Intersection signal status prediction method and system based on intelligent connected vehicle group perception

Abstract

The present invention relates to a method and system for predicting the short-term signal status of intersections based on the perception of intelligent connected vehicle groups. The signal status prediction method is based on the collected traffic light status of the intelligent connected vehicle group perception data, and the traffic light status is input into the prediction model, including the identification of the signal control scheme of a single intersection; the prediction of the signal status of a single intersection; and the joint prediction of the signal status of multiple intersections. The subsequent short-term traffic signal status is adaptively predicted and inferred by a machine learning method. The present invention provides different prediction and inference processes for the two modes of independent control of a single intersection and collaborative control of multiple intersections, respectively, and is applicable to various scenarios. It provides a reliable short-term traffic signal status data foundation for the planning and decision-making of intelligent connected vehicle groups.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic signal state prediction and inference, and in particular to a method and system for predicting intersection signal state based on perception of a group of intelligent connected vehicles. Background Art

[0002] With the rapid development of technologies such as vehicle networking, single-vehicle intelligence, and multi-vehicle collaboration, the coordinated control of intelligent connected vehicle swarms will become a future trend. Under certain road conditions, intelligent connected vehicles adapt and adjust their driving state to maintain a relatively stable position and speed relative to other intelligent connected vehicles, thereby achieving coordinated control of intelligent connected vehicle swarms. A prerequisite for achieving coordinated control of intelligent connected vehicle swarms is strong vehicle perception capabilities, which necessitates the installation of sensors such as cameras, radar, and ultrasonic sensors. Combined with roadside sensing equipment and communication technologies, these vehicles can capture multiple sources of information, including the vehicle's surroundings, blind spot information, surrounding vehicles, and traffic light status.

[0003] Traffic signal status is crucial information for planning and decision-making in intelligent connected vehicle swarms. When a single vehicle perceives a real-time traffic signal, it can be obstructed by preceding vehicles and roadside infrastructure. This makes it difficult to accurately infer real-time traffic signals based solely on single-vehicle perception. Furthermore, intelligent connected vehicle swarm planning and decision-making also require consideration of future conditions, such as whether the remaining green light time is sufficient for the swarm to pass and whether control instructions to reduce energy consumption should be executed before the next green light is illuminated.

[0004] Existing technologies have already achieved relevant results in how to obtain real-time signal status, but most of them are technologies for independent control scenarios of a single intersection, and there are few technologies for predicting short-term signal status. Obtaining real-time signal status through intersection signals is very difficult, but intelligent connected vehicles with powerful perception capabilities provide a new approach. Existing intersection signal status data is stored by signal manufacturers, and different signal manufacturers have different data storage formats and communication protocols, which makes it difficult to transmit real-time signal status. On the other hand, traffic signal control has different modes of independent control of a single intersection and coordinated control of multiple intersections, and traffic signal control schemes are updated after a period of time, making it particularly difficult to predict traffic signal status. In addition, the storage space of signal equipment is limited, and its historical data is insufficient to support data-driven short-term signal status prediction. To address the above problems, it is necessary to invent a data-driven intersection short-term signal status prediction method and system to improve the accuracy of real-time signal status inference and provide short-term signal status suitable for intelligent connected vehicles. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method and system for predicting the intersection signal status based on the perception of intelligent connected vehicle groups.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] As a first aspect of the present invention, a method for predicting the short-term signal state of an intersection based on perception by a group of intelligent connected vehicles is provided. The method collects traffic light states based on perception data from a group of intelligent connected vehicles, inputs the traffic light states into a prediction model, and obtains a prediction result from the prediction model. The prediction model includes the following components:

[0008] Identification of signal control schemes for a single intersection;

[0009] Predict the signal status of a single intersection;

[0010] Joint prediction of signal status of multiple intersections.

[0011] Furthermore, the traffic light status is based on real-time perception data and is extracted from the real-time perception data transmitted by the intelligent connected vehicle group through a semantic segmentation model. The traffic light status includes whether the green light, yellow light and red light at each entrance of the intersection are on or off.

