Traffic state estimation method, system and device

Abstract

The invention discloses a traffic state estimation method, system and device, and relates to the technical field of traffic state estimation, and the method comprises the steps: obtaining a historical traffic flow data set of a current traffic road section; fuzzy logic is introduced into a Poisson process model, and the arrival rates of vehicles at different road sections are modeled; modeling a transition probability of a traffic flow state by using a Markov chain, introducing a dynamic transition matrix to capture changes of the traffic flow state in different time periods, and further constructing a fuzzy state transition model based on the Markov chain; a fuzzy joint probability density function is obtained by combining a fuzzy Poisson distribution model and a fuzzy state transition model based on a Markov chain, and the traffic flow state is estimated; according to the method, the vehicle track and the traffic state can be estimated in various complex traffic scenes by combining the two models.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic state estimation, and in particular to a traffic state estimation method, system and device. Background Art

[0002] With the acceleration of modern urbanization, traffic congestion is becoming increasingly serious, and urban traffic management and optimization are facing huge challenges.

[0003] Traditional traffic state estimation methods, such as models based on classical traffic flow theory, often rely on idealized assumptions and are difficult to accurately reflect the complex dynamic characteristics of traffic. For example, based on a fixed Poisson distribution model or Markov process, although it can effectively simulate vehicle arrival rates and traffic state transitions, it is difficult to fully handle the complexity, ambiguity and uncertainty in actual traffic flows. Specifically, vehicle arrival rates, vehicle speeds and other states fluctuate under different traffic conditions, and these fluctuations are usually random and fuzzy. The traditional Poisson distribution assumes that vehicle arrivals are completely random and fails to fully consider the correlation between the time changes of traffic flow and road conditions; while the Markov model assumes that state transitions have a fixed transition probability matrix and cannot effectively capture the ambiguity in traffic states.

[0004] In summary, most traditional traffic state estimation methods are based on deterministic models, which usually assume that various aspects of traffic flow are predictable and stable. However, traffic flow in the real world is often highly random and volatile. Many existing traffic state estimation methods fail to fully consider the randomness and uncertainty in traffic flow, and it is difficult to accurately describe and predict the evolution of traffic flow. When dealing with complex dynamic traffic scenarios, they often show large errors. Summary of the invention

[0005] In view of the shortcomings of the prior art that the randomness and uncertainty in traffic flow are not fully considered and it is difficult to accurately describe and predict the evolution process of traffic flow, the present invention proposes a traffic state estimation method, system and device, which introduces fuzzy logic into the traditional Poisson process model to build a fuzzy Poisson distribution model, and uses the Markov process to model the transition of traffic state. By combining these two models, traffic state estimation is achieved, thereby solving the problems existing in the prior art.

[0006] A traffic state estimation method comprises the following steps:

[0007] Acquire a historical traffic flow data set of the current traffic section; the traffic flow data set includes vehicle arrival rate, vehicle position and speed;

[0008] According to the arrival rate of vehicles, fuzzy logic is introduced into the Poisson process model to estimate the arrival rate of vehicles within the fuzzy time interval, and a fuzzy Poisson distribution model is built; according to the position and speed of vehicles in different time periods, a traffic flow state transfer model is established using Markov chain, and the transition probability between traffic flow states is adjusted by introducing fuzzy logic into the model, and the dynamic transfer matrix is ​​introduced to capture the changes of traffic flow states in different time periods, and a fuzzy state transfer model based on Markov chain is built; by combining the fuzzy Poisson distribution model with the fuzzy state transfer model based on Markov chain, a fuzzy joint probability density model of traffic flow state and vehicle arrival rate within a set time interval is constructed;

[0009] The vehicle arrival rate and the transition probability of the traffic flow state of the traffic section to be tested in the current time period are taken as the input of the fuzzy joint probability density model, and the probability distribution of the traffic flow state of the traffic section in a set time interval adjacent to the current time period is output. The traffic state of the traffic section in the set time interval is estimated based on the probability distribution.

[0010] Furthermore, after obtaining the historical traffic flow dataset of the current traffic section, the traffic flow dataset is preprocessed using Bayesian reasoning, and the processing process includes the following steps:

[0011] Traffic flow data sets are collected through multi-source sensors. The traffic flow state is set to λ, and Bayesian estimation is performed on it. Assuming that the prior probability is P(λ) and the observed data is N, the following is obtained according to the Bayesian formula:

[0012]

[0013] Where λ represents the vehicle arrival rate; N represents the number of observed vehicles; P(λ|N) represents the posterior distribution of the traffic flow state λ given the observed data N; P(N|λ) is the observation model, which represents the probability of the observed data given λ; P(λ) is the prior distribution, and P(N) is the normalization constant;

[0014] Bayesian reasoning is used to perform multi-source data fusion on the data collected by each sensor. For each sensor i, its observation model P(N i |λ) is:

[0015]

[0016] in, Indicates the observation value N collected by sensor i under a given λ i Subject to λ as the mean, The variance is normally distributed, and the data collected by different sensors have different variances.

[0017] The observation data of multiple sensors are jointly modeled and expressed as:

[0018]

[0019] Among them, N1, N2, ..., N n are observation data from different sensors;

[0020] Based on Bayesian reasoning, the posterior distribution of the fused traffic flow state λ is expressed as:

[0021]

[0022] The maximum a posteriori probability estimate MAP of the arrival rate λ is expressed as:

[0023]

[0024] The specific process of removing noise from traffic flow data by Bayesian inference is as follows: define the relationship between the observed value N and the actual traffic flow state λ, expressed as:

[0025] N=λ+ξ,ξ~m(0,σ 2 );

[0026] Where ξ represents Gaussian noise. Introducing the observation model into the Bayesian formula, we get:

[0027]

[0028] The noise is removed through the prior distribution P(λ), and the estimated value of the traffic state is dynamically adjusted.

