A signal lamp countdown prediction method based on multi-source data fusion

Abstract

The application relates to a signal lamp countdown prediction method based on multi-source data fusion, which comprises the following steps: step 101: extracting characteristic values based on signal system historical data; step 102: constructing a signal lamp cycle prediction model based on signal system historical data; step 103: performing intersection signal lamp data calculation based on trajectory data; step 104: constructing a signal lamp cycle prediction model based on trajectory data; step 105: constructing a fusion prediction model, adopting a model fusion algorithm to comprehensively operate and fuse the prediction results of steps 102 and 104, and generating a final signal lamp change prediction result; step 106: optimizing and adjusting the prediction model according to real-time data; and step 107: using the optimized prediction model for real-time countdown data prediction. The application can improve the traffic signal lamp cycle and countdown prediction precision, so that the traffic flow can be accurately controlled in traffic operation.

Description

Technical Field

[0001] This invention relates to the technical field of traffic light countdown prediction, and in particular to a traffic light countdown prediction method based on multi-source data fusion. Background Technology

[0002] Traffic lights are signals that control the movement and timing of different phases at intersections. With advancements in smart technology, many navigation maps, such as Gaode Maps and Baidu Maps, provide information on the status and location of traffic lights, as well as countdown timers. These countdowns help vehicles and pedestrians better plan their routes, thereby reducing traffic congestion, improving traffic efficiency, and enhancing traffic safety.

[0003] There are two existing methods for displaying traffic light data:

[0004] 1. Cooperate with the traffic police department to directly obtain the traffic light status and countdown data from the traffic control system. This method requires the traffic control brand's equipment to support the real-time push of traffic light status and countdown data, and to ensure the accuracy of the traffic light data.

[0005] 2. Traffic light prediction based on trajectory data: Navigation providers like Gaode Maps use the distance between the user and traffic lights and vehicle speed information from car navigation systems as basic data. They then use data mining to predict the duration and timing of red and green lights. This method requires a large amount of real-time data, and its accuracy is difficult to guarantee when users do not have navigation software running or when there are few users.

[0006] Due to the instability of navigation data, using traffic signal data for display is more stable. However, the traffic light countdown data pushed out by the traffic control system suffers from inaccurate values ​​and offline traffic signal equipment, making it unreliable to display countdown data externally. To achieve stable and accurate display of data on the navigation map, this solution adopts a multi-source data fusion approach, combining traffic control platform data with trajectory data to achieve accurate prediction of traffic light countdowns. Summary of the Invention

[0007] To address the aforementioned technical problems, the purpose of this invention is to provide a traffic light countdown prediction method based on multi-source data fusion.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A traffic light countdown prediction method based on multi-source data fusion.

[0010] Includes the following steps:

[0011] Step 101: Extract feature values ​​based on historical data of the signal system;

[0012] Step 102: Construct a traffic light cycle prediction model based on historical data of the signal system. By extracting features from historical signal data, construct an ARMA time series prediction model.

[0013] Step 103: Calculate traffic light data at intersections based on trajectory data;

[0014] Step 104: Construct a traffic light cycle prediction model based on trajectory data;

[0015] Step 105: Construct a fusion prediction model and use a model fusion algorithm to perform comprehensive calculations and fusion of the prediction results from Step 102 and Step 104 to generate the final traffic light change prediction results.

[0016] Step 106: Optimize and adjust the prediction model based on real-time data;

[0017] Step 107: Use the optimized prediction model to predict real-time countdown data.

[0018] Preferably, in the traffic light countdown prediction method based on multi-source data fusion, the feature value extraction in step 101 is based on historical traffic light data and uses a CART decision tree model to extract the feature values ​​affecting the traffic light cycle.

[0019] Includes the following steps:

[0020] 1) Define the light states of the traffic lights: green light, flashing green light, yellow light, flashing yellow light, and red light. During the model building process, flashing green light is regarded as the green light state, while flashing yellow light is regarded as a separate state.

[0021] 2) Define the state cycle of the traffic lights:

[0022] [Green light cycle] Green -> Yellow; [Yellow light cycle] Yellow -> Red; [Red light cycle] Red -> Green;

[0023] 3) Based on the changes in the phase of the intersection lights, record the time point of each light state and calculate the duration of each light state. Store the phase cycle data obtained from each calculation.

