Traffic state determination method based on multi-source heterogeneous data

Abstract

The application provides a traffic state determination method based on multi-source heterogeneous data, which can be applied to the technical field of intelligent transportation. The determination method comprises the following steps: performing anomaly detection on a plurality of initial traffic data in an initial traffic data sequence according to the distance between each initial traffic data arranged according to time information and a preset data and the abnormal value of each initial traffic data, to obtain abnormal traffic data; performing denoising on the initial traffic data sequence according to the abnormal traffic data, to obtain a target traffic data sequence; performing feature extraction on the target traffic data sequence based on the time information of the target traffic data sequence and the traffic correlation relationship between a plurality of traffic intersections, to obtain the time feature and the space feature of a target traffic intersection; and determining the traffic state of the target traffic intersection based on the time feature and the space feature.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent transportation, more particularly to a traffic state determination method based on multi-source heterogeneous data. BACKGROUND

[0002] Automatic driving, intelligent transportation management, and intelligent vehicle as a representative of intelligent transportation system have been widely concerned and applied. In traffic flow prediction, traffic facility monitoring, and working environment of traffic management objects, such intelligent transportation system has problems of multi-source data heterogeneity, noise interference, and information redundancy, and needs to improve the data quality of such intelligent transportation system through data cleaning and fusion technology, and ensure the real-time and accuracy of traffic flow.

[0003] Designing an efficient and intelligent data cleaning and fusion control method is an important link for optimization and management of intelligent transportation system. Current traffic data information sources are more, and the identification of abnormal data is not accurate, resulting in inaccurate traffic prediction results. SUMMARY

[0004] In view of the above problems, the present application provides a traffic state determination method based on multi-source heterogeneous data to improve the accuracy of the state of the traffic intersection.

[0005] According to one aspect of the present application, a traffic state determination method based on multi-source heterogeneous data is provided, comprising: performing abnormality detection on a plurality of initial traffic data in an initial traffic data sequence arranged according to time information and a plurality of abnormal values of the initial traffic data, to obtain abnormal traffic data, wherein the initial traffic data represents the traffic state of a target traffic intersection at a target location, the preset data is determined according to the Lagrange multiplier obtained by inputting a historical traffic data set of a plurality of traffic intersections at the target location into a quadratic programming solver, and the abnormal value is obtained by inputting the initial traffic data sequence into an isolation forest algorithm; denoising the initial traffic data sequence according to the abnormal traffic data to obtain a target traffic data sequence; extracting features of the target traffic data sequence based on time information of the target traffic data sequence and traffic correlation between a plurality of traffic intersections to obtain time features and spatial features of the target traffic intersection; and determining the traffic state of the target traffic intersection based on the time features and the spatial features.

[0006] According to the embodiments of the present application, the above abnormality detection on the plurality of initial traffic data according to the distance between each of the plurality of initial traffic data and the preset data and the abnormal value of each of the plurality of initial traffic data, to obtain abnormal traffic data, comprises: determining a first abnormal data set including at least one first abnormal data from the plurality of initial traffic data according to the distance between each of the plurality of initial traffic data and the preset data; determining a second abnormal data set including at least one second abnormal data from the plurality of initial traffic data according to the abnormal value of each of the plurality of initial traffic data; and determining the abnormal traffic data according to the repeated data between the first abnormal data set and the second abnormal data set.

[0007] According to the embodiments of the present application, the above determining a first abnormal data set including at least one first abnormal data from the plurality of initial traffic data according to the distance between each of the plurality of initial traffic data and the preset data, comprises: determining the initial traffic data as the first abnormal data when the distance between the initial traffic data and the preset data is greater than a preset distance, wherein the preset distance is determined according to the Lagrange multiplier obtained by inputting the historical traffic data set into the quadratic programming solver.

[0008] According to the embodiments of the present application, the above determining a second abnormal data set including at least one second abnormal data from the plurality of initial traffic data according to the abnormal value of each of the plurality of initial traffic data, comprises: determining the initial traffic data as the second abnormal data when the abnormal value is less than a preset abnormal value.

