A traffic signal fuzzy control method and system considering individual prior data

Abstract

The application relates to a traffic signal fuzzy control method considering individual prior data, which comprises the following steps: step 1, cleaning and analyzing license plate recognition data; step 2, performing support vector machine driver personalized travel time prediction based on Bayesian optimization; step 3, performing Bayesian driver personalized path prediction based on a Markov chain; step 4, fusing vehicle historical travel information and real-time detection information; and step 5, obtaining phase arrival rate and turning rate through vehicle travel time and path prediction, and constructing a traffic signal fuzzy control method. The traffic flow arrival rate calculation idea provided by the application can effectively estimate the traffic flow arrival rate under four-phase control, the calculation result can be used as a data source for traffic signal control, the average vehicle delay can be significantly reduced, the vehicle travel time can be highly accurately predicted, and the service quality of the intersection can be significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of traffic engineering technology, specifically relating to a fuzzy control method and system for traffic signals that considers individual prior data. Background Technology

[0002] Fuzzy logic signal control, which uses fuzzy logic to time traffic signals, is one of the most classic and effective methods in the field of traffic signal control. It can effectively optimize signal timing results, significantly alleviating intersection congestion and improving intersection operational efficiency. However, most existing fuzzy logic-based signal control methods only optimize signal timing based on real-time detection data, lacking the use of historical and predictive data. This results in long computation times, leading to severe lag in control strategies and causing traffic congestion, thus failing to achieve real-time control.

[0003] With the development of intelligent traffic detection technology, spatiotemporal data of individual vehicle trips can be more easily obtained. Mining individual travel data provides new solutions and methods for improving traffic signal control. Therefore, there is an urgent need for a fuzzy control method and system for traffic signals that considers individual prior data. By comprehensively utilizing historical and real-time vehicle data, the system can predict the arrival time of vehicles at intersections, effectively improving intersection traffic efficiency. Summary of the Invention

[0004] The purpose of this invention is to address the current problem that adaptive signal control methods at intersections lack comprehensive utilization of historical and real-time vehicle data, leading to a serious lag in control strategies and causing traffic congestion. Therefore, this invention proposes a fuzzy control method for traffic signals that considers individual prior data.

[0005] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a fuzzy control method for traffic signals that considers individual prior data, comprising the following steps:

[0006] Step 1: Clean and analyze the license plate recognition data;

[0007] Step 2: Perform personalized travel time prediction for drivers using support vector machines based on Bayesian optimization;

[0008] Step 3: Perform Bayesian personalized route prediction for drivers based on Markov chains.

[0009] Step 4: Integrate and utilize historical vehicle travel information and real-time detection information;

[0010] Step 5: Obtain phase arrival rate and turning rate through vehicle travel time and path prediction, and construct a fuzzy control method for traffic signals.

[0011] Furthermore, in step 1, historical travel data of individual vehicles is obtained through traffic detection technology, a historical travel database is constructed, and the historical travel data is cleaned and analyzed.

[0012] Furthermore, in step 2, a support vector machine model is used to predict driver travel time, utilizing features such as weather, time of day, average travel time per road segment, and average driver travel time to predict future driver travel times; i Factors influencing travel time prediction, x j The radial basis function kernel function and decision function of the support vector machine travel time prediction model are as follows: (The values ​​are the predicted travel times.)

[0013] k(x i ,x j )=exp(-γ||x i -x j || 2 (1)

[0014]

[0015] In the formula, α i The solution to the dual problem transformed using the Lagrange multiplier method is given by γ, where γ is the bandwidth of the kernel function and b is the bias term.

[0016] In the process of using the above model to predict travel time, the Bayesian optimization method is used to seek the optimal combination of parameters such as the radial basis kernel function bandwidth parameter γ, the penalty coefficient C, and the insensitivity coefficient ε. That is, the parameters are optimized by using prior probability and observation data to calculate posterior probability.

[0017] The proxy model is as follows:

[0018] The mean function is μ(x), the covariance function is k(x,x′), and the Gaussian process is represented as:

[0019] f(x)~GP(μ(x),k(x,x′)) (3)

[0020] The sampling strategy is used to determine the values ​​of the next set of hyperparameters. The sampling strategy used is the desired improvement, and the formula is as follows:

[0021]

[0022] In the formula, μ(x) is the posterior distribution mean, σ(x) is the covariance, and f(x) is the mean of the distribution. t+1 ) is the optimal objective function.

[0023] Furthermore, in step 3, a Bayesian method is used to predict the vehicle path. Based on real-time license plate recognition achieved by the checkpoint monitoring equipment, a portion of the vehicle's travel path is determined, and the vehicle's preceding path T is obtained. q To predict the location of a vehicle at the next checkpoint, based on a Bayesian model, the problem is transformed into calculating the known preceding path T. q Under these conditions, the vehicle passes through the next intersection. j The probability of:

[0024]

[0025] In the formula: P(l j Let be the prior probability, and let l be the location of the vehicle at the next checkpoint. j The probability of passing through checkpoint l is... j The ratio of the number of paths to the amount of historical path data;

[0026] In predicting individual vehicle travel routes, the travel route is treated as a Markov chain, assuming that the driver's next position is related to their current position. A single travel route T = {t} i ,L,t j},t k ∈R decomposed into

[0027] T={t i ,t i+1}{t i+1 ,t i+2}L{t j-1 ,t j},t k ∈R

[0028] Individual vehicle travel path T = {t i ,t i+1}{t i+1 ,t i+2}L{t j-1 ,t j},t k After ∈R, the probability of traveling from checkpoint li to checkpoint lj is as follows:

[0029] P(T)=p i,i+1 p i+1,i+2 L p j-1,j (6)

[0030] In the formula: p ij For two adjacent checkpoints l i , l j The state transition probability is equivalent to that of the checkpoint l i Heading towards the checkpoint j The probability of containing l is i -l jNumber of historical paths and the number of paths containing l i The ratio of the number of historical paths;

[0031] Construct the state transition probability matrix M between all adjacent checkpoints:

[0032]

[0033] The full path probability between the two checkpoints is as follows:

[0034]

[0035] In the formula: s is the shortest distance between checkpoints i and j; k is the detour coefficient.