[0012] Furthermore, the identification of a single intersection signal control scheme includes the following steps:

[0013] S1: Single light duration stage division:

[0014] Based on the historical time series of traffic light states, the duration phase is divided into the start time, duration time and end time of each traffic light;

[0015] S2: Single traffic direction stage division:

[0016] Based on the duration phase output by S1, the start time, duration and interval time of the green, yellow and red lights in each direction of travel are obtained; the duration phases of the green, yellow and red lights in each direction of travel are merged into the merged phase of that direction of travel; the discontinuous data caused by the missing yellow light is processed; and the phase results of the travel direction in each direction are divided according to the order of green, yellow and red lights.

[0017] S3: Signal cycle division:

[0018] Based on the single traffic direction stage output by S2, the stages of all traffic directions of a single intersection are merged and arranged in chronological order to find the largest set of the same traffic direction. The elements between the rows with the same traffic direction in adjacent chronological order in the set constitute a signal cycle.

[0019] Furthermore, the step of dividing the single traffic direction into stages includes:

[0020] Each time, the data corresponding to one direction is taken from the 12 traffic directions of a single intersection, and the duration of the green light, yellow light and red light in that direction is merged into the merged phase of that traffic direction, and then sorted in ascending order according to the start time timestamp;

[0021] If the first row of the merged phase for the traffic direction is not a green light, select the moment when the first green light appears and remove the lighting phase data before that moment; if the last row is not a red light, select the moment when the last red light appears and remove the continuous phase data after that moment;

[0022] Starting from the first green light, yellow light, and red light, extract data every 2 lines; if the data extracted is not the corresponding green light, yellow light, or red light, record the index of the data and jump out of the loop;

[0023] If the yellow light index is less than the green light index and the yellow light index is less than the red light index, the yellow light is considered missing. A new row is constructed and added to the duration phase. The start time timestamp of this row is equal to the start time timestamp corresponding to the yellow light index, the duration is 0, and the interval time is 0. The row is re-sorted in ascending order by the start time timestamp, and a first-order difference is performed on the start time timestamp column to obtain the interval time between each row. Three rows of data are taken each time, and a new row containing the start time, duration, and interval time of each of the three light colors is constructed and added to the directional phase division result set.

[0024] Output the phase division results for single-traffic direction.

[0025] Furthermore, the single intersection signal status prediction includes the following steps:

[0026] S4: Current signal status identification

[0027] Input the real-time status time series of the signal light, identify the current signal status, and output the real-time duration of the light color and the lighting start time;

[0028] S5: Signal period matching

[0029] Plan 1: refers to the most recent complete signal control plan before the real-time transmission moment. Plan 2: refers to the most recent complete signal control plan before Plan 1;

[0030] If the phases of Scheme 1 and Scheme 2 are the same, Scheme 1 is used as the predicted value and Scheme 2 is used as the true value, and the error between the two is calculated;

[0031] Different levels of credibility are divided according to the size of the error value;

[0032] When the credibility is high, Scheme 1 is used as the signal control scheme Cs for subsequent inference; when the credibility is low, the gradient descent decision tree model is used to give the best signal control scheme Cs.

[0033] S6: Subsequent signal state inference

[0034] Based on the credibility, signal control scheme and current signal status, considering the overlap phase scenario, the signal phase corresponding to the real-time signal status in the signal control scheme is obtained, and the remaining lighting time and next phase of the lighting lamp are inferred based on the signal phase.

[0035] Furthermore, the error includes but is not limited to indicators such as the absolute deviation ε1 of the cycle duration and the maximum value ε2 of the absolute deviation of the duration of each light color in each traffic direction within a cycle.

[0036] Specifically, the absolute deviation of the cycle duration

[0037]

[0038] Among them: T and are the actual and predicted durations of the cycle, in seconds;

[0039] The maximum absolute deviation of the duration of each light color in each direction of traffic within a cycle

[0040]

[0041] Where: t i,j is the duration of the i-th traffic direction light color j being on in this cycle, Then t i,j The predicted value of , in seconds;

[0042] The calculation formula of the credibility is:

[0043] Credibility = (the proportion of errors less than threshold 1 + the proportion of errors less than threshold 2) * 0.5.

[0044] Furthermore, the gradient descent decision tree model is obtained by training data of all low-credibility scenarios in history, with its input being the current signal state and its output being a number according to the signal control scheme.