[0024] 4) The calculation of the intersection phase cycle is completed when the intersection phase completes one cycle in the same light state;

[0025] 5) Training the CART decision tree

[0026] The dataset is divided into a training set and a test set by random sampling. 70% of the data is used as the training set and the remaining data is used as the test set. A CART decision tree model is trained using the training set data.

[0027]

[0028] Where D represents the training dataset and Y represents the class set. This represents the proportion of samples of class k in the training dataset;

[0029] Let the training set of the node be D. Calculate the Gini coefficient of the existing feature attributes for the dataset. According to the value of feature attribute A, divide the training set into two categories, namely, according to feature attribute A = off-peak period, divide the training set into A = off-peak period and A ≠ off-peak period, denoted as D1 and D2. Calculate the Gini coefficient of D1 and D2 and the Gini coefficient of attribute A.

[0030] Among all possible attributes A and all their possible split points a, select the feature attribute with the smallest Gini coefficient and its corresponding split point as the optimal feature attribute and optimal split point. Based on the optimal feature and optimal split point, generate two child nodes from the current node, distribute the training set to the two child nodes according to the feature attribute, and recursively call the above steps on the two child nodes until the stopping condition is met.

[0031] 6) Obtain feature values. After training, the decision tree model calculates the importance of each feature. Feature importance indicates the degree to which the feature contributes to the model's predictive ability. Features with higher importance are selected based on feature importance.

[0032] Preferably, in the traffic light countdown prediction method based on multi-source data fusion, the stopping condition is: no more feature attributes or the number of samples in the node is less than a predetermined threshold or the Gini coefficient in the training set is less than a predetermined threshold.

[0033] Preferably, in the traffic light countdown prediction method based on multi-source data fusion, the feature importance includes intersection type, light status, control mode, time characteristics, and traffic flow characteristic values.

[0034] Preferably, the traffic light countdown prediction method based on multi-source data fusion is described above.

[0035] Constructing a traffic light cycle prediction model

[0036] Based on the feature values ​​extracted in step 101, the random forest and moving average models are used together to construct a prediction model, including the following steps:

[0037] 1) Divide the dataset into a training set and a test set, with 60% of the data used as training data and 40% of the data used as the test set;

[0038] 2) Use a moving average model to fit the training set to obtain the trend of periodic changes. The periodic model is mainly affected by the intersection type and time feature values.

[0039]

[0040] in, These are predicted values, where m represents the moving average window size, and y... i This represents the actual value of time i;

[0041] 3) Use a random forest model to fit the training set to obtain the trend of non-periodic changes. The non-periodic model is mainly affected by the control mode and traffic flow characteristic values.

[0042]

[0043] in, f(x) represents the predicted value at time t. t ) represents the prediction result of the random forest regression model, x t Represents historical data and characteristic values ​​for time t;

[0044] 4) Add the results obtained in steps 2) and 3) to obtain the final prediction result;

[0045]

[0046] in, This represents the predicted value for time t.

[0047] Preferably, the traffic light countdown prediction method based on multi-source data fusion, which calculates the intersection cycle based on trajectory data, includes the following steps:

[0048] 1) Match the trajectory data with the map data to map the vehicle trajectory to the corresponding roads and intersections;

[0049] 2) Classify the trajectory data, using trajectory direction as the main classification element;

[0050] 3) For each category of vehicle trajectory points, sort them in chronological order;

[0051] 4) For each category of vehicle trajectory points, calculate the speed of each point, which is the distance between two adjacent points divided by the time difference;

[0052] 5) For each category of vehicle trajectory points, find the points where the speed is zero or close to zero, i.e., the points near the stop line;

[0053] 6) For each category of vehicle trajectory points, find the point where the speed changes from zero or close to zero to a positive value, i.e. the point where the vehicle begins to pass through the intersection;

[0054] 7) For each category of vehicle trajectory points, find the point where the speed changes from a positive value to zero or close to zero, i.e. the point where the vehicle ends at the intersection.

[0055] 8) Calculate the duration of the traffic light cycle using the timestamp information corresponding to the trajectory points.

[0056] Preferably, the traffic light countdown prediction method based on multi-source data fusion is described above.

[0057] A traffic light cycle prediction model is constructed based on trajectory data. Using historical trajectory cycle data as a foundation, a double exponential smoothing method is employed to build the model. The double exponential smoothing formula is as follows:

[0058] S i =αx i +(1-α)(S i-1 +t i-1 )

[0059] Among them, S i It is the smoothed signal light cycle value for the i-th cycle time, which is based on historical signal light cycle data x1, x2, ... x. i x is calculated using the smoothing coefficient α; i This is the actual traffic light cycle value within this cycle time.