[0029] Furthermore, the method of introducing fuzzy logic into the Poisson process model to estimate the vehicle arrival rate within the fuzzy time interval and constructing a fuzzy Poisson distribution model specifically includes the following steps:

[0030] The vehicle arrival rate within the time interval t is defined as:

[0031]

[0032] Where λ(t) is the vehicle arrival rate within time t; λ0 is the vehicle arrival rate at the beginning of a specific time period; φ k is the ARIMA model state, δ t is the noise term, f(t) is the correction function based on external factors, and β is the weight;

[0033] The number of vehicles arriving within the time interval t, N(t), satisfies the Poisson distribution:

[0034]

[0035] Where P(N(t)=k) represents the probability of k vehicles arriving within time interval t; N(t) represents the number of vehicles arriving within time interval t; k represents the number of vehicles;

[0036] Then the cumulative distribution function of vehicle arrival is:

[0037]

[0038] Where F(N(t)≤k) represents the cumulative probability that the number of arriving vehicles is less than or equal to k; i is the summation index, which represents the number of vehicles that can arrive;

[0039] The superposition formula of multiple Poisson processes in different regions is:

[0040]

[0041] Where λtotal(t) represents the total vehicle arrival rate, which is the sum of the arrival rates of multiple regions; λi(t) is the vehicle arrival rate of the i-th region; n is the number of regions;

[0042] The arrival probability distribution in each area is:

[0043]

[0044] Where P(N i (t) = k i ) is the number of nodes that reach k in the i-th region i The probability of a car; N i (t) is the number of vehicles arriving in the ith area; ki is the number of vehicles arriving;

[0045] The conditional probability distribution based on the previous adjacent time period of time interval t is:

[0046]

[0047] Where P(N(t)|N(t-Δt)) is the probability of the number of vehicles arriving within time interval t given the arrival rate in the previous time period; Δt is the step size.

[0048] Furthermore, it also includes using fuzzy set A to represent the uncertainty of the vehicle arrival rate λ and state transition probability; assuming that λ and state transition probability P ij is a fuzzy number, and its membership functions are μ λ (x), μ Pij (x), for fuzzy numbers Its membership function is expressed as:

[0049]

[0050] where λ cis the central value of the arrival rate, σ λ is the standard deviation;

[0051] Then, when the vehicle arrival rate λ is fuzzy, the probability distribution of the number of vehicle arrivals n is expressed as:

[0052]

[0053] Among them, the membership function μ λ (λ) represents the fuzziness of the vehicle arrival rate, It means to sum all lambdas;

[0054] In order to further process the fuzzy Poisson distribution process, the fuzzy probability generating function of the fuzzy Poisson distribution process is defined. The fuzzy probability generating function is expressed as:

[0055]

[0056] Using the fuzzy probability generating function, the mean and variance of the fuzzy Poisson process are calculated, which are expressed as:

[0057]

[0058] in, represents the mean and S of the fuzzy Poisson process 2 Represents the variance of the fuzzy Poisson process.

[0059] Furthermore, the process of building the Markov chain-based fuzzy state transfer model specifically includes the following steps:

[0060] The state space S is defined as the position and velocity of the vehicle at any time: S i =(x i ,y i ,v i );where S i is the i-th state, including the vehicle's position (x i ,y i ) and speed v i ; xi is the horizontal coordinate of the vehicle in the i-th state; y i is the ordinate of the vehicle in the i-th state; v i is the speed of the vehicle in the i-th state;

[0061] State transition probability P(S i |S i-1 ) is expressed as:

[0062]

[0063] Where P(S i |S i-1) is from state S i-1 Transfer to S i The probability of P(v i |v i-1 ) is the speed from v i-1 The probability of transferring to vi;

[0064] Fuzzy state transition probability Assume that each element P in the state transition matrix ij is a fuzzy number, then the fuzzy state transition probability is expressed as:

[0065]

[0066] Furthermore, the construction of the state transfer matrix P comprises the following steps:

[0067] Assume there is a traffic state vector X and a state transfer matrix P. The traffic state is represented by the feature X = [x1, x2, ..., x n ], and the state transfer matrix is ​​represented by P = [p ij ] indicates that:

[0068] i=1,2,...,n,where μ i , are the mean and variance of the i-th feature respectively;

[0069] State transfer matrix P = [p ij ] is a row normalized matrix, each element represents the probability of transitioning from state i to state j:

[0070]

[0071] Among them, θ ij is the state between state i and state j;

[0072] A deep neural network is used to learn the state transition matrix, where the calculation process of each layer is expressed as:

[0073] h( l )=f(W( l )h( l-1 )+b( l ));

[0074] where h( l ) is the output of the lth layer; W( l ) and b( l ) are the weight matrix and bias vector of the lth layer respectively; f(·) is the activation function; the output of the final output layer is expressed as:

[0075] pij(t)=softmax(W·hij(t)+b+γ·g(Ot));

[0076] Among them, p ij (t) represents the state transition probability estimated by the deep neural network, h ij (t) is the output of the hidden layer of the deep neural network, W and b are weights and biases respectively, and g(Ot) is the current observation value O t Function for nonlinear transformation, γ is its weight;

[0077] The activation function softmax is defined as:

[0078]

[0079] The cross entropy loss function is used to measure the difference between the predicted state transfer matrix and the true matrix, which is expressed as:

[0080]

[0081] where y ij is the true state transition probability, is the predicted state transition probability matrix;

[0082] The state transfer matrix P is predicted by the neural network, and the output of the neural network is P = [p ij ], construct the traffic state estimation model, P = [p ij ] represents the transition probability from state i to state j; assuming that the state transition probability of step t is predicted, the state probability distribution is expressed as:

[0083] P(t) = Pt·P(0);

[0084] Where P(t) is the state transition probability matrix of the tth step, and P(0) is the probability distribution of the initial state.