[0045] Furthermore, the joint prediction of the signal status of multiple intersections includes the following steps:

[0046] S7: Cooperative Control Intersection Identification:

[0047] By analyzing the spatiotemporal distribution of historical signal control schemes at multiple intersections, multiple intersections that adopt coordinated control are identified based on clustering methods.

[0048] S8: Collaborative control scheme formation:

[0049] Based on the cooperatively controlled intersection identified in S7, determine whether there is a green wave band for each direction of travel, and form a complete green wave cooperative signal control scheme Cm based on the green wave bands of all directions of travel;

[0050] S9: Joint inference of signal states:

[0051] Based on the S6 subsequent signal state inference, the credibility is input, and according to the signal control scheme Cs and the current signal state, the inferred phase result of the subsequent signal state is obtained. If the traffic direction of the inferred phase exists in the collaborative control signal scheme Cm, the remaining green light time of the intersection downstream of the intersection is increased, and the joint inference result is output.

[0052] Furthermore, the specific steps of determining whether a green wave band exists for each traffic direction include:

[0053] For each traffic direction, sort the intersections in ascending order according to the green light start time, and traverse the sorted intersections;

[0054] If the green light start time of an intersection is greater than the green light end time of the previous intersection, it means there is no green wave band, that is, the linear intersections are released continuously on the time axis;

[0055] Otherwise, there is a green wave band in the traversed direction.

[0056] As a second aspect of the present invention, a system for implementing the above-mentioned method for predicting the short-term signal state of an intersection based on perception of a group of intelligent connected vehicles is provided, the system comprising:

[0057] Intelligent connected vehicles acquire real-time perception data and transmit it to a cloud database via the network;

[0058] Cloud database, storing real-time perception data;

[0059] The data processing engine processes data in the cloud database, including extracting traffic light status from the real-time perception data of the intelligent connected vehicle group and training pre-pushing models; and publishing the trained pre-pushing models to each intelligent connected vehicle;

[0060] A single intelligent connected vehicle extracts the traffic light status from the shared real-time perception data and inputs it into the local prediction model to obtain the real-time intersection signal status.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] The present invention accurately collects the signal status of traffic lights based on the perception capabilities of a group of intelligent connected vehicles, avoiding collection errors caused by obstructions by vehicles and roadside facilities. For the independent control mode of a single intersection, a data-driven approach is used to capture the changing patterns of the intersection's signal control scheme, and the subsequent short-term signal status is inferred and predicted based on real-time perception data. For the collaborative control mode of multiple intersections, based on the historical signal control schemes of each intersection, a collaborative control scheme is formed by clustering and identifying the collaboratively controlled intersections. The subsequent signal status of the collaborative intersection is jointly predicted based on the inference logic of the subsequent signal status of a single intersection and the collaborative control signal scheme, providing a reliable short-term traffic signal data foundation for the planning and decision-making of intelligent connected vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flow chart of the intersection signal status prediction method based on the perception of intelligent connected vehicle groups in the present invention;

[0064] Figure 2 A flow chart for identifying a single intersection signal control scheme of the present invention;

[0065] Figure 3 A flowchart for predicting the signal status of a single intersection of the present invention;

[0066] Figure 4 A flowchart of the present invention for jointly predicting the signal states of multiple intersections;

[0067] Figure 5 Schematic diagram of the short-term signal state prediction system of the present invention. DETAILED DESCRIPTION

[0068] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0069] Example 1

[0070] As one aspect of the present invention, a method for predicting intersection signal states based on perception by intelligent connected vehicles is provided. This method performs signal state prediction based on a prediction model that includes identifying a single intersection signal control scheme, predicting a single intersection signal state, and jointly predicting the signal states of multiple intersections.

[0071] Figure 1 This is a flow chart of the prediction method. The steps of the prediction method include: collecting traffic light status based on the perception data of the intelligent connected vehicle group; inputting the traffic light status into the prediction model, and using the prediction model to identify the signal control plan of a single intersection, predict the signal status of a single intersection, and predict the signal status of multiple intersections.

[0072] Traffic light status is collected based on the semantic segmentation model of a neural network, such as YOLO, which extracts the traffic light status from the real-time perception data of the intelligent connected vehicle group, including whether the green, yellow and red lights of each entrance are on or off.