[0060] t i =β(S i -S i-1 )+(1-β)t i-1

[0061] t i This is the smoothed trend value of the traffic light cycle over the i-th cycle time. It is based on historical traffic light cycle data x1, x2, ... x i α and β are calculated from the smoothing coefficient β, which reflects the direction and amplitude of the change in the traffic light cycle. α and β are smoothing coefficients with values ​​between [0, 1].

[0062] The prediction formula for the double exponential smoothing method is as follows:

[0063] x i+h =S i +ht i

[0064] Where, x i+hThis is the predicted signal light cycle time for the h-th cycle in the future, based on the current smoothing value S. i and trend value t i And the predicted compensation h is calculated.

[0065] Preferably, the traffic light countdown prediction method based on multi-source data fusion constructs a fusion prediction model, including the following steps:

[0066] 1) The periodic prediction models based on historical data of the information control system and trajectory data are integrated, and a binary tree approach is used to select different prediction models for periodic prediction based on different feature values;

[0067] 2) Based on the phase cycle prediction, the traffic light countdown is predicted according to the static information of the intersection phase and direction. The prediction includes the green light countdown of the current operating phase and the red light countdown of other phases.

[0068] 3) Match the intersection phase cycle prediction results based on the current intersection number;

[0069] 4) Calculation of green light countdown for the current phase: Take the current phase light cycle directly. The time period from the start of the phase cycle to the start of the yellow light cycle is the duration of the green light countdown.

[0070] Green light countdown = Phase cycle – Yellow light duration – All-red light duration

[0071] a. When the current phase is green, consider the case where consecutive phases have the same direction. If so, the green light countdown for the same direction needs to be accumulated.

[0072] Green light countdown = current phase cycle + (next phase cycle – yellow light duration – all-red light duration);

[0073] b. When the current phase is green, consider the case where consecutive phases exist in the same direction. If the direction exists in every phase of the entire cycle, then the direction is always green, and there is no need to calculate the traffic light countdown.

[0074] 5) Calculation of red light countdown for other phases: Starting from the next phase after the current phase, calculate the red light countdown one by one. The red light countdown is the duration from the start time of the current phase to the start time of the phase in which the red light countdown is calculated.

[0075] Red light countdown = Σ phase period

[0076] When other phases are in the same continuous direction, the countdown of the red light in the same direction should be calculated using the countdown of the previous phase.

[0077] Preferably, the traffic light countdown prediction method based on multi-source data fusion, which optimizes the prediction model using real-time data, includes the following steps:

[0078] 1) The prediction model incorporates real-time trajectory data and traffic light status data into its calculations to perform real-time corrections on the prediction cycle model;

[0079] 2) When the traffic light prediction cycle is inconsistent with the actual trajectory data, the actual trajectory data will be used as the primary feedback, and the model will stop predicting in the next round and correct the erroneous prediction model.

[0080] By means of the above-described solution, the present invention has at least the following advantages:

[0081] This invention improves the accuracy of traffic light cycle and countdown prediction through various prediction methods, enabling precise control of traffic flow and alleviating traffic congestion.

[0082] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0083] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0084] Figure 1 This is a flowchart of the traffic light countdown prediction method of the present invention;

[0085] Figure 2 This is a flowchart of the construction of the fusion prediction model of the present invention;

[0086] Figure 3 This is a traffic diagram illustrating the traffic light countdown of the present invention. Detailed Implementation

[0087] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0088] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0089] Example

[0090] like Figures 1 to 3 As shown, a traffic light countdown prediction method based on multi-source data fusion includes the following steps:

[0091] Step 101: Extract feature values ​​based on historical data of the signal system;

[0092] Step 102: Construct a traffic light cycle prediction model based on historical data of the signal system. By extracting features from historical signal data, construct an ARMA (Autoregressive Moving Average) time series prediction model.

[0093] Step 103: Calculate traffic light data at intersections based on trajectory data;

[0094] Step 104: Construct a traffic light cycle prediction model based on trajectory data;

[0095] Step 105: Construct a fusion prediction model and use a model fusion algorithm to perform comprehensive calculations and fusion of the prediction results from Step 102 and Step 104 to generate the final traffic light change prediction results.