[0036] When a vehicle passes through checkpoint j, its preceding path is T. q The probability of:

[0037]

[0038] In the formula, P(T) q ) represents the path probability; p c→j p represents the probability that a vehicle will move from its current location to the next checkpoint j. s→j Let be the probability of a vehicle moving from its starting position s to the next checkpoint j;

[0039] According to formula (1), the vehicle's path T is known. q Under the given conditions, the probability of passing through different checkpoints at the next moment is used to select the checkpoint with the highest probability as the path prediction result.

[0040] Further, in step 4, the traffic detection equipment extracts the vehicle's historical travel information to construct a personalized travel feature database, which includes the vehicle's travel mode and route preference information. The real-time detected vehicle information is compared with the historical database to identify and verify the vehicle's identity and behavior. Combining current weather, time of day, average travel time of road segment, and average driver travel time, a travel time prediction model is used to accurately predict the vehicle's travel time from upstream to the target intersection. Based on the current travel path trajectory, the vehicle's next location is predicted using a vehicle path prediction model, and the vehicle's turning behavior is analyzed to obtain the phase required for the vehicle to pass through the intersection, predicting the vehicle's next location and arrival time. The vehicle's arrival prediction information is correlated with the intersection's signal cycle and phase. The arrival vehicles for each phase in each cycle are summed and divided by the cycle duration to obtain the predicted arrival rate of each phase at the intersection in each cycle. After the trip is completed, the real-time detected data is used to update the personalized travel feature database to analyze the individual vehicle's behavior patterns.

[0041] Furthermore, in step 5, based on the phase setting of the four phases, and combined with the data from the upstream detectors at the intersection to predict the traffic demand for each flow direction, a traffic signal control method based on a two-level fuzzy controller is constructed with the goal of reducing vehicle travel time.

[0042] The signal control system is initialized. Based on the arrival rate prediction for the green light phase, a first-level fuzzy controller is used to output the initial green light delay. The first-level fuzzy controller is the traffic intensity determination module for the current green light phase. This module outputs the initial green light extension time based on the vehicle arrival situation for the green light phase. The design steps are as follows:

[0043] (1) Input and output of fuzzy controller

[0044] Input variable: Predicted arrival rate λ of the current green light phase gi

[0045] Output variable: Green light delay DT1

[0046] (2) Domain of variables and scaling factor

[0047] The upstream detector is set at the beginning of the road segment connecting the upstream intersection and the target intersection, and the downstream detector is set at the stop line of the approach lane of the target intersection; based on the actual traffic flow in the study area and relevant research, the predicted arrival rate λ of the current green light phase is... gi The basic universe of discourse is set to [0, 0.3], the scaling factor is 1, and the fuzzy universe of discourse is [0, 0.3]; the basic universe of discourse for the green light delay DT1 is set to [0, 30], the scaling factor is 1, and the fuzzy universe of discourse is [0, 30].

[0048] (3) Fuzzy subsets of input and output variables

[0049] The predicted arrival rate λ of the current green light phase gi The fuzzy controller is divided into seven fuzzy subsets, denoted as {Very Low, Low, Lower, Medium, Higher, High, Very High}, and further subdivided into {VL, L, RL, M, RH, H, VH}. In this invention, all input and output variables of the fuzzy controller are also divided into these seven fuzzy subsets.

[0050] (4) Membership function

[0051] Choose the triangle membership function

[0052] (5) Control rules

[0053] The first-level fuzzy controller is a single-input, single-output fuzzy controller. The traffic intensity of the current green light phase, i.e., the predicted vehicle arrival rate λ, is used. gi The larger the value, the longer the corresponding green light phase extension time; conversely, the smaller the value, the lower the predicted vehicle arrival rate λ for the current green light phase. giThe smaller the value, the shorter the corresponding green light phase extension time; based on the fuzzy subset division of input and output variables, 7 corresponding rules are formulated;

[0054] (6) Defuzzy

[0055] The centroid method is used to resolve fuzziness. The functions of a first-level fuzzy controller are implemented through programming, with the input variable set as follows: the predicted arrival rate λ of the current green light phase. gi After fuzzification, fuzzy reasoning, and defuzzification, the output variable, the value of green light delay DT1, is obtained.

[0056] After the target phase turns red, the signal enters the second-level fuzzy controller. Based on the arrival rate predicted for the red light phase, the second-level fuzzy controller outputs green light delay compensation. The second-level fuzzy controller is the traffic intensity determination module for the current red light phase. This module, based on the vehicle arrival status of the three current red light phases and the green light extension time output by the first-level fuzzy controller, uses fuzzy inference to determine the traffic intensity from the current phase to the red light phase with the initial green light extension time, and outputs the green light delay compensation time. The design steps are as follows:

[0057] (1) Input and output of fuzzy controller

[0058] Input variables: Green light delay DT1, maximum predicted arrival rate λ for the current red light phase. ri

[0059] Output variable: Green light delay compensation DT2

[0060] (2) Domain of variables and scaling factor

[0061] Maximum predicted arrival rate λ for the current red light phase ri The basic universe of discourse is set to [0, 0.3], the scaling factor is 1, and the fuzzy universe of discourse is [0, 0.3]. The basic universe of discourse for the green light delay DT1 is set to [0, 30]. Since the delay compensation time DT2 cannot exceed the green light delay DT1, the scaling factor for the green light delay DT1 is 2, and the fuzzy universe of discourse is [0, 60]. The basic universe of discourse for the green light delay compensation DT2 is set to [0, 30], the scaling factor is 1, and the fuzzy universe of discourse is [0, 30].