[0073] Figure 2 The flowchart of the signal control scheme identification method for a single intersection provided by an embodiment of the present invention includes the following steps: dividing the signal control scheme into a single light duration phase, dividing the signal control scheme into a single traffic direction phase, and dividing the signal cycle. The specific steps are as follows:

[0074] S1: Single light duration phase division

[0075] Input is the time series of the historical status of the traffic lights, that is, at each time point, whether the green, yellow, and red lights of each entrance are on (1) or off (0). Take the intersection as an example, it has four entrances and each entrance has three directions: left turn, straight ahead, and right turn. Therefore, there are a total of 12 traffic directions and 36 different traffic lights.

[0076] Each time, the time series data of the on or off state of a traffic light is taken out, and the first-order difference of the light data is performed. The column after the difference selects the index (row number) with a value equal to 1 or -1. If the index does not exist, skip the light. If the original value corresponding to the first index is not 1, remove the index. If the number of indexes is not a multiple of 2, remove the last index. Take the timestamp corresponding to the index in the original data at intervals of one. This timestamp is the start time of each light color. Starting from the second index, take the timestamp corresponding to the index at intervals of one. This timestamp is the end time of each light color. Subtract two adjacent timestamps to get the duration of each light. Finally, output multiple duration stages of the traffic light containing the start time, duration and end time.

[0077] S2: Single traffic direction phase division

[0078] Each time, data corresponding to one of the 12 traffic directions is collected, that is, one direction is collected from three directions (left, straight, and right) of the four directions (east, south, west, and north). The durations of the green, yellow, and red lights in that direction are combined into the combined phase for that traffic direction, and then sorted in ascending order by the start time timestamp.

[0079] If the first row of the merged phase for that direction of traffic is not a green light, select the moment of the first green light and remove the data for the lighting phase before that moment. If the last row of traffic is not a red light, select the moment of the last red light and remove the data for the continuous phase after that moment.

[0080] Starting from the first green / yellow / red light, extract data every two rows. If the data is not green / yellow / red, record the index of the data and exit the loop. If the yellow light index is less than the green light index and the yellow light index is less than the red light index, the yellow light is missing. Construct a new row with a start time timestamp equal to the start time timestamp of the yellow light index, a duration of 0, and an interval of 0. Add this new row to the duration phase.

[0081] Re-sort the data in ascending order by the start timestamp column and perform a first-order difference on the start timestamp column to obtain the interval time between each row. Take three rows of data at a time: the first row for green, the second for yellow, and the third for red. Construct a new row containing the start time, duration, and interval time for each of the three light colors. Add this new row to the stage classification result set. Finally, output the stage classification results for the 12 traffic directions.

[0082] S3: Signal cycle division

[0083] The phase division results for the 12 traffic directions are sequentially connected according to the green light start time for each direction to form an intermediate data set. Rows in the intermediate data set with the same traffic direction are found. The data between adjacent rows constitutes one signal cycle. For example, row 1 has two directions: north-left and north-straight, row 2 has one direction: south-straight, and row 6 has two directions: north-left and north-straight. Therefore, rows 1 and 6 have the same traffic direction, and rows 1 to 5 constitute one signal cycle.

[0084] Figure 3 The flowchart of the signal state prediction for a single intersection provided by an embodiment of the present invention includes the following steps: identifying the current signal state, matching the signal cycle, and inferring the subsequent signal state. The specific steps are as follows:

[0085] S4: Current signal status identification

[0086] Input the real-time status time series of the traffic light. The time range must be less than or equal to that used in S1, but the time length must be greater than two signal cycles (based on engineering experience, this can be set to 30 minutes, for example). Extract the time series data for each light's on and off state, and perform a first-order difference on that column. For the differenced column, select the index with a value equal to 1 or -1. If the index does not exist, skip the column. Select the row corresponding to the last index in the original data. If the light in that row is off, skip the column.

[0087] The start time of the light color is equal to the timestamp of the row; the duration of the light color is equal to the timestamp of the last row of the original data minus the timestamp of the row. The final output includes multiple current signal states including the start time, duration, and traffic direction.

