Real-time identification method, device and system for adverse traffic conditions

Abstract

The present invention discloses a method, device and system for real-time identification of adverse traffic conditions, the method comprising: sending the acquired historical road traffic data to a pre-trained traffic condition identification model to obtain the traffic condition evaluation result of each monitoring point; the traffic condition identification model is a multi-layer perceptron model based on the Swish activation function, including an input layer, multiple hidden layers and an output layer, each layer is connected by a full connection, and a cross-validation technique is used to adjust the hyperparameters during the training of the traffic condition identification model; based on the traffic condition evaluation result of each monitoring point, a threshold judgment method is used to determine the area with congestion or potential risks. The present invention can realize high-precision real-time identification in a complex and changeable urban traffic environment, timely discover and warn of adverse traffic conditions, and improve the efficiency and safety of traffic management.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent traffic technology, and specifically relates to a method, device and system for real-time identification of adverse traffic conditions. Background Art

[0002] In recent years, with the acceleration of urbanization, traffic congestion and safety issues have become increasingly prominent. Intelligent Transportation System (ITS) has realized the real-time collection, processing, analysis and release of traffic information by integrating information technology, communication technology, sensor technology and automation technology, thereby significantly improving traffic efficiency and safety. However, traditional traffic state identification methods rely on fixed thresholds and rules, which are static, lagging and inaccurate when dealing with dynamically changing traffic environments. Although machine learning technologies, such as decision trees, random forests and support vector machines, have improved the identification effect to a certain extent through historical data learning, they still have limitations when processing high-dimensional and nonlinear data, and it is difficult to meet the needs of real-time identification.

[0003] In contrast, deep learning technology, especially the Multilayer Perceptron (MLP) combined with the Swish activation function, can provide higher prediction accuracy and real-time performance when processing complex traffic data due to its powerful automatic feature learning ability and smooth and non-monotonic characteristics, making it an ideal choice for real-time identification of adverse traffic conditions. Summary of the invention

[0004] In response to the above problems, the present invention proposes a method, device and system for real-time identification of adverse traffic conditions, which can achieve high-precision real-time identification in a complex and changeable urban traffic environment, timely discover and warn of adverse traffic conditions, and improve the efficiency and safety of traffic management.

[0005] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions:

[0006] In a first aspect, the present invention provides a method for real-time identification of adverse traffic conditions, comprising:

[0007] The acquired historical road traffic data is sent to a pre-trained traffic state recognition model to obtain the traffic state evaluation result of each monitoring point; the traffic state recognition model is a multi-layer perceptron model based on the Swish activation function, including an input layer, a hidden layer and an output layer, and each layer is connected by a full connection. The cross-validation technology is used to adjust the hyperparameters during the training process of the traffic state recognition model;

[0008] Based on the traffic status assessment results of each monitoring point, the threshold judgment method is used to determine areas with congestion or potential risks.

[0009] In combination with the first aspect, optionally, the method for acquiring historical road traffic data includes:

[0010] Obtaining original road traffic data;

[0011] Performing data fusion on the original road traffic data, integrating data from different sources into a unified data stream, and obtaining first road traffic data;

[0012] Cleaning the first road traffic data, removing invalid data points, and filling missing values ​​to obtain second road traffic data;

[0013] Feature extraction is performed on the second road traffic data to obtain historical road traffic data.

[0014] In combination with the first aspect, optionally, when removing invalid data points, the calculation formula used is:

[0015] ;

[0016] in, It is the standardized data points, It is The original value of the data point, It is Data points The weight of is the weighted mean, is the weighted standard deviation; if , then it is believed that is an invalid data point. is the threshold value, is the total number of data points.

[0017] In combination with the first aspect, optionally, when filling missing values, a calculation formula is used:

[0018] ;

[0019] in, is the timestamp of the previous valid data point, is the timestamp of the next valid data point, is the timestamp of the data point to be interpolated, is the weight of the previous valid data point, is the weight of the next valid data point, is the adjusted weight of the previous valid data point, is the weight of the next valid data point after adjustment, is the value of the previous valid data point, is the value of the next valid data point, is the interpolated value.

