A traffic incident prediction method based on spatiotemporal Hawkes process

Abstract

The present invention discloses a traffic event prediction method based on a spatiotemporal Hawkes process, comprising the following steps: step 1, collecting historical spatiotemporal data of all types of traffic events; step 2, establishing a spatiotemporal Hawkes process model, which can describe the correlation and probability intensity of spatiotemporal data; step 3, estimating the parameters of the spatiotemporal Hawkes process model by training the spatiotemporal data; step 4, using the trained model to predict traffic events. The above technical solution can use the spatiotemporal Hawkes process model to effectively solve the problems and challenges faced by existing traffic event prediction methods, effectively capture spatiotemporal correlation based on spatiotemporal data, and accurately predict the occurrence of traffic events.

Description

Technical Field

[0001] The present invention relates to the field of traffic event prediction, and in particular to a traffic event prediction method based on a spatiotemporal Hawkes process, which utilizes the spatiotemporal Hawkes process to model the correlation between historical events in the traffic field and provide predictions of the probability of future events occurring. Background Art

[0002] The occurrence of traffic incidents has a significant impact on traffic flow and traffic management. Accurately predicting the occurrence of traffic incidents can help traffic management departments take corresponding measures to reduce traffic congestion, improve traffic safety, and provide more efficient traffic services.

[0003] Currently, many studies have explored methods for traffic incident prediction, but there are still some challenges, as follows: (1) Incomplete and inaccurate data. Traffic incident prediction usually relies on a large amount of real-time and historical traffic data, including traffic flow, speed, road status, etc. However, there may be problems with data acquisition and quality, such as data missing, noise, and errors, which will affect the accuracy and reliability of the prediction; (2) Challenges in dealing with spatiotemporal correlation. Traffic incidents usually have obvious spatiotemporal correlation, that is, events at a certain location may be affected by events at surrounding locations. However, traditional prediction methods often have difficulty in accurately capturing and modeling spatiotemporal correlation, resulting in inaccurate or unreliable prediction results; (3) Limited by the complexity of the model and computational efficiency, some existing traffic incident prediction methods use complex statistical models or machine learning algorithms, which require a lot of computing resources and time. This limits the feasibility and real-time performance of these methods in practical applications. Summary of the invention

[0004] The present invention provides a traffic incident prediction method based on the spatiotemporal Hawkes process, aiming to overcome the shortcomings of the prior art. The method can effectively solve the problems and challenges faced by the existing methods, accurately capture the spatiotemporal correlation based on spatiotemporal data and predict the occurrence of traffic incidents.

[0005] The method of the present invention provides a traffic incident prediction method based on the spatiotemporal Hawkes process, and the specific steps include the following steps:

[0006] Step 1: Collect historical spatiotemporal data of all types of traffic events;

[0007] Step 2: Establish a spatiotemporal Hawkes process model, which can describe the correlation and probability intensity of spatiotemporal data;

[0008] Step 3: Estimate the parameters of the spatiotemporal Hawkes process model by training the spatiotemporal data;

[0009] Step 4: Use the trained model to predict traffic events.

[0010] Preferably, in step 1, the historical spatiotemporal data E of n traffic events = {E 1 ,E 2 ,…,E k ,…,E n}, the historical spatiotemporal sequence of the kth traffic event is defined as in and Traffic incidents The timestamp information and spatial longitude and latitude information of the occurrence ( For events Longitude information, For events dimension information).

[0011] Preferably, in step 2, given the historical spatiotemporal data E of all n types of traffic events, a spatiotemporal Hawkes process model is constructed to calculate the kth target traffic event. Occurs at target timestamp and the target space longitude and latitude The probability intensity of is modeled as:

[0012]

[0013] Where: μ k is the kth traffic event e k The basic probability strength of occurrence, It is used to indicate similar historical traffic events. For target traffic events The impact of traffic events of the same type is shown in Figure 2. Other types of historical traffic events For target traffic events f2(·,·) is the impact function of different types of traffic events, f l (x)=1 / (1+exp(-x)) is the logistic function, which is used to ensure the non-negativity of the probability intensity.

[0014] Similar traffic incidents Target traffic incident The influence function of is defined as:

[0015]

[0016] in, Indicates that the impact of the same type of historical traffic events decays exponentially as the time interval increases, α kis the time attenuation coefficient of the impact of events of the same type (k), It means that the impact of historical traffic events of the same type decays exponentially as the spatial interval becomes larger. and is the spatial latitude and longitude information of the target traffic event, and Similar historical traffic events The longitude and latitude information, β k is the spatial attenuation coefficient of the impact of events of the same type (k). k It is a real number parameter indicating whether the relationship between events of the same type is negatively correlated (inhibitory effect) or positively correlated (stimulatory effect).

[0017] Other types of historical traffic events Target traffic incident The influence function of is defined as:

[0018]

[0019] in, It means that the impact of different types of historical traffic events decays exponentially as the time interval increases. represents the time attenuation coefficient of the influence of the k′th historical event on the kth target event, It means that the impact of historical traffic events of the same type decays exponentially as the spatial interval becomes larger. and are the latitude and longitude information of the target space, and For different types of historical traffic events The longitude and latitude information, is the spatial attenuation coefficient of the impact of the k′ event on the k event. is a real number parameter indicating whether the relationship between different types of events is negatively correlated (inhibitory effect) or positively correlated (stimulatory effect).