[0088] S5: Signal period matching

[0089] Plan 1: Refers to the most recent complete signal control plan before the real-time transmission. Plan 2: Refers to the most recent complete signal control plan before Plan 1. If Plans 1 and 2 are in phase, Plan 1 is used as the predicted value and Plan 2 as the actual value. The error between the two is calculated, including but not limited to the absolute deviation ε1 of the cycle duration and the maximum absolute deviation ε2 of the duration of each light color in each direction within a cycle.

[0090] Specifically, the absolute deviation of the cycle duration

[0091]

[0092] Among them: T and They are the actual and predicted duration of the cycle, in seconds.

[0093] The maximum absolute deviation of the duration of each light color in each direction of traffic within a cycle

[0094]

[0095] Where: t i,j is the duration of the i-th traffic direction light color j being on in this cycle, Then t i,j The predicted value in seconds.

[0096] Different levels of credibility are divided according to the size of the error value, for example, credibility = (the proportion of errors less than threshold 1 + the proportion of errors less than threshold 2) * 0.5. When the credibility is high, plan 1 is used as the signal control plan Cs for subsequent inference. When the credibility is low, the gradient descent decision tree model is used to give the best signal control plan Cs. The gradient descent decision tree model is trained with data from all low-credibility scenarios in history. Its input is the current signal state, and its output is the number of the signal control plan, for example: including but not limited to dimensions such as 0 for the previous cycle, 1 for the previous day, and 2 for the previous week.

[0097] S6: Subsequent signal state inference

[0098] Input the confidence level, the signal control scheme, and the current signal state. Traverse the list of phases in the signal control scheme to obtain the phase object and phase sequence. If the phase direction corresponds to the direction in the current signal state, the current phase is that phase; otherwise, the phase is skipped.

[0099] In the overlapping phase scenario, the direction of travel will cross the phase stage. Based on the traversal of a single light, find the phase to which each light belongs. According to the found phase and the real-time lighting duration of each light, calculate the remaining lighting time and give the inferred phase result of the subsequent signal state. For example: the credibility is 95%, the current signal state is North Left Green for 15 seconds and North Straight Green for 15 seconds. According to the corresponding phase of the signal control plan, North Left Green for 30 seconds and North Straight Green for 30 seconds, then the output credibility is 95%, North Left Green for 15 seconds and North Straight Green for 15 seconds.

[0100] Figure 4 The flowchart of the joint prediction of the signal status of multiple intersections provided by the embodiment of the present invention includes the following steps: identification of cooperatively controlled intersections, formation of a cooperative control plan, and joint inference of signal status. The specific steps are as follows:

[0101] S7: Cooperative Control Intersection Identification

[0102] By analyzing the spatiotemporal distribution of historical signal control schemes for multiple intersections, we identify multiple intersections that utilize collaborative control. We input the signal control schemes for all intersections within the same time period and use clustering methods, including K-means, density clustering, and hierarchical clustering, to identify each cluster as a potential set of collaboratively controlled intersections. We search for identical intersection clusters along the timeline. When the number of clusters reaches a threshold (this threshold is dependent on the collected data and requires calibration), we consider the cluster to be a collaboratively controlled intersection within the searched time period.

[0103] S8: Collaborative control scheme formation

[0104] Based on the cooperatively controlled intersections identified in S7, the green light start time and green light end time of each traffic direction at each intersection are extracted. For each traffic direction, the green light start time is sorted in ascending order, and the sorted intersections are traversed. If there is an intersection whose green light start time is greater than the green light end time of the previous intersection, it means that there is no green wave band, that is, the linear intersections are released continuously in succession on the time axis. Otherwise, there is a green wave band in the traversed traffic direction. The green wave bands of all traffic directions form a complete green wave cooperative signal control scheme Cm. For example: the green light at intersection 1 north straight starts at 09:00 and lasts for 50 seconds, the green light at intersection 2 north straight starts at 09:01 and lasts for 60 seconds, ..., the green light at intersection n north left starts at 09:03 and lasts for 30 seconds.