[0096] Step 106: Optimize and adjust the prediction model based on real-time data;

[0097] Step 107: Use the optimized prediction model to predict real-time countdown data.

[0098] The feature value extraction in step 101 of this invention is based on historical traffic light data and uses a CART decision tree model to extract the feature values ​​that affect the traffic light cycle.

[0099] Includes the following steps:

[0100] 1) Define the light states of the traffic lights: green light, flashing green light, yellow light, flashing yellow light, and red light. During the model building process, flashing green light is regarded as the green light state, while flashing yellow light is regarded as a separate state.

[0101] 2) Define the state cycle of the traffic lights:

[0102] [Green light cycle] Green -> Yellow; [Yellow light cycle] Yellow -> Red; [Red light cycle] Red -> Green;

[0103] 3) Based on the changes in the phase of the intersection lights, record the time point of each light state and calculate the duration of each light state. Store the phase cycle data obtained from each calculation.

[0104] 4) The calculation of the intersection phase cycle is completed when the intersection phase completes one cycle in the same light state;

[0105] 5) Training the CART decision tree

[0106] The dataset is divided into a training set and a test set by random sampling. 70% of the data is used as the training set and the remaining data is used as the test set. A CART decision tree model is trained using the training set data.

[0107]

[0108] Where D represents the training dataset and Y represents the class set. This represents the proportion of samples of class k in the training dataset;

[0109] Let the training set of the node be D. Calculate the Gini coefficient of the existing feature attributes for the dataset. According to the value of feature attribute A, divide the training set into two categories, namely, according to feature attribute A = off-peak period, divide the training set into A = off-peak period and A ≠ off-peak period, denoted as D1 and D2. Calculate the Gini coefficient of D1 and D2 and the Gini coefficient of attribute A.

[0110] Among all possible attributes A and all their possible split points a, select the feature attribute with the smallest Gini coefficient and its corresponding split point as the optimal feature attribute and optimal split point. Based on the optimal feature and optimal split point, generate two child nodes from the current node, distribute the training set to the two child nodes according to the feature attribute, and recursively call the above steps on the two child nodes until the stopping condition is met.

[0111] 6) Obtain feature values. After training, the decision tree model calculates the importance of each feature. Feature importance indicates the degree to which the feature contributes to the model's predictive ability. Features with higher importance are selected based on feature importance.

[0112] The stopping conditions are: no more feature attributes or the number of samples in the node is less than a predetermined threshold or the Gini coefficient in the training set is less than a predetermined threshold; the feature importance includes intersection type, light status, control mode, time features, and traffic flow feature values.

[0113] In this invention, a traffic light cycle prediction model (prediction model 1) is constructed.

[0114] Based on the feature values ​​extracted in step 101, the random forest and moving average models are used together to construct a prediction model, including the following steps:

[0115] 1) Divide the dataset into a training set and a test set, with 60% of the data used as training data and 40% of the data used as the test set;

[0116] 2) Use a moving average model to fit the training set to obtain the trend of periodic changes. The periodic model is mainly affected by the intersection type and time feature values.

[0117]

[0118] in, These are predicted values, where m represents the moving average window size, and y... i This represents the actual value of time i;

[0119] 3) Use a random forest model to fit the training set to obtain the trend of non-periodic changes. The non-periodic model is mainly affected by the control mode and traffic flow characteristic values.

[0120]

[0121] in, f(x) represents the predicted value at time t. t ) represents the prediction result of the random forest regression model, x t Represents historical data and characteristic values ​​for time t;

[0122] 4) Add the results obtained in steps 2) and 3) to obtain the final prediction result;

[0123]

[0124] in, This represents the predicted value for time t.

[0125] The intersection cycle calculation based on trajectory data in this invention includes the following steps:

[0126] 1) Match the trajectory data with the map data to map the vehicle trajectory to the corresponding roads and intersections;

[0127] 2) Classify the trajectory data, using trajectory direction as the main classification element;

[0128] 3) For each category of vehicle trajectory points, sort them in chronological order;

[0129] 4) For each category of vehicle trajectory points, calculate the speed of each point, which is the distance between two adjacent points divided by the time difference;

[0130] 5) For each category of vehicle trajectory points, find the points where the speed is zero or close to zero, i.e., the points near the stop line;

[0131] 6) For each category of vehicle trajectory points, find the point where the speed changes from zero or close to zero to a positive value, i.e. the point where the vehicle begins to pass through the intersection; these points are the trajectory points of the first vehicle.