[0020] In combination with the first aspect, optionally, extracting features from the second road traffic data includes:

[0021] Based on the second road traffic data, the traffic density of each monitoring point is calculated. , ,in, It is Monitoring points at time Traffic density reflects the density of vehicles on the road; It is The monitoring point The weight of each type of vehicle; It's in time No. The monitoring point Number of vehicles of each type; It's in time No. The monitoring point Effective road capacity for different types of vehicles;

[0022] Based on the second road traffic data and the traffic density of each monitoring point , using a recursive feature elimination algorithm, gradually removing the most irrelevant features, and generating feature vectors as historical road traffic data; the selection formula used by the recursive feature elimination algorithm is: ,in, is the initial feature set, is the number of features selected.

[0023] In combination with the first aspect, optionally, the input layer of the multilayer perceptron model receives a feature vector , where the eigenvector Include Features

[0024] The hidden layer contains a preset number of neurons, and the formula used by the hidden layer to process data is: ,in, Represents the hidden layer The output of a neuron, is the input feature vector Features, is the Swish activation function, is the weight matrix representing the weight from the input layer nodes to the hidden layer The connection weights of neurons, is the hidden layer The bias term of each neuron; Swish activation function The expression is: , , , is the Sigmoid function;

[0025] The output of the hidden layer is equipped with a Batch Normalization layer, which converts the output of each neuron into Normalized, then passed through learnable parameters and Scaling and translation are performed to obtain the final normalized output , the calculation formula used by the Batch Normalization layer is:

[0026] ;

[0027] in, is the hidden layer The output of a neuron, and are the mean and variance respectively, is a constant, which can be , and is a learnable parameter;

[0028] The output layer contains one neuron that directly outputs the traffic density using a linear activation function:

[0029]

[0030] in, is the traffic density, is the weight matrix of the output layer, which represents the connection weight from each neuron in the last hidden layer to the output layer; is the output of the hidden layer, is the bias term of the output layer.

[0031] In combination with the first aspect, optionally, when performing multi-layer perceptron model training, the following steps are included:

[0032] Initialize parameters, initialize weights and bias ;

[0033] Forward propagation, calculate the predicted probability of the output layer ;

[0034] Calculate the loss, using mean squared error as the loss function:

[0035]

[0036] in, is the true value, is the predicted value, is the sample size;

[0037] Back propagation, calculation of gradients, and update of parameters:

[0038]

[0039] in, is the learning rate;

[0040] Iterative training, repeating forward propagation and back propagation until convergence or reaching the maximum number of iterations;

[0041] The hyperparameters of the multilayer perceptron model are obtained by the following steps:

[0042] Divide the data set into K subsets;

[0043] K-fold cross validation, taking each subset as the validation set and the remaining subsets as the training set; training and validation are performed on each subset;

[0044] Calculate the accuracy evaluation indicators of the multi-layer perceptron model and continuously adjust the hyperparameters.

[0045] In combination with the first aspect, optionally, a method for calculating the threshold of traffic density in the threshold determination method includes:

[0046] Calculate the mean of traffic density and standard deviation :

[0047] ;

[0048] in, It is Monitoring points at time The traffic density, is the total number of monitoring points;

[0049] Mean based on traffic density and standard deviation Determining the threshold of traffic density , the threshold The calculation formula is:

[0050] ;

[0051] in, is an empirical coefficient.

[0052] In a second aspect, the present invention provides a device for real-time identification of adverse traffic conditions, comprising:

[0053] The traffic state evaluation result prediction module is used to send the acquired historical road traffic data to a pre-trained traffic state recognition model to obtain the traffic state evaluation result of each monitoring point; the traffic state recognition model is a multi-layer perceptron model based on the Swish activation function, including an input layer, multiple hidden layers and an output layer, and the layers are connected in a fully connected manner. The cross-validation technology is used to adjust the hyperparameters during the training process of the traffic state recognition model;

[0054] The risk identification module is used to determine areas with congestion or potential risks based on the traffic status assessment results of each monitoring point and using a threshold judgment method.

[0055] In a third aspect, the present invention provides a real-time identification system for adverse traffic conditions, including a storage medium and a processor;

[0056] The storage medium is used to store instructions;

[0057] The processor is configured to operate according to the instructions to execute the method according to any one of the first aspects.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] By using the Swish activation function, the model can better fit complex nonlinear relationships, thereby improving the prediction accuracy of adverse traffic conditions; the non-monotonicity and smoothness of the Swish function help capture subtle changes in traffic data, making the model more accurate in identifying traffic congestion, accidents and the like.

