and the correlation may depend on environmental factors such as distances between intersections, so that when traffic flow data of a target intersection is predicted, it is not accurate and strict to simply consider the environment of the target intersection. Therefore, the invention can directly take the flow in the target road network into the prediction consideration range under the premise that the actual application environment allows (for example, the distance between the target road network and the adjacent road network and the peripheral road network does not exceed a certain distance), namely, the flow in the target road network is predicted. And then determining the flow of the target intersection from the predicted flow of the target road network.
The target road network may be divided and determined by a technician in the traffic environment around the target intersection according to the actual situation, which is not limited in the present invention.
The traffic flow data of the target intersection in the target prediction time period can be the traffic flow data of the target intersection required to be predicted by the method.
The historical period may be a period adjacent to the target prediction period and before the target prediction period, and the period may include a corresponding time length. For example, if the current time period required for traffic flow data prediction is a time period within 21 minutes to 30 minutes from 12 hours, the historical time period may be a time period within 11 minutes to 20 minutes from 12 hours; for another example, if the current time period in which traffic flow data prediction is required is a time period within 21 to 30 minutes from 12 hours, the historical time period may be a time period within 01 to 20 minutes from 12 hours.
It should be noted that the time length included in the history period and the time length included in the target prediction period may be equal to each other or may not be equal to each other.
Specifically, the present invention can obtain traffic flow data of each intersection in the target road network in the above historical time period. For example, when the target road network includes a first intersection and a second intersection, the present invention may obtain the traffic flow data of the first intersection in the historical time period, and obtain the traffic flow data of the second intersection in the historical time period.
Optionally, in other signal lamp timing methods provided in this embodiment, each intersection is provided with a flow sensor for monitoring traffic flow data of the intersection; at this time, step S101 may include:
and respectively obtaining the traffic flow data of each road junction in the historical time period, which is sent by each flow sensor.
The flow sensor can be a sensor for monitoring and recording the traffic flow of the intersection where the flow sensor is located. Alternatively, the flow sensor may be a radar sensor. Specifically, each intersection can be provided with one or more flow sensors, and the traffic flow data of the intersection sent by the traffic flow sensors can be obtained from the flow sensors arranged at each intersection.
Specifically, the present invention can obtain traffic flow data of each intersection returned by each flow sensor in the same cycle. For example, the traffic flow sensors arranged at each intersection can return the monitored traffic flow data every 5 minutes.
S102, inputting traffic flow data of each intersection in the historical time period into a trained graph volume network model for predicting the traffic flow data;
it is understood that the network architecture of the graph convolution network model may be a graph convolution neural network.
Specifically, the graph convolution network model can be used for predicting traffic flow data of each intersection in the target road network in a target prediction time period. The graph convolution network model can be input into the traffic flow data of each road junction in the target road network in the historical time period, and output into the traffic flow data of each road junction in the target road network in the target prediction time period.
It should be noted that the convolutional network model may be a machine learning model trained by using a training set, a validation set, and a test set and a machine learning model training mode, with the convolutional neural network as a model architecture.
S103, obtaining traffic flow data of each intersection output by the graph convolution network model in a target prediction time period;
specifically, the present invention may obtain the traffic flow data of each intersection in the target road network within the target prediction time period output by the graph-convolution network model after inputting the traffic flow data of each intersection in the target road network within the history time period into the graph-convolution network model.
S104, obtaining traffic flow data of the target crossing in the target prediction time period from the traffic flow data of each crossing in the target prediction time period;
specifically, the present invention may determine and record the position of the traffic flow data of the target intersection in the traffic flow data of each intersection to be input into the graph-convolution network model in the above-mentioned history period. Then, when the traffic flow data of each intersection in the target prediction time period output by the graph convolution network model is obtained, the traffic flow data of the target intersection in the target prediction time period can be searched and determined from the traffic flow data of each intersection in the target prediction time period according to the recorded positions.
S105, constructing a signal lamp timing model based on traffic flow data of the target intersection in a target prediction time period;
specifically, after traffic flow data of the target intersection in the target prediction time period are obtained, the traffic data of each traffic phase of the target intersection in the target prediction time period are determined, and a signal lamp timing model is constructed based on the traffic data of each traffic phase in the target prediction time period.
The parameters to be solved in the signal lamp timing model may include green lamp display duration, signal lamp period duration, and the like.
S106, determining the optimal signal lamp timing information matched with the traffic flow data of the target intersection in the target prediction time period by using a signal lamp timing model; the optimal signal timing information includes at least: an optimal green light configuration duration and/or an optimal signal light period.
Specifically, the invention can determine the corresponding optimal signal lamp timing information by using the signal lamp timing model after obtaining the signal lamp timing model.
Specifically, the optimal signal light timing information may further include timing information such as a red light timing duration and a yellow light timing duration.
Optionally, in the second signal light timing method provided in this embodiment, the allowable driving directions at the target intersection include a first number of phase directions; the traffic flow data of the target crossing in the target prediction time period comprises the following steps: traffic flow data within the target prediction period in each phase direction. At this time, the step S105 may include:
constructing a first number of single-phase congestion models, wherein the single-phase congestion models are the same;
correspondingly weighting each single-phase congestion model based on the traffic flow ratio of the traffic flow data of each phase direction in the target prediction time period at the target intersection;
and integrating the weighted single-phase congestion models into a signal lamp timing model.