[0085] Furthermore, the fuzzy joint probability density model is expressed as:

[0086]

[0087] in, is the fuzzy number of vehicle arrival rate within time interval t; is the probability of observing the state sequence within the time interval t; p ij (t) is the state transition probability estimated by the neural network; h ij (t) is the output of the hidden layer of the neural network, and the membership function μ λ (λ) represents the fuzziness of the vehicle arrival rate, represents the fuzzy membership function associated with the state transition probability, represents the probability distribution of the state at time t-1, h ij (t) is the hidden layer output of the neural network.

[0088] The present invention also includes a traffic state estimation system, comprising:

[0089] An acquisition module is used to acquire a historical traffic flow data set of a current traffic section; the traffic flow data set includes a vehicle arrival rate, a vehicle position and a speed;

[0090] The probability density model building module is used to introduce fuzzy logic into the Poisson process model to estimate the vehicle arrival rate within the fuzzy time interval according to the vehicle arrival rate, and build a fuzzy Poisson distribution model; according to the position and speed of the vehicle in different time periods, a traffic flow state transition model is established using the Markov chain, and the transition probability between traffic flow states is adjusted by introducing fuzzy logic into the model, and the dynamic transfer matrix is ​​introduced to capture the changes in traffic flow states in different time periods, so as to build a fuzzy state transition model based on the Markov chain; by combining the fuzzy Poisson distribution model with the fuzzy state transition model based on the Markov chain, a fuzzy joint probability density model of the traffic flow state and the vehicle arrival rate within a set time interval is constructed;

[0091] The estimation module is used to take the vehicle arrival rate and the transition probability of the traffic flow state of the traffic section to be tested in the current time period as the input of the fuzzy joint probability density model, output the probability distribution of the traffic flow state of the traffic section in a set time interval adjacent to the current time period, and estimate the traffic state of the traffic section in the set time interval based on the probability distribution.

[0092] The present invention also includes a traffic state estimation computer device, including: a memory, a processor, and a computer program stored in the memory, and the processor implements the steps of the traffic state estimation method when executing the computer program.

[0093] The present invention also includes a readable storage medium, wherein the readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, they are used to execute the steps of the traffic state estimation method.

[0094] The traffic state estimation method provided by the present invention has the following beneficial effects:

[0095] The present invention introduces fuzzy logic into the traditional Poisson process model to cope with the uncertainty and ambiguity in traffic flow and estimates the number of vehicle arrivals within the fuzzy time interval; uses Markov chain to model the transition of traffic states, uses fuzzy logic to adjust the transition probability between traffic states, and introduces a dynamic transition matrix to capture the changes in traffic flow states in different time periods, such as the possibility of changing from a smooth state to a congested state, and establishes a fuzzy state transition model based on Markov chain; by combining these two models, the estimation of vehicle trajectory and traffic status can be achieved in a variety of complex traffic scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] Figure 1 A flow chart of traffic state estimation in an embodiment of the present invention;

[0097] Figure 2 A flowchart of Bayesian reasoning in an embodiment of the present invention;

[0098] Figure 3 This is a simulation diagram of the Poisson distribution of vehicle arrival in an embodiment of the present invention;

[0099] Figure 4 It is a simulation diagram of the state transfer process in an embodiment of the present invention;

[0100] Figure 5 This is a flow chart of a fuzzy Poisson distribution modeling module in an embodiment of the present invention;

[0101] Figure 6 It is a fuzzy Poisson distribution diagram of vehicle arrival in an embodiment of the present invention;

[0102] Figure 7 This is a heat map of fuzzy inference rules in an embodiment of the present invention;

[0103] Figure 8 This is a flow chart of a state transition modeling module in an embodiment of the present invention;

[0104] Fig. 9 It is a prediction state transfer matrix diagram in an embodiment of the present invention;

[0105] Fig.10 Schematic diagram of state transition matrix of different samples in the embodiment of the present invention; Fig.10 (a) is the state transfer matrix of sample 1, Fig.10 (b) is the state transfer matrix of sample 2, Fig.10 (c) is the state transfer matrix of sample 3;

[0106] Fig.11 A comparison diagram of the predicted value and the true value of the neural network in the embodiment of the present invention;

[0107] Fig.12 This is a joint modeling flow chart in an embodiment of the present invention;

[0108] Fig.13 It is a distribution diagram of data and joint probability density in an embodiment of the present invention; Fig.13 (a) is the Poisson distribution data graph, Fig.13 (b) is the fuzzy Poisson distribution data graph. Fig.13 (c) is the Markov chain state transfer matrix, Fig.13 (d) is the state probability distribution radar chart of the joint model;

[0109] Fig.14 A comparison diagram of the results of Poisson distribution, fuzzy Poisson and Markov chain in an embodiment of the present invention; Fig.14 (a) is a schematic diagram of traffic state estimation based on fuzzy Poisson distribution and Markov chain. Fig.14 (b) is a schematic diagram of traffic state estimation based on Poisson distribution. DETAILED DESCRIPTION

[0110] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0111] The present invention proposes a traffic state estimation method, such as Figure 1 As shown, the method specifically comprises the following steps:

[0112] S1. Data collection and preprocessing module: First, traffic flow data is collected through multi-source sensors (such as traffic cameras, radars, GPS, etc.), including raw data such as vehicle arrival time, vehicle speed, and road occupancy rate. For the collected discrete data, fuzzy logic is used to preprocess it to establish the basis for uncertainty and fuzziness processing.