[0060] The present invention utilizes an efficient forward propagation algorithm to complete the processing and analysis of a large amount of real-time data in a short period of time, thereby realizing real-time identification of adverse traffic conditions; this helps traffic management departments to promptly discover and respond to emergencies, reduce traffic delays and safety hazards.

[0061] The Swish multi-layer perceptron in the present invention can automatically learn and extract useful features from high-dimensional traffic data, reduce the dependence on artificial feature engineering, and enhance the generalization ability of the model; this means that the model can not only perform well on the training data, but also maintain a high recognition accuracy on new data that has not been seen.

[0062] The present invention simplifies the feature engineering process and reduces the workload of data preprocessing through the automatic feature learning capability of the multi-layer perceptron (MLP); the model can automatically extract features that are valuable for traffic status identification from the original data, thereby improving the overall work efficiency.

[0063] The deep learning-based model of the present invention can dynamically adjust the prediction model according to real-time data to adapt to the ever-changing traffic environment; this enables the system to maintain a high recognition accuracy at different times and places, enhancing the flexibility and robustness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor, among which:

[0065] Figure 1 A framework diagram of a method for real-time identification of adverse traffic conditions according to an embodiment of the present invention;

[0066] Figure 2 Schematic diagram of the process of using the threshold judgment method. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0069] Example 1

[0070] An embodiment of the present invention provides a method for real-time identification of adverse traffic conditions, comprising the following steps:

[0071] (1) The acquired historical road traffic data is sent to a pre-trained traffic state recognition model to obtain the traffic state evaluation result of each monitoring point; the traffic state recognition model is a multi-layer perceptron model based on the Swish activation function, including an input layer, multiple hidden layers and an output layer, and the layers are connected in a fully connected manner. The cross-validation technology is used to adjust the hyperparameters during the training process of the traffic state recognition model;

[0072] (2) Based on the traffic status assessment results of each monitoring point, a threshold judgment method is used to determine areas with congestion or potential risks.

[0073] In a specific implementation of the embodiment of the present invention, the method for acquiring historical road traffic data includes:

[0074] Obtaining original road traffic data;

[0075] Performing data fusion on the original road traffic data, integrating data from different sources into a unified data stream, and obtaining first road traffic data;

[0076] Cleaning the first road traffic data, removing invalid data points, and filling missing values ​​to obtain second road traffic data;

[0077] Feature extraction is performed on the second road traffic data to obtain historical road traffic data.

[0078] Among them, when removing invalid data points, the calculation formula used is:

[0079] ;

[0080] in, is the normalized data point, It is The original value of the data point, It is Data points The weight of is the weighted mean, is the weighted standard deviation; if , then it is believed that is an invalid data point. is the threshold value, is the total number of data points.

[0081] When filling missing values, the calculation formula is:

[0082] ;

[0083] in, is the timestamp of the previous valid data point, is the timestamp of the next valid data point, is the time point that needs to be interpolated. is the weight of the previous valid data point, is the weight of the next valid data point, is the adjusted weight of the previous valid data point, is the weight of the next valid data point after adjustment, is the value of the previous valid data point, is the value of the next valid data point, is the interpolated value.

[0084] The extracting features from the second road traffic data comprises:

[0085] Based on the second road traffic data, the traffic density of each monitoring point is calculated. , ,in, It is Monitoring points at time Traffic density reflects the density of vehicles on the road; It is The monitoring point The weight of each type of vehicle; It's in time Neidi The monitoring point Number of vehicles of each type; It's in time Neidi The monitoring point Effective road capacity for different types of vehicles;

[0086] Based on the second road traffic data and the traffic density of each monitoring point , using a recursive feature elimination algorithm, gradually removing the most irrelevant features, and generating feature vectors as historical road traffic data; the selection formula used by the recursive feature elimination algorithm is: ,in, is the initial feature set, is the number of features selected.