The first number may be the number of allowed traffic phases at the target intersection, for example, the first number is 4 in fig. 3.
The single-phase congestion model may be a traffic congestion index model corresponding to a single traffic phase at the target intersection. It should be noted that the single-phase congestion model may be obtained based on the Webster delay model. Optionally, the construction process of the single-phase congestion model may consider normal phase delay, and may also consider random delay and supersaturation delay. The present invention may use, among other things, the oversaturation lagged vehicle multiplied by saturation to represent random delay and oversaturation delay.
The parameters to be solved in the single-phase congestion model can include green light display time and signal light cycle time. It should be noted that, when the single-phase congestion models corresponding to the traffic phases are the same, the green light display duration and the signal lamp period duration at each traffic phase of the target intersection may be the same, and the green light display duration and the signal lamp period duration solved in the signal lamp timing model may be used as the green light display duration and the signal lamp period duration at each traffic phase of the target intersection. At the moment, the method can simplify the operation process of the green light display time length and the signal lamp period time length on each traffic phase, reduce the operation amount and improve the calculation efficiency.
Optionally, the single-phase congestion model may be:
wherein the content of the first and second substances,congestionthe congestion index to be finally calculated;delaydelay for parking;capthe unit is the traffic capacity of the road;his the parking rate, with the unit of vehicle; wherein:
wherein the content of the first and second substances,qthe traffic flow is unit vehicle/min;gdisplaying the duration for the green light;cis the signal lamp period duration;sthe method is based on the prior knowledge and can be used for measuring the saturation flow of the general road on the spot if the data accuracy is pursued and the engineering condition allows;q/sto represent the degree of saturation of the road, with the result that more than 1 is in an oversaturated state and less than 1 is underA saturated state;lagthe neural reflex delay for the driver, including the green light start delay and the yellow light stop delay, can be set to 5 seconds according to general prior knowledge.
Specifically, the traffic flow duty ratio of each traffic phase in the target intersection can be determined based on the traffic flow data of each traffic phase in the target intersection, the single-phase congestion models of each traffic phase are respectively weighted according to the traffic flow duty ratio, and then the weighted single-phase congestion models are integrated into the signal lamp timing model. For example, if the target intersection includes intersections with four traffic phases, the single-phase congestion models corresponding to the four traffic phases are respectively f (q)1)、f(q2)、f(q3) And f (q)4) The traffic flow on the four traffic phases is q1、q2、q3And q is4If the sum of the traffic flows of the target intersection is S = q1 +q2 +q3 +q4Then, the signal lamp timing model of the target intersection may be:
F(q)=(q1/S)*f(q1)+ (q2/S)*f(q2)+ (q3/S)*f(q3)+ (q4/S)*f(q4)；
it should be noted that, after each single-phase congestion model is weighted based on the traffic occupancy, the weight may be correspondingly corrected based on the experience of the staff in observing the traffic at the target intersection, and then the signal lamp timing model may be obtained based on each single-phase congestion model after the weight is corrected.
It can be understood that, in the signal lamp timing model integrated by the weighted single-phase congestion models, the parameters to be solved can be the green light display time length and the signal lamp period time length.
Optionally, in other signal lamp timing methods provided in this embodiment, the single-phase congestion models corresponding to the traffic phases at the target intersection may also be different.
Optionally, in the second signal light timing method, step S106 may include:
determining a signal lamp timing model as a target function;
determining a variable space for searching an optimal solution for the objective function based on a preset green light display duration range and a preset signal light period range;
and (3) carrying out space search in a variable space by adopting a genetic algorithm of the self-attenuation variation rate, and determining the optimal signal lamp timing information when the target function obtains the optimal solution.
Specifically, the signal lamp timing model is used as a target function by utilizing a genetic algorithm, and an optimal solution is searched in a specified space.
It should be noted that, according to the priori knowledge in the traffic control field, it is known that the green light display duration may be not less than a certain duration a but not more than a certain duration b, and the signal lamp cycle duration (the sum of the display durations of the green light, the yellow light, and the red light) may not be less than a certain duration c but not more than a certain duration d, i.e., the numerical ranges of the green light display duration and the signal lamp cycle duration may be within the specified intervals [ a, b ] and [ c, d ]. It should be noted that the values of a, b, c and d can be set by the staff according to the priori knowledge and the actual situation, and the invention is not limited to this. Specifically, the optimal solution can be found by taking [ a, b ] and [ c, d ] as the solution space of the signal lamp timing model.
Specifically, the overall process of the genetic algorithm used in the present invention can be as shown in fig. 4, i.e., performing gene coding, initializing a population, calculating the individual fitness in the population, selecting elimination, cross-inheritance, adding new variant individuals, and then returning to the step of calculating the individual fitness in the population for cyclic evolution.