[0132] 7) For each category of vehicle trajectory points, find the point where the speed changes from a positive value to zero or close to zero, i.e. the point where the vehicle ends at the intersection; these points are the trajectory points of the last vehicle.

[0133] 8) Calculate the duration of the traffic light cycle using the timestamp information corresponding to the trajectory points.

[0134] In this invention, a traffic light cycle prediction model (prediction model 2) is constructed based on trajectory data. Using historical trajectory cycle data as a foundation, a double exponential smoothing method is employed to construct the traffic light cycle prediction model based on trajectory data. The double exponential smoothing formula is as follows:

[0135] S i =αx i +(1-α)(S i-1 +t i-1 )

[0136] Among them, S i It is the smoothed signal light cycle value for the i-th cycle time, which is based on historical signal light cycle data x1, x2, ... x. i x is calculated using the smoothing coefficient α; i This is the actual traffic light cycle value within this cycle time.

[0137] t i =β(S i -S i-1 )+(1-β)t i-1

[0138] t i This is the smoothed trend value of the traffic light cycle over the i-th cycle time. It is based on historical traffic light cycle data x1, x2, ... x i α and β are calculated from the smoothing coefficient β, which reflects the direction and amplitude of the change in the traffic light cycle. α and β are smoothing coefficients with values ​​between [0, 1].

[0139] The prediction formula for the double exponential smoothing method is as follows:

[0140] x i+h =S i +ht i

[0141] Where, x i+h This is the predicted signal light cycle time for the h-th cycle in the future, based on the current smoothing value S. i and trend value t i And the predicted compensation h is calculated.

[0142] like Figure 2 and Figure 3 As shown, constructing a fusion prediction model includes the following steps:

[0143] 1) The periodic prediction models based on historical data of the information control system and trajectory data are integrated, and a binary tree approach is used to select different prediction models for periodic prediction based on different feature values;

[0144] 2) Based on the phase cycle prediction, the traffic light countdown is predicted according to the static information of the intersection phase and direction. The prediction includes the green light countdown of the current operating phase and the red light countdown of other phases.

[0145] 3) Match the intersection phase cycle prediction results based on the current intersection number;

[0146] 4) Calculation of green light countdown for the current phase: Take the current phase light cycle directly. The time period from the start of the phase cycle to the start of the yellow light cycle is the duration of the green light countdown.

[0147] Green light countdown = Phase cycle – Yellow light duration – All-red light duration

[0148] a. When the current phase is green, consider the case where consecutive phases have the same direction. If so, the green light countdown for the same direction needs to be accumulated.

[0149] Green light countdown = current phase cycle + (next phase cycle – yellow light duration – all-red light duration);

[0150] For example, during time period 1 of phase 1, the green light countdown for direction B = phase 1 cycle + (phase 2 cycle - phase 2 yellow light duration - phase 2 all-red light duration)

[0151] b. When the current phase is green, consider the case where consecutive phases exist in the same direction. If the direction exists in every phase of the entire cycle, then the direction is always green, and there is no need to calculate the traffic light countdown.

[0152] 5) Calculation of red light countdown for other phases: Starting from the next phase after the current phase, calculate the red light countdown one by one. The red light countdown is the duration from the start time of the current phase to the start time of the phase in which the red light countdown is calculated.

[0153] Red light countdown = Σ phase period

[0154] When other phases are in the same continuous direction, the countdown of the red light in the same direction should be calculated using the countdown of the previous phase.

[0155] For example, during time period 3 of phase 3, the red light countdown for direction A is equal to the period of phase 3.

[0156] Real-time data optimization prediction model

[0157] 1) The prediction model incorporates real-time trajectory data and traffic light status data into its calculations to perform real-time corrections on the prediction cycle model;

[0158] 2) When the traffic light prediction cycle is inconsistent with the actual trajectory data, the actual trajectory data will be used as the primary feedback, and the model will stop predicting in the next round and correct the erroneous prediction model.

[0159] For example, such as Figure 3 As shown in Table 1, during the current operating phase 3 time period 3, the red light countdown for direction A is equal to the period of phase 3.