[0087] In a specific implementation of the embodiment of the present invention, the input layer of the multi-layer perceptron model receives the feature vector , where the eigenvector Include Features

[0088] The hidden layer contains a preset number of neurons, and the formula used by the hidden layer to process data is: ,in, Represents the hidden layer The output of a neuron, is the input feature vector Features, is the Swish activation function, is the weight matrix representing the nodes to the hidden layer The connection weights of neurons, is the hidden layer The bias term of each neuron; Swish activation function The expression is: , , , is the Sigmoid function;

[0089] The output of the hidden layer is equipped with a Batch Normalization layer to convert the output of each neuron Normalized, then passed through learnable parameters and Scaling and translation are performed to obtain the final normalized output , the calculation formula used by the BatchNormalization layer is:

[0090] ;

[0091] in, is the hidden layer The output of a neuron, and are the mean and variance of the small batch data, is a small constant , and is a learnable parameter;

[0092] The output layer contains one neuron that directly outputs the traffic density using a linear activation function:

[0093]

[0094] in, is the traffic density, is the weight matrix of the output layer, which represents the connection weight from each neuron in the last hidden layer to the output layer; is the output of the last hidden layer, is the bias term of the output layer.

[0095] In a specific implementation of the embodiment of the present invention, when performing multi-layer perceptron model training, the following steps are included:

[0096] Initialize parameters, initialize weights and bias ;

[0097] Forward propagation, calculate the predicted probability of the output layer ;

[0098] Calculate the loss and use the cross entropy loss function:

[0099]

[0100] in, is the true value, is the predicted value, is the sample size;

[0101] Back propagation, calculation of gradients, and update of parameters:

[0102]

[0103] in, is the learning rate;

[0104] Iterate training, repeating forward propagation and backpropagation until convergence or reaching the maximum number of iterations.

[0105] In a specific implementation of the embodiment of the present invention, the hyperparameters of the multilayer perceptron model are obtained by the following steps:

[0106] Divide the dataset into K subsets.

[0107] K-fold cross validation, taking each subset as the validation set and the remaining subsets as the training set; training and validation are performed on each subset.

[0108] Calculate the accuracy evaluation indicators of the multi-layer perceptron model and continuously adjust the hyperparameters.

[0109] The method in the embodiment of the present invention is described in detail below in conjunction with a specific implementation manner.

[0110] like Figure 1 As shown, the method for real-time identification of adverse traffic conditions includes the following steps:

[0111] Step 1: Data collection, collecting real-time traffic flow through various sensors installed along the highway (such as cameras, radars, infrared detectors, etc.) , average speed , Lane occupancy and other information; using the vehicle communication system V2X technology to obtain vehicle Location ,speed and acceleration Dynamic data such as rainfall; get weather information from weather stations , wind speed , Monitoring Point Visibility and other environmental parameters; obtain construction information, accident reports and other non-periodic event information that affects traffic mobility from the road monitoring center; aggregate the above multi-source data through the Internet of Things (IoT) platform to form original road traffic data.

[0112] Step 2: Perform data fusion on the original road traffic data, integrate the data from different sources into a unified data stream, and obtain the first road traffic data. Specifically, use data fusion technology to integrate the data from different sources into a unified data stream, and adopt the Kalman Filter algorithm: , smooth the sensor data to improve the data quality. The process of processing data by the Kalman filter algorithm is a prior art and will not be described in detail in this application;

[0113] Cleaning the first road traffic data, removing invalid data points, and filling missing values ​​to obtain second road traffic data;

[0114] When removing invalid data points, the calculation formula used is:

[0115] ;

[0116] in, is the normalized data point, It is The original value of the data point, It is Data points The weight of is the weighted mean, is the weighted standard deviation; if , then it is believed that is an invalid data point. is the threshold value, is the total number of data points;

[0117] When filling missing values, the calculation formula is:

[0118] ;

[0119] in, is the timestamp of the previous valid data point, is the timestamp of the next valid data point, is the time point that needs to be interpolated. is the weight of the previous valid data point, is the weight of the next valid data point, is the adjusted weight of the previous valid data point, is the weight of the next valid data point after adjustment, is the value of the previous valid data point, is the value of the next valid data point, is the interpolated value.

[0120] The second road traffic data is subjected to feature extraction to obtain historical road traffic data. The feature selection method in the library (recursive feature elimination RFE) is used to select the most representative features and reduce redundant features; specifically, the traffic density of each monitoring point is calculated , ,in, It is Monitoring points at time Traffic density reflects the density of vehicles on the road; It is The monitoring point The weight of each type of vehicle; It's in time Neidi The monitoring point Number of vehicles of each type; It's in time Neidi The monitoring point Based on the second road traffic data and the traffic density of each monitoring point , using a recursive feature elimination algorithm, gradually removing the most irrelevant features, and generating feature vectors as historical road traffic data; the selection formula used by the recursive feature elimination algorithm is: ,in, is the initial feature set, is the number of features selected.

[0121] As shown in Table 1, the following is the display of feature extraction results based on the key road sections of a city and the data of a certain year, month and day:

[0122] Table 1 Display of feature extraction results

[0123] Timestamp Traffic density Average speed (km / h) Ratio of vehicle types (passenger cars / trucks / motorcycles) Road occupancy rate Temperature (℃) Visibility (m) Wind speed (m / s) 14:00 0.75 30 0.6 / 0.3 / 0.1 0.8 20 10000 5 14:01 0.78 28 0.55 / 0.35 / 0.1 0.82 19 10000 5 14:02 0.8 26 0.5 / 0.4 / 0.1 0.85 20 8000 6 14:03 0.82 25 0.45 / 0.45 / 0.1 0.88 19 6000 7

[0124] Step 3: Establish a traffic state recognition model: Build a Swish-based multi-layer perceptron (MLP) model, which includes an input layer, a hidden layer (each hidden layer contains a certain number of neurons) and an output layer. The multi-layer perceptron (MLP) model uses Activation function: , , , It is a Sigmoid function; and Batch Normalization technology is used to prevent overfitting: ,in, is the hidden layer The output of a neuron, and are the mean and variance of the small batch data, is a small constant , and is a learnable parameter;

[0125] Training traffic state recognition model: Through the back propagation algorithm ( ) Adjust the weights to minimize the loss function (cross entropy loss function: ); Model validation: The accuracy and robustness of the model are evaluated through cross-validation technology (K-fold cross-validation), and the hyperparameters of the traffic state recognition model are adjusted according to the validation results.

[0126] The acquired historical road traffic data is sent to a pre-trained traffic state recognition model to obtain the traffic state evaluation result of each monitoring point, that is, the traffic density of each monitoring point;

[0127] Step 4: Use the threshold judgment method (set the congestion index to be greater than a certain threshold as an unfavorable traffic state). The threshold is determined by statistically analyzing the collected traffic density data and then calculating the mean of the traffic density. and standard deviation :

[0128]

[0129] in, It is Monitoring points at time The traffic density, is the total number of monitoring points;

[0130] Mean based on traffic density and standard deviation Determining the threshold of traffic density , the threshold The calculation formula is:

[0131] ;

[0132] in, It is an empirical coefficient, usually ranging from 1 to 3;

[0133] If the traffic density at the monitoring point Greater than or equal to threshold , then the monitoring point is considered to be in an unfavorable traffic state and all areas with congestion or potential risks are determined. For details, see Figure 2 .

[0134] Step 5: Output all areas with congestion or potential risks for further analysis or early warning. Display all identified areas with congestion or potential risks through visualization tools, send early warning information to the driver's infotainment system or mobile device through the wireless communication module, and record important decision-making basis in each identification process.

[0135] Example 2

[0136] Based on the same inventive concept as that of Embodiment 1, an embodiment of the present invention provides a real-time identification device for adverse traffic conditions, comprising:

[0137] The traffic state evaluation result prediction module is used to send the acquired historical road traffic data to a pre-trained traffic state recognition model to obtain the traffic state evaluation result of each monitoring point; the traffic state recognition model is a multi-layer perceptron model based on the Swish activation function, including an input layer, multiple hidden layers and an output layer, and the layers are connected in a fully connected manner. The cross-validation technology is used to adjust the hyperparameters during the training process of the traffic state recognition model;

[0138] The risk identification module is used to determine areas with congestion or potential risks based on the traffic status assessment results of each monitoring point and using a threshold judgment method.

[0139] Example 3

[0140] Based on the same inventive concept as that of Embodiment 1, a real-time identification system for adverse traffic conditions is provided in an embodiment of the present invention, including a storage medium and a processor;

[0141] The storage medium is used to store instructions;

[0142] The processor is configured to operate according to the instructions to execute the method according to any one of the embodiments 1.