It should be noted that the objective function of the general intelligent algorithm is a unitary function, and the signal lamp timing model of the present invention is a binary objective function. Therefore, in the encoding step, binary encoding can be carried out on both variables to participate in the evolution process; in the elimination step, the invention can adopt a roulette mode, and the ratio of the individual fitness to the population accumulated fitness is used as the probability of elimination, and at the moment, the less desirable the fitness is, the more easily the fitness is selected to eliminate. Therefore, the genetic algorithm used by the invention can gradually eliminate irrelevant individuals in iteration to breedThe whole group evolves towards an expected space, and the unstable influence of the algorithm on the result due to the random characteristic is effectively reduced. In addition, for the cross rate and the variation rate of two important parameters of the genetic algorithm, the cross rate can determine the solving speed of the algorithm, and the variation rate determines the solving precision and the stability of the result. The crossover rate can be set to a fixed value where circumstances permit, as long as the runtime of the algorithm is accepted, and can be fine-tuned based on the results when implemented. The variation rate influences the result of the algorithm to a great extent, and the randomness of the algorithm is increased due to the excessively large variation rate, so that the result is unstable; if the variation rate is too small, the search precision of the algorithm is reduced, and the algorithm is easy to fall into a local optimal trap. The variable variation rate can achieve a certain degree of self-adaptive effect. The invention can set the initial variation rate of the algorithm to be a larger value, increase the search step length, accelerate the convergence of the algorithm, gradually reduce the variation rate along with the iteration of the algorithm, and improve the search precision, so that the algorithm can more accurately search the optimal solution in the interval close to the optimal solution. The invention can adopt an exponential decay mode to control the variation rate, and the decay formula is as follows:
；
wherein crossvar may be the rate of variation; dec _ rate may be the decay rate of the variability rate; step may be the number of iterations; crossvar _ s may be the initial variation rate, which may be set to 0.09; crossvar _ e may be the end rate and may be set to 0.001 to ensure that the rate does not decay below the normal range, preventing meaningless inheritance due to too small a rate.
It should be noted that, the inventor of the present invention has performed an experimental test on the above method by using actual traffic data, the test result shows that the prediction effect of the traffic flow is accurate, the calculated green light display duration and the calculated signal lamp period duration can effectively and automatically change along with the predicted traffic flow, the timing efficiency and timing accuracy of the signal lamp are improved, manual signal lamp timing is performed without observing the intersection traffic condition by manpower, the consumption of human resources is avoided, and the signal lamp timing cost is reduced.
The signal lamp timing method provided by the embodiment respectively obtains traffic flow data of a plurality of intersections in a target road network in a historical time period corresponding to a target prediction time period; inputting traffic flow data of each intersection in a historical time period into a trained graph volume network model for predicting the traffic flow data to obtain the traffic flow data of each intersection in a target prediction time period, which is output by the graph volume network model; acquiring traffic flow data of the target intersection in the target prediction time period from the traffic flow data of each intersection in the target prediction time period; constructing a signal lamp timing model based on traffic flow data of the target intersection in a target prediction time period; determining optimal signal lamp timing information matched with traffic flow data of a target intersection in a target prediction time period by using a signal lamp timing model; the optimal signal timing information includes at least: an optimal green light configuration duration and/or an optimal signal light period. The method can predict the traffic flow data of the target intersection, construct the signal lamp timing model based on the traffic flow data of the target intersection, determine the corresponding optimal signal lamp timing information by using the signal lamp timing model, automatically determine and adjust the signal lamp timing according to the traffic flow data, improve the timing efficiency and timing accuracy of the signal lamp, avoid the consumption of human resources and reduce the timing cost of the signal lamp without observing the traffic condition of the intersection through manpower to perform manual signal lamp timing.
Based on fig. 1, the present embodiment proposes a third signal timing method. In the method, after traffic flow data of each intersection in a historical time period are obtained by a graph convolution network model, the traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters and a graph matrix and an adjacent matrix corresponding to road network information of a target road network are calculated, and the traffic flow data of each intersection in a target prediction time period are obtained and output.
The graph volume network model may have input of traffic flow data of each intersection in the target road network in the historical time period, and output of traffic flow data of each intersection in the target road network in the target prediction time period.
The graph matrix and the adjacency matrix corresponding to the road network information of the target road network may be set in advance based on the road network information of the target road network.
Optionally, the traffic flow data sent by each flow sensor arranged in the target road network may be used in advance to generate a training set, a verification set, and a test set for training the graph convolution network, the graph convolution network is trained by using the training set, the verification set, and the test set to obtain a trained graph convolution network model, and then the traffic flow data in the target prediction time period is predicted by using the trained graph convolution network model.
Specifically, the historical traffic flow data returned by each flow sensor arranged in the target road network within a short period of time (such as the previous months) can be placed in the same data table, the data table is utilized to count the historical traffic flow data returned by each flow sensor, and a training set, a verification set and a test set for training a graph convolution network model are obtained based on the data table. Specifically, each line of data in the data table may be traffic flow data returned by one flow sensor at different times, and each line of data in the data table may be traffic flow data returned by each flow sensor at the same time. Wherein, each flow sensor can return data every x minutes, and the continuous x/60 row data in the data sheet is the traffic flow data collected in one hour. In this case, each row of data in the data table may be time-series data sorted in time. Optionally, the present invention may divide the training set, the validation set, and the test set by row range in the data table.
Optionally, in the process of training the graph convolution network model, the invention iteratively updates the weight parameters of each layer in the graph convolution network model by using an interlayer information forward propagation mode and a backward propagation loss until the trained graph convolution network model is obtained.