[0062] (3) Fuzzy subsets of input and output variables

[0063] The maximum predicted arrival rate λ of the current red light phase ri Green light delay DT1 and green light delay compensation DT2 are both divided into 7 fuzzy subsets.

[0064] (4) Membership function

[0065] Consistent with the first-level fuzzy controller, a triangular membership function is selected.

[0066] (5) Control rules

[0067] The second-level fuzzy controller is a multi-input single-output fuzzy controller, predicting the arrival rate λ. ri The larger the value of λ, the longer the green light delay DT1, and the longer the delay compensation time DT2; conversely, the smaller the value of λ, the longer the predicted arrival rate λ. ri The smaller the value, the shorter the green light delay DT1, and the shorter the delay compensation time DT2; based on the fuzzy subset division of input and output variables, 49 corresponding rules are formulated.

[0068] (6) Defuzzy

[0069] The centroid method is used to resolve fuzziness, and the resolved output surface is obtained through programming; for λ ri For any value of DT1, a unique correspondence can be found in the unfuzzy surface, and as λ increases... ri As DT1 decreases, the green light delay compensation DT2 also decreases, and with λ ri As DT1 increases, the green light delay compensation DT2 also increases accordingly.

[0070] Based on the actual situation at the intersection, output a green light delay and proceed to the next phase.

[0071] The present invention also relates to a system for a fuzzy control method for traffic signals that considers individual prior data. The system includes a computer module incorporating the fuzzy control method for traffic signals that considers individual prior data. The computer module includes a first-level fuzzy controller, a second-level fuzzy controller, a decision module, and a phase switching module.

[0072] Beneficial effects

[0073] This invention proposes a fuzzy traffic signal control method that considers individual prior data. By predicting the travel time of vehicles from upstream to the target intersection, the method matches vehicles with signal cycles and predicts vehicle travel paths to obtain the phase required for vehicles to pass through the intersection. Furthermore, it proposes a method for calculating the predicted arrival rate of intersection approach lanes and constructs a traffic signal control method based on a two-level fuzzy controller.

[0074] The traffic flow arrival rate calculation method proposed in this invention can effectively estimate the traffic flow arrival rate under four-phase control, and the calculation results can be used as a data source for traffic signal control. In addition, this invention can significantly reduce the average vehicle delay. Compared with traditional fuzzy control, the average vehicle delay is reduced by 18.75% during off-peak hours and by 16.11% during peak hours, which significantly improves the service quality of intersections and provides an important reference for traffic signal control in the future environment of all individual vehicle samples. Attached Figure Description

[0075] Figure 1This is a schematic diagram of the control method framework proposed in this invention;

[0076] Figure 2 This is a flowchart of the vehicle travel time prediction model algorithm based on Bayesian parameter optimization support vector machine in this invention;

[0077] Figure 3 This is a flowchart of the vehicle route prediction algorithm based on the Bayesian method in this invention;

[0078] Figure 4 This is a flowchart illustrating the design of fuzzy control for traffic signals at a single intersection in this invention.

[0079] Figure 5 This is a schematic diagram of the intersection used in the simulation process of this invention;

[0080] Figure 6a This invention provides a timing signal control scheme for off-peak period simulation.

[0081] Figure 6b This invention provides a timing signal control scheme for peak-hour simulation.

[0082] Figure 7a This is a schematic diagram of the detector placement in Scheme 2 during the simulation process of this invention;

[0083] Figure 7b This is a schematic diagram of the detector placement in Scheme 3 during the simulation process of this invention. Detailed Implementation

[0084] The following combination Figures 1 to 4 The embodiments of the present invention will be described in detail below.

[0085] This invention is a fuzzy control method for traffic signals that considers individual prior data, mainly including the following steps:

[0086] Step 1: Clean and analyze the license plate recognition data.

[0087] First, data fields are extracted from the license plate recognition data, including intersection name, road segment name, intersection latitude and longitude, passage time, and driving direction. Then, data reduction is performed, data anomalies are analyzed, and data preprocessing is carried out. Next, the preprocessed data is analyzed, including the number of days inspected, the number of checkpoints passed by each trajectory, and the time distribution of vehicles passing through two checkpoints. Finally, the temporal and spatial characteristics of vehicle travel in the study area are studied, the morning and evening peak periods of vehicle travel are divided, and areas with high vehicle traffic and roads and intersections with high traffic volume are identified.

[0088] Step 2: Perform personalized driver trip time prediction using Bayesian-optimized support vector machine.

[0089] like Figure 2 As shown, based on the historical travel database obtained in step 1, vehicle segment travel time data is extracted, and a Bayesian optimization support vector machine is used. Support vector machine regression (SVR) maps the data to a Gaussian space and finds an optimal hyperplane that separates samples of different categories as much as possible in that space. By combining influencing factors such as weather, time of day, average segment travel time, and average driver travel time, a high-precision prediction model for vehicle travel time from upstream to the target intersection is generated.

[0090] Among them, the Bayesian optimization-based support vector machine (BO-SVM) method:

[0091] Assume x i Factors influencing travel time prediction, x j The radial basis function kernel function and decision function of the support vector machine travel time prediction model are as follows: (The values ​​are the predicted travel times.)

