Data-driven highway traffic operation state evaluation method and system

Abstract

The application discloses a data-driven highway traffic operation state evaluation method and system, which comprises the following steps: calculating highway evaluation indexes and performing multi-level division to obtain highway evaluation index grades, performing dynamic incremental learning to obtain first index weights, performing fuzzy adjustment to obtain second index weights and constructing a fuzzy consistent matrix, establishing an extension matter element, performing deep learning on the highway evaluation indexes to determine a classic field and a section field, calculating the classic field distance and the section field distance of an extension matter element of a highway to be evaluated, calculating the evaluation correlation degree of the highway evaluation indexes and the highway evaluation index grades, determining a weight vector of the highway to be evaluated according to the fuzzy consistent matrix, and performing extension evaluation according to the weight vector and the evaluation correlation degree to obtain a highway traffic operation state evaluation grade. The method can not only improve the efficiency and accuracy of highway traffic operation state evaluation, but also has good interpretability and can be directly applied to a highway traffic operation state evaluation system.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of state evaluation, and in particular to a highway traffic operation state evaluation method and system based on data driving. BACKGROUND

[0002] With the continuous acceleration of urbanization and the rapid growth of motor vehicle ownership, the highway transportation system is facing unprecedented pressure. Traffic congestion, frequent accidents, and low operational efficiency seriously restrict the high-quality development of regional economy and society, and also affect social safety. Therefore, it is of great significance to scientifically, accurately and timely evaluate the highway traffic operation state, so as to improve the traffic capacity of the highway network, ensure the safe and stable operation of the transportation system, and promote the construction of intelligent transportation system.

[0003] Traditional highway traffic operation state evaluation methods mostly rely on experience-based index selection, static weight distribution or simple statistical analysis models, which have many limitations. On the one hand, the index system is often not comprehensive, and the determination of index weight lacks dynamic adaptability, which cannot respond to real-time changes in traffic flow. On the other hand, the evaluation model does not consider the uncertainty and fuzziness of traffic data, and has weak generalization ability, which is difficult to cope with complex and variable traffic scenarios, resulting in poor accuracy and timeliness of the evaluation results. In recent years, with the rapid development of big data, artificial intelligence and other technologies, new ideas have been provided to solve the above problems. The present application proposes a highway traffic operation state evaluation method and system based on data driving, which divides data and index levels by decision tree, determines index weight by dynamic incremental learning and fuzzy adjustment, introduces extension matter element and deep learning to construct evaluation correlation degree, and realizes final grade determination by fuzzy consistent matrix and extension evaluation. This method not only can comprehensively integrate key indicators such as flow, speed and density, improve the dynamic and rationality of weight determination, but also can enhance the adaptability and anti-interference ability of the evaluation model to complex traffic environment, effectively improve the accuracy and reliability of the evaluation results, and provide stronger technical support for highway traffic management and decision-making, which has important practical value for promoting the intelligent and fine management of highway transportation system. SUMMARY

[0004] The purpose of the present application is to provide a highway traffic operation state evaluation method and system based on data driving.

[0005] In order to achieve the above purpose, the present application is implemented according to the following technical solutions:

[0006] The present application comprises the following steps:

[0007] The highway traffic running data is divided by using a decision tree, and highway evaluation indexes are calculated. The highway evaluation indexes are divided into multiple levels to obtain highway evaluation index levels. The highway evaluation indexes include flow evaluation indexes, speed evaluation indexes, and density evaluation indexes.

[0008] The first index weight is obtained by dynamically incrementally learning the highway evaluation indexes. The second index weight is obtained by fuzzy adjustment of the first index weight. A fuzzy consistent matrix is constructed according to the second index weight.

[0009] According to the highway evaluation index levels and the highway evaluation indexes, an extension matter element is established. The classic field and the section field are determined by deep learning of the highway evaluation indexes. The classic field distance and the section field distance of the extension matter element of the highway to be evaluated are calculated. The evaluation correlation degree of the highway evaluation indexes and the highway evaluation index levels is calculated according to the classic field distance and the section field distance.

[0010] The weight vector of the highway to be evaluated is determined according to the fuzzy consistent matrix. The highway traffic running state evaluation level is obtained by extension evaluation according to the weight vector and the evaluation correlation degree.

[0011] Further, the method for obtaining highway evaluation index levels comprises:

[0012] The highway traffic running data is divided according to flow, speed, and density by using a decision tree. The flow evaluation indexes, the speed evaluation indexes, and the density evaluation indexes are calculated according to the division results. The flow evaluation indexes include flow growth rate, saturation, and saturation balance coefficient. The speed evaluation indexes include speed change rate, free flow speed ratio, and travel time stability. The density evaluation indexes include space occupancy rate and space-time occupancy rate.

[0013] The basic three-level threshold of each highway evaluation index is determined according to the industry standard. The correlation coefficients between indexes are calculated respectively. Two highway evaluation indexes with a correlation coefficient greater than 0.8 are taken as strong correlation indexes, and the remaining indexes are taken as weak correlation indexes. The time series fluctuation coefficient is obtained by time series analysis of the historical data of the strong correlation indexes. The historical data of the strong correlation indexes is divided into four-level indexes according to the basic three-level threshold. The data skewness and the data standard deviation of the same strong correlation index at different levels are calculated. The product of the data skewness mean and the data standard deviation mean of the same strong correlation index at different levels is taken as the data fluctuation coefficient. The basic three-level threshold of the strong correlation indexes is adjusted to obtain the optimized three-level threshold according to the time series fluctuation coefficient and the data fluctuation coefficient. The weak correlation indexes are divided according to the basic three-level threshold, and the strong correlation indexes are divided according to the optimized three-level threshold to obtain the highway evaluation index levels. Further, the method for obtaining the first index weight comprises:

[0014] The initial weight is obtained by using the entropy weight method to process the historical highway evaluation indexes. The first index weight is obtained according to the initial weight and the highway evaluation index levels. Time traffic data sample constructs feature matrix , is a feature matrix set, is a sample number, is a data dimension; the data dimension includes vehicle type, highway evaluation index, spatial position and time stamp;

[0015] Calculate the current data sample and the historical sample reference distribution Wasserstein distance, expression is:

[0016]

[0017]

[0018] Wherein is the time Wasserstein distance, is the historical reference data sample, is a set of all possible joint distributions, is the feature mapping function of the current data , is the joint distribution Lower feature mapping and the expectation of the square of the Euclidean distance, is a spatial feature weight matrix, is a time weight matrix, is the spatial coordinate Encoding as String, is the time stamp;

[0019] Drift determination: when Trigger weight update, wherein is a dynamic threshold, taking the 90% quantile of the historical difference, is the noise tolerance;

[0020] The initial weight is updated by a double-channel learning mechanism to obtain the first weight; the double-channel learning mechanism includes short-term incremental adjustment and long-term incremental adjustment, expression is:

[0021]

[0022]

[0023]

[0024] Wherein is a highway evaluation index exist The weight of the primary indicator is updated in real time. Highway evaluation indicators The initial weights, For short-term weighting, for Short-term adjustment volume at any time. For long-term weighting, Highway evaluation indicators exist Continuous and long-term adjustment of quantity For learning rate, For symbolic functions, Take 1 at a time Time takes -1, Take 0 at time, for Cross-entropy loss function at different times The gradient vector of the weights, As a smoothing factor, Highway evaluation indicators With state Time-varying mutual information, Highway evaluation indicators at the previous moment The first indicator weight.

[0025] Furthermore, the method for constructing the fuzzy consistency matrix includes:

[0026] The kernel function between the weights of the first indicator is calculated using the radial basis function-Gaussian mixture fuzzy kernel algorithm. The weights of the second indicator are obtained by adjusting the weights of the first indicator through kernel function mapping. The expression is as follows:

[0027]

[0028]

[0029] in Highway evaluation indicators exist The weight of the second indicator is updated in real time. To adjust the intensity, The number of highway evaluation indicators, Highway evaluation indicators and The radial basis-Gaussian mixture fuzzy kernel function, As a spatiotemporal adjustment factor, This is the weighted similarity sensitivity coefficient. Let Gaussian components be the number of components. For component weights, It is a multivariate Gaussian distribution. The mean of the Gaussian components. It is the covariance matrix;

[0030] Construct a fuzzy consistency matrix based on the weight of the second indicator. The expression is:

[0031]

[0032] in For fuzzy consistency matrix element, , These are the two largest weights among the weights of the second indicator.

[0033] Furthermore, the method for calculating the classical domain distance and nodal domain distance of the highway extension element to be evaluated includes:

[0034] Based on the highway evaluation index levels and the definition of highway evaluation index extension elements ,in This is a set of highway evaluation index levels. For highway evaluation indicators, for about The evaluation set;

[0035] Construct a boundary generation network and input multi-source traffic data tensors into the boundary generation network to obtain classical domain intervals. and the interval The multi-source traffic data tensor includes highway evaluation indicators, time step, and number of road segments.

[0036] The boundary generation network includes an encoder and a decoder. The encoder receives multi-source traffic data tensors and uses a bidirectional LSTM to extract temporal features to obtain a fixed-length context vector. The decoder processes the context vector based on the domain-generated signal and outputs the classical domain and segment domain corresponding to the domain-generated signal level. The boundary generation network is trained using a two-layer loss function. The two-layer loss function includes a classical domain boundary loss function and a segment domain boundary loss function, the expressions of which are:

[0037]

[0038]

[0039]

[0040] in It is a two-layer loss function. For level The sample set of highway evaluation indicators For the classical domain boundary loss function, For the node boundary loss function, For the loss weight of the section domain, For level The lower bound of the classical domain, For level The upper bound of the classical domain, This is the interval width penalty coefficient. The lower boundary of the domain. This is the upper boundary of the domain;

[0041] The classical domain distance of the highway extension element to be evaluated is obtained by adaptive weighted distance calculation based on the output classical domain interval, and the nodal domain distance of the highway extension element to be evaluated is obtained by dynamic compensation of the output nodal domain interval. The expression is as follows:

[0042]

[0043]

[0044] in , The indicator values ​​are respectively Highway evaluation indicators Classical domain distance and section domain distance, The interval width weight is used. Attention sensitivity coefficient For level Cluster centers This is the drift compensation coefficient. The time window length, for Highway evaluation indicators The value of .

[0045] Furthermore, the method for obtaining the highway traffic operation status evaluation level through extension evaluation includes:

[0046] The evaluation correlation between highway evaluation indicators and highway evaluation indicator levels is calculated based on the classical domain distance and the section domain distance. The expression is as follows:

[0047]

[0048] in The index value is Highway evaluation indicators Corresponding level The degree of correlation of evaluation For variance sensitivity, As an indicator The time series variance, It is a smoothing factor;

[0049] According to the evaluation correlation degree of the highway evaluation index grade, the correlation degree matrix is constructed with the index as the row and the grade as the column, the weight vector of the to-be-evaluated highway is determined according to the fuzzy consistent matrix, the weighted correlation degree of the same highway evaluation index under different grades is calculated by the weight vector and the correlation degree matrix, the grade corresponding to the maximum weighted correlation degree of the highway evaluation index is taken as the evaluation grade, and the evaluation grade of the same type of highway evaluation index is taken as the average value to obtain the highway traffic operation state evaluation grade; the highway traffic operation state evaluation grade includes the highway flow grade, the highway speed grade and the highway density grade.

[0050] In a second aspect, a data-driven highway traffic operation state evaluation system includes:

[0051] A grading module is configured to divide highway traffic operation data using a decision tree and calculate highway evaluation indexes, and perform multi-level division on the highway evaluation indexes to obtain highway evaluation index grades.

[0052] A weight module is configured to perform dynamic incremental learning on the highway evaluation indexes to obtain first index weights, perform fuzzy adjustment on the first index weights to obtain second index weights, and construct a fuzzy consistent matrix according to the second index weights.

[0053] A correlation degree module is configured to establish an extension matter element according to the highway evaluation index grades and the highway evaluation indexes, determine a classic field and a node field by deep learning on the highway evaluation indexes, calculate the classic field distance and the node field distance of the to-be-evaluated highway extension matter element, and calculate the evaluation correlation degree of the highway evaluation indexes and the highway evaluation index grades according to the classic field distance and the node field distance.

[0054] An evaluation module is configured to determine the weight vector of the to-be-evaluated highway according to the fuzzy consistent matrix, and perform extension evaluation according to the weight vector and the evaluation correlation degree to obtain a highway traffic operation state evaluation grade.

[0055] An intelligent supervision module is configured to store, view and manage the fuzzy consistent matrix, the evaluation correlation degree and the highway traffic operation state evaluation grade, and perform traffic control according to the highway traffic operation state evaluation grade.

[0056] The present application has the following advantages:

[0057] Compared with the prior art, the present application has the following technical effects:

[0058] The application can improve the data preprocessing capability and enhance the model adaptability in the precision evaluation of the highway traffic operation state evaluation, thereby improving the efficiency and precision of the highway traffic operation state evaluation, optimizing the highway traffic operation state evaluation technology, greatly saving resources, improving work efficiency, realizing the evaluation of the highway traffic operation state, providing stronger technical support for highway traffic management and decision-making, having important practical value for promoting the intelligent and fine management of the highway traffic system, being suitable for different highway traffic operation state evaluation systems and the highway traffic operation state evaluation needs of different users, and having certain universality. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 The step flowchart of the highway traffic operation state evaluation method based on data driving of the application. DETAILED DESCRIPTION

[0060] The application will be further described below through specific embodiments, and the illustrative embodiments of the application and the description are used to explain the application but do not limit the application.

[0061] The highway traffic operation state evaluation method and system based on data driving of the application include the following steps:

[0062] As shown in the figure, in the embodiment, the following steps are included: Figure 1

[0063] The decision tree is used to divide the highway traffic operation data and calculate the highway evaluation index, the highway evaluation index is divided into multiple levels to obtain the highway evaluation index level; the highway evaluation index includes the flow evaluation index, the speed evaluation index and the density evaluation index;

[0064] The dynamic incremental learning is performed on the highway evaluation index to obtain the first index weight, the fuzzy adjustment is performed on the first index weight to obtain the second index weight, and the fuzzy consistent matrix is constructed according to the second index weight;

[0065] The extension matter element is established according to the highway evaluation index level and the highway evaluation index, the deep learning is performed on the highway evaluation index to determine the classic field and the node field, the classic field distance and the node field distance of the extension matter element of the highway to be evaluated are calculated, and the evaluation correlation degree of the highway evaluation index and the highway evaluation index level is calculated according to the classic field distance and the node field distance;

[0066] The weight vector of the highway to be evaluated is determined according to the fuzzy consistent matrix, the extension evaluation is performed according to the weight vector and the evaluation correlation degree to obtain the highway traffic operation state evaluation level.

[0067] ​In the embodiment, the method for obtaining the highway evaluation index level comprises:

[0068] The highway traffic operation data is divided according to the flow class, the speed class and the density class by using the decision tree, and the flow class evaluation index, the speed class evaluation index and the density class evaluation index are calculated according to the division result; the flow class evaluation index comprises the flow growth rate, the saturation degree and the saturation degree balance coefficient; the speed class evaluation index comprises the speed change rate, the free flow speed ratio and the travel time stability; and the density class evaluation index comprises the space occupancy rate and the space-time occupancy rate.

[0069] The basic three-level threshold values of the highway evaluation indexes are determined according to the industry standard, the correlation coefficients between the indexes are calculated respectively, two highway evaluation indexes with the correlation coefficient greater than 0.8 are taken as the strong correlation indexes, and the rest indexes are taken as the weak correlation indexes; the time sequence analysis is performed on the historical data of the strong correlation indexes to obtain the time sequence fluctuation coefficient; the historical data of the strong correlation indexes are divided into four indexes according to the basic three-level threshold values; the data skewness and the data standard deviation of the same strong correlation index at different levels are calculated; the product of the data skewness mean value and the data standard deviation mean value of the same strong correlation index at different levels is taken as the data fluctuation coefficient; the basic three-level threshold values of the strong correlation indexes are adjusted according to the time sequence fluctuation coefficient and the data fluctuation coefficient to obtain the optimized three-level threshold values; the weak correlation indexes are divided according to the basic three-level threshold values, and the strong correlation indexes are divided according to the optimized three-level threshold values to obtain the highway evaluation index level.

[0070] In the actual evaluation, when the flow class evaluation index is calculated, the flow growth rate comprises the flow year-on-year and month-on-month growth rates of the road section / channel / road network, the saturation degree comprises the road section saturation degree = cross section traffic capacity / traffic capacity, the channel saturation degree = (∑road section saturation degree*road section length) / channel length, and the road network saturation degree = (∑road section saturation degree*road section length) / road network length; and the saturation degree balance coefficient comprises the channel saturation degree balance coefficient = standard deviation of channel each road section saturation degree / average saturation degree of channel each road section, and the road network saturation degree balance coefficient = standard deviation of road network each road section saturation degree / average saturation degree of road network each road section.

[0071] When the speed class evaluation index is calculated, the speed change rate (by vehicle type) comprises the average speed year-on-year and month-on-month change rates of the road section / channel / road network, the free flow speed ratio (by vehicle type) comprises the road section free flow speed ratio = vehicle road section average speed / road section speed limit, the channel free flow speed ratio = (∑road section free flow speed ratio*road section mileage) / channel total mileage, and the road network free flow speed ratio = (∑road section free flow speed ratio*road section mileage) / road network total mileage; and the travel time stability comprises the time dimension travel time stability (standard deviation of multi-day average travel time of the road section or the channel) and the space dimension travel time stability (standard deviation of average travel time of each road section on a channel).

[0072] When calculating density-based evaluation indicators, spatial occupancy includes road segment spatial occupancy = ∑ vehicle length / road segment mileage, and road network spatial occupancy = (∑ road segment spatial occupancy * road segment length) / total road network mileage. Spatiotemporal occupancy includes road segment spatiotemporal occupancy = (∑ vehicle length * vehicle on-road time during the study period) / (road segment mileage * study time), corridor spatiotemporal occupancy = (∑ road segment spatiotemporal occupancy * road segment length) / total corridor mileage, and road network spatiotemporal occupancy = (∑ road segment spatiotemporal occupancy * road segment length) / total road network mileage.

[0073] The time series volatility coefficient was obtained by using an autoregressive moving average model for time series analysis. The basic third-level thresholds of strongly correlated indicators are scaled to obtain optimized third-level thresholds, and the highway evaluation indicators are divided into 4 levels based on the optimized third-level thresholds.

[0074] In this embodiment, the method for obtaining the first indicator weight includes:

[0075] The entropy weight method is used to process historical highway evaluation indicators to obtain initial weights, based on... Constructing a feature matrix from traffic data samples at different times , For the set of characteristic matrices, For the sample size, The data dimensions include vehicle type, highway evaluation indicators, spatial location, and timestamp.

[0076] Calculate the current data sample Compared with historical sample reference distribution The Wasserstein distance is expressed as:

[0077]

[0078]

[0079] in For time Wasserstein distance at time 10:00 For historical reference data samples, The set of all possible joint distributions. For current data Feature mapping function, For joint distribution Lower feature mapping and The expectation of the square of the Euclidean distance. The spatial feature weight matrix, This is the time weight matrix. To convert spatial coordinates Encoded as string, is a timestamp;

[0080] drift determination is performed: when a weight update is triggered, where is a dynamic threshold, taking the 90th percentile of the historical difference, is a noise tolerance;

[0081] the initial weight is updated by a double-channel learning mechanism to obtain a first weight; the double-channel learning mechanism includes short-term incremental adjustment and long-term incremental adjustment, and the expression is:

[0082]

[0083]

[0084]

[0085] wherein is a highway evaluation index the first index weight is updated at the initial weight of the highway evaluation index is updated at the short-term weight, is the short-term adjustment amount at the long-term weight, is the long-term adjustment amount of the highway evaluation index at is a learning rate, is a sign function, is 1 when is -1 when is 0 when is the cross-entropy loss function at is a gradient vector with respect to the weight, is a smoothing factor, is the time-varying mutual information of the highway evaluation index and the state is the first index weight of the highway evaluation index at the last time; In actual evaluation, the cross-entropy loss function

[0086] is expressed as:

[0087]

[0088] wherein ​for the number of rating grades, for the real state label at the moment, is a model prediction function for outputting the probability of the grade, is an index weight vector.

[0089] evaluation index and state time-varying mutual information The expression is:

[0090]

[0091] wherein is a state label set, i.e. 1-4 index grades, is a highway evaluation index set, is the joint probability distribution at the moment, is the index marginal distribution, is the state marginal distribution.

[0092] Take the short-term weight 0.5, the long-term weight 0.3, the learning rate 0.01, the smoothing factor 0.9 to calculate the first index weight.

[0093] In the embodiment, the method for constructing the fuzzy consistent matrix comprises:

[0094] According to the radial basis-Gaussian mixed fuzzy kernel algorithm, the kernel function between the first index weights is calculated, and the second index weight is obtained by adjusting the first index weight through the kernel function mapping, and the expression is:

[0095]

[0096]

[0097] wherein is a highway evaluation index the second index weight updated at the moment, is the adjustment intensity, is the number of highway evaluation indexes, is a highway evaluation index and the radial basis-Gaussian mixed fuzzy kernel function of is the space-time adjustment factor, is the weight similarity sensitivity coefficient, is the number of Gaussian components, a component weight, a multivariate Gaussian distribution, a mean of a Gaussian component, a covariance matrix;

[0098] constructing a fuzzy consistent matrix according to the second index weight , the expression is:

[0099]

[0100] wherein is a fuzzy consistent matrix element, , are the two largest second index weights in the second index weight;

[0101] In actual evaluation, the number of highway evaluation indexes is 28, the adjustment intensity is 0.1, the space-time adjustment factor is 0.5, the weight similarity sensitivity coefficient is 0.1, the number of Gaussian components is 5, the kernel function between the first index weights is calculated, and the second index weight is obtained by adjusting the first index weight through the kernel function mapping.

[0102] In the embodiment, the method for calculating the classical domain distance and the node domain distance of the extension matter-element of the road to be evaluated comprises:

[0103] defining the extension matter-element according to the highway evaluation index level and the highway evaluation index , wherein is a set of highway evaluation index levels, is a set of highway evaluation indexes, is an evaluation set about ;

[0104] constructing a boundary generation network, inputting a multi-source traffic data tensor into the boundary generation network to obtain a classical domain interval and a node domain interval ; the multi-source traffic data tensor comprises a highway evaluation index, a time step and a number of road sections;

[0105] The boundary generation network comprises an encoder and a decoder, the encoder receives the multi-source traffic data tensor, and adopts a bidirectional LSTM to extract time sequence features to obtain a fixed-length context vector, the decoder processes the context vector according to a domain generation signal to output a classical domain and a node domain corresponding to the level of the domain generation signal; the boundary generation network is trained through a double-layer loss function; the double-layer loss function comprises a classical domain boundary loss function and a node domain boundary loss function, and the expression is:

[0106]

[0107]

[0108]

[0109] wherein is a double-layer loss function, is a highway evaluation index sample set of level , is a classic domain boundary loss function, is a section domain boundary loss function, is a section domain loss weight, is a classic domain lower bound of level , is a classic domain upper bound of level , is an interval width penalty coefficient, is a section domain lower bound, is a section domain upper bound;

[0110] The classic domain distance of the highway extension matter element to be evaluated is obtained by adaptive weighted distance calculation according to the output classic domain interval, and the section domain distance of the highway extension matter element to be evaluated is obtained by dynamic compensation on the output section domain interval, and the expression is:

[0111]

[0112]

[0113] wherein , are respectively the classic domain distance and the section domain distance of the highway evaluation index with the index value , is an interval width weight, is an attention sensitive coefficient, is a clustering center of level , is a drift compensation coefficient, is a time window length, is the value of the highway evaluation index at the moment ;

[0114] In actual evaluation, the highway evaluation index level set respectively represents unobstructed, mild congestion, serious congestion and paralysis, and the evaluation set takes “ ” as an example to represent the value interval of the highway evaluation index of level 3;

[0115] In the boundary generation network, 28-dimensional highway evaluation indicators for 120 road segments of a certain highway over a period of 1440 minutes are input, and the segment loss weights are taken. The interval width penalty coefficient is 0.8. The value is 0.3, and the canonical domain and section domain of each highway evaluation index are output.

[0116] Take the interval width weight Attention sensitivity coefficient: 1.2 The drift compensation coefficient is 0.8. The time window length is 0.15. For a period of 24 hours, calculate the classical domain distance and nodal domain distance of the highway extension element to be evaluated.

[0117] In this embodiment, the method for obtaining the highway traffic operation status evaluation level through extension evaluation includes:

[0118] The evaluation correlation between highway evaluation indicators and highway evaluation indicator levels is calculated based on the classical domain distance and the section domain distance. The expression is as follows:

[0119]

[0120] in The index value is Highway evaluation indicators Corresponding level The degree of correlation of evaluation For variance sensitivity, As an indicator The time series variance, It is a smoothing factor;

[0121] Based on the evaluation correlation of highway evaluation index levels, a correlation matrix is ​​constructed with the index as rows and the level as columns. The weight vector of the highway to be evaluated is determined according to the fuzzy consistency matrix. The weighted correlation of the same highway evaluation index at different levels is calculated by the weight vector and the correlation matrix. The level corresponding to the maximum weighted correlation of the highway evaluation index is taken as the evaluation level. The average evaluation level of the highway traffic operation status is obtained by taking the average evaluation level of the same type of highway evaluation index. The highway traffic operation status evaluation level includes highway flow level, highway speed level and highway density level.

[0122] In actual evaluation, variance sensitivity is taken. Smoothing factor is 0.15. The correlation between highway evaluation indicators and highway evaluation indicator levels is calculated with a value of 0.1, and a 28×4 correlation matrix is ​​constructed.

[0123] Taking the weighted correlation degree calculation of link saturation as an example, the four-level weight vector of link saturation is determined by the fuzzy consistent matrix as [0.35, 0.35, 0.4, 0.6], and the evaluation correlation degrees of link saturation and each level are determined according to the correlation matrix as [0.85, 0.72, 0.65, 0.45], and the weighted correlation degrees of link saturation under different levels are calculated as 0.2975, 0.252, 0.26, and 0.27, and the level (level 1) corresponding to the maximum weighted correlation degree 0.2975 of link saturation is taken as the evaluation level;

[0124] The evaluation level mean values of the traffic class, speed class and density class highway evaluation indexes are calculated respectively, the level mean value of 11 traffic class indexes is 1.81, the level mean value of 11 speed class indexes is 1.18, and the level mean value of 6 speed class indexes is 2, that is, the traffic flow level of highway traffic operation is level 2 (traffic flow is stable, and the influence is slight), the speed level of highway traffic operation is level 1 (the speed is fast, and the traffic is smooth), and the density level of highway traffic operation is level 2 (the density is moderate, and the traffic is slightly affected).

[0125] In a second aspect, the data-driven highway traffic operation state evaluation system comprises:

[0126] The grading module is used for dividing highway traffic operation data by using a decision tree and calculating highway evaluation indexes, and the highway evaluation indexes are divided into multiple levels to obtain highway evaluation index levels;

[0127] The weight module is used for dynamically incrementally learning the highway evaluation indexes to obtain first index weights, fuzzy adjusting the first index weights to obtain second index weights, and constructing a fuzzy consistent matrix according to the second index weights;

[0128] The correlation degree module is used for establishing an extension matter element according to the highway evaluation index levels and the highway evaluation indexes, determining a classic field and a section field by deeply learning the highway evaluation indexes, calculating the classic field distance and the section field distance of the extension matter element to be evaluated, and calculating the evaluation correlation degrees of the highway evaluation indexes and the highway evaluation index levels according to the classic field distance and the section field distance;

[0129] The evaluation module is used for determining the weight vector of the highway to be evaluated according to the fuzzy consistent matrix, and performing extension evaluation according to the weight vector and the evaluation correlation degrees to obtain a highway traffic operation state evaluation level;

[0130] The intelligent supervision module is used for storing, viewing and managing the fuzzy consistent matrix, the evaluation correlation degrees and the highway traffic operation state evaluation level, and performing traffic control according to the highway traffic operation state evaluation level.

Claims

1. A data-driven based method for evaluating highway traffic operation condition, characterized in that, The method comprises the following steps: S1, adopting a decision tree to divide highway traffic operation data and calculate highway evaluation indexes, and performing multi-level division on the highway evaluation indexes to obtain highway evaluation index grades; the highway evaluation indexes comprise flow type evaluation indexes, speed type evaluation indexes and density type evaluation indexes; S2, performing dynamic incremental learning on the highway evaluation indexes to obtain first index weights, performing fuzzy adjustment on the first index weights to obtain second index weights, and constructing a fuzzy consistent matrix according to the second index weights; S3, establishing an extension matter element according to the highway evaluation index grades and the highway evaluation indexes, determining a classic field and a node field by performing deep learning on the highway evaluation indexes, calculating a classic field distance and a node field distance of an extension matter element of a highway to be evaluated, and calculating an evaluation correlation degree of the highway evaluation indexes and the highway evaluation index grades according to the classic field distance and the node field distance; S4, determining a weight vector of the highway to be evaluated according to the fuzzy consistent matrix, and performing extension evaluation according to the weight vector and the evaluation correlation degree to obtain a highway traffic operation state evaluation grade; The method for obtaining the first index weights comprises: The initial weight is obtained by using an entropy weight method to process historical highway evaluation indexes, and the initial weight is adjusted according to The feature matrix is constructed by using the real-time traffic data sample , The feature matrix set is The sample number is The data dimension includes vehicle type, highway evaluation index, spatial position and time stamp; Computing the current data sample the Wasserstein distance to the historical sample reference distribution expressed by the formula: wherein is time Wasserstein distance at time is a historical reference data sample, is a set of all possible joint distributions, is current data a feature mapping function, is a joint distribution a feature mapping with expectation of squared Euclidean distance, is a spatial feature weight matrix, is a temporal weight matrix, is a function that encodes spatial coordinates into a string, is a timestamp; Drift determination is performed: when a weight update is triggered, where is a dynamic threshold, taking the 90th percentile of the historical difference, is a noise tolerance; updating the initial weights by a double-channel learning mechanism to obtain first weights; the double-channel learning mechanism comprises short-term incremental adjustment and long-term incremental adjustment, and the expression is as follows: in Highway evaluation indicators exist The weight of the primary indicator is updated in real time. Highway evaluation indicators The initial weights, For short-term weighting, for Short-term adjustment volume at any time. For long-term weighting, Highway evaluation indicators exist Continuous and long-term adjustment of quantity For learning rate, For symbolic functions, Take 1 at a time Time takes -1, Take 0 at time, for Cross-entropy loss function at different times The gradient vector of the weights, As a smoothing factor, Highway evaluation indicators With state Time-varying mutual information, Highway evaluation indicators at the previous moment The first indicator weight; The method for constructing the fuzzy consistent matrix comprises: calculating a kernel function between the first index weights according to a radial basis-Gaussian mixed fuzzy kernel algorithm, adjusting the first index weights by the kernel function mapping to obtain second index weights, and the expression is as follows: wherein is a road evaluation index at is a second index weight updated in real time, is an adjustment intensity, is a number of road evaluation indexes, is a road evaluation index and is a radial basis-Gaussian mixed fuzzy kernel function, is a space-time adjustment factor, is a weight similarity sensitivity coefficient, is a number of Gaussian components, is a component weight, is a multivariate Gaussian distribution, is a mean of a Gaussian component, is a covariance matrix; According to the second index weight, a fuzzy consistent matrix is constructed The expression is: wherein is a fuzzy consistent matrix element, , are the two largest second index weights among the second index weights.

2. The method according to claim 1, wherein, The method for obtaining the highway evaluation index grades comprises: dividing highway traffic operation data according to flow type, speed type and density type by a decision tree, and calculating flow type evaluation indexes, speed type evaluation indexes and density type evaluation indexes according to the division results; the flow type evaluation indexes comprise a flow growth rate, a saturation degree and a saturation degree balance coefficient; the speed type evaluation indexes comprise a speed change rate, a free flow speed ratio and a travel time stability; and the density type evaluation indexes comprise a space occupancy rate and a space-time occupancy rate; determining basic three-level thresholds of each highway evaluation index according to an industry standard, respectively calculating correlation coefficients between the indexes, taking two highway evaluation indexes with a correlation coefficient greater than 0.8 as strong correlation indexes, and taking the remaining indexes as weak correlation indexes, performing time series analysis on historical data of the strong correlation indexes to obtain a time series fluctuation coefficient, dividing the historical data of the strong correlation indexes into four indexes according to the basic three-level thresholds, calculating data skewness and data standard deviation of the same strong correlation index at different levels, taking a product of a data skewness mean and a data standard deviation mean of the same strong correlation index at different levels as a data fluctuation coefficient, adjusting the basic three-level thresholds of the strong correlation indexes to obtain optimized three-level thresholds according to the time series fluctuation coefficient and the data fluctuation coefficient, dividing the weak correlation indexes according to the basic three-level thresholds, and dividing the strong correlation indexes according to the optimized three-level thresholds to obtain the highway evaluation index grades.

3. The method according to claim 1, wherein, The method for calculating the classic field distance and the node field distance of the extension matter element of the highway to be evaluated comprises: According to the highway evaluation index grade and the highway evaluation index definition extension matter element Wherein The highway evaluation index grade set is The highway evaluation index set is The highway evaluation index grade set is The highway evaluation index set is The evaluation set of The boundary generation network is constructed, and a multi-source traffic data tensor is input into the boundary generation network to obtain a classical domain interval and a section domain interval ; the multi-source traffic data tensor comprises a highway evaluation index, a time step and a number of road sections. The boundary generation network comprises an encoder and a decoder, the encoder receives a multi-source traffic data tensor and extracts time sequence features by using a bidirectional LSTM to obtain a fixed-length context vector, and the decoder processes the context vector according to a domain to output a classical domain and a section domain corresponding to a level of a domain generation signal; the boundary generation network is trained by using a double-layer loss function; the double-layer loss function comprises a classical domain boundary loss function and a section domain boundary loss function, and the expressions are as follows: in It is a two-layer loss function. For level The sample set of highway evaluation indicators For the classical domain boundary loss function, For the node boundary loss function, For the loss weight of the section domain, For level The lower bound of the classical domain, For level The upper bound of the classical domain, This is the interval width penalty coefficient. The lower boundary of the domain. This is the upper boundary of the domain; A classical domain distance of the to-be-evaluated road extension matter element is obtained by performing adaptive weighted distance calculation according to the output classical domain interval, and a section domain distance of the to-be-evaluated road extension matter element is obtained by performing dynamic compensation on the output section domain interval, and the expressions are as follows: wherein , are the highway evaluation indexes with index values are the classic domain distance and section distance, is the interval width weight, is the attention sensitivity coefficient, is the cluster center of the grade, is the drift compensation coefficient, is the time window length, is the value of the highway evaluation index at the moment.​ 4. The method according to claim 1, wherein, The method for performing the extension evaluation to obtain the evaluation level of the road traffic running state comprises the following steps: An evaluation correlation degree of the road evaluation index and the road evaluation index level is calculated according to the classical domain distance and the section domain distance, and the expression is as follows: wherein is an index value of a road evaluation index corresponding to a grade of an evaluation correlation degree, is a variance sensitivity, is a time series variance of an index is a smoothing factor; A correlation degree matrix is constructed according to the evaluation correlation degree of the road evaluation index level, taking the index as the row and the level as the column, a weight vector of the to-be-evaluated road is determined according to a fuzzy consistent matrix, a weighted correlation degree under different levels of the same road evaluation index is calculated by using the weight vector and the correlation degree matrix, the level corresponding to the maximum weighted correlation degree of the road evaluation index is taken as the evaluation level, and the evaluation level of the same type of road evaluation index is taken as the mean value to obtain the evaluation level of the road traffic running state; the evaluation level of the road traffic running state comprises a road flow level, a road speed level and a road density level.

5. A data-driven based highway traffic operational state evaluation system, characterized in that, A device for performing the method of any one of claims 1-4, comprising: a grading module for dividing road traffic running data and calculating road evaluation indexes by using a decision tree, and performing multi-level division on the road evaluation indexes to obtain road evaluation index levels; a weight module for performing dynamic incremental learning on the road evaluation indexes to obtain a first index weight, performing fuzzy adjustment on the first index weight to obtain a second index weight, and constructing a fuzzy consistent matrix according to the second index weight; a correlation degree module for establishing an extension matter element according to the road evaluation index levels and the road evaluation indexes, performing deep learning on the road evaluation indexes to determine a classical domain and a section domain, calculating a classical domain distance and a section domain distance of the to-be-evaluated road extension matter element, and calculating an evaluation correlation degree of the road evaluation index and the road evaluation index level according to the classical domain distance and the section domain distance; an evaluation module for determining a weight vector of the to-be-evaluated road according to the fuzzy consistent matrix, and performing extension evaluation according to the weight vector and the evaluation correlation degree to obtain the evaluation level of the road traffic running state; an intelligent supervision module for storing, viewing and managing the fuzzy consistent matrix, the evaluation correlation degree and the evaluation level of the road traffic running state, and performing traffic control according to the evaluation level of the road traffic running state. an intelligent supervision module for storing, viewing and managing the fuzzy consistent matrix, the evaluation correlation degree and the evaluation level of the road traffic running state, and performing traffic control according to the evaluation level of the road traffic running state.

Images