Optionally, an activation function is set in the graph convolution network model; the calculating the traffic flow data of each intersection in the historical time period, the trained and iterated weight parameter, and the graph matrix and the adjacency matrix corresponding to the road network information of the target road network may include:
and the graph convolution network model inputs traffic flow data of each intersection in a historical period, trained and iterated weight parameters and a graph matrix and an adjacent matrix corresponding to the road network information of the target road network into the activation function for calculation.
The activation function may be an activation function used in an inter-layer information forward propagation manner. And the interlayer information forward propagation mode can be a core training mode adopted in the training process of the graph convolution network model. This approach may follow the formula:
wherein the content of the first and second substances,
input data which may be of a previous layer;
may be a weight parameter of an upper layer;
the activation function may be a conventional activation function such as RELU, and may be set by a technician according to a framework required by actual use.
Specifically, in the process of training the graph convolution network model, the weight parameters may be randomly initialized according to the dimensionality of the input data at the initial time of training, the MAE, that is, the average value of the absolute values of the errors between the predicted value and the true value, is calculated as the loss value of the network according to the prediction result in the training iteration, the weight parameters of each layer are updated through the back propagation loss, and the update formula may be:
wherein the content of the first and second substances,
any extremely small number 1e-6 can be taken to prevent the denominator in the formula from being 0;learning_ ratemay be a learning rate; the gradient can be automatically calculated from the back-propagated loss value through a framework-internal mechanism.m (t) In order to be the momentum of the gradient,v (t) in order to change the amount of the gradient,m (t) andv (t) can be obtained by the following formula:
wherein the content of the first and second substances,
and
may be defined attenuation parameters inside the frame to control the gradient momentum and the weight of the gradient changes.
Optionally, in the training of the invention, an attenuation formula can be added to the learning rate to control the change of the learning rate. Because the training effect is not well improved by the fixed learning rate adjustment, the variable learning rate can be set to be larger at the beginning of training, the convergence of the model is accelerated, and then the learning rate is gradually reduced to improve the precision of searching the optimal solution interval and prevent the model from falling into the local optimal trap. Specifically, the present invention can set the attenuation ratio
And number of attenuation steps
(i.e. every other fixed step according to the number of iteration stepsNumber decays once by decay rate), setting the current iteration step number
At this time, the learning rate of each iteration
Can be as follows:
it should be noted that the present invention utilizes the graph convolution network to predict the traffic flow of all global intersections of the target road network, and effectively considers the correlation between each intersection in the process of applying the graph matrix and the adjacent matrix corresponding to the road network information of the target road network, and does not need to perform the complicated process of dividing sub-areas and the like in the prior art, thereby effectively improving the accuracy and efficiency of traffic flow data prediction.
It can be understood that the number of layers of the graph convolution network structure is small, the algorithm complexity is simple, and the requirement on the implementation environment is low, so that the traffic flow data is predicted by utilizing the graph convolution network structure, the time distribution cost of the signal lamp can be effectively saved, the prediction accuracy is improved, and the time distribution accuracy of the signal lamp is ensured.
Optionally, the present invention may use an adjacent matrix corresponding to the road network information of the target road network and traffic data of each intersection as the input of the graph convolution network. The traffic data may include, among other things, traffic volume, speed, occupancy, and the like. Specifically, the number of data types included in the traffic data at each intersection is the dimension of the input features learned by the graph convolution network, and the graph convolution network can perform feature extraction based on the laplacian matrix and perform full connection with the traffic data to obtain one-dimensional predicted traffic data through fusion.
Optionally, in the process of training the graph convolution network, the present invention may use three consecutive rows of data in the data table as one input to train the graph convolution network. Of course, the invention can also adjust the data amount required by the input by the technicians according to the characteristics of the data used actually, the characteristics of the network model and the actual training effect, for example, the continuous five-line data in the data table can also be used as the input of the primary graph convolution network model.
When the cycle of returning the traffic flow data by the flow sensor is 5 minutes, the traffic flow data is predicted every 15 minutes, which is equivalent to inputting the data of three consecutive rows in the data table once. For different prediction requirements, such as long-term prediction, the model parameters of the graph convolution network can be modified by the method. And, the traffic flow data itself has a periodic characteristic, and it is difficult for immediate change to occur in a short time unless a sudden time such as a car accident occurs to a link, so that prediction for a too short time such as less than 5-10 minutes has no practical application value, and 10-15 minutes may be a relatively suitable short-time prediction range.
Specifically, the laplacian matrix adopted in this embodiment may be:
wherein the content of the first and second substances,
the laplace matrix is used;
is a common Laplace matrix, and can be defined as:
wherein the content of the first and second substances,
is a degree matrix of a graph corresponding to the road network information of the target road network,
is an adjacent matrix corresponding to the road network information of the target road network. It should be noted that how many edges are connected to the intersection node in the undirected graph, how many degrees the intersection node has; in the directed graph, the number of edges taking an intersection node as a terminal is used as the incoming degree of the intersection node, and the number of edges taking the intersection node as a starting point is used as the outgoing degree of the intersection node. Therefore, in the actual implementation process, if strong correlation exists between the nodes of the intersection in the direction, the entry matrix or the exit matrix can be used for adding calculation, and the entry matrix or the exit matrix can also be used simultaneously, so that the degree of polymerization of the nodes of the intersection on the characteristics of the nodes of the neighboring intersections is enhanced. Furthermore, for the adjacency matrix
Since the diagonal line of the adjacency matrix is 0 regardless of whether the adjacency or directed graph is an undirected graph, but the diagonal line elements represent the nodes themselves, the aggregation result of the calculation cannot contain the characteristics of the nodes themselves. The adjacency matrix is thus self-loop calculated, summing the adjacency matrix with an identity matrix (only diagonal elements are non-0 and all 1's).