[0105] S9: Joint Inference of Signal States

[0106] Based on the inference of the subsequent signal state of S6, the credibility is input, according to the signal control scheme Cs and the current signal state, and the inferred phase result of the subsequent signal state is obtained. If the direction of travel of the inferred phase exists in the collaborative control signal scheme Cm, the remaining green light time of the intersection downstream of the intersection is added, and the joint inference result is output. For example: the credibility is 95%, the current signal state of Intersection 2 is North Direct Green for 15 seconds, and the corresponding phase according to the signal control scheme Cs is North Direct Green for 30 seconds. The collaborative control signal scheme Cm is North Direct Green for 30 seconds at Intersection 1, North Direct Green for 60 seconds at Intersection 2, and North Direct Green for 60 seconds at Intersection 3. Then the output credibility is 95%, North Direct Green for Intersection 2 has 15 seconds left, and the next intersection, that is, Intersection 3, has 60 seconds left.

[0107] Example 2

[0108] As another embodiment of the present invention, a short-term signal state prediction system structure diagram is provided. Figure 5 As shown. First, the intelligent connected vehicle group transmits real-time perception data through the network, including but not limited to image and point cloud data. In addition, the cloud database stores real-time perception data. The data in the cloud database is processed by a unified data processing engine, including semantic segmentation and training of the pre-push model as described in the above embodiment. A semantic segmentation model based on a neural network, such as YOLO, extracts the traffic light status from the real-time perception data of the intelligent connected vehicle group, including whether the green, yellow and red lights of each entrance are on or off. Then, the pre-push model is published to each intelligent connected vehicle, and a single vehicle inputs the shared real-time perception data into the local pre-push model to obtain the real-time intersection signal status.

[0109] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A method for predicting the short-term signal status of an intersection based on the perception of a group of intelligent connected vehicles, characterized in that: The signal state prediction method collects traffic light states based on data sensed by a group of intelligent connected vehicles, inputs the traffic light states into a prediction model, and obtains prediction results of the prediction model. The prediction model includes the following parts: The specific steps for identifying a single intersection signal control scheme are as follows: S1: Single light duration stage division: Based on the historical time series of traffic light states, the duration phase is divided into the start time, duration time and end time of each traffic light; S2: Single traffic direction stage division: Based on the duration phase output by S1, the start time, duration, and interval of the green, yellow, and red lights in each direction are obtained; the duration phases of the green, yellow, and red lights in each direction are combined into the combined phase for that direction; the discontinuous data caused by the missing yellow light is processed; and the phase results of the direction of travel in each direction are divided according to the order of green, yellow, and red lights. S3: Signal cycle division: Based on the single traffic direction phase output by S2, merge the phases of all traffic directions of a single intersection, arrange them in chronological order, and find the largest set of the same traffic direction. The elements between the rows with the same traffic direction in the adjacent chronological order in the set constitute a signal cycle. The specific steps for predicting the signal status of a single intersection are as follows: S4: Current signal status identification Input the real-time status time series of the signal light, identify the current signal status, and output the real-time duration of the light color and the lighting start time; S5: Signal period matching Plan 1: refers to the most recent complete signal control plan before the real-time transmission moment; Plan 2: refers to the most recent complete signal control plan before Plan 1; If the phases of Scheme 1 and Scheme 2 are the same, Scheme 1 is used as the predicted value and Scheme 2 is used as the true value, and the error between the two is calculated; Different levels of credibility are divided according to the size of the error value; When the credibility is high, Scheme 1 is used as the signal control scheme Cs for subsequent inference; when the credibility is low, the gradient descent decision tree model is used to give the best signal control scheme Cs; S6: Subsequent signal state inference Based on the credibility, signal control scheme and current signal status, considering the overlap phase scenario, the corresponding signal phase of the real-time signal status in the signal control scheme is obtained, and the remaining lighting time and the next phase of the lighting lamp are inferred based on the signal phase; The specific steps for joint prediction of signal status of multiple intersections are as follows: S7: Cooperative Control Intersection Identification: By analyzing the spatiotemporal distribution of historical signal control schemes at multiple intersections, multiple intersections that adopt coordinated control are identified based on clustering methods. S8: Collaborative control scheme formation: Based on the cooperatively controlled intersection identified in S7, determine whether there is a green wave band for each direction of travel, and form a complete green wave cooperative signal control scheme Cm based on the green wave bands of all directions of travel; S9: Joint inference of signal states: Based on the S6 subsequent signal state inference, the credibility is input, and according to the signal control scheme Cs and the current signal state, the inferred phase result of the subsequent signal state is obtained. If the traffic direction of the inferred phase exists in the collaborative control signal scheme Cm, the remaining green light time of the intersection downstream of the intersection is increased, and the joint inference result is output.