[0160] Table 1

[0161] Phase 1 Phase 2 Phase 3 Time period Direction B\C Direction A\B Direction D Period 1 Green light red light red light Period 2 red light Green light red light Time period 3 red light red light Green light

[0162] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0163] In the description of this application, it should be noted that the terms "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product is in use. They are used only for the convenience of describing this application and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0164] Furthermore, terms such as "horizontal" and "vertical" do not imply that components must be absolutely horizontal or vertical, but rather that they can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0165] In the description of this application, it should also be noted that, unless otherwise expressly specified and limited, the terms "set up," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0166] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A signal light countdown prediction method based on multi-source data fusion, Characterized in that, The method comprises the following steps: Step 101: extracting feature values based on signal system historical data; Step 102: constructing a signal light cycle prediction model based on signal system historical data, and constructing an ARMA time series prediction model through the features extracted from the historical signal data; Step 103: performing intersection signal light data calculation based on trajectory data; Step 104: constructing a signal light cycle prediction model based on trajectory data; Step 105: constructing a fusion prediction model, and performing comprehensive operation and fusion on the prediction results of steps 102 and 104 by using a model fusion algorithm to generate a final signal light change prediction result; Step 106: optimizing and adjusting the prediction model according to real-time data; Step 107: using the optimized prediction model for real-time countdown data prediction; The signal light cycle prediction model based on trajectory data is constructed based on historical trajectory cycle data, and a double exponential smoothing method is used to construct the signal light cycle prediction model based on trajectory data, wherein the double exponential smoothing formula is as follows: S i =αx i +(1-α)(S i-1 +t i-1 ) where S i is the smoothed signal cycle value at the i-th cycle time, which is calculated according to the historical signal cycle data x1, x2,.... x i and the smoothing coefficient a; x i is the actual signal cycle value at the cycle time. t i = β (S i - S i-1 ) + (1 - β) t i-1 t i is the smoothed signal cycle trend value in the i th cycle time, which is calculated according to historical signal cycle data x1, x2,.... xi and smoothing coefficient β, and reflects the change direction and amplitude of the signal cycle; α and β are smoothing coefficients, and the values are between 0 and 1. i is the smoothed signal cycle trend value in the i th cycle time, which is calculated according to historical signal cycle data x1, x2,.... xi and smoothing coefficient β, and reflects the change direction and amplitude of the signal cycle; α and β are smoothing coefficients, and the values are between 0 and 1. The prediction formula of the double exponential smoothing method is as follows: x i+h =S i +ht i wherein x i+h is a traffic light cycle prediction value for a future hth cycle time, which is calculated from a current smoothed value S i and a trend value t i and a prediction offset h; The fusion prediction model is constructed, comprising the following steps: 1) fusing the cycle prediction models constructed based on the signal control system historical data and the trajectory data, and using a binary tree to select different prediction models for cycle prediction according to different feature values; 2) on the basis of phase cycle prediction, predicting the signal light countdown according to the intersection phase and direction static information, the prediction including the green light countdown of the current operating phase and the red light countdown of other phases; 3) matching the intersection phase cycle prediction result according to the current intersection number; 4) green light countdown calculation of the current phase: directly taking the current phase light state cycle, and the time period from the beginning of the phase cycle to the beginning of the yellow light state as the time length of the green light countdown; Green light countdown = phase cycle - yellow light time length - all red time length a. When the current phase is green light, consider the case that the continuous phase exists in the same direction, and if it exists, the green light countdown of the same direction needs to be accumulated; Green light countdown = current phase cycle + (next phase cycle - yellow light time length - all red time length); b. When the current phase is green light, consider the case that the continuous phase exists in the same direction, and if the direction exists in each phase in the entire cycle, the direction is always green and the signal light countdown does not need to be calculated; 5) red light countdown calculation of other phases: starting from the next phase of the current phase, the red light countdown is calculated one by one, and the red light countdown is the time length from the start time of the current phase to the start time of the calculated red light countdown phase; Red light countdown = Σ phase cycle If the other phases exist in the same direction continuously, the red light countdown of the same direction needs to be calculated by using the red light countdown of the previous phase.