[0092] k(x i ,x j )=exp(-γ||x i -x j || 2 (10)

[0093]

[0094] In the formula, α i The solution to the dual problem transformed using the Lagrange multiplier method is given by γ, where γ is the bandwidth of the kernel function and b is the bias term.

[0095] In the process of using the above model to predict travel time, the Bayesian optimization method is used to seek the optimal combination of parameters, namely the radial basis kernel function bandwidth parameter γ, the penalty coefficient C, and the insensitivity coefficient ε. That is, the parameters are optimized by using prior probability and observation data to calculate posterior probability.

[0096] (1) Proxy Model

[0097] Assuming the mean function is μ(x) and the covariance function is k(x,x′), then the Gaussian process can be represented as:

[0098] f(x)~GP(μ(x),k(x,x′)) (12)

[0099] (2) Sampling strategy

[0100] The sampling strategy is used to determine the values ​​of the next set of hyperparameters. The sampling strategy used is the desired improvement, and the formula is as follows:

[0101]

[0102] In the formula, μ(x) is the posterior distribution mean, σ(x) is the covariance, and f(x) is the mean of the distribution. t+1 ) is the optimal objective function.

[0103] Step 3: Perform Bayesian personalized route prediction for drivers based on Markov chains

[0104] like Figure 3 As shown, based on the historical travel database obtained in step 1, individual vehicle travel path information is extracted. By utilizing the state transition characteristics of Markov chains, a Markov transition matrix is ​​constructed to capture the transition patterns of vehicles between different roads and establish a Bayesian vehicle path prediction model.

[0105] Among them, the Bayesian method based on Markov chains:

[0106] By using checkpoint monitoring equipment to achieve real-time license plate recognition and determine part of the vehicle's travel path, the vehicle's preceding path T can be obtained. q Based on this, predict the next checkpoint location that the vehicle is most likely to pass through in space. Based on the Bayesian model, the problem can be transformed into calculating the known preceding path T. q Under the condition that the vehicle passes through the next possible intersection j The probability of:

[0107]

[0108] In the formula: P(l j Let be the prior probability, and let l be the location of the vehicle at the next checkpoint. j The probability of passing through checkpoint l is... j The ratio of the number of paths to the amount of historical path data.

[0109] In predicting individual vehicle travel routes, the travel route is viewed as a Markov chain, assuming that the driver's next position is related to their current position. Therefore, a single travel route T = {t} i ,L,t j},t k ∈R decomposed into T={t i ,t i+1}{t i+1 ,t i+2}L{t j-1 ,t j},t k ∈R.

[0110] Individual vehicle travel path T = {t i ,t i+1}{t i+1 ,ti+2}L{t j-1 ,t j},t k After ∈R, by the checkpoint l i Heading towards the checkpoint j The probabilities are as follows:

[0111] P(T)=p i,i+1 p i+1,i+2 L p j-1,j (15)

[0112] In the formula: p ij For two adjacent checkpoints l i , l j The state transition probability, i.e., determined by the checkpoint l i Heading towards the checkpoint j The probability of containing l is i -l j Number of historical paths and the number of paths containing l i The ratio of the number of historical paths.

[0113] Construct the state transition probability matrix M between all adjacent checkpoints:

[0114]

[0115] The full path probability between the two checkpoints is as follows:

[0116]

[0117] In the formula: s is the shortest distance between checkpoints i and j; k is the detour coefficient.

[0118] Therefore, when a vehicle passes through checkpoint j, its preceding path is T. q The probability of:

[0119]

[0120] In the formula, P(T) q ) represents the path probability; p c→j p represents the probability that a vehicle will move from its current location to the next checkpoint j. s→j Let be the probability of a vehicle moving from its starting position s to the next checkpoint j.

[0121] Therefore, according to formula (1), the vehicle's position on the known path T can be obtained. q Under the given conditions, the probability of passing through different checkpoints at the next moment is used to select the checkpoint with the highest probability as the path prediction result.

[0122] To analyze vehicle turning direction selection results at intersections and clarify the positional relationship between intersections and road segments, road segments are named based on checkpoint names and in sequence. The relative positions of adjacent checkpoints and the direction of the road segment are analyzed to construct a road network turning position lookup table. After completing path prediction, the path T = {t} i ,L,t j},t k ∈R is split to form a sequence of turning paths {t i ,t i+1 ,t i+2}{t i+1 ,t i+2 ,t i+3}L{t j-2 ,t j-1 ,t j},t k ∈R, where {t i ,t i+1} represents the path the vehicle is on. i -l i+1 And we are about to reach the intersection. i+1 , l i+2 To predict the next position, the vehicle's intersection turning behavior is analyzed based on the road network turning position lookup table.

[0123] Step 4: Integrate and utilize historical vehicle travel information and real-time detection information

[0124] Traffic detection equipment monitors vehicle traffic information in real time, matches it with historical travel data, and combines factors such as current weather, time of day, average travel time per road segment, and average driver travel time. A travel time prediction model is used to accurately predict vehicle travel time, forecasting the time it takes for a vehicle to reach the target intersection from upstream. Simultaneously, based on the current travel path trajectory, a vehicle path prediction model predicts the next location of the vehicle and analyzes vehicle turning behavior to obtain the phase required for the vehicle to pass through the intersection. This yields the predicted next location and arrival time of the vehicle. The predicted vehicle arrival information is then correlated with the signal cycle and phase of the intersection. The arrival rates for each phase in each cycle are summed and divided by the cycle duration to obtain the predicted arrival rate λ for each phase of the intersection in each cycle. gi .

[0125] Step 5: Obtain phase arrival rate and turning rate through vehicle travel time and path prediction, and construct a fuzzy control method for traffic signals.

[0126] Step 5-1: Initialize the signal control system and set the minimum green light time g for each phase. min Maximum green light time g maxThe green light extension time is Δg; let the current green light phase i at the intersection be the initial phase, and give the initial phase a minimum green light time g. i =g min ;

[0127] Step 5-2: As Figure 4 As shown, when a vehicle arrives at the controlled intersection, it is determined whether the green light of the target phase has ended. If the green light has not ended, the vehicle is included in the predicted vehicle arrival rate λ of the current green light phase. gi The system then enters the first-level fuzzy controller within 2 seconds before the green light ends, inputting the predicted vehicle arrival rate λ for the current green light phase. gi Output the green light delay DT1; otherwise, include the vehicle in the predicted vehicle arrival rate λ for the current red light phase. gi DT1 is based on changes in arrival rate.

[0128] The first-level fuzzy controller is the traffic intensity determination module for the current green light phase. This module mainly considers the vehicle arrival situation for the current green light phase and initially outputs the green light extension time. The design steps are as follows:

[0129] (1) Input and output of fuzzy controller

[0130] Input variable: Predicted arrival rate λ of the current green light phase gi .

[0131] Output variable: Green light delay DT1.

[0132] (2) Domain of variables and scaling factor

[0133] In this invention, the upstream detector is set at the starting point of the road segment connecting the upstream intersection and the target intersection, and the downstream detector is set at the stop line of the approach lane of the target intersection. Based on the actual traffic flow in the study area and relevant research, the predicted arrival rate λ of the current green light phase is... gi The basic universe of discourse is set to [0, 0.3], the scaling factor is 1, and the fuzzy universe of discourse is [0, 0.3]. The basic universe of discourse for the green light delay DT1 is set to [0, 30], the scaling factor is 1, and the fuzzy universe of discourse is [0, 30].

[0134] (3) Fuzzy subsets of input and output variables

[0135] The predicted arrival rate λ of the current green light phase gi The fuzzy controller is divided into 7 fuzzy subsets, namely {very low, low, lower, medium, higher, high, very high}, denoted as {VL, L, RL, M, RH, H, VH}; all input and output variables of the fuzzy controller in this invention are divided into 7 fuzzy subsets.

[0136] (4) Membership function

[0137] Triangle membership functions have advantages such as low computational complexity and flexible and easy-to-adjust parameters, making them the preferred choice.

[0138] (5) Control rules

[0139] The first-level fuzzy controller is a single-input, single-output fuzzy controller. The traffic intensity of the current green light phase, i.e., the predicted vehicle arrival rate λ, is used. gi A larger value corresponds to a longer green light phase extension time; conversely, a smaller value corresponds to a shorter predicted vehicle arrival rate λ for the current green light phase. gi The smaller the value, the shorter the corresponding green light phase extension time. Therefore, based on the fuzzy subset division of input and output variables, seven corresponding rules are formulated.

[0140] (6) Defuzzy

[0141] The centroid method is used to resolve fuzziness. The functions of a first-level fuzzy controller are implemented through programming, with the input variable set as follows: the predicted arrival rate λ of the current green light phase. gi After fuzzification, fuzzy reasoning, and defuzzification, the output variable, the value of green light delay DT1, is obtained.

[0142] Step 5-3: After the target phase turns red, enter the secondary fuzzy controller and input the maximum vehicle arrival rate λ predicted for the current red light phase. gi Output green light delay compensation DT2. The minimum green light time g in the target phase. i Before the end, output the final green light delay Δg = DT1 - DT2.

[0143] The second-level fuzzy controller is the traffic intensity determination module for the current red light phase. This module mainly considers the vehicle arrival situation of the three current red light phases and the green light extension time output by the first-level fuzzy controller. Through fuzzy inference, it determines the traffic intensity of the current phase up to the red light phase with the initial green light extension time, and finally outputs the green light delay compensation time. The design steps are as follows:

[0144] (1) Input and output of fuzzy controller

[0145] Input variables: Green light delay DT1, maximum predicted arrival rate λ for the current red light phase. ri

[0146] Output variable: Green light delay compensation DT2

[0147] (2) Domain of variables and scaling factor

[0148] Maximum predicted arrival rate λ for the current red light phase riThe basic universe of discourse for green light delay DT1 is set to [0, 0.3], the scaling factor is 1, and the fuzzy universe of discourse is [0, 0.3]. The basic universe of discourse for green light delay DT1 is set to [0, 30]. Since the delay compensation time DT2 cannot exceed the green light delay DT1, the scaling factor for green light delay DT1 is 2, and the fuzzy universe of discourse is [0, 60]. The basic universe of discourse for green light delay compensation DT2 is set to [0, 30], the scaling factor is 1, and the fuzzy universe of discourse is [0, 30].

[0149] (3) Fuzzy subsets of input and output variables

[0150] The maximum predicted arrival rate λ of the current red light phase ri Green light delay DT1 and green light delay compensation DT2 are both divided into 7 fuzzy subsets.

[0151] (4) Membership function

[0152] Consistent with the first-level fuzzy controller, the triangular membership function is selected here.

[0153] (5) Control rules

[0154] The second-level fuzzy controller is a multi-input single-output fuzzy controller, predicting the arrival rate λ. ri The larger the value of λ, the longer the green light delay DT1, and the longer the delay compensation time DT2; conversely, the smaller the value of λ, the longer the predicted arrival rate λ. ri The smaller the value, the shorter the green light delay DT1, and the shorter the delay compensation time DT2. Based on the fuzzy subset division of input and output variables, 49 corresponding rules are established.

[0155] (6) Defuzzy

[0156] The centroid method is used for fuzzy resolution, and the resolved output surface can be obtained through programming. For λ... ri For any value of DT1, a unique correspondence can be found in the unfuzzy surface, and as λ increases... ri As DT1 decreases, the green light delay compensation DT2 also decreases, and with λ ri As DT1 increases, the green light delay compensation DT2 also increases.

[0157] Step 5-4: Determine whether all four phases have run within the signal cycle. If not, select the red light phase with the highest vehicle arrival rate λ that has not yet run. ri The corresponding red light phase is set as the next green light phase; if both are already in operation, the signal cycle ends, and the system switches to the phase with the highest red light arrival rate among the remaining three phases, setting the maximum vehicle arrival rate λ during the red light period as the threshold. ri The corresponding red light phase is set as the next green light phase.

[0158] Effect verification

[0159] Choose as Figure 5 The intersection shown is used as a real-world case study to verify the effectiveness of this invention in improving intersection traffic efficiency. Traffic input schemes for off-peak and peak periods are designed, as shown in Table 1.

[0160] Table 1 Flow Input Scheme

[0161]

[0162] To verify the effectiveness of this method, SUMO simulation software was used to compare the proposed fuzzy traffic signal control based on vehicle travel time and path prediction with timed signal control and real-time arrival rate detection fuzzy control. The saturation flow rate of the left-turn lane was set to 1550 (pch / s), and the saturation flow rate of the straight lane was set to 1650 (pch / s), with a uniform four-phase configuration. The experiment ensured that all control methods had the same flow input during the same time period. Three effectiveness evaluation indicators—average vehicle delay, queue length, and number of stops—were selected to evaluate the signal control methods. The experimental scheme is as follows:

[0163] (1) Timing signal control scheme

[0164] Scheme 1 employs a time-segmented signal control method to conduct simulation experiments on the research intersection. The control parameters for the time-segmented signal control are calculated using Webster's optimal cycle length formula based on off-peak and peak traffic volume data. The minimum green light time for each phase is set to 10 seconds, the yellow light time to 3 seconds, the all-red light time to 2 seconds, and the start-up loss time to 3 seconds. The timing scheme is referenced... Figure 6a , 6b Where gi represents the green light phase, i = 1, 2, 3, 4 represent four phases, y represents the yellow light phase, and r represents the red light phase.

[0165] (2) Fuzzy control scheme based on detection arrival rate

[0166] Scheme 2 employs a fuzzy traffic signal control method based on real-time arrival rate detection using coil detection. A detector is installed at the beginning of the guide lane of each approach lane, 50 meters upstream of the stop line. After detecting a vehicle, the detector determines the vehicle's direction based on its lane and calculates the arrival rate for that direction. Based on the real-time arrival rate for each direction, a structurally similar two-level fuzzy controller proposed in this invention is used to control the simulated intersection.

[0167] (3) Fuzzy control scheme based on vehicle travel time and route prediction

[0168] Scheme 3 employs the fuzzy traffic signal control method based on predicted arrival rate proposed in this invention. A detector is installed 300 meters upstream of the stop line at each approach lane. When a vehicle arrives at the detector, the detector reads the vehicle identification information and invokes the travel time prediction algorithm and path prediction algorithm. Based on the vehicle's historical travel information, the arrival time at the intersection and the vehicle's direction of travel are predicted. The arrival prediction information of individual vehicles is correlated with the signal cycle and phase of the intersection to obtain the predicted arrival rate for each phase of the intersection in each cycle. Based on the predicted arrival rate, the fuzzy traffic signal control method proposed in this invention is used to control the intersection.

[0169] Schemes 2 and 3 use the same parameters as the timing signal control for minimum green light time, saturation flow rate, and phase settings, ensuring that control effects are compared under identical conditions. Detector settings for Schemes 2 and 3 are referenced in the appendix. Figure 7a , 7b In addition, the maximum green light time is set to 50 seconds, and the output green light extension time is set to 0 to 40 seconds.

[0170] To simulate real off-peak and peak traffic flow, all three schemes generate the corresponding number of traffic flows as shown in Table 1. Schemes 1 and 2 simply define the number of vehicles in each direction. Scheme 3 extracts the vehicle travel time dataset and vehicle path dataset of the research intersection based on the preprocessed license plate recognition data, inputs vehicles that appear twice or more into the simulation environment, and randomly generates them according to the corresponding traffic volume.

[0171] Under a unified traffic flow input, simulation experiments were first conducted on intersections under three schemes: Scheme 1, Scheme 2, and Scheme 3, during both off-peak and peak traffic conditions. The simulation warm-up time was 200 seconds, and the effective simulation time was set to 1800 seconds.

[0172] Simulation experiments were conducted for two traffic conditions, and the average vehicle delay at the intersection was continuously recorded. The statistical results of average vehicle delay, queue length, and number of stops were finally output. The experiment was repeated 10 times, and the average value of the three evaluation indicators obtained by the three methods in the multiple simulation experiments was calculated. The results are shown in Table 2.

[0173] Table 2 Comparison of Signal Control Effect Evaluation Indicators for Three Schemes

[0174]

[0175] The average delays of this method are 30.29 seconds and 36.43 seconds, respectively, representing improvements of 33.81% and 35.34% compared to timed signal control, and 22.87% and 19.98% compared to traditional fuzzy control. Regarding queue length, this method improves efficiency by 30.16% and 34.15% compared to timed signal control during off-peak and peak hours, respectively, and by 19.89% and 14.59% compared to traditional fuzzy control. This indicates that this method can reduce queue length by predicting the time it takes for vehicles to travel from upstream to the intersection and rationally allocating green light extension times. In terms of the number of stops, this method improves efficiency by 26.74% and 21.98% compared to timed signal control during off-peak and peak hours, respectively, and by 18.18% and 16.47% compared to traditional fuzzy control. This demonstrates that this method can effectively reduce the number of times vehicles stop and minimize secondary stops. By predicting vehicle arrival times at the intersection, it can reasonably extend green light times, allowing vehicles that would otherwise have to wait for the next cycle to pass through in the current cycle, thus reducing the number of stops. Therefore, this method can effectively improve intersection traffic efficiency.

[0176] The above examples of this invention are merely illustrative of the computational model and process of this invention, and are not intended to limit the implementation of this invention. For ease of description, this invention names intersections according to their orientation, but this does not limit the application of this invention. This statement applies to all orientation-based descriptions of this invention. Those skilled in the art can make other variations or modifications based on the above description. It is impossible to exhaustively list all implementation methods here. Any obvious variations or modifications derived from the technical solutions of this invention are still within the protection scope of this invention.

Claims

1. A fuzzy control method for traffic signals that considers individual prior data, characterized in that, Includes the following steps: Step 1: Clean and analyze the license plate recognition data; Step 2: Perform personalized travel time prediction for drivers using support vector machines based on Bayesian optimization; Step 3: Perform Bayesian personalized route prediction for drivers based on Markov chains. Step 4: Integrate and utilize historical vehicle travel information and real-time detection information; Historical travel information of vehicles is extracted by traffic detection equipment to build a personalized travel feature database for vehicles, which includes the vehicle's travel mode and route preference information; The real-time detected vehicle information is compared with the historical database to identify and verify the vehicle's identity and behavior. Combined with the current weather, time of day, average travel time of road segment, and average travel time of driver, the travel time prediction model is used to make a high-precision prediction of the vehicle's travel time from upstream to the target intersection. Based on the travel path trajectory, the vehicle path prediction model is used to predict the next location of the vehicle, and the vehicle's turning behavior is analyzed to obtain the phase required for the vehicle to pass through the intersection, and to predict the next location information and the time of appearance of the vehicle. The arrival prediction information of vehicles is correlated with the signal cycle and phase of the intersection. The arrival vehicles of each phase in each cycle are summed and divided by the cycle duration to obtain the predicted arrival rate of each phase of the intersection in each cycle. After the trip is completed, the personalized travel feature database of vehicles is updated with the real-time detected data to analyze the behavior patterns of individual vehicles. Step 5: Obtain phase arrival rate and turning rate through driver personalized travel time and driver personalized route prediction, and construct a fuzzy control method for traffic signals.

2. The fuzzy traffic signal control method considering individual prior data according to claim 1, characterized in that, In step 1, historical travel data of individual vehicles is obtained through traffic detection technology, a historical travel database is constructed, and the historical travel data is cleaned and analyzed.

3. The fuzzy traffic signal control method considering individual prior data according to claim 1, characterized in that, In step 2, a support vector machine model is used to predict the driver's personalized travel time. The model uses weather, time of day, average travel time of road segment, and average travel time of driver to predict the driver's travel time in the future. Factors affecting travel time prediction The radial basis function kernel function and decision function of the support vector machine travel time prediction model are as follows: (The values ​​are the predicted travel times.) In the formula, The solution to the dual problem transformed using the Lagrange multiplier method is given by γ, where γ is the bandwidth of the kernel function and b is the bias term. In the process of using the above model to predict travel time, the Bayesian optimization method is used to seek the optimal combination of parameters such as the radial basis kernel function bandwidth parameter γ, the penalty coefficient C, and the insensitivity coefficient ε. That is, the parameters are optimized by using prior probability and observation data to calculate posterior probability. The proxy model is as follows: The mean function is The covariance function is The Gaussian process is represented as The sampling strategy is used to determine the values ​​of the next set of hyperparameters. The sampling strategy used is the desired improvement, and the formula is as follows: In the formula, The mean of the posterior distribution. For covariance, This is the optimal objective function.

4. The fuzzy traffic signal control method considering individual prior data according to claim 1, characterized in that, In step 3, a Bayesian method is used to predict the driver's personalized route. Based on real-time license plate recognition by the checkpoint monitoring equipment, a portion of the vehicle's driving path is determined, and the vehicle's preceding path is obtained. Predicting the location of a vehicle at the next checkpoint, based on a Bayesian model, transforms the problem into calculating the known preceding path. Under these conditions, the vehicle passes through the next intersection. j The probability of: In the formula: As a priori probability, the next checkpoint location for the vehicle is l. j The probability of passing through checkpoint l is... j The ratio of the number of paths to the amount of historical path data; In predicting individual vehicle travel routes, the travel route is treated as a Markov chain, assuming that the driver's next location is related to their current location, and a single travel route is considered as a whole. Decomposed into Individual vehicle passage routes Then, the probability of driving from checkpoint li to checkpoint lj is as follows: In the formula: For two adjacent checkpoints l i , l j The state transition probability is equivalent to that of the checkpoint l i Driving towards the checkpoint j The probability of containing l is i -l j Number of historical paths and the number of paths containing l i The ratio of the number of historical paths; Construct the state transition probability matrix M between all adjacent checkpoints: The full path probability between the two checkpoints is as follows: In the formula: s is the shortest distance between checkpoints i and j; k is the detour coefficient. When a vehicle passes through checkpoint j, its preceding path is T. q The probability of: In the formula, For path probability; Let be the probability that the vehicle will move from its current location to the next checkpoint j. Let be the probability of a vehicle moving from its starting position s to the next checkpoint j; According to formula (5), the vehicle's path T is known. q Under the given conditions, the probability of passing through different checkpoints at the next moment is used to select the checkpoint with the highest probability as the driver's personalized route prediction result.

5. The fuzzy traffic signal control method considering individual prior data according to claim 1, characterized in that, In step 5, based on the phase settings of the four phases, and combined with the data from the upstream detectors at the intersection, the traffic demand for each flow direction is predicted. With the goal of reducing vehicle travel time, a traffic signal control method based on a two-level fuzzy controller is constructed.

6. The fuzzy traffic signal control method considering individual prior data according to claim 5, characterized in that, In step 5, the signal control system is initialized. Based on the predicted arrival rate of the green light phase, the initial green light delay is output using the first-level fuzzy controller. The first-level fuzzy controller is the traffic intensity determination module for the current green light phase. This module outputs the initial green light extension time based on the vehicle arrival situation of the green light phase. The design steps are as follows: (1) Input and output of fuzzy controller Input variable: Predicted arrival rate λ of the current green light phase gi Output variable: Green light delay DT1 (2) Variable domain and scaling factor The upstream detector is set at the beginning of the road segment connecting the upstream intersection and the target intersection, and the downstream detector is set at the stop line of the approach lane of the target intersection; based on the actual traffic flow in the study area and relevant research, the predicted arrival rate λ of the current green light phase is... gi The basic universe of discourse is set to [0, 0.3], the scaling factor is 1, and the fuzzy universe of discourse is [0, 0.3]; the basic universe of discourse for the green light delay DT1 is set to [0, 30], the scaling factor is 1, and the fuzzy universe of discourse is [0, 30]. (3) Fuzzy subsets of input and output variables The predicted arrival rate λ of the current green light phase gi The fuzzy controller is divided into seven fuzzy subsets, denoted as {Very Low, Low, Lower, Medium, Higher, High, Very High}, and further subdivided into {VL, L, RL, M, RH, H, VH}. In this invention, all input and output variables of the fuzzy controller are also divided into these seven fuzzy subsets. (4) Membership function Choose the triangle membership function (5) Control rules The first-level fuzzy controller is a single-input, single-output fuzzy controller. The traffic intensity of the current green light phase, i.e., the predicted vehicle arrival rate λ, is used. gi The larger the value, the longer the corresponding green light phase extension time; conversely, the smaller the value, the lower the predicted vehicle arrival rate λ for the current green light phase. gi The smaller the value, the shorter the corresponding green light phase extension time; based on the fuzzy subset division of input and output variables, 7 corresponding rules are formulated; (6) Defuzzification The centroid method is used to resolve fuzziness. The functions of a first-level fuzzy controller are implemented through programming, with the input variable set as follows: the predicted arrival rate λ of the current green light phase. gi After fuzzification, fuzzy reasoning, and defuzzification, the output variable, the value of green light delay DT1, is obtained.

7. The fuzzy traffic signal control method considering individual prior data according to claim 5, characterized in that, In step 5, after the target phase turns red, it enters the second-level fuzzy controller. Based on the arrival rate predicted by the red phase, the second-level fuzzy controller outputs green light delay compensation. The second-level fuzzy controller is the traffic intensity determination module for the current red light phase. Based on the vehicle arrival status of the three current red light phases and the green light extension time output by the first-level fuzzy controller, this module uses fuzzy inference to determine the traffic intensity from the current phase to the red light phase with the initial green light extension time, and outputs the green light delay compensation time. The design steps are as follows: (1) Input and output of fuzzy controller Input variables: Green light delay DT1, maximum predicted arrival rate λ for the current red light phase. ri Output variable: Green light delay compensation DT2 (2) Variable domain and scaling factor Maximum predicted arrival rate λ for the current red light phase ri The basic universe of discourse is set to [0, 0.3], the scaling factor is 1, and the fuzzy universe of discourse is [0, 0.3]. The basic universe of discourse for the green light delay DT1 is set to [0, 30]. Since the delay compensation time DT2 cannot exceed the green light delay DT1, the scaling factor for the green light delay DT1 is 2, and the fuzzy universe of discourse is [0, 60]. The basic universe of discourse for the green light delay compensation DT2 is set to [0, 30], the scaling factor is 1, and the fuzzy universe of discourse is [0, 30]. (3) Fuzzy subsets of input and output variables The maximum predicted arrival rate λ of the current red light phase ri Green light delay DT1 and green light delay compensation DT2 are both divided into 7 fuzzy subsets. (4) Membership function Consistent with the first-level fuzzy controller, a triangular membership function is selected. (5) Control rules The second-level fuzzy controller is a multi-input single-output fuzzy controller, predicting the arrival rate λ. ri The larger the value of λ, the longer the green light delay DT1, and the longer the delay compensation time DT2; conversely, the smaller the value of λ, the longer the predicted arrival rate λ. ri The smaller the value, the shorter the green light delay DT1, and the shorter the delay compensation time DT2; based on the fuzzy subset division of input and output variables, 49 corresponding rules are formulated. (6) Defuzzification The centroid method is used to resolve fuzziness, and the resolved output surface is obtained through programming; for λ ri For any value of DT1, a unique correspondence can be found in the unfuzzy surface, and as λ increases... ri As DT1 decreases, the green light delay compensation DT2 also decreases, and with λ ri As DT1 increases, the green light delay compensation DT2 also increases accordingly. Based on the actual situation at the intersection, output a green light delay and proceed to the next phase.

8. A system for implementing the fuzzy traffic signal control method considering individual prior data as described in any one of claims 1 to 7, characterized in that, The system includes a computer module containing a fuzzy control method for traffic signals that takes into account individual prior data.

9. The system of the fuzzy traffic signal control method considering individual prior data according to claim 8, characterized in that, The computer module includes a first-level fuzzy controller, a second-level fuzzy controller, a decision module, and a phase switching module.

Images