The signal lamp timing method provided by the embodiment can utilize the graph convolution network to predict the traffic flow of all global intersections of the target road network, effectively considers the correlation among all the intersections in the process of applying the graph matrix and the adjacent matrix corresponding to the road network information of the target road network, does not need to perform the complicated process of dividing sub-areas and the like in the prior art, and can further improve the accuracy and efficiency of traffic flow data prediction.
Optionally, in another signal lamp timing method proposed in this embodiment, after step S101, the method may further include:
carrying out normalization processing on traffic flow data of each intersection in a historical time period to obtain normalized data;
at this time, step S102 may include:
and inputting the data after the normalization processing into a graph convolution network model.
It should be noted that there may be differences in the traffic flow data returned by each flow sensor, and there may be some abnormal values and extreme values (reasonable phenomena due to measurement errors). In order to avoid the adverse effects of abnormal values and extreme values, the traffic flow data can be normalized in advance after the traffic flow data returned by each flow sensor is obtained.
Specifically, the present invention may employ a normalization processing method such as (0, 1) normalization, Z-score normalization, Sigmoid, or the like. When the normalization processing is performed by adopting (0, 1) standardization, the method can traverse all traffic flow data, make a difference max-min between a maximum value max and a minimum value min in the traffic flow data, and perform the normalization processing on each traffic flow data by taking the difference max-min as a normalization base number. Optionally, after normalization processing, the present invention may call a data loading mode provided by the framework to divide all data into tenor stream data required by neural network input.
At this time, the third signal timing method may further include, after step S103:
carrying out reverse normalization processing on traffic flow data of each intersection in a target prediction time period to obtain reverse normalization processed data of each intersection;
at this time, step S104 may include:
obtaining the data after the inverse normalization processing of the target intersection from the data after the inverse normalization processing of each intersection;
at this time, step S105 may include:
and constructing a signal lamp timing model based on the data after the inverse normalization processing of the target intersection.
Specifically, after traffic flow data of each intersection in the target road network in the target prediction time period are obtained, corresponding inverse normalization processing is carried out on the traffic flow data, and data after the return normalization processing of the target road network are obtained. And then, searching the data after the inverse normalization processing of the target intersection from the data after the inverse normalization processing of the target road network, and constructing a signal lamp timing model by using the data after the inverse normalization processing to obtain the optimal signal lamp timing information.
The signal lamp timing method provided by the embodiment can avoid the adverse effects of abnormal values and extreme values, and further improve the prediction accuracy of traffic flow data and the reliability of signal lamp timing information.
Corresponding to the method shown in fig. 1, as shown in fig. 5, the present embodiment proposes a signal lamp timing apparatus. The apparatus may include: a first obtaining unit 101, a first input unit 102, a second obtaining unit 103, a third obtaining unit 104, a first constructing unit 105, and a first determining unit 106; wherein:
a first obtaining unit 101, configured to obtain traffic flow data of multiple intersections in a target road network in a historical time period corresponding to a target prediction time period;
the first input unit 102 is used for inputting traffic flow data of each intersection in a historical time period into a trained graph volume network model for predicting the traffic flow data;
a second obtaining unit 103, configured to obtain traffic flow data of each intersection output by the graph convolution network model in the target prediction time period;
a third obtaining unit 104, configured to obtain traffic flow data of the target intersection in the target prediction time period from the traffic flow data of each intersection in the target prediction time period;
the first construction unit 105 is used for constructing a signal lamp timing model based on traffic flow data of the target intersection in a target prediction time period;
a first determining unit 106, configured to determine, by using a signal lamp timing model, optimal signal lamp timing information that matches traffic flow data of a target intersection in a target prediction time period; the optimal signal timing information includes at least: an optimal green light configuration duration and/or an optimal signal light period.
It should be noted that specific processing procedures of the first obtaining unit 101, the first input unit 102, the second obtaining unit 103, the third obtaining unit 104, the first constructing unit 105, and the first determining unit 106 and technical effects brought by the processing procedures may refer to related descriptions of steps S101, S102, S103, S104, S105, and S106 in the method corresponding to fig. 1 in this embodiment, and are not described herein again.
Optionally, after obtaining the traffic flow data of each intersection in the historical time period, the graph convolution network model calculates the traffic flow data of each intersection in the historical time period, the trained and iterated weight parameter, and the graph matrix and the adjacency matrix corresponding to the road network information of the target road network, and obtains and outputs the traffic flow data of each intersection in the target prediction time period.
Optionally, an activation function is set in the graph convolution network model; calculating traffic flow data of each intersection in a historical period, trained and iterated weight parameters and a graph matrix and an adjacent matrix corresponding to road network information of a target road network, and setting as follows:
and the graph convolution network model inputs traffic flow data of each intersection in a historical period, trained and iterated weight parameters and a graph matrix and an adjacent matrix corresponding to the road network information of the target road network into the activation function for calculation.