2. The method for predicting the short-term signal status of an intersection based on perception of an intelligent connected vehicle group according to claim 1 is characterized in that: The traffic light status is based on real-time perception data and is extracted from the real-time perception data transmitted by the intelligent connected vehicle group through a semantic segmentation model. The traffic light status includes whether the green, yellow and red lights at each entrance of the intersection are on or off.

3. The method for predicting the short-term signal status of an intersection based on perception of an intelligent connected vehicle group according to claim 1 is characterized in that: The single traffic direction stage division step includes: Each time, the data corresponding to one direction is taken from the 12 traffic directions of a single intersection, and the duration of the green light, yellow light and red light in that direction is merged into the merged phase of that traffic direction, and then sorted in ascending order according to the start time timestamp; If the first row of the merged phase for the traffic direction is not a green light, select the moment when the first green light appears and remove the lighting phase data before that moment; if the last row is not a red light, select the moment when the last red light appears and remove the continuous phase data after that moment; Starting from the first green light, yellow light, and red light, extract data every 2 lines; if the data extracted is not the corresponding green light, yellow light, or red light, record the index of the data and jump out of the loop; If the yellow light index is less than the green light index and the yellow light index is less than the red light index, the yellow light is considered missing. A new row is constructed and added to the duration phase. The start time timestamp of this row is equal to the start time timestamp corresponding to the yellow light index, the duration is 0, and the interval time is 0. The row is re-sorted in ascending order by the start time timestamp, and a first-order difference is performed on the start time timestamp column to obtain the interval time between each row. Three rows of data are taken each time, and a new row containing the start time, duration, and interval time of each of the three light colors is constructed and added to the directional phase division result set. Output the phase division results for single-traffic direction.

4. The method for predicting the short-term signal status of an intersection based on perception of an intelligent connected vehicle group according to claim 1 is characterized in that: The error includes but is not limited to the absolute deviation of the cycle duration The maximum absolute deviation of the duration of each light color in each direction of traffic within a cycle indicators; Specifically, the absolute deviation of the cycle duration in: and are the actual and predicted durations of the cycle, in seconds; The maximum absolute deviation of the duration of each light color in each direction of traffic within a cycle in: For this cycle Traffic direction light colors The duration of lighting, Then The predicted value of , in seconds; The calculation formula of the credibility is: Credibility = (the proportion of errors less than threshold 1 + the proportion of errors less than threshold 2) * 0.

5.

5. The method for predicting the short-term signal status of an intersection based on perception of an intelligent connected vehicle group according to claim 1 is characterized in that: The gradient descent decision tree model is obtained by training data of all low-credibility scenarios in history. Its input is the current signal state, and its output is the number of the signal control scheme.

6. The method for predicting the short-term signal status of an intersection based on perception of an intelligent connected vehicle group according to claim 1 is characterized in that: The specific steps of determining whether a green wave band exists for each traffic direction include: For each traffic direction, sort the intersections in ascending order according to the green light start time, and traverse the sorted intersections; If the green light start time of an intersection is greater than the green light end time of the previous intersection, it means there is no green wave band, that is, the linear intersections are released continuously on the time axis; Otherwise, there is a green wave band in the traversed direction.

7. A system for implementing the method for predicting the short-term signal status of an intersection based on perception of an intelligent connected vehicle group as described in any one of claims 1 to 6, characterized in that: The system comprises: Intelligent connected vehicles acquire real-time perception data and transmit it to a cloud database via the network; Cloud database, storing real-time perception data; The data processing engine processes data in the cloud database, including extracting traffic light status from the real-time perception data of the intelligent connected vehicle group and training pre-pushing models; and publishing the trained pre-pushing models to each intelligent connected vehicle; A single intelligent connected vehicle extracts the traffic light status from the shared real-time perception data and inputs it into the local prediction model to obtain the real-time intersection signal status.

Images