2. The signal countdown prediction method based on multi-source data fusion according to claim 1, characterized in that: The feature value extraction in step 101 is based on signal light historical data, and a CART decision tree model is used to extract the feature values affecting the signal light cycle, Comprising the following steps: 1) setting the light states of the signal light: green light, green flash, yellow light, yellow flash and red light, and in the model construction process, green flash is regarded as a green light state, and yellow flash is regarded as a separate state; 2) Define the signal light state cycle: [Green light cycle] Green -> Yellow; [Yellow light cycle] Yellow -> Red; [Red light cycle] Red -> Green; 3) According to the change of the light state of the intersection phase, record the time point of each signal light state in turn, and calculate the duration of each light state. The phase cycle data obtained each time is stored; 4) The calculation of the intersection phase cycle is completed when the intersection phase runs a complete cycle in the same light state; 5) Train the CART decision tree Divide the data set into training set and test set, realize by random sampling method, extract 70% data as training set, the remaining data as test set, use training set data to train a CART decision tree model; wherein D represents a training data set, Y represents a category set, represents the proportion of the kth category sample in the training data set; Set the training set of the node as D, calculate the Gini coefficient of the existing feature attribute on the data set, according to the value of the feature attribute A, the training set is divided into two categories, that is, according to the feature attribute A = flat peak period, the training set is divided into A = flat peak period and A ≠ flat peak period, recorded as D1, D2, calculate the Gini coefficient of D1, D2 and the Gini coefficient of attribute A; Among all possible attributes A and all possible split points a, select the feature attribute with the smallest Gini coefficient and the corresponding split point as the optimal feature attribute and the optimal split point, and generate two child nodes from the current node according to the optimal feature and the optimal split point. The training set is allocated to two child nodes according to the feature attribute, and the above steps are recursively called for the two child nodes until the stop condition is met; 6) Obtain feature value, after training is completed, the decision tree model calculates the importance of each feature, the feature importance represents the contribution degree of the feature to the prediction ability of the model, and the features with high importance are selected according to the feature importance.

3. The signal countdown prediction method based on multi-source data fusion according to claim 2, characterized in that: The stop condition: there are no more feature attributes or the number of samples in the node is less than a predetermined threshold or the Gini coefficient in the training set is less than a predetermined threshold.

4. The signal countdown prediction method based on multi-source data fusion according to claim 2, characterized in that: The feature importance includes intersection type, light state, control mode, time feature, and vehicle flow feature.

5. The signal light countdown prediction method based on multi-source data fusion according to claim 1, characterized in that: constructing a signal light cycle prediction model Based on the feature values extracted in step 101, the random forest and moving average model are used to construct the prediction model, including the following steps: 1) Divide the data set into training set and test set, 60% of the data as training data, 40% of the data as test set; 2) Use the moving average model to fit the training set to obtain the periodic trend, and the periodic model is mainly affected by the intersection type, time feature value; wherein, is the predicted value, m represents the moving average window size, y i represents the actual value at time i; 3) Use the random forest model to fit the training set to obtain the non-periodic trend, and the non-periodic model is mainly affected by the control mode and vehicle flow feature value; wherein, represents a prediction value of time t, represents a prediction result of the random forest regression model, represents historical data and feature values of time t; 4) Add the results obtained in steps 2) and 3) to obtain the final prediction result; wherein, represents a prediction value of time t.

6. The signal countdown prediction method based on multi-source data fusion according to claim 1, characterized in that: Based on the trajectory data, the intersection cycle is calculated, including the following steps: 1) Match the trajectory data with the map data, and map the vehicle trajectory to the corresponding road and intersection; 2) Classify the trajectory data according to the trajectory direction as the main element of classification; 3) For each classified vehicle driving trajectory point, sort them in time sequence; 4) For each classified vehicle driving trajectory point, calculate the speed of each point, which is the distance between two adjacent points divided by the time difference; 5) For each classified vehicle driving trajectory point, find the point with speed of zero or close to zero, which is the point near the stop line; 6) For each classified vehicle driving trajectory point, find the point where the speed changes from zero or close to zero to positive, which is the point where the vehicle starts to pass through the intersection; 7) For each classified vehicle driving trajectory point, find the point where the speed changes from positive to zero or close to zero, which is the point where the vehicle ends to pass through the intersection; 8) Calculate the signal light cycle length through the time stamp information of the trajectory points.

7. The signal countdown prediction method based on multi-source data fusion according to claim 1, characterized in that: Real-time data optimization prediction model, including the following steps: 1) The prediction model includes real-time trajectory data and signal light state data in the prediction model calculation, and corrects the prediction period model in real time; 2) When the signal light prediction period is inconsistent with the actual trajectory data, the actual trajectory data feedback is given priority, and at the same time the model will stop predicting in the next round and correct the wrong prediction model.

Images