[0113] In this module, the idea of ​​Bayesian reasoning is combined to dynamically update the estimation of traffic flow status through multi-source data fusion and uncertainty processing. In this module, Bayesian reasoning is used to process information from different data sources (such as radar, camera, GPS, etc.) and deal with noise, uncertainty and real-time requirements, such as Figure 2 shown.

[0114] Definition 1: Bayesian inference: Bayesian inference is the ratio of the posterior probability to the prior probability, which indicates how likely it is that a hypothesis is true when data Z is observed, relative to the case where the hypothesis is false.

[0115] Set the traffic flow state to λ and perform Bayesian estimation on it. Assuming the prior probability is P(λ) and the observed data is N, according to the Bayesian formula:

[0116]

[0117] Where λ represents the arrival rate, N represents the number of observed vehicles, P(λ|N) represents the posterior distribution of the traffic flow state λ given the observation data N, P(N|λ) is the observation model, which represents the probability of the observation data given λ, P(λ) is the prior distribution, which reflects the prior knowledge of λ, and P(N) is the normalization constant.

[0118] For the data collected by different sensors, multi-source data fusion is performed and regarded as multiple observation values ​​N1, N2, ..., N n , the measurement data of each sensor has different noise characteristics and sampling frequencies, so Bayesian reasoning is needed for multi-source data fusion. For each sensor i, define its observation model P(N i |λ) is:

[0119]

[0120] in, Indicates the observation value N collected by sensor i under a given λ i Subject to λ as the mean, The variance is normal distribution, and the data of different sensors have different variances. It reflects the observation noise of different data sources. The observation data of multiple sensors can be jointly modeled by the following formula:

[0121]

[0122] Among them, N1, N2, ..., N n are observation data from different sensors, is the noise variance of each sensor.

[0123] Based on Bayesian reasoning, the posterior distribution of the fused traffic flow state λ can be expressed as:

[0124]

[0125] In order to obtain an estimate of the arrival rate λ, it is necessary to calculate its maximum a posteriori probability estimate MAP, as follows:

[0126]

[0127] Traffic state data inevitably has noise during the collection process, especially when the sensor hardware is limited and the external environment is affected. However, Bayesian reasoning can effectively eliminate the impact of noise and improve the quality of data by dynamically adjusting the model state. Assuming that the sensor noise follows a Gaussian distribution, Bayesian reasoning can handle uncertainty by constructing a probability distribution of the observation noise. The relationship between the observation value N and the actual traffic flow state λ is defined as follows:

[0128] N=λ+ξ,ξ~m(0,σ 2 )(5);

[0129] where ξ represents Gaussian noise, σ 2 Represents the noise variance and introduces the observation model into the Bayesian formula:

[0130]

[0131] Through the prior distribution P(λ), the noise can be effectively removed and the estimated value of the traffic state can be dynamically adjusted.

[0132] In order to further deal with uncertainty and ambiguity, Bayesian reasoning is combined with fuzzy logic, and fuzzy membership functions are used to modify prior distributions and observation models. Assume that the traffic state λ belongs to a fuzzy set A, and its fuzzy membership function is μ A (λ), use Bayesian inference to adjust the prior and posterior distributions according to the membership degree, then the new Bayesian inference formula is:

[0133]

[0134] Among them, the fuzzy membership μ A (λ) plays a role in correcting model uncertainty in the Bayesian framework and dynamically adjusts the estimation of traffic status according to real-time data.

[0135] S2, fuzzy Poisson distribution modeling module, such as Figure 5 As shown in Figure 1, in Poisson distribution, the vehicle arrival rate λ is a key state, reflecting the expected value of vehicle arrival per unit time. In the traditional model, λ is a fixed value, while in the present invention, by introducing fuzzy logic, λ is processed into a fuzzy number, and its fuzziness is represented by a membership function. The number of vehicle arrivals in a fuzzy time interval is estimated by fuzzy Poisson distribution, as shown in Figure 1. Figure 3 As shown, a joint probability model of fuzzy Poisson distribution is constructed.

[0136] Definition 1: Fuzzy Poisson process introduces fuzzy logic into the traditional Poisson process model to deal with the uncertainty and ambiguity in traffic flow. This model can handle the inherent ambiguity in actual traffic data, such as the ambiguity of vehicle arrival rate and time interval, thus providing realistic results for traffic state estimation.

[0137] Since the arrival of vehicles is discrete and random, Poisson distribution can be used to describe these events. At the same time, since the arrival of vehicles is disturbed by the traffic flow, its uncertainty needs to be considered, that is, the state (usually the event rate) is no longer an exact value, but a fuzzy set. Discrete random variable fuzzy Poisson distribution is a probability distribution model that combines the concepts of traditional Poisson distribution and fuzzy set theory.