[0020] Preferably, in step 3, the objective function for estimating the parameters of the space-time Hawkes process model is formally defined as:

[0021]

[0022] in: Given n types of historical spatiotemporal data E of traffic events, the kth target traffic event Occurs at time and space The probability of

[0023] Then, the above objective function O is maximized by using the gradient descent optimization algorithm to obtain the optimal values ​​of all parameters and the final space-time Hawkes process model.

[0024] Preferably, the process of predicting traffic events using the trained model in step 4 is defined as given time information t and spatial information l, calculating the occurrence probabilities of all n types of traffic events and sorting them in descending order according to probability values, and finally outputting the sorted event list and probability values.

[0025] The present invention has the following characteristics and beneficial effects:

[0026] The spatiotemporal Hawkes process proposed in the present invention can more accurately model the spatiotemporal correlation between traffic events, capture the triggering and propagation effects between traffic events, and help predict future events. In addition, the method proposed in the present invention can process real-time traffic data and make event predictions based on the latest observations. This enables it to capture changes in traffic events in a timely manner and provide real-time prediction results. Finally, the method can make full use of existing spatiotemporal historical data, and can model discontinuous traffic event sequence data, providing accurate and efficient traffic event predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 The present invention provides a flow chart of a traffic incident prediction method based on a spatiotemporal Hawkes process. DETAILED DESCRIPTION

[0028] The traffic incident prediction method based on the spatiotemporal Hawkes process provided by the present invention is described in detail below. The flowchart of the method proposed by the present invention is as follows: Figure 1 As shown, the following steps are included:

[0029] Step 1: Collect the historical spatiotemporal data of all types of traffic events; specifically, the historical spatiotemporal data of all n types of traffic events are expressed as E = {E 1 ,E 2 ,…,E k ,…,E n}, the historical spatiotemporal sequence of the kth traffic event is defined as the time series in and Traffic incidents The timestamp information and spatial longitude and latitude information of the occurrence ( For events Longitude information, For events dimension information).

[0030] Step 2: Establish a spatiotemporal Hawkes process model, which can describe the correlation of various events in spatiotemporal data and the probability intensity of a certain traffic event; specifically, given the historical spatiotemporal data E of all n traffic events, construct a spatiotemporal Hawkes process model for the kth target traffic event Occurs at target timestamp and the target space longitude and latitude The probability intensity of is modeled as:

[0031]

[0032] Where: μ k is the kth traffic event e k The specific basic probability intensity of occurrence. Each type of traffic event has a specific basic probability intensity; It is used to indicate similar historical traffic events. For target traffic events The sum of the impact of , f1(·,·) is the spatial impact function of the same type of traffic events, Other types of historical traffic events For target traffic events f2(·,·) is the time-space impact function of different types of traffic events, and fl(x)=1 / (1+exp(-x)) is the logistic function, which is used to improve the nonlinear feature extraction capability of the model and ensure the non-negativity of the probability intensity.

[0033] It should be noted that in actual applications, all or part of the historical event data can be selected for modeling and solving according to actual needs.

[0034] Specifically, the same type of traffic incidents Target traffic incident The influence function of is defined as:

[0035]

[0036] in, Indicates that the impact of the same type of historical traffic events decays exponentially as the time interval increases, α k is the time attenuation coefficient of the impact of the same type (k) event, α k The larger the value is, the higher the corresponding attenuation degree is; It means that the impact of historical traffic events of the same type decays exponentially as the spatial interval becomes larger. and is the spatial latitude and longitude information of the target traffic event, and Similar historical traffic events Longitude and latitude information, β k is the spatial attenuation coefficient of the same type (k) event, β k The larger the value of , the higher the corresponding attenuation degree. k is a real number parameter.

[0037] In this embodiment, γ k <0 indicates that the relationship between events of the same type is negatively correlated (inhibitory effect), γ k >0 indicates that the relationship between events of the same type is positively correlated (stimulation effect). If the time or space information of a historical event is missing, the corresponding function is assigned a value of 1.

[0038] Similarly, other types of historical traffic events Target traffic incident The influence function of is defined as:

[0039]

[0040] in, It means that the impact of different types of historical traffic events decays exponentially as the time interval increases. represents the time attenuation coefficient of the influence of the k′th historical event on the kth target event, It means that the impact of historical traffic events of the same type decays exponentially as the spatial interval becomes larger. and are the latitude and longitude information of the target space, and For different types of historical traffic events The longitude and latitude information, is the spatial attenuation coefficient of the impact of the k′ event on the k event. is a real number parameter.

[0041] In this embodiment, Indicates that the relationship between different types of events is negatively correlated (inhibitory effect), Indicates that the relationship between different types of events is positively correlated (stimulatory effect).