Optionally, each intersection is provided with a flow sensor for monitoring traffic flow data of the intersection;
a first obtaining unit 101, configured to obtain traffic flow data of each intersection in a history period sent by each flow sensor, respectively.
Optionally, the allowable travel directions in the target intersection include a first number of phase directions; the traffic flow data of the target crossing in the target prediction time period comprises the following steps: traffic flow data within the target prediction period in each phase direction.
Optionally, the first building unit 105 includes: the system comprises a model building unit, a weighting unit and an integration unit; wherein:
the model building unit is used for building a first number of single-phase congestion models, and all the single-phase congestion models are the same;
the weighting unit is used for correspondingly weighting each single-phase congestion model based on the traffic flow ratio of the traffic flow data of each phase direction in the target prediction time period at the target intersection;
and the integration unit is used for integrating the weighted single-phase congestion models into a signal lamp timing model.
Optionally, the first determining unit 106 includes: a second determining unit, a third determining unit and a fourth determining unit; wherein:
the second determining unit is used for determining the signal lamp timing model as an objective function;
the third determining unit is used for determining a variable space for searching an optimal solution for the objective function based on a preset green light display duration range and a preset signal light period range;
and the fourth determining unit is used for performing space search in the variable space by adopting a genetic algorithm of the self-attenuation variation rate and determining the optimal signal lamp timing information when the objective function obtains the optimal solution.
Optionally, the apparatus further comprises: a fourth obtaining unit;
the fourth obtaining unit is used for normalizing the traffic flow data of each road junction in the historical time period after traffic flow data of a plurality of road junctions in the target road network in the historical time period corresponding to the target prediction time period are respectively obtained, and data after normalization processing are obtained;
the first input unit 102 is configured to input the normalized data into the graph convolution network model.
Optionally, the apparatus further comprises: a fifth obtaining unit;
a fifth obtaining unit, configured to, after obtaining traffic flow data of each intersection in the target prediction time period output by the graph convolution network model, perform inverse normalization processing on the traffic flow data of each intersection in the target prediction time period, and obtain data after inverse normalization processing of each intersection;
a third obtaining unit 104, configured to obtain data after inverse normalization processing of the target intersection from the data after inverse normalization processing of each intersection;
the first construction unit 105 is configured to construct a signal lamp timing model based on the data after the inverse normalization processing of the target intersection.
The signal lamp timing device provided by the embodiment respectively obtains traffic flow data of a plurality of intersections in a target road network in a historical time period corresponding to a target prediction time period; inputting traffic flow data of each intersection in a historical time period into a trained graph volume network model for predicting the traffic flow data to obtain the traffic flow data of each intersection in a target prediction time period, which is output by the graph volume network model; acquiring traffic flow data of the target intersection in the target prediction time period from the traffic flow data of each intersection in the target prediction time period; constructing a signal lamp timing model based on traffic flow data of the target intersection in a target prediction time period; determining optimal signal lamp timing information matched with traffic flow data of a target intersection in a target prediction time period by using a signal lamp timing model; the optimal signal timing information includes at least: an optimal green light configuration duration and/or an optimal signal light period. The method can predict the traffic flow data of the target intersection, construct the signal lamp timing model based on the traffic flow data of the target intersection, determine the corresponding optimal signal lamp timing information by using the signal lamp timing model, automatically determine and adjust the signal lamp timing according to the traffic flow data, improve the timing efficiency and timing accuracy of the signal lamp, avoid the consumption of human resources and reduce the timing cost of the signal lamp without observing the traffic condition of the intersection through manpower to perform manual signal lamp timing.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in the process, method, article, or apparatus that comprises the element.
The above are merely examples of the present application and are not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.
Claims (10)
1. A signal lamp timing method, comprising:
respectively obtaining traffic flow data of a plurality of intersections in a target road network in a historical time period corresponding to a target prediction time period;
inputting the traffic flow data of each intersection in the historical time period into a trained graph volume network model for predicting the traffic flow data to obtain the traffic flow data of each intersection in a target prediction time period, which is output by the graph volume network model;
obtaining traffic flow data of a target intersection in the target prediction time period from the traffic flow data of each intersection in the target prediction time period; the allowable travel directions in the target intersection include a first number of phase directions; the traffic flow data of the target crossing in the target prediction time period comprises: traffic flow data in each of the phase directions within the target prediction period;
constructing the first number of single-phase congestion models, wherein each single-phase congestion model is the same;
correspondingly weighting each single-phase congestion model based on the traffic flow ratio of the traffic flow data of each phase direction in the target prediction time period at the target intersection;
integrating the weighted single-phase congestion models into a signal lamp timing model;
determining the optimal signal lamp timing information matched with the traffic flow data of the target intersection in the target prediction time period by utilizing the signal lamp timing model; the optimal signal timing information at least comprises: an optimal green light configuration duration and/or an optimal signal light period.
2. The signal light timing method according to claim 1, wherein after the graph convolution network model obtains the traffic flow data of each intersection in the historical time period, the graph matrix and the adjacency matrix corresponding to the traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters, and the road network information of the target road network are calculated, and the traffic flow data of each intersection in the target prediction time period is obtained and output.