[0138] In the fuzzy Poisson distribution, the event rate is a fuzzy number that can be expressed as a set of probability density functions, each function corresponding to a possible Poisson state value. In this way, for a given data observation, the likelihood of all possible states can be calculated, thus forming a more comprehensive probability interpretation, such as Figure 7 shown.

[0139] Define the vehicle arrival rate within the time interval:

[0140]

[0141] Where λ(t) is the vehicle arrival rate within time t; λ0 is the initial vehicle arrival rate; φ k is the ARIMA model state, δ t is the noise term, f(t) is the correction function based on external factors (such as weather, time, etc.), and β is its weight.

[0142] The number of vehicles arriving within the time interval t, N(t), satisfies the Poisson distribution:

[0143]

[0144] Where P(N(t)=k) represents the probability of k vehicles arriving within time t; N(t) represents the number of vehicles arriving within time t; k represents the specific number of vehicles.

[0145] Cumulative distribution function of vehicle arrivals:

[0146]

[0147] Where F(N(t)≤k) represents the cumulative probability that the number of arriving vehicles is less than or equal to k; i is the summation index, representing each possible number of arriving vehicles.

[0148] In addition, the superposition formula of multiple Poisson processes in different regions is as follows:

[0149]

[0150] where λ total (t) is the total vehicle arrival rate, which is the sum of the arrival rates of multiple areas;i (t) is the vehicle arrival rate in the i-th zone; n is the number of zones.

[0151] The arrival probability distribution in each area is:

[0152]

[0153] Where P(N i (t) = k i ) is the number of nodes that reach k in the i-th region i The probability of a car; N i (t) is the number of vehicles arriving in the i-th area; k i is the specific number of vehicle arrivals.

[0154] The conditional probability distribution based on the previous time period is:

[0155]

[0156] Where P(N(t)|N(t-Δt)) is the probability of the number of vehicles arriving in the current time period given the arrival rate in the previous time period; Δt is the step size.

[0157] Since the arrival of vehicles is uncertain, and fuzzy logic allows the introduction of uncertainty when dealing with traffic conditions, the uncertainty of the vehicle arrival rate λ and the state transition probability can be expressed by the fuzzy set A. So, assuming that the λ of this module and the P of the state transition module are ij is a fuzzy number, and its membership functions are μ λ (x),μ Pij (x), the membership function can be triangular, trapezoidal or Gaussian. Its membership function can be expressed as:

[0158]

[0159] where λ c is the central value of the arrival rate, σ λ is the standard deviation.

[0160] After fuzzification, the Poisson distribution can be expressed as a fuzzy Poisson distribution, and the formula is as follows:

[0161]

[0162] The formula represents the probability distribution of the number of vehicle arrivals n when the vehicle arrival rate λ is fuzzy, and the membership function μ λ (λ) represents the fuzziness of the vehicle arrival rate, and the integral represents the sum of all possible λ, such as Figure 6 shown.

[0163] In order to further process the fuzzy Poisson distribution process, the fuzzy probability generating function of the fuzzy Poisson distribution process is defined as follows:

[0164]

[0165] Using the fuzzy probability generating function, the mean and variance of the fuzzy Poisson process can be calculated as follows:

[0166]

[0167] The above two formulas represent the mean and variance of the fuzzy Poisson process respectively. The mean represents the expected number of vehicle arrivals after considering the fuzziness, and the variance reflects the volatility of the arrival number.

[0168] S3, state transition modeling and Markov process modules, such as Figure 8 As shown in the figure: Markov chain is used to model the transition of traffic state, and the transition probability between states is adjusted using fuzzy logic. By introducing a dynamic transfer matrix to capture the changes in traffic flow state in different time periods, combining historical data with real-time traffic information, a fuzzy state transition model based on Markov chain is established, as shown in the figure. Figure 4 shown.

[0169] The state space S is defined as the position and velocity of the vehicle at a certain moment:

[0170] S i =(x i ,y i ,v i )(19);

[0171] Where Si is the i-th state, including the position of the vehicle (x i ,y i ) and speed vi; xi is the horizontal coordinate of the vehicle in the i-th state; y i is the ordinate of the vehicle in the i-th state; v i is the speed of the vehicle in the i-th state.

[0172] State transition probability P(S i |S i-1 ):

[0173]

[0174] Where P(S i |S i-1 ) is the probability of transitioning from state Si-1 to Si; P(v i |v i-1 ) is the probability of velocity transfer from vi-1 to vi.

[0175] Fuzzy state transition probability Assume that each element P in the state transition matrix ij is a fuzzy number, and the fuzzy state transition probability can be expressed as:

[0176]

[0177] The state transfer matrix P is constructed as follows:

[0178] The learning-based state transition matrix is ​​a dynamic matrix automatically generated by a machine learning model, which is used to describe the transition probability between traffic system states. The core of this method is to use complex machine learning algorithms (such as deep neural networks, random forests, etc.) to train a large amount of historical traffic data to generate a state transition matrix that can accurately reflect the current traffic conditions. The present invention uses a deep learning neural network to learn the state transition matrix.

[0179] Use random variables and probability distribution to represent the data generation part. Suppose there is a traffic state vector X and a state transition matrix P. The traffic state is represented by the feature X = [x1, x2, ..., x n ], and the state transfer matrix is ​​represented by P = [p ij ] indicates. It can be assumed as follows:

[0180]

[0181] where μ i , are the mean and variance of the ith feature.

[0182] State transfer matrix P = [p ij ] is a row normalized matrix, each element represents the probability of transitioning from state i to state j:

[0183]

[0184] Among them, θ ij is the state between state i and state j.