[0143] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0144] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0145] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0146] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0147] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

[0148] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for real-time identification of adverse traffic conditions, characterized in that: include: The acquired historical road traffic data is sent to a pre-trained traffic status recognition model to obtain the traffic status evaluation results of each monitoring point; The traffic state recognition model is a multi-layer perceptron model based on the Swish activation function, including an input layer, a hidden layer and an output layer, and each layer is connected in a fully connected manner. The cross-validation technology is used to adjust the hyperparameters during the training process of the traffic state recognition model; Based on the traffic status evaluation result of each monitoring point, a threshold judgment method is used to determine the area with congestion or potential risk, wherein the traffic status evaluation result is traffic density; The method for acquiring historical road traffic data comprises: Obtaining original road traffic data; Performing data fusion on the original road traffic data, integrating data from different sources into a unified data stream, and obtaining first road traffic data; Cleaning the first road traffic data, removing invalid data points, and filling missing values ​​to obtain second road traffic data; performing feature extraction on the second road traffic data to obtain historical road traffic data; When removing invalid data points, the calculation formula used is: Among them, z i is the i-th data point after standardization, x i is the original value of the ith data point, w i is the i-th data point x i The weight, μ w is the weighted mean, σ w is the weighted standard deviation; if |z i ∣>Z, then x i is an invalid data point, Z is the threshold, and N1 is the total number of data points; When filling missing values, the calculation formula is: Among them, t1 is the timestamp of the previous valid data point, t2 is the timestamp of the next valid data point, t is the timestamp of the data point to be interpolated, ω1 is the weight of the previous valid data point, ω2 is the weight of the next valid data point, ω'1 is the adjusted weight of the previous valid data point, ω'2 is the adjusted weight of the next valid data point, y1 is the value of the previous valid data point, and y2 is the value of the next valid data point. is the interpolated value; The extracting features from the second road traffic data comprises: Based on the second road traffic data, the traffic density D of each monitoring point is calculated. k (t), Among them, D k (t) is the traffic density of the kth monitoring point at time t, reflecting the density of vehicles on the road; a k,j is the weight of the j-th type of vehicle at the k-th monitoring point; V k,j (t) is the number of vehicles of the jth type at the kth monitoring point at time t; F k,j (t) is the effective road capacity for the jth type of vehicle at the kth monitoring point at time t; Based on the second road traffic data and the traffic density D of each monitoring point k (t), using a recursive feature elimination algorithm to gradually remove the most irrelevant features and generate a feature vector as the historical road traffic data; the selection formula used by the recursive feature elimination algorithm is: RFE(S, N2) = select N2 features from S, where S is the initial feature set and N2 is the number of selected features; The calculation method of the traffic density threshold in the threshold judgment method includes: Calculate the mean μ of traffic density D and standard deviation σ D : Among them, D k (t) is the traffic density of the kth monitoring point at time t, k max is the total number of monitoring points; Mean μ based on traffic density D and standard deviation σ D Determine the threshold D of traffic density th , the threshold D th The calculation formula is: D th =μ D +h D ·s D ; Among them, h D is an empirical coefficient.

2. A method for real-time identification of adverse traffic conditions according to claim 1, characterized in that: The input layer of the multi-layer perceptron model receives a feature vector X, wherein the feature vector X contains N2 features; The hidden layer contains a preset number of neurons. The formula used by the hidden layer to process data is: d =f(W e,d X e +b d ), where h d represents the output of the dth neuron in the hidden layer, X e is the e-th feature of the input feature vector, f is the Swish activation function, W e,d is the weight matrix representing the connection weight from the e-th node in the input layer to the d-th neuron in the hidden layer, b d is the bias term of the dth neuron in the hidden layer; the expression of the Swish activation function f is: Swish(z d )=z d ·σ(z d ), σ(z d ) is the Sigmoid function; the output end of the hidden layer is provided with a BatchNormalization layer, which converts the output h of each neuron d Normalized, then scaled and translated by learnable parameters γ and β to obtain the final normalized output The calculation formula used by the Batch Normalization layer is: Among them, h d is the output of the dth neuron in the hidden layer, μ B and are the mean and variance respectively, ∈ is a constant, γ and β are learnable parameters; The output layer contains one neuron that directly outputs the traffic density using a linear activation function: in, is the traffic density, W out is the weight matrix of the output layer, which represents the connection weight from each neuron in the last hidden layer to the output layer; h is the output of the hidden layer, b out is the bias term of the output layer.

3. The method for real-time identification of adverse traffic conditions according to claim 1, characterized in that: When training a multi-layer perceptron model, the following steps are included: Initialize parameters, initialize weight W and bias b; Forward propagation, calculate the predicted probability of the output layer Calculate the loss, using mean squared error as the loss function: Among them, y is the true value, is the predicted value, m is the number of samples; Back propagation, calculation of gradients, and update of parameters: Among them, α is the learning rate; Iterative training, repeating forward propagation and back propagation until convergence or reaching the maximum number of iterations; The hyperparameters of the multilayer perceptron model are obtained by the following steps: Divide the data set into K subsets; K-fold cross validation, taking each subset as the validation set and the remaining subsets as the training set; training and validation are performed on each subset; Calculate the accuracy evaluation indicators of the multi-layer perceptron model and continuously adjust the hyperparameters.

4. A device for real-time identification of adverse traffic conditions, characterized in that: include: The traffic state evaluation result prediction module is used to send the acquired historical road traffic data to a pre-trained traffic state recognition model to obtain the traffic state evaluation result of each monitoring point; the traffic state recognition model is a multi-layer perceptron model based on the Swish activation function, including an input layer, multiple hidden layers and an output layer, and the layers are connected in a fully connected manner. The cross-validation technology is used to adjust the hyperparameters during the training process of the traffic state recognition model; A risk identification module, for determining areas with congestion or potential risks based on a traffic status assessment result of each monitoring point using a threshold judgment method, wherein the traffic status assessment result is a traffic density; The method for acquiring historical road traffic data comprises: Obtaining original road traffic data; Performing data fusion on the original road traffic data, integrating data from different sources into a unified data stream, and obtaining first road traffic data; Cleaning the first road traffic data, removing invalid data points, and filling missing values ​​to obtain second road traffic data; performing feature extraction on the second road traffic data to obtain historical road traffic data; When removing invalid data points, the calculation formula used is: Among them, z i is the i-th data point after standardization, x i is the original value of the ith data point, w i is the i-th data point x i The weight, μ w is the weighted mean, σ w is the weighted standard deviation; if |z i ∣>Z, then x i is an invalid data point, Z is the threshold, and N1 is the total number of data points; When filling missing values, the calculation formula is: Among them, t1 is the timestamp of the previous valid data point, t2 is the timestamp of the next valid data point, t is the timestamp of the data point to be interpolated, ω1 is the weight of the previous valid data point, ω2 is the weight of the next valid data point, ω'1 is the adjusted weight of the previous valid data point, ω'2 is the adjusted weight of the next valid data point, y1 is the value of the previous valid data point, and y2 is the value of the next valid data point. is the interpolated value; The extracting features from the second road traffic data comprises: Based on the second road traffic data, the traffic density D of each monitoring point is calculated. k (t), Among them, D k (t) is the traffic density of the kth monitoring point at time t, reflecting the density of vehicles on the road; a k,j is the weight of the j-th type of vehicle at the k-th monitoring point; V k,j (t) is the number of vehicles of the jth type at the kth monitoring point at time t; F k,j (t) is the effective road capacity for the jth type of vehicle at the kth monitoring point at time t; Based on the second road traffic data and the traffic density D of each monitoring point k (t), using a recursive feature elimination algorithm to gradually remove the most irrelevant features and generate a feature vector as the historical road traffic data; the selection formula used by the recursive feature elimination algorithm is: RFE(S, N2) = select N2 features from S, where S is the initial feature set and N2 is the number of selected features; The calculation method of the traffic density threshold in the threshold judgment method includes: Calculate the mean μ of traffic density D and standard deviation σ D : Among them, D k (t) is the traffic density of the kth monitoring point at time t, k max is the total number of monitoring points; Mean μ based on traffic density D and standard deviation σ D Determine the threshold D of traffic density th , the threshold D th The calculation formula is: D th =μ D +h D ·s D ; Among them, h D is an empirical coefficient.

5. A real-time identification system for adverse traffic conditions, characterized in that: including storage media and processors; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the method according to any one of claims 1-3.

Images