3. The signal timing method according to claim 2, wherein an activation function is provided in the graph convolution network model; the calculating of the traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters, and the graph matrix and the adjacency matrix corresponding to the road network information of the target road network comprises:
and the graph convolution network model inputs traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters and a graph matrix and an adjacent matrix corresponding to the road network information of the target road network into the activation function for calculation.
4. The signal lamp timing method according to claim 1, wherein each intersection is provided with a flow sensor for monitoring traffic flow data of the intersection; the method for respectively obtaining the traffic flow data of a plurality of intersections in the target road network in the historical time period corresponding to the target prediction time period comprises the following steps:
and respectively obtaining the traffic flow data of each intersection sent by each flow sensor in the historical time period.
5. The signal light timing method according to claim 1, wherein said determining optimal signal light timing information that matches traffic flow data of the target intersection within the target prediction period using the signal light timing model comprises:
determining the signal lamp timing model as an objective function;
determining a variable space for searching an optimal solution for the objective function based on a preset green light display duration range and a preset signal light period range;
and performing space search in the variable space by adopting a genetic algorithm of self-attenuation variation rate, and determining the optimal signal lamp timing information when the objective function obtains the optimal solution.
6. The signal light timing method according to claim 1, wherein after said respectively obtaining traffic flow data of a plurality of intersections in the target road network within a history period corresponding to the target prediction period, the method further comprises:
carrying out normalization processing on traffic flow data of each intersection in the historical time period to obtain normalized data;
inputting the traffic flow data of each intersection in the historical time period into a trained graph volume network model for predicting the traffic flow data, wherein the graph volume network model comprises the following steps:
and inputting the data after the normalization processing into the graph convolution network model.
7. The signal light timing method according to claim 6, wherein after obtaining traffic flow data of each intersection output by said graph-volume network model within a target prediction time period, said method further comprises:
carrying out reverse normalization processing on the traffic flow data of each intersection in the target prediction time period to obtain reverse normalization processed data of each intersection;
the obtaining of the traffic flow data of the target intersection in the target prediction time period from the traffic flow data of each intersection in the target prediction time period comprises:
obtaining the data after the inverse normalization processing of the target intersection from the data after the inverse normalization processing of each intersection;
the method for constructing the signal lamp timing model based on the traffic flow data of the target intersection in the target prediction time period comprises the following steps:
and constructing a signal lamp timing model based on the data after the inverse normalization processing of the target intersection.
8. A signal lamp timing apparatus, comprising: the system comprises a first obtaining unit, a first input unit, a second obtaining unit, a third obtaining unit, a model building unit, a weighting unit, an integration unit and a first determining unit; wherein:
the first obtaining unit is used for respectively obtaining traffic flow data of a plurality of intersections in the target road network in a historical time period corresponding to the target prediction time period;
the first input unit is used for inputting traffic flow data of each intersection in the historical time period into a trained graph volume network model for predicting the traffic flow data;
the second obtaining unit is used for obtaining traffic flow data of each intersection output by the graph convolution network model in a target prediction time period;
the third obtaining unit is used for obtaining the traffic flow data of the target intersection in the target prediction time period from the traffic flow data of each intersection in the target prediction time period;
the model construction unit is used for constructing the first number of single-phase congestion models, and each single-phase congestion model is the same;
the weighting unit is used for correspondingly weighting each single-phase congestion model based on the traffic flow ratio of the traffic flow data of each phase direction in the target prediction time period at the target intersection;
the integration unit is used for integrating the weighted single-phase congestion models into a signal lamp timing model;
the first determining unit is used for determining the optimal signal lamp timing information matched with the traffic flow data of the target intersection in the target prediction time period by using the signal lamp timing model; the optimal signal timing information at least comprises: an optimal green light configuration duration and/or an optimal signal light period.
9. The signal lamp timing device according to claim 8, wherein the graph convolution network model calculates traffic flow data of each intersection in the history time period, trained and iterated weight parameters, and a graph matrix and an adjacent matrix corresponding to the road network information of the target road network after obtaining the traffic flow data of each intersection in the history time period, and obtains and outputs the traffic flow data of each intersection in the target prediction time period.