[0042] Step 3: Estimate the parameters of the space-time Hawkes process model by training the space-time data. Specifically, the objective function for estimating the parameters of the space-time Hawkes process model is formally defined as:

[0043]

[0044] in: Given the historical spatiotemporal data E of all n traffic events and the kth target traffic event Occurs at time and space The probability of , formally defined as:

[0045]

[0046] in, Indicates that the kth target traffic event occurs at time and space The probability strength of Indicates other types of target traffic events occurring at time and space The probability strength.

[0047] Then, the above objective function O is maximized by using the gradient descent optimization algorithm to obtain the optimal values ​​of all parameters and the final space-time Hawkes process model.

[0048] Step 4: Use the trained model to predict traffic events.

[0049] Specifically, the process of predicting traffic events using the trained model is defined as given time information t and spatial information l, calculating the occurrence probabilities of all n types of traffic events and sorting them in descending order according to the probability values, and finally outputting the sorted event list and probability values.

[0050] In this embodiment, for a given time information t and space information l, the occurrence probability of the kth type of event (k∈[1,2,3,…,n]) is defined as:

[0051]

[0052]

[0053]

[0054] On this basis, the occurrence probabilities of all n types of traffic events are sorted in descending order according to the probability values, and finally the sorted event list and probability values ​​are output.

[0055] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments including components are made without departing from the principles and spirit of the present invention, and still fall within the scope of protection of the present invention.

Claims

1. A traffic incident prediction method based on spatiotemporal Hawkes process, characterized in that: The following steps are involved: Step 1: Collect the historical spatiotemporal data of all types of traffic events and define the historical spatiotemporal data of n types of traffic events E = {E 1 ,E 2 ,…,E k ,…,E n }, the historical spatiotemporal sequence of the kth traffic event is defined as in and Traffic incidents The timestamp information and spatial longitude and latitude information of the occurrence, among which, For events Longitude information, For events Dimension information of Step 2: Establish the time-space Hawkes process model, the expression is as follows: Among them, μ k is the kth traffic event e k The basic probability strength of occurrence, Indicates similar historical traffic events For target traffic events The impact of traffic events of the same type is shown in Figure 2. Other types of historical traffic events For target traffic events The impact of traffic incidents, f2(·,·) is the impact function of different types of traffic incidents, fl(x)=1 / (1+exp(-x)) is the logistic function; Step 3: Estimate the parameters of the spatiotemporal Hawkes process model by training the spatiotemporal data; The modeling method is: First, the objective function for estimating the parameters of the space-time Hawkes process model is formally defined as: in: Given n types of historical spatiotemporal data E of traffic events, the kth target traffic event In time and space The probability of occurrence, Then, the objective function O is maximized by using the gradient descent optimization algorithm to obtain the optimal values ​​of all parameters and the final time-space Hawkes process model. Step 4: Use the trained model to predict traffic events.

2. The traffic incident prediction method based on the spatiotemporal Hawkes process according to claim 1, characterized in that: The influence of the same type of traffic events on the target traffic event includes negative correlation and positive correlation, the negative correlation plays an inhibitory role, and the positive correlation plays a stimulating role.

3. The traffic incident prediction method based on the spatiotemporal Hawkes process according to claim 2 is characterized in that: In the step 2, Traffic incidents of the same type Target traffic incident The influence function of is defined as: in, Indicates that the impact of the same type of historical traffic events decays exponentially as the time interval increases, α k is the time attenuation coefficient of the impact of events of the same type (k), It means that the impact of historical traffic events of the same type decays exponentially as the spatial interval becomes larger. and is the spatial latitude and longitude information of the target traffic event, and Similar historical traffic events The longitude and latitude information, β k is the spatial attenuation coefficient of the same type of events, γ k is a real number parameter.

4. The traffic incident prediction method based on the spatiotemporal Hawkes process according to claim 3 is characterized in that: Traffic incidents of the same type Target traffic incident In the influence function of k Indicates whether the relationship between different types of events is negative or positive.

5. The traffic incident prediction method based on the spatiotemporal Hawkes process according to claim 2 is characterized in that: In the step 2, Other types of historical traffic events Target traffic incident The influence function of is defined as: in, It means that the impact of different types of historical traffic events decays exponentially as the time interval increases. represents the time attenuation coefficient of the influence of the k′th historical event on the kth target event, It means that the impact of historical traffic events of the same type decays exponentially as the spatial interval becomes larger. and Target traffic events Spatial latitude and longitude information, and For different types of historical traffic events Spatial latitude and longitude information, is the spatial attenuation coefficient of the effect of the k′ event on the k event, is a real number parameter.

6. The traffic incident prediction method based on spatiotemporal Hawkes process according to claim 5, characterized in that: Other types of historical traffic events Target traffic incident In the influence function of Indicates whether the relationship between different types of events is negative or positive.

7. The traffic incident prediction method based on the spatiotemporal Hawkes process according to claim 1, characterized in that: In step 4: the process of predicting traffic events using the trained model is defined as given time information t and space information l, calculating the occurrence probabilities of all n types of traffic events and sorting them in descending order according to probability values, and finally outputting the sorted event list and probability values.

Images