[0185] Use a deep neural network to learn the state transfer matrix. Specifically, the calculation of each layer can be expressed as:

[0186] h( l )=f(W( l )h( l-1 )+b( l ))(twenty four);

[0187] where h (l) is the output of layer l; W (l) and b (l)are the weight matrix and bias vector of the lth layer respectively; f(·) is the activation function, commonly used are ReLU and Sigmoid functions, and the activation function used in this invention is the sigmoid function. A neural network is used to estimate the state transition probability in the Markov process, and the complex relationship between states is enhanced through deep learning. The final output layer is as follows:

[0188] p ij (t) = softmax(W·h ij (t)+b+γ·g(O t ))(25);

[0189] Among them, p ij (t) represents the state transition probability estimated by the deep neural network, h ij (t) is the output of the hidden layer of the deep neural network, W and b are weights and biases respectively, and g(O t ) is a function for nonlinear transformation of the current observation value Ot, and γ is its weight.

[0190] Among them, the softmax function is defined as follows:

[0191]

[0192] The cross entropy loss function is used to measure the difference between the predicted state transfer matrix and the true matrix. The formula is as follows:

[0193]

[0194] where y ij is the true state transition probability, is the predicted state transition probability matrix, such as Fig. 9 , Fig.11 shown.

[0195] Finally, the state transfer matrix P is predicted by the neural network prediction model. According to the output of the neural network P = [p ij ], construct a traffic state estimation model. This matrix represents the transition probability from state i to state j. Assuming that t-step prediction is required, the probability distribution of the state can be calculated by the following formula:

[0196] P(t)=P t ·P(0)(28);

[0197] Where P(t) is the state transition probability matrix of step t, and P(0) is the probability distribution of the initial state. In this way, the output vector P can become a matrix representing the state transition probability, such as Fig.10 The figure shows the transition matrix diagram of different sample states.

[0198] S4, Fuzzy Poisson Distribution and Markov Joint Probability Density Model Module: This module combines the fuzzy Poisson distribution with the Markov process model and proposes a traffic state estimation method based on the joint probability density function. By taking the vehicle arrival rate generated by the fuzzy Poisson distribution as the input of the probability density model and combining it with the state transition characteristics of the Markov process, it can not only effectively handle the randomness of vehicle arrival, but also deal with uncertainty through fuzzy logic, capture the randomness and state dependence of traffic flow, and thus provide more accurate traffic flow prediction results and improve the accuracy of traffic state estimation.

[0199] Combining the fuzzy Poisson distribution and the Markov process, we construct the joint probability density function of the vehicle state and arrival rate at time t. Combining the fuzzy Poisson distribution and the fuzzy state transition probability, we can get the fuzzy joint probability density function. The formula is as follows:

[0200]

[0201] in, is the fuzzy number of vehicle arrival rate within the time interval; P|X(t) is the state sequence observation probability at time t; p ij (t) is the state transition probability estimated by deep neural network obtained by module 2; h ij (t) is the output of the hidden layer of the deep neural network, which represents the fuzzy vehicle arrival rate λ and the fuzzy state transition probability P ij Under this condition, the number of vehicle arrivals n and the state transition S t The joint probability of ,the formula reflects the uncertainty of the data by integrating the fuzzy variables.

[0202] like Fig.12 As shown in the figure, Arabic numbers 1-9 refer to the order of relationships between nodes, such as from "data preprocessing" to "Poisson distribution modeling" to "fuzzy processing" and other steps. They are used to identify the process connections between steps in the flowchart. Fig.12 A to F stand for: data preprocessing, Poisson distribution modeling, fuzzy processing, Markov chain modeling, joint model solving, and result visualization. Fig.13 shown.

[0203] Compare the estimated values ​​of Poisson distribution and fuzzy Poisson distribution with the true state (assumed Poisson mean) respectively. Error calculation: Use mean square error to measure the deviation of 'Poisson estimation' and 'fuzzy Poisson + Markov chain estimation' from the true state. Error comparison: Output the estimation errors of the two models, and compare the error sizes to illustrate the superiority of the fuzzy Poisson + Markov model. Fig.14The estimation curves of Poisson distribution and fuzzy Poisson + Markov chain are shown to facilitate intuitive comparison of their accuracy, and the estimation results of Poisson distribution and fuzzy Poisson + Markov chain are plotted respectively. The results are: fuzzy Poisson + Markov estimation error: 0.001600000000000003, Poisson distribution estimation error: 21.777777777777782, which shows that the method proposed in the present invention is superior to the traditional method.

[0204] Based on the same inventive concept, the present invention also proposes a traffic state estimation system, comprising:

[0205] The acquisition module is used to obtain the historical traffic flow dataset of the current traffic section; the traffic flow dataset includes vehicle arrival rate, vehicle position and speed.

[0206] The probability density model construction module is used to introduce fuzzy logic into the Poisson process model according to the vehicle arrival rate to estimate the vehicle arrival rate within the fuzzy time interval, and build a fuzzy Poisson distribution model; according to the position and speed of the vehicle in different time periods, a traffic flow state transfer model is established using the Markov chain, and the transition probability between traffic flow states is adjusted by introducing fuzzy logic into the model, and the dynamic transfer matrix is ​​introduced to capture the changes in traffic flow states in different time periods, and a fuzzy state transfer model based on Markov chain is built; by combining the fuzzy Poisson distribution model with the fuzzy state transfer model based on Markov chain, a fuzzy joint probability density model of the traffic flow state and vehicle arrival rate within a set time interval is constructed.

[0207] The estimation module is used to take the vehicle arrival rate and the transition probability of the traffic flow state of the traffic section to be tested in the current time period as the input of the fuzzy joint probability density model, output the probability distribution of the traffic flow state of the traffic section in a set time interval adjacent to the current time period, and estimate the traffic state of the traffic section in the set time interval based on the probability distribution.

[0208] The present invention also proposes a traffic state estimation computer device, comprising: a memory, a processor, and a computer program stored in the memory, and the processor implements the steps of the traffic state estimation method when executing the computer program.

[0209] The present invention further proposes a readable storage medium, which stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of the traffic state estimation method.

[0210] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A traffic state estimation method, characterized in that: The following steps are involved: Acquire a historical traffic flow data set of the current traffic section; the traffic flow data set includes vehicle arrival rate, vehicle position and speed; According to the arrival rate of vehicles, fuzzy logic is introduced into the Poisson process model to estimate the arrival rate of vehicles within the fuzzy time interval, and a fuzzy Poisson distribution model is built; according to the position and speed of vehicles in different time periods, a traffic flow state transfer model is established using Markov chain, and the transition probability between traffic flow states is adjusted by introducing fuzzy logic into the model, and the dynamic transfer matrix is ​​introduced to capture the changes of traffic flow states in different time periods, and a fuzzy state transfer model based on Markov chain is built; by combining the fuzzy Poisson distribution model with the fuzzy state transfer model based on Markov chain, a fuzzy joint probability density model of traffic flow state and vehicle arrival rate within a set time interval is constructed; The vehicle arrival rate and the transition probability of the traffic flow state of the traffic section to be tested in the current time period are taken as the input of the fuzzy joint probability density model, and the probability distribution of the traffic flow state of the traffic section in a set time interval adjacent to the current time period is output. The traffic state of the traffic section in the set time interval is estimated based on the probability distribution.

2. A traffic state estimation method according to claim 1, characterized in that: After obtaining the historical traffic flow dataset of the current traffic section, the traffic flow dataset is preprocessed using Bayesian reasoning, and the processing process includes the following steps: Traffic flow data sets are collected through multi-source sensors. The traffic flow state is set to λ, and Bayesian estimation is performed on it. Assuming that the prior probability is P(λ) and the observed data is N, the following is obtained according to the Bayesian formula: Where λ represents the vehicle arrival rate; N represents the number of observed vehicles; P(λ|N) represents the posterior distribution of the traffic flow state λ given the observed data N; P(N|λ) is the observation model, which represents the probability of the observed data given λ; P(λ) is the prior distribution, and P(N) is the normalization constant; Bayesian reasoning is used to perform multi-source data fusion on the data collected by each sensor. Sensor i, defines its observation model P(N i |λ) is: in, Indicates the observation value N collected by sensor i under a given λ i Subject to λ as the mean, The variance is normally distributed, and the data collected by different sensors have different variances. The observation data of multiple sensors are jointly modeled and expressed as: Among them, N1, N2, ..., N n are observation data from different sensors; Based on Bayesian reasoning, the posterior distribution of the fused traffic flow state λ is expressed as: The maximum a posteriori probability estimate MAP of the arrival rate λ is expressed as: The specific process of removing noise from traffic flow data by Bayesian inference is as follows: define the relationship between the observed value N and the actual traffic flow state λ, expressed as: N=λ+ξ,ξ~m(0,σ 2 ); Where ξ represents Gaussian noise. Introducing the observation model into the Bayesian formula, we get: The noise is removed through the prior distribution P(λ), and the estimated value of the traffic state is dynamically adjusted.

3. A traffic state estimation method according to claim 1, characterized in that: The method of introducing fuzzy logic into the Poisson process model to estimate the vehicle arrival rate within the fuzzy time interval and constructing a fuzzy Poisson distribution model specifically includes the following steps: The vehicle arrival rate within the time interval t is defined as: Where λ(t) is the vehicle arrival rate within time t; λ0 is the vehicle arrival rate at the beginning of a specific time period; φ k is the ARIMA model state, δ t is the noise term, f(t) is the correction function based on external factors, and β is the weight; The number of vehicles arriving within the time interval t, N(t), satisfies the Poisson distribution: Where P(N(t)=k) represents the probability of k vehicles arriving within time interval t; N(t) represents the number of vehicles arriving within time interval t; k represents the number of vehicles; Then the cumulative distribution function of vehicle arrival is: Where F(N(t)≤k) represents the cumulative probability that the number of arriving vehicles is less than or equal to k; i is the summation index, which represents the number of vehicles that can arrive; The superposition formula of multiple Poisson processes in different regions is: where λ total (t) represents the total vehicle arrival rate, which is the sum of the arrival rates of multiple areas; λ i (t) is the vehicle arrival rate in the ith zone; n is the number of zones; The arrival probability distribution in each area is: Where P(N i (t) = k i ) is the number of nodes that reach k in the i-th region i The probability of a car; N i (t) is the number of vehicles arriving in the i-th area; k i is the number of vehicle arrivals; The conditional probability distribution based on the previous adjacent time period of time interval t is: Where P(N(t)|N(t-Δt)) is the probability of the number of vehicles arriving within time interval t given the arrival rate in the previous time period; Δt is the step size.

4. A traffic state estimation method according to claim 3, characterized in that: It also includes using fuzzy set A to represent the uncertainty of vehicle arrival rate λ and state transition probability; assuming that λ and state transition probability P ij is a fuzzy number, and its membership functions are μ λ (x), μ Pij (x), for fuzzy numbers Its membership function is expressed as: where λ c is the central value of the arrival rate, σ λ is the standard deviation; Then, when the vehicle arrival rate λ is fuzzy, the probability distribution of the number of vehicle arrivals n is expressed as: Among them, the membership function μ λ (λ) represents the fuzziness of the vehicle arrival rate, It means to sum all lambdas; In order to further process the fuzzy Poisson distribution process, the fuzzy probability generating function of the fuzzy Poisson distribution process is defined. The fuzzy probability generating function is expressed as: Using the fuzzy probability generating function, the mean and variance of the fuzzy Poisson process are calculated, which are expressed as: in, represents the mean and S of the fuzzy Poisson process 2 Represents the variance of the fuzzy Poisson process.

5. A traffic state estimation method according to claim 4, characterized in that: The process of building the fuzzy state transfer model based on the Markov chain specifically includes the following steps: The state space S is defined as the position and velocity of the vehicle at any time: S i =(x i ,y i ,v i );where S i is the i-th state, including the vehicle's position (x i ,y i ) and speed v i ;x i is the horizontal coordinate of the vehicle in the i-th state; y i is the ordinate of the vehicle in the i-th state; v i is the speed of the vehicle in the i-th state; State transition probability P(S i |S i-1 ) is expressed as: Where P(S i |S i-1 ) is from state S i-1 Transfer to S i The probability of P(v i |v i-1 ) is the speed from v i-1 Transfer to v i probability; Fuzzy state transition probability Assume that each element P in the state transition matrix ij is a fuzzy number, then the fuzzy state transition probability is expressed as:

6. A traffic state estimation method according to claim 5, characterized in that: The construction of the state transfer matrix P includes the following steps: Assume there is a traffic state vector X and a state transfer matrix P. The traffic state is represented by the feature X = [x1, x2, ..., x n ], and the state transfer matrix is ​​represented by P = [p ij ] indicates that: where μ i , are the mean and variance of the i-th feature respectively; State transfer matrix P = [p ij ] is a row normalized matrix, each element represents the probability of transitioning from state i to state j: Among them, θ ij is the state between state i and state j; A deep neural network is used to learn the state transition matrix, where the calculation process of each layer is expressed as: h (l) =f(W (l) h (l-1) +b (l) ); where h (l) is the output of layer l; W (l) and b (l) are the weight matrix and bias vector of the lth layer respectively; f(·) is the activation function; the output of the final output layer is expressed as: p ij (t)=softmax(W·h ij (t)+b+γ·g(O t )); Among them, p ij (t) represents the state transition probability estimated by the deep neural network, h ij (t) is the output of the hidden layer of the deep neural network, W and b are weights and biases respectively, and g(O t ) is the current observation value O t Function for nonlinear transformation, γ is its weight; The activation function softmax is defined as: The cross entropy loss function is used to measure the difference between the predicted state transfer matrix and the true matrix, which is expressed as: where y ij is the true state transition probability, is the predicted state transition probability matrix; The state transfer matrix P is predicted by the neural network, and the output of the neural network is P = [p ij ], construct the traffic state estimation model, P = [p ij ] represents the transition probability from state i to state j; assuming that the state transition probability of step t is predicted, the state probability distribution is expressed as: P(t)=P t ·P(0); Where P(t) is the state transition probability matrix of the tth step, and P(0) is the probability distribution of the initial state.

7. A traffic state estimation method according to claim 6, characterized in that: The fuzzy joint probability density model is expressed as: in, is the fuzzy number of vehicle arrival rate within time interval t; is the probability of observing the state sequence within the time interval t; p ij (t) is the state transition probability estimated by the neural network; h ij (t) is the output of the hidden layer of the neural network, and the membership function μ λ (λ) represents the fuzziness of vehicle arrival rate, μ Pij represents the fuzzy membership function associated with the state transition probability, represents the probability distribution of the state at time t-1, h ij (t) is the hidden layer output of the neural network.

8. A traffic state estimation system, characterized in that: include: An acquisition module is used to acquire a historical traffic flow data set of a current traffic section; the traffic flow data set includes a vehicle arrival rate, a vehicle position and a speed; The probability density model building module is used to introduce fuzzy logic into the Poisson process model to estimate the vehicle arrival rate within the fuzzy time interval according to the vehicle arrival rate, and build a fuzzy Poisson distribution model; according to the position and speed of the vehicle in different time periods, a traffic flow state transition model is established using the Markov chain, and the transition probability between traffic flow states is adjusted by introducing fuzzy logic into the model, and the dynamic transfer matrix is ​​introduced to capture the changes in traffic flow states in different time periods, so as to build a fuzzy state transition model based on the Markov chain; by combining the fuzzy Poisson distribution model with the fuzzy state transition model based on the Markov chain, a fuzzy joint probability density model of the traffic flow state and the vehicle arrival rate within a set time interval is constructed; The estimation module is used to take the vehicle arrival rate and the transition probability of the traffic flow state of the traffic section to be tested in the current time period as the input of the fuzzy joint probability density model, output the probability distribution of the traffic flow state of the traffic section in a set time interval adjacent to the current time period, and estimate the traffic state of the traffic section in the set time interval based on the probability distribution.

9. A traffic state estimation computer device, characterized in that: include: A memory, a processor, and a computer program stored in the memory, wherein the processor implements the steps of the traffic state estimation method according to any one of claims 1 to 7 when executing the computer program.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of the traffic state estimation method according to any one of claims 1 to 7.

Images