10. The signal lamp timing device as claimed in claim 9, wherein an activation function is provided in the graph convolution network model; calculating the traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters and the graph matrix and the adjacency matrix corresponding to the road network information of the target road network, and setting as follows:
and the graph convolution network model inputs traffic flow data of each intersection in the historical time period, the trained and iterated weight parameters and a graph matrix and an adjacent matrix corresponding to the road network information of the target road network into the activation function for calculation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210159646.9A CN114333361B (en) | 2022-02-22 | 2022-02-22 | Signal lamp timing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210159646.9A CN114333361B (en) | 2022-02-22 | 2022-02-22 | Signal lamp timing method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114333361A true CN114333361A (en) | 2022-04-12 |
| CN114333361B CN114333361B (en) | 2022-06-28 |
Family
ID=81031400
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210159646.9A Active CN114333361B (en) | 2022-02-22 | 2022-02-22 | Signal lamp timing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114333361B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114707560A (en) * | 2022-05-19 | 2022-07-05 | 北京闪马智建科技有限公司 | Data signal processing method and device, storage medium and electronic device |
| CN114822037A (en) * | 2022-06-01 | 2022-07-29 | 浙江大华技术股份有限公司 | Traffic signal control method and device, storage medium and electronic device |
| CN114973699A (en) * | 2022-05-10 | 2022-08-30 | 阿波罗智联(北京)科技有限公司 | Traffic control signal generation method, edge calculation unit and road side unit |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105279967A (en) * | 2015-10-15 | 2016-01-27 | 深圳市城市交通规划设计研究中心有限公司 | System and method for traffic operation index calculation |
| CN110111592A (en) * | 2019-06-25 | 2019-08-09 | 浪潮软件集团有限公司 | Method based on traffic signal controlling machine Dynamic Matching Optimal Signals timing scheme |
| CN112419726A (en) * | 2020-11-20 | 2021-02-26 | 华南理工大学 | Urban traffic signal control system based on traffic flow prediction |
| CN113538910A (en) * | 2021-07-14 | 2021-10-22 | 李丹丹 | Self-adaptive full-chain urban area network signal control optimization method |
| CN113706860A (en) * | 2021-07-14 | 2021-11-26 | 石家庄铁道大学 | Intelligent timing traffic light control method based on raspberry group |
-
2022
- 2022-02-22 CN CN202210159646.9A patent/CN114333361B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105279967A (en) * | 2015-10-15 | 2016-01-27 | 深圳市城市交通规划设计研究中心有限公司 | System and method for traffic operation index calculation |
| CN110111592A (en) * | 2019-06-25 | 2019-08-09 | 浪潮软件集团有限公司 | Method based on traffic signal controlling machine Dynamic Matching Optimal Signals timing scheme |
| CN112419726A (en) * | 2020-11-20 | 2021-02-26 | 华南理工大学 | Urban traffic signal control system based on traffic flow prediction |
| CN113538910A (en) * | 2021-07-14 | 2021-10-22 | 李丹丹 | Self-adaptive full-chain urban area network signal control optimization method |
| CN113706860A (en) * | 2021-07-14 | 2021-11-26 | 石家庄铁道大学 | Intelligent timing traffic light control method based on raspberry group |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114973699A (en) * | 2022-05-10 | 2022-08-30 | 阿波罗智联(北京)科技有限公司 | Traffic control signal generation method, edge calculation unit and road side unit |
| CN114707560A (en) * | 2022-05-19 | 2022-07-05 | 北京闪马智建科技有限公司 | Data signal processing method and device, storage medium and electronic device |
| CN114707560B (en) * | 2022-05-19 | 2024-02-09 | 北京闪马智建科技有限公司 | Data signal processing method and device, storage medium and electronic device |
| CN114822037A (en) * | 2022-06-01 | 2022-07-29 | 浙江大华技术股份有限公司 | Traffic signal control method and device, storage medium and electronic device |
| CN114822037B (en) * | 2022-06-01 | 2023-09-08 | 浙江大华技术股份有限公司 | Traffic signal control method and device, storage medium and electronic device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114333361B (en) | 2022-06-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114333361B (en) | Signal lamp timing method and device | |
| US11555706B1 (en) | Processing graph representations of tactical maps using neural networks | |
| KR101385057B1 (en) | Prediction of urban congestion using ITS based data | |
| CN106887141B (en) | Queuing theory-based continuous traffic node congestion degree prediction model, system and method | |
| CN113257016B (en) | Traffic signal control method and device and readable storage medium | |
| CN109345832B (en) | Urban road overtaking prediction method based on deep recurrent neural network | |
| Odeh et al. | A hybrid fuzzy genetic algorithm for an adaptive traffic signal system | |
| CN113362600A (en) | Traffic state estimation method and system | |
| CN115862322A (en) | Vehicle variable speed limit control optimization method, system, medium and equipment | |
| CN109800908A (en) | Signalized intersections operating status prediction technique and system based on LSTM model | |
| CN113821547B (en) | Rapid and efficient short-time prediction method, system and storage medium for occupancy of parking lot | |
| Zhao et al. | Traffic signal control with deep reinforcement learning | |
| WO2022258943A1 (en) | Traffic control system | |
| CN113140108B (en) | Cloud traffic situation prediction method in internet-connected intelligent traffic system | |
| WO2021042401A1 (en) | Method and device for traffic light control | |
| CN116386317A (en) | Traffic intersection traffic flow prediction method and device | |
| CN115759329A (en) | Scenic spot tourist number prediction method based on convolutional neural network | |
| CN112308434A (en) | Traffic safety risk assessment method and system | |
| Bouriachi et al. | Traffic Signal Control Model on Isolated Intersection Using Reinforcement Learning: A Case Study on Algiers City, Algeria. | |
| Kareem et al. | Integrated tripartite modules for intelligent traffic light system | |
| CN113823096B (en) | Random traffic flow obstacle object arrangement method for simulation test | |
| CN115578854B (en) | Method, device, equipment and storage medium for predicting queuing length of toll station vehicle | |
| Wu et al. | The Optimal Path Planning for Emergency Vehicles Based on Path Saturation Analysis | |
| CN114626292A (en) | Signal control intersection queuing length estimation method based on ensemble learning | |
| UDOFIA | ANFIS-based Intelligent Traffic Signal Control of Two Interconnected Junctions |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |