Traffic state oriented time-varying sector dynamic classification method

Abstract

The application discloses a kind of traffic state-oriented time-varying sector dynamic classification methods, steps are as follows: sector historical operation data are collected, dynamic operation characteristic variable data are calculated and data preprocessing is carried out, and clustering input feature vector is obtained;Sector traffic operation state clustering analysis;Time-varying sector dynamic classification model based on the neural network of multilayer perceptron is constructed;Develop time-varying sector dynamic classification for sector traffic operation state.The method of the application optimizes the traffic distribution of air route according to the traffic state of sector in different time periods, fully utilizes air traffic control resources, and helps to create a strong performance civil aviation system with dynamic resource sharing, capacity elasticity adjustment and autonomous and efficient operation.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of air traffic management and airspace planning, and particularly relates to a time-varying sector dynamic classification method for traffic state. BACKGROUND

[0002] With the rapid development of China's air transportation, flight demand is increasing, airspace environment is more complex, and airspace usage is diversified, complex and flexible. The existing sector classification is basically based on the classification of airspace inherent structure and function, and is divided into regional sectors, approach sectors, etc. This classification method is a physical classification according to the flight process of the aircraft, and is only suitable for the basic sector classification in the control process. It lacks comprehensive consideration of the structural characteristics of the sector and the traffic flow characteristics, and cannot meet the needs of flow management and airspace management.

[0003] From the domestic and foreign patents and research literature related to sector classification, there are few studies on sector classification based on sector operation characteristics. At the same time, overall classification of sectors based on operation characteristics is beneficial to macroscopically understand the sector and implement corresponding management measures. However, sector classification research should also consider the dynamic operation state of the sector, i.e. dynamic category, which is given to the sector by the time-varying traffic flow. The past sector classification research analyzes and mines historical data to give a fixed category to each time period of the sector, which seems to be not flexible and objective. When the traffic flow changes due to weather influence and flow control strategy in a certain time period, it may not match the given fixed category. Therefore, such sector classification research results have little significance in helping the decision of air traffic managers at the pre-tactical level. In addition, the sector classification research based on sector complexity and capacity by domestic and foreign scholars mostly uses hard clustering (such as system clustering method, K-means, etc.) in clustering algorithm. The results of classification and the interpretability of clustering have certain degree of irrationality. SUMMARY

[0004] In view of the deficiencies of the prior art, the purpose of the present application is to provide a time-varying sector dynamic classification method for traffic state, so as to solve the problems that the sector is given a fixed category in the prior art, the dynamic characteristics of sector traffic operation cannot be reflected, and the control strategy of the controller at the pre-tactical level cannot be formulated, and the hard clustering algorithm in the traditional classification is not reasonable in category division. The method of the present application optimizes the en route route flow distribution according to the traffic state of the sector at different time periods, fully utilizes the air traffic control resources, and helps to build a strong performance civil aviation airspace system with dynamic resource sharing, capacity elasticity adjustment and autonomous and efficient operation.

[0005] To achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0006] A traffic state-oriented time-varying sector dynamic classification method, steps are as follows:

[0007] 1) Collect sector historical operation data, calculate dynamic operation characteristic variable data and perform data preprocessing, and obtain clustering input feature vectors;

[0008] 2) Sector traffic operation state clustering analysis; the GA-KFCM clustering algorithm is adopted to perform clustering division of the clustering input feature vectors obtained in step 1) on the sector traffic operation state, and the data set of each type of traffic operation state of the sector is obtained through the division result, which is used as the corresponding traffic operation state measurement standard;

[0009] 3) Construct a time-varying sector dynamic classification model based on a multilayer perceptron neural network;

[0010] 4) Develop a time-varying sector dynamic classification for sector traffic operation states.

[0011] Further, the step 1) specifically includes:

[0012] 11) Analyze the real ASDB radar track data and the sector historical operation data obtained by the AirTop computer simulation with 15 minutes as a time sample, obtain time-varying state representation index data related to the sector traffic operation state as a basic data set, including the static structure factors and dynamic traffic factors of the sector;

[0013] 12) Normalize the sample data by using Z-Score, the mean value of the normalized data is 0 and the standard deviation is 1; perform correlation analysis on the index, and then perform reduction processing on the data by PCA analysis to obtain the clustering input feature vectors.

[0014] Further, the step 2) specifically includes:

[0015] 21) Sector traffic operation state category number determination: perform multiple comparison experiments on different values of the cluster number c of the GA-KFCM clustering algorithm to determine the best natural cluster number; analyze the overall silhouette coefficient value and V KXB Index value of the traffic flow sample data set clustering under different cluster number values, and select the best cluster number c value to obtain the best clustering effect;

[0016] 22) Sector traffic operation state clustering analysis: the parameters of the GA-KFCM clustering algorithm are initialized and set, the clustering input feature vector is input into the clustering algorithm, and the traffic operation states of each sector are divided; the traffic flow characterization index distribution reflected by the kernel density estimation graph of part of the indexes under different traffic states is combined to analyze the traffic operation states of each sector; the historical state data set of each traffic operation state in the sector is obtained through the division result, which is used as the corresponding traffic operation state measurement standard.

[0017] Further, the step 3) specifically comprises:

[0018] 31) Separating the historical state data set into a training data set and an evaluation data set; verifying the split parameter setting to be 0.2 of the original data set size to realize the splitting of the data set, i.e. 80% of the original data as a training set and 20% of the original data as a test set;

[0019] 32) Selecting an activation function and an error function, wherein the activation function is divided into an implicit layer activation function and an output layer activation function;

[0020] 33) Setting the number of neurons in the input layer to be the number of feature variables of the input data set, and the number of neurons in the output layer to be the expected target dimension of the sector dynamic classification; designing multiple groups of comparative optimization experiments under different parameter combinations of the number of hidden layers and the number of neurons, monitoring the training of the model through the TensorBoard machine learning visualization tool provided by TensorFlow, and obtaining the influence of different numbers of hidden layers on the model training; reading the comparative experiment logs, outputting the comparison chart of the accuracy and the iterative change of the loss function in each group of experiments, and determining the number of hidden layers and the number of neurons contained in each hidden layer in the sector dynamic classification model; the experience formula for determining the number of hidden layer neurons is as follows:

[0021]

[0022] In the formula, N i is the number of input signals; N0 is the expected target output dimension; N s is the number of samples contained in the training set; and a is an arbitrary variable that can be taken by itself, and the range can be taken as 2-10.

[0023] Further, the step 32) specifically comprises:

[0024] 321) Selecting a Relu function as the implicit layer activation function, and the expression is:

[0025]

[0026] 322) Selecting a softmax function as the output layer activation function, and the expression is:

[0027]

[0028] where z is a vector, z i and z j is one of the elements;

[0029] 323) using cross-entropy error as the loss measure of the sector dynamic classification model; the formula of cross-entropy is as follows: the loss function parameter is set to the classification cross-entropy logarithmic loss function, and the expression is:

[0030]

[0031] where log represents the natural logarithm with e as the base; y k is the model output, t k is each label, and t k only the label of the correct solution is 1, and the rest are 0; therefore, the greater the output corresponding to the correct solution label, the closer the value of the cross-entropy to 0; when the output is 1, the cross-entropy error is 0; on the contrary, if the output corresponding to the correct solution label is smaller, the value of the cross-entropy is greater.

[0032] Further, the step 4) specifically comprises: training the time-varying sector dynamic classification model according to the neural network topology structure determined in the step 3), and the selected excitation function and loss function, to predict the sector dynamic category.

[0033] Further, the step 4) specifically comprises:

[0034] 41) setting the multilayer perceptron algorithm parameters and format encoding; performing one-hot encoding processing on the dynamic classification data, when there are y categories, the output target value of the sample is encoded into a y-dimensional vector, and the identification in the position of the vector indicating that the sample belongs to a certain category is 1, and the identifications in the remaining positions are all 0; the multilayer perceptron algorithm parameters are set as follows:

[0035] Optimizer: using the adam algorithm in the efficient gradient descent method as the optimization algorithm for searching the optimal weight;

[0036] Metric: measurement index, specified as accuracy in the multivariate classification problem;

[0037] Epochs: setting the number of cycles for training the overall training data set to i times;

[0038] Batchsize: specifying the batch size for updating the weight;

[0039] 42) The segmented sample verification evaluates the sector classification model performance, uses the time-varying state characterization index data related to the sector traffic operation state as the training sample to generate the model, and tests on the reserved sample; according to the test set and the verification set that have been segmented, the optimal network topology structure and the corresponding multi-layer perceptron algorithm parameter setting are determined to start training the model and perform performance evaluation; and the output log contains the data observation trend of each training cycle to determine whether the model is normally trained and whether the training needs to be terminated;

[0040] 43) The k-fold cross-validation is used to evaluate the classification algorithm performance; the data set to be learned is randomly divided into m non-intersecting subsets, each subset has the same number of instances; 1 subset is randomly selected as the evaluation data set, and the other m-1 subsets are used to train the model, and each subsequent subset is used as the test set of the model trained by the other m-1 subsets to evaluate the effect of the model;

[0041] 44) The time-varying state characterization index data related to the sector traffic operation state is input, and the sector dynamic classification model is used to perform time-varying sector dynamic classification for the sector traffic flow index data of 96 time periods in a day in 15-minute time segments, to obtain the sector category every 15 minutes; the obtained sector 15-minute time period dynamic category is compared with the actual category of the sector 15-minute time period in the sample data, to illustrate the accuracy and availability of the sector dynamic classification model.

[0042] The beneficial effects of the present application are:

[0043] 1) Based on the fuzziness of the time-varying traffic state of the sector, the GA-KFCM soft clustering algorithm is introduced to overcome the unreasonable classification of the traditional hard clustering algorithm and ensure the division result of the sector traffic state. The massive sector traffic flow operation index data is used for clustering analysis to obtain the sector historical traffic state sample data set, which can provide reliable historical sample data input for the sector dynamic classification model.

[0044] 2) The existing technology analyzes the historical data and only gives a fixed category of sector classification method for each time period of the sector. The present application studies the sector dynamic classification facing the time-varying traffic state by constructing a sector traffic state classification and discrimination algorithm. The traffic flow characteristics and complex situation corresponding to different sector traffic states obtained by clustering are analyzed. Further, a high-efficiency and accurate sector traffic operation state discrimination prediction model is established by combining the deep learning neural network algorithm. Finally, the sector operation data is input to predict the output sector pre-tactical level traffic state, realize the time-varying sector dynamic classification facing the traffic state, and overcome the defect that the classification effect of the existing technology is small on the pre-tactical level.

[0045] 3, The sector dynamic category division realized by the application can provide refined sector dynamic operation information with 15 minutes as a time period, and can assist controllers in formulating control strategies under different traffic operation states, so as to strengthen the judgment of future traffic scenes. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 The method flowchart of the application;

[0047] Figure 2 The linear correlation analysis visualization result schematic diagram of the sector time-varying state characteristic indexes obtained by the correlation analysis adopted by the application;

[0048] Figure 3a The kernel density estimation diagram of the traffic flow index under different dynamic categories;

[0049] Figure 3b The kernel density estimation diagram of the potential conflict index under different dynamic categories;

[0050] Figure 3c The kernel density estimation diagram of the mean flight time index under different dynamic categories;

[0051] Figure 3d The kernel density estimation diagram of the descending traffic flow index under different dynamic categories;

[0052] Figure 4a The accuracy change trend diagram of the hidden layer number optimization experiment;

[0053] Figure 4b The loss function change trend diagram of the hidden layer number optimization experiment;

[0054] Figure 5 The comparison diagram of the sector historical operation period category and the sector dynamic classification model prediction category. DETAILED DESCRIPTION

[0055] In order to facilitate the understanding of those skilled in the art, the application will be further described below in combination with embodiments and drawings. The content mentioned in the embodiments is not a limitation of the application.

[0056] Referring to Figure 1 The traffic state-oriented time-varying sector dynamic classification method of the application includes the following steps:

[0057] Step 1: Collect sector historical operation data, calculate dynamic operation characteristic variable data and perform data preprocessing;

[0058] Historical sector operation data is selected and analyzed in 15-minute intervals to obtain basic data describing the dynamic traffic status of the sector. This includes dynamic factors characterizing the sector traffic status such as the number of aircraft entering and leaving the rack, the total number of aircraft occupying the sector, the maximum number of aircraft occupying the sector within 15 minutes, the average flight time of aircraft, the average flight distance, the proportion of aircraft with different flight attitudes, the entropy value of airway pairs, and the number of potential conflicts within the sector.

[0059] Correlation analysis refers to the analysis of two or more correlated variables; the Pearson correlation coefficient rho(a,b), introduced based on covariance, is defined as:

[0060]

[0061] Where X in matrix X a Columns and matrix Y b List, n is the length of each column; when |rho|≤0.3, it indicates no linear correlation; 0.3<|rho|≤0.5, it indicates a low linear correlation between variables; 0.5<|rho|≤0.8, it indicates a significant linear correlation between variables; |rho|>0.8, it indicates a high linear correlation between variables; based on the sign of the value, it is divided into positive and negative correlations, and 0 represents mutual independence; the visualization results of the linear correlation analysis between sector time-varying state characterization indicators are as follows: Figure 2 As shown.

[0062] Correlation analysis of the selected indicator variables revealed that different indicators exhibit varying degrees of linear correlation, which can lead to overlap in the information reflected by the sample data.

[0063] Furthermore, principal component analysis was used to extract principal components with a cumulative contribution rate of over 85%, which were then used as the feature vectors for cluster analysis of sector traffic status.

[0064] Step 2: Cluster analysis of sector traffic operation status;

[0065] Multiple comparative experiments were conducted on different values ​​of the cluster number c in the clustering algorithm to determine the optimal number of natural clusters. Too many or too few categories of sector traffic operation status are not suitable. Too few categories cannot reflect the variability of sector traffic status and cannot accurately distinguish between different categories. Too many categories also have the problem of unclear distinction, and at the same time increase the complexity for controllers to formulate control strategies and make decisions on opening and closing sectors when using sector dynamic categories.

[0066] The cluster number is optimized in a proper interval, the overall silhouette coefficient value of the traffic flow sample data set clustering under different cluster number values is obtained and analyzed, and the V KXB index value is obtained; when the cluster number is c, the overall silhouette coefficient takes the maximum value MAX, V KXB the index value takes the minimum value MIN, and according to the cluster number selection standard, the cluster number c is determined.

[0067] The GA-KFCM algorithm parameters are initialized and set, and the sector traffic state clustering analysis is started.

[0068] a. Genetic algorithm parameter setting, population size n, initial crossover probability p c0 , initial mutation probability p m0 , genetic evolution generation T, genetic algorithm convergence threshold δ;

[0069] b. Set the basic parameters of the KFCM algorithm, the scale parameter σ of the Gaussian kernel function, the termination tolerance ε of the objective function, the fuzzy clustering algebra k, the cluster number c, and the fuzzy index m;

[0070] The feature vector is input into the clustering algorithm, and c classes of sector traffic states are obtained by clustering division. The kernel density estimation diagram of part of the indexes under the c classes of traffic states is shown in Figures 3a-3d The clustering results are analyzed. The c classes of different traffic states in which the sectors operate in different time periods have obvious differences, and the distribution of the traffic flow, the number of potential conflicts, the average flight time and the proportion of descending traffic flow in the sectors corresponding to the different traffic states reflect the operating characteristics of the sectors in different traffic states, as well as the traffic concentration of different route pairs in the sectors and the workload of controllers. In summary, the low-load traffic state at night is identified in the c classes of traffic states, in which the traffic flow in the sector is low, and there is no potential conflict between aircraft; the traffic state is concentrated on aircraft descent, in which the proportion of traffic flow descending in height is higher than the average and the median of the whole; the flight flow characteristics are obvious, in which the traffic flow is high, there are more traffic flows changing in height and more dispersed route pair flow distribution in the sector, and the sector capacity is close to saturation; the sector overload operating state, in which the route pair flow entropy, the number of 15-minute entries into the frame, and the number of potential conflicts between aircraft are higher than those in other types of traffic states, and a large number of aircraft have changed in vertical direction; the workload of controllers basically exceeds the 70% threshold.

[0071] To further verify the rationality of traffic state division, according to historical operation data, the sector flight flow is counted at 15-minute time intervals throughout the day (24 hours), and the temporal and spatial distribution characteristics of the sector flow are analyzed. The proportion of c-type traffic states in the historical typical day operation is compared with the proportion of c-type traffic state clustering samples obtained by clustering, providing reliable baseline data for subsequent experiments.

[0072] Step 3: Constructing a time-varying sector dynamic classification model based on a neural network of multilayer perceptron;

[0073] Obtain a data sample at 15-minute intervals, a total of N data samples; divide 80% of the original data as the training set and 20% of the original data as the test set.

[0074] Select the relu function that facilitates backpropagation as the activation function in the hidden layer; the sector operation has c different states, so the output layer contains c neurons. For multi-classification problems, the softmax function is used as the activation function of the output layer, and the loss function parameter is set to the categorical cross-entropy logarithmic loss function.

[0075] As shown in Figure 4a , Figure 4b , design multiple comparison experiments with different parameter combinations of hidden layer number and neuron number, and output the accuracy and loss function iteration changes in each group of experiments; according to the comparison experiment results, determine the hidden layer number j of the sector traffic state discrimination prediction model constructed, and the number of neurons k contained in each hidden layer.

[0076] Step 4: According to the neural network topology structure determined in step 3, as well as the selection of related activation functions and loss functions; set other parameters of the multilayer perceptron algorithm, and then train the sector dynamic classification model to perform traffic state-oriented time-varying sector dynamic classification.

[0077] Input the time-varying state representation index data related to the sector traffic operation state, use the sector dynamic classification model to predict and discriminate the sector category, and compare it with the actual sector period category to illustrate the accuracy and availability of the sector dynamic classification model. Select the traffic flow data sample within 24 hours to obtain the sector dynamic category for display. As shown in Figure 5As shown, the sector categories of 96 time periods in a day are discriminatively predicted with 15 minutes as a time segment, and in the classification results, the sector categories of 3 time periods are not consistent with the true categories, but the traffic states and complexities in the categories that are not consistent with the actual situation are not greatly deviated from the true categories. It is illustrated that the sector dynamic classification model constructed by the application has strong prediction and discrimination accuracy, and the extracted sector dynamic classification indexes have good universality. The sector dynamic information of 15 minutes as a time period can be provided, and the regulation strategy under different traffic operation states can be formulated to assist the controller to strengthen the research and judgment of the future traffic scene. Meanwhile, as management assistance, when the sector category fluctuates little in the future time period, and the corresponding traffic flow under the state is low and the regulation difficulty is small, the sectors under the same state can be considered to be combined for operation. In addition, efficient and safe sector operation configuration schemes and human resource configuration schemes can also be proposed according to different sector dynamic categories, and new post modes are innovatively developed.

[0078] The application has many specific application ways, and the above description is only the preferred embodiment of the application. It should be pointed out that, for ordinary skilled persons in the technical field, some improvements can be made without departing from the principle of the application, and these improvements should also be regarded as the protection scope of the application.

Claims

1. A traffic state oriented time-varying sector dynamic classification method, characterized by, The steps are as follows: 1) Collect sector historical operation data, calculate dynamic operation characteristic variable data and perform data preprocessing to obtain clustering input feature vectors; 2) Sector traffic operation state clustering analysis; the clustering input feature vectors obtained in step 1) are clustered and divided into sector traffic operation state, and the data set of each type of sector traffic operation state is obtained through the division result, which is used as the corresponding traffic operation state measurement standard; 3) Construct a time-varying sector dynamic classification model based on a multilayer perceptron neural network; 4) Develop a time-varying sector dynamic classification for sector traffic operation state; The step 3) specifically comprises: 31) Separate the historical state data set into a training data set and an evaluation data set; verify the segmentation parameter setting to be 0.2 of the original data set size to realize the segmentation of the data set, that is, 80% of the original data as the training set and 20% of the original data as the test set; 32) Select an activation function and an error function, wherein the activation function is divided into an implicit layer activation function and an output layer activation function; 33) Set the number of neurons in the input layer to be the number of feature variables in the input data set, and the number of neurons in the output layer to be the expected target dimension of the sector dynamic classification; design a plurality of groups of comparison optimization experiments under different parameter combinations of the number of hidden layers and the number of neurons, monitor the training of the model through the TensorBoard machine learning visualization tool provided by TensorFlow, and obtain the influence of different numbers of hidden layers on the model training; read the comparison experiment log, output the comparison graph of the iteration changes of the accuracy and the loss function in each group of experiments, and determine the number of hidden layers and the number of neurons contained in each hidden layer in the sector dynamic classification model; the experience formula for determining the number of hidden layer neurons is as follows: In the formula, N i is the number of input signals; N0 is the expected target output dimension; N s is the number of samples contained in the training set; and α is an arbitrary variable that can be self-selected. The step 32) specifically comprises: 321) Select the Relu function as the implicit layer activation function, and the expression is: 322) Select the softmax function as the output layer activation function, and the expression is: where z is a vector, z i and z j is one of the elements; 323) Use cross-entropy error as the loss measurement of the sector dynamic classification model; the formula of cross-entropy is as follows: the loss function parameter is set to the classification cross-entropy logarithmic loss function, and the expression is: wherein log represents a natural logarithm with e as a base number; y k is a model output, t k is each label, t k only the label of a correct solution is 1, and the rest are 0; the greater the output corresponding to the correct solution label, the closer the value of the cross-entropy to 0; when the output is 1, the cross-entropy error is 0; conversely, if the output corresponding to the correct solution label is smaller, the value of the cross-entropy is greater; The step 4) specifically comprises: 41) Set the multilayer perceptron algorithm parameters and format encoding; perform one-hot encoding processing on the dynamic classification data, when there are y classes, the output target value of the sample is encoded into a y-dimensional vector, and the position indicating that the sample belongs to a certain class in the vector is 1, and the positions of the rest are 0; the multilayer perceptron algorithm parameters are set as follows: Optimizer: use the adam algorithm in the efficient gradient descent method as the optimization algorithm for searching the optimal weight; Metric: measurement index, specified as accuracy in the multivariate classification problem; Epochs: set the number of cycles for training the entire training data set to i times; Batchsize: specify the batch size for updating the weight; 42) The segmented sample verification evaluates the sector classification model performance, uses the time-varying state characterization index data related to the sector traffic operation state as training sample to generate the model, and tests on the reserved sample; according to the test set and the verification set which have been segmented, the optimal network topology structure and the corresponding multi-layer perceptron algorithm parameter setting are determined to start training the model and perform performance evaluation; and the output log contains the data observation trend of each training cycle to determine whether the model is normally trained and whether the training needs to be terminated; 43) The k-fold cross-validation is used to evaluate the performance of the classification algorithm; the data set to be learned is randomly divided into m non-intersecting subsets, each subset has the same number of instances; 1 subset is randomly selected as the evaluation data set, and the other m-1 subsets are used to train the model, and each subsequent subset is used as the test set of the model trained by the other m-1 subsets to evaluate the effect of the model; 44) The time-varying sector dynamic classification model is used to input the time-varying state characterization index data related to the sector traffic operation state, and the sector dynamic classification is carried out for the sector traffic flow index data of 96 time periods in a day with 15 minutes as a time segment, and the sector category every 15 minutes is obtained; the obtained sector 15-minute time period dynamic category is compared with the actual category of the sector 15-minute time period in the sample data, which illustrates the accuracy and availability of the sector dynamic classification model.

2. The traffic state oriented time-varying sector dynamic classification method according to claim 1, characterized in that, The step 1) specifically comprises: 11) The real radar track data and the simulated sector historical operation data are analyzed with 15 minutes as a time sample to obtain the time-varying state characterization index data related to the sector traffic operation state as the basic data set, including the static structure factors and dynamic traffic factors of the sector; 12) The Z-Score normalized sample data is used, and the mean value of the normalized data is 0 and the standard deviation is 1; the correlation analysis is carried out on the index, and then the data is reduced by PCA analysis to obtain the clustering input feature vector.

3. The traffic state oriented time-varying sector dynamic classification method according to claim 1, characterized in that, The step 2) specifically comprises: 21) Sector traffic operation state category number determination: multiple sets of comparative experiments are conducted on the clustering number c of the GA-KFCM clustering algorithm to determine the optimal natural clustering number; the overall silhouette coefficient value of the traffic flow sample data set is analyzed under different clustering cluster number value conditions, and the V KXB index value is selected to select the optimal clustering number c value and obtain the best clustering effect; 22) Sector traffic operation state clustering analysis: the GA-KFCM clustering algorithm parameters are initialized and set, the clustering input feature vector is input into the clustering algorithm, and the divided sector traffic operation states are obtained; the traffic flow characterization index distribution reflected by the kernel density estimation graph of part of the index under different traffic states is analyzed to analyze the sector traffic operation states; the historical state data set of each type of traffic operation state in the sector is obtained through the division result, which is used as the corresponding traffic operation state measurement standard.

4. The traffic state oriented time-varying sector dynamic classification method of claim 1, wherein, The step 4) specifically comprises: training the time-varying sector dynamic classification model according to the neural network topology structure determined in step 3), and selecting the excitation function and the loss function to predict the sector dynamic category.

Images