[0009] According to the embodiments of the present application, the above determining the abnormal traffic data according to the repeated data between the first abnormal data set and the second abnormal data set, comprises: obtaining an abnormal weighted value of the repeated data by weighted sum of the repeated data according to a first abnormal weight and a second abnormal weight; and determining the repeated data as the abnormal traffic data when the abnormal weighted value is greater than a preset weighted value.

[0010] According to an embodiment of the present application, the feature extraction of the target traffic data sequence based on the time information of the target traffic data sequence and the traffic correlation between the plurality of traffic intersections includes: processing the plurality of target traffic data in the target traffic data sequence based on the time information of the target traffic data sequence to obtain the time feature of the target traffic intersection; and inputting the target traffic data sequence and the traffic correlation into an attention graph convolution network to obtain the spatial feature of the target traffic intersection and a traffic network graph, the traffic network graph including a plurality of nodes and edges between the plurality of nodes, the plurality of nodes representing the plurality of traffic intersections at the target location, and the edges representing the traffic correlation.

[0011] According to an embodiment of the present application, the processing of the plurality of target traffic data in the target traffic data sequence based on the time information of the target traffic data sequence to obtain the time feature of the target traffic intersection includes: processing the plurality of target traffic data in the target traffic data sequence in the time order of the target traffic data sequence by using a forward long short-term memory network to obtain a forward hidden feature of the target traffic intersection; processing the target traffic data sequence in the reverse time order of the target traffic data sequence by using a reverse long short-term memory network to obtain a reverse hidden feature of the target traffic intersection; and performing feature fusion on the forward hidden feature and the reverse hidden feature to obtain the time feature of the target traffic intersection.

[0012] According to an embodiment of the present application, the inputting of the target traffic data sequence and the traffic correlation into the attention graph convolution network to obtain the spatial feature of the target traffic intersection includes: determining a neighborhood data set of each of the plurality of nodes based on the edges between the plurality of nodes in the traffic network graph, the neighborhood data set including at least one neighborhood data subset, and the neighborhood data subset being a target traffic data sequence of a neighborhood node connected to the node by l edges, l being an integer greater than or equal to 1; performing weighted aggregation on the plurality of neighborhood data subsets based on a learnable projection matrix and an adaptive weighting factor corresponding to each of the plurality of nodes to obtain an aggregated feature; and processing the aggregated feature by using a nonlinear activation function to obtain the spatial feature.

[0013] According to an embodiment of the present application, the determination of the traffic state of the target traffic intersection based on the time feature and the spatial feature includes: inputting the time feature, the spatial feature, and the traffic network graph into a hierarchical progressive attention module in a cascaded progressive attention network to obtain a spatio-temporal feature; and inputting the spatio-temporal feature into a feedforward network in the cascaded progressive attention network to obtain a traffic prediction result of the target traffic intersection, the traffic prediction result including the traffic state.

[0014] According to the embodiment of the present application, the above-mentioned inputting the above-mentioned time feature, the above-mentioned space feature and the above-mentioned traffic network graph into the hierarchical progressive attention module of the cascaded progressive attention network to obtain the space-time feature includes: extracting a space dependency relationship between a plurality of nodes from the above-mentioned traffic network graph, the space dependency relationship representing a topological relationship between the plurality of nodes; fusing the above-mentioned space feature into the above-mentioned time feature based on a learnable coefficient to obtain an initial fusion feature, the learnable coefficient being used to control the fusion strength between the space feature and the time feature; inputting the initial fusion feature and the space dependency relationship into the hierarchical progressive attention module to obtain the space-time feature.

[0015] According to the embodiment of the present application, the initial traffic data is subjected to anomaly detection according to the distance between the initial traffic data and the preset data and the abnormal value of the initial traffic data. Since the preset data is determined according to the Lagrange multiplier obtained by inputting the historical traffic data set into the quadratic programming solver, and the abnormal value is obtained by inputting the traffic data sequence into the isolation forest algorithm, the abnormal traffic data obtained through the two anomaly detection methods is more accurate. The initial traffic data set is denoised according to the abnormal traffic data to obtain a target traffic data set. The result of traffic prediction using the target traffic data set is also more accurate. Based on the space feature, the time feature and the traffic network graph extracted from the target traffic data set, the state of the target traffic intersection is determined. Since the traffic information is more abundant, the state of the traffic intersection is more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0016] The above content and other purposes, features and advantages of the present application will be more apparent through the following description of the embodiments of the present application with reference to the accompanying drawings.

[0017] Figure 1 The application scenario diagram of the traffic state determination method based on multi-source heterogeneous data according to the embodiment of the present application is shown.

[0018] Figure 2 The flowchart of the traffic state determination method based on multi-source heterogeneous data according to the embodiment of the present application is shown.

[0019] Figure 3 The flowchart of the traffic state determination method based on multi-source heterogeneous data according to another embodiment of the present application is shown.

[0020] Figure 4 The abnormal data diagram detected by one sensor according to an embodiment is shown.

[0021] Figure 5 The abnormal data diagram detected by another sensor according to an embodiment is shown.

[0022] Figure 6 A comparison chart of predicted traffic flow and actual traffic flow according to an embodiment.

[0023] Figure 7 A scatter plot of predicted and actual traffic data according to an embodiment.

[0024] Figure 8 A structure block diagram of a traffic state determination apparatus based on multi-source heterogeneous data according to an embodiment of the present application.

[0025] Figure 9 A block diagram of an electronic device adapted to implement a traffic state determination method based on multi-source heterogeneous data according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. It should be understood, however, that the description which follows is merely illustrative and is not intended to limit the scope of the present application. In the following detailed description of embodiments of the present application, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that one or more embodiments of the present application can be practiced without these specific details. In other instances, well-known structures and functions have not been described in detail in order to avoid obscuring aspects of the present application.

[0027] The terms used herein are merely used to describe specific embodiments and are not intended to limit the present application. The terms "include" and "have" and the like used herein indicate the presence of the described features, steps, operations, and / or components but do not preclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein, including technical and scientific terms, have the same meanings as those generally understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having meanings consistent with the context of the present description, and should not be interpreted in an idealized or overly formal way.

[0029] In the case of using expressions similar to "at least one of A, B, and C, etc.", it should generally be interpreted to include any of the possibilities of one, two, or all of the items enumerated, unless otherwise specifically limited. In other words, "at least one of A, B, and C" should be interpreted to mean that A alone, B alone, C alone, a combination of A and B, a combination of A and C, a combination of B and C, or a combination of A and B and C are all possible.

[0030] Traffic data processing methods in intelligent transportation systems can be divided into model-based methods (such as predictive planning) and data-based methods (such as real-time traffic scheduling). The former requires advance planning and relatively complex model building, and is therefore not suitable for real-time traffic scenarios with frequent dynamic changes; the latter, because it can flexibly respond to real-time changes in traffic flow and is easy to combine with data-driven methods such as machine learning, is more suitable for real-time data processing and prediction.

[0031] Typical data-driven approaches are based on deep learning and graph convolutional networks. However, these methods cannot guarantee real-time performance and accuracy in complex environments, especially in the optimization of multi-source heterogeneous data fusion, which requires further improvement. In recent years, the integration of multimodal data fusion and deep learning technologies has provided a novel approach to solving this problem, achieving efficient data cleaning and fusion by leveraging deep neural networks and spatiotemporal modeling methods.

[0032] The models used in deep learning-based traffic flow optimization and data fusion schemes are conservative and lack clear mathematical definitions, resulting in room for improvement in optimization performance. Furthermore, effectively integrating data from different sources, external uncertainties, and historical data is a critical issue that urgently needs to be addressed.

[0033] This application provides a method for determining traffic state based on multi-source heterogeneous data, comprising: performing anomaly detection on multiple initial traffic data in an initial traffic data sequence according to time information and preset data, and anomalies of each initial traffic data, to obtain abnormal traffic data. The initial traffic data represents the traffic state of a target intersection at a target location. The preset data is determined by inputting historical traffic datasets of multiple intersections at the target location into a quadratic programming solver to obtain Lagrange multipliers. The anomalies are obtained by inputting the traffic data sequence into an isolated forest algorithm. The initial traffic data sequence is denoised based on the abnormal traffic data to obtain a target traffic data sequence. The target traffic data sequence is feature extracted based on the time information of the target traffic data sequence and the traffic correlation between multiple intersections to obtain the time and spatial features of the target intersection. The traffic state of the target intersection is determined based on the time and spatial features.

[0034] According to the embodiment of the present application, the initial traffic data is subjected to anomaly detection according to the distance between the initial traffic data and the preset data and the outliers of the initial traffic data, and since the preset data is determined according to the Lagrange multiplier obtained by inputting the historical traffic data set into the quadratic programming solver, and the outliers are obtained by inputting the traffic data sequence into the isolation forest algorithm, the abnormal traffic data obtained through the two anomaly detection methods is more accurate. The initial traffic data set is denoised according to the abnormal traffic data to obtain a target traffic data set, and the result of traffic prediction using the target traffic data set is also more accurate. Based on the spatial features, time features and traffic network graph extracted from the target traffic data set, the state of the target traffic intersection is determined, and since the traffic information is more abundant, the state of the traffic intersection is more accurate.

[0035] Figure 1 An application scenario diagram of the traffic state determination method and device based on multi-source heterogeneous data according to the embodiment of the present application is shown.

[0036] As shown in Figure 1 The application scenario 100 according to the embodiment can include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104 and a server 105. The network 104 is a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103 and the server 105. The network 104 can include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0037] A user can use the first terminal device 101, the second terminal device 102, the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, the third terminal device 103, such as web browser applications, search applications, instant messaging tools, email clients, etc. (only as examples).

[0038] The first terminal device 101, the second terminal device 102, the third terminal device 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smartphones, tablet computers, laptop computers and desktop computers, etc.

[0039] The server 105 can be a server providing various services, such as a background management server supporting the website browsed by the user using the first terminal device 101, the second terminal device 102, the third terminal device 103 (only as an example). The background management server can analyze and process the received user requests and other data, and feed back the processing results (such as web pages, information or data obtained or generated according to the user requests) to the terminal device.

[0040] It should be noted that the traffic state determination method based on multi-source heterogeneous data provided in the embodiments of the present application can be generally executed by the server 105. Accordingly, the traffic state determination apparatus based on multi-source heterogeneous data provided in the embodiments of the present application can be generally arranged in the server 105. The traffic state determination method based on multi-source heterogeneous data provided in the embodiments of the present application can also be executed by a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the traffic state determination apparatus based on multi-source heterogeneous data provided in the embodiments of the present application can also be arranged in a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0041] It should be understood that Figure 1 The number of terminal devices, networks and servers in the above scenario is only illustrative. According to the implementation needs, there can be any number of terminal devices, networks and servers.

[0042] The traffic state determination method based on multi-source heterogeneous data of the disclosed embodiments will be described in detail below based on the scenario described above. Figure 1 Figures 2-7 The traffic state determination method based on multi-source heterogeneous data of the disclosed embodiments will be described in detail below based on the scenario described above.

[0043] Figure 2 The traffic state determination method based on multi-source heterogeneous data according to the embodiments of the present application is shown in the flowchart.

[0044] As shown in Figure 2 The traffic state determination method based on multi-source heterogeneous data of the embodiments includes operations S210-S240, which can be executed by a server.

[0045] In operation S210, according to the distance between each of the plurality of initial traffic data arranged according to time information in the initial traffic data sequence and the preset data, and the abnormal value of each of the plurality of initial traffic data, the plurality of initial traffic data is subjected to abnormality detection to obtain abnormal traffic data.

[0046] In the embodiments of the present application, the initial traffic data represents the traffic state of the target traffic intersection at the target location, the preset data is determined according to the Lagrange multiplier obtained by inputting the historical traffic data set of the plurality of traffic intersections at the target location into a quadratic programming solver, and the abnormal value is obtained by inputting the initial traffic data sequence into an isolation forest algorithm. The time information can be the acquisition time of the initial traffic data.

[0047] ​The dual problem of the historical traffic data set of the multiple traffic intersections at the target position is solved by a quadratic programming (QP) solver to obtain a Lagrange multiplier, and further, according to the Lagrange multiplier, a kernel function is used to implicitly represent the ball center and radius of the historical traffic data set, and the ball center is taken as preset data and the radius is taken as a preset distance.

[0048] The initial traffic data can be collected by a road loop detector; for example, a road loop detector is placed at a target traffic intersection at a target position to collect initial traffic data, and the initial traffic data is taken as traffic flow data; the traffic flow data can include traffic volume, traffic speed, and waiting time, etc.; the road loop detector can record 3 sets of traffic flow data at the target traffic intersection, each set of traffic flow data can include 288 data points, each data point is obtained by the road loop detector sampling every 5 minutes, and the road loop detector collects 864 data points per day.

[0049] In the embodiments of the present application, the historical traffic data set is collected by placing road loop detectors at multiple traffic intersections at the target position; for example, 23 detection points of road loop detectors are set at the multiple traffic intersections at the target position, the detection points can be set according to corresponding rules or actual needs, and for example, the detection points are set according to corresponding rules, the interval between adjacent detection points can be fixed as a preset distance. For example, the interval between adjacent detection points can be fixed as 1 kilometer. 3 sets of traffic flow data recorded by each detection point are taken as traffic data, and the traffic flow data collected from the 23 detection points in February, May, August and December are selected as the historical traffic data set.

[0050] In operation S220, the initial traffic data sequence is denoised according to the abnormal traffic data to obtain a target traffic data sequence.

[0051] In the embodiments of the present application, the abnormal traffic data can be taken as noise data, and the abnormal traffic data in the initial traffic data sequence can be removed. The target traffic data sequence can also be subjected to data cleaning operations such as missing value filling and / or redundant data elimination.

[0052] In operation S230, the target traffic data sequence is subjected to feature extraction based on time information of the target traffic data sequence and a traffic correlation relationship between multiple traffic intersections to obtain time features and spatial features of a target traffic intersection.

[0053] In the embodiments of the present application, the plurality of traffic intersections can be a plurality of traffic intersections at the target position, including the target traffic intersection and a plurality of neighborhood traffic intersections of the target traffic intersection. The traffic correlation between the plurality of traffic intersections refers to the connection relationship between the target traffic intersection and the plurality of neighborhood traffic intersections of the target traffic intersection. Since the initial traffic data sequence includes a plurality of initial traffic data arranged according to time information, and the target traffic data sequence is obtained by denoising the initial traffic data sequence according to the abnormal traffic data, the target traffic data sequence has time information. Since the initial traffic data represents the traffic state of the target traffic intersection at the target position, the time feature of the target traffic intersection can be determined according to the time information of the target traffic data sequence, and the spatial feature of the target traffic intersection can be determined according to the traffic correlation.

[0054] In operation S240, the traffic state of the target traffic intersection is determined based on the time feature and the spatial feature.

[0055] In the embodiments of the present application, the traffic state can include traffic flow, congestion condition, waiting time, vehicle flow and vehicle flow rate, etc. According to the traffic state of the target traffic intersection, the signal light duration can be dynamically adjusted to give priority to the congestion section for evacuation; traffic control can be implemented in a specific area to guide the reasonable distribution of vehicles and ensure the smoothness of the road and the efficiency of traffic operation.

[0056] According to the embodiments of the present application, the initial traffic data is detected for abnormality according to the distance between the initial traffic data and the preset data and the abnormal value of the initial traffic data. Since the preset data is determined according to the Lagrange multiplier obtained by inputting the historical traffic data set into the quadratic programming solver, and the abnormal value is obtained by inputting the traffic data sequence into the isolation forest algorithm, the abnormal traffic data obtained by the two abnormality detection methods is more accurate. The initial traffic data set is denoised according to the abnormal traffic data to obtain a target traffic data set; the result of traffic prediction using the target traffic data set is also more accurate. Since the spatial feature, the time feature and the traffic network graph extracted from the target traffic data set have rich traffic information, the state of the target traffic intersection obtained is more accurate.

[0057] According to an embodiment of the present application, the initial traffic data is detected according to the distance between each of the initial traffic data arranged according to time information in the initial traffic data sequence and the preset data, and the abnormal value of each of the initial traffic data, to obtain abnormal traffic data, including: determining a first abnormal data set including at least one first abnormal data from the initial traffic data according to the distance between each of the initial traffic data and the preset data; determining a second abnormal data set including at least one second abnormal data from the initial traffic data according to the abnormal value of each of the initial traffic data; and determining the abnormal traffic data according to the repeated data between the first abnormal data set and the second abnormal data set.

[0058] In an embodiment of the present application, the distance between each of the initial traffic data and the preset data can adopt Euclidean distance, Chebyshev distance or Mahalanobis distance. The distance between each of the initial traffic data and the preset data is calculated to obtain a plurality of distances, an abnormal distance in the plurality of distances is determined, and the first abnormal data is determined according to the initial traffic data corresponding to the abnormal distance. In the process of determining the abnormal distance in the plurality of distances, the first abnormal data can be determined according to the relationship between each of the plurality of distances and the preset abnormal distance by a preset abnormal distance, for example, the relationship between each of the plurality of distances and the preset abnormal distance is determined by applying percentage deviation method, standard deviation method, interval judgment method and trend change method, and then the first abnormal data is determined.

[0059] In the process of determining the second abnormal data, the initial traffic data satisfying a preset condition in the comparison result of the initial traffic data and the preset abnormal data threshold can be selected as the second abnormal data from the initial traffic data by a preset abnormal data threshold.

[0060] According to an embodiment of the present application, through the double screening mechanism of distance analysis and abnormal value detection, the data distribution characteristics and statistical characteristics are effectively combined, and the accuracy and reliability of the abnormal traffic data identification are significantly improved. The first abnormal data is determined based on distance analysis, which can reflect systematic deviation; the local anomaly can be captured based on statistical analysis of abnormal value, and the abnormal traffic data can be determined based on two different ways, which reduces the misjudgment rate of a single method, avoids processing a large number of edge cases, and improves the accuracy and processing efficiency of the abnormal detection.

[0061] According to an embodiment of the present application, the first abnormal data set including at least one first abnormal data is determined from the initial traffic data according to the distance between each of the initial traffic data and the preset data, including: determining the initial traffic data as the first abnormal data in the case that the distance between the initial traffic data and the preset data is greater than a preset distance, and the preset distance is determined according to the Lagrange multiplier obtained by inputting the historical traffic data set into a quadratic programming solver.

[0062] Exemplarily, the initial traffic data is taken as a sample, and the initial traffic data sequence is taken as a sample set, and a minimum boundary can be constructed based on a support vector data description algorithm to detect an abnormal point. It is considered that a data sample boundary is divided by a high-dimensional hypersphere, and is represented by a target function as follows:

[0063] (1).

[0064] Wherein F represents a target function, is a center point of a sample, represents a radius of a boundary, is a regularization parameter. is a sample to the center of the square Euclidean distance, and the formula (1) requires that the sample falls within a hypersphere with as the center and a radius of , is a slack variable, represents that the sample exceeds the radius , The greater the sample point is considered to be an abnormal point, N is the number of samples, and T represents transposition.

[0065] Taking R as a preset distance, the distance between the initial traffic data and the preset data is calculated by .

[0066] For the out-of-bound strength of the support vector data description algorithm, the original distance deviation of the sample is: . represents that the sample is outside the sphere, and the greater it is, the more suspicious it is; represents that the sample is within the sphere, and the greater the absolute value is, the more normal it is. The normalized distance abnormal score is: wherein and represent the median and the absolute median difference of . Usually, 1 is taken, is the final out-of-bound strength score of the i-th sample.

[0067] In the embodiments of the present application, the threshold comparison method is applied to determine the first abnormal data from the plurality of initial traffic data, which can quickly and accurately identify the obvious outliers. And the way of applying the quadratic programming solver to determine the preset distance can more accurately reflect the real distribution characteristics of the historical data, thereby improving the accuracy and adaptability of the anomaly detection.

[0068] According to an embodiment of this application, determining a second abnormal dataset including at least one second abnormal data from multiple initial traffic data based on the abnormal values ​​of each of the multiple initial traffic data includes: determining the initial traffic data as the second abnormal data when the abnormal value is less than a preset abnormal value.

[0069] For example, an outlier detection is achieved by designing the number of partitions traversed from the root node to the leaf node based on the isolation forest algorithm. Each sample is randomly divided into a group of isolation trees. Within each isolation tree, the path length of a sampling point is defined as the number of partitions traversed from the root node to the leaf node, and the expected value of this path length represents:

[0070] (2).

[0071] Where: h(x) represents the number of partitions that sample x passes through in the isolation tree from the root node to the leaf node, i.e., the path length. It is the average path length, often approximated by , It is the number of sample points. It is a sample The expected path length. Equation (1) It is the first in the dataset For each test sample, equation (2) This represents any sample to be evaluated within the isolated forest. On the same dataset, With a certain They can be the same sample object.

[0072] For the short path strength and path anomaly score of the Isolation Forest algorithm: , It is a sample The expected path length, It is a sample Path anomaly score. The smaller the sample size, the shorter the sample path, the easier it is to isolate, and the more suspicious it is.

[0073] In the embodiments of this application, a threshold comparison method is applied to determine the second abnormal data from multiple initial traffic data sets, which can quickly and accurately identify obvious outliers. Since the outliers are obtained by inputting the initial traffic data sequence into the isolation forest algorithm, this method automatically learns the anomaly threshold based on data distribution characteristics, effectively identifying traffic data anomaly patterns that are significantly different from the majority of patterns, thus improving the accuracy and adaptability of anomaly detection.

[0074] According to an embodiment of the present application, the abnormal traffic data is determined according to the repeated data between the first abnormal data set and the second abnormal data set, including: performing weighted summation on the repeated data according to the first abnormal weight and the second abnormal weight to obtain an abnormal weighted value of the repeated data; in the case that the abnormal weighted value is greater than a preset weighted value, the repeated data is determined as the abnormal traffic data.

[0075] According to an embodiment of the present application, in order to improve the prediction accuracy, the initial traffic data sequence collected is subjected to outlier detection and analysis, and a support vector data description algorithm and an isolation forest algorithm combined model are used for outlier detection of the original data. The preset weighted value is used as the outlier detection score, and the outlier detection score is pre-set, for example, the outlier detection score is set to 0.01-0.1, and the abnormal data is filtered.

[0076] For example, when the proportion of the abnormal traffic data corresponding to the abnormal points in the sample is not clear, only one algorithm may not be able to completely identify all the abnormal points. Therefore, the support vector data description algorithm and the isolation forest algorithm are combined to improve the detection accuracy.

[0077] According to the size of the data sample, the outlier score is set . The outliers detected by the two methods are compared to find the most overlapping outliers in the two results. The data outliers are filtered out to ensure the accuracy of the model sample training. Among them, The abnormal score of the out-of-bound strength of the support vector data description algorithm and the short path strength of the isolation forest algorithm is normalized and weighted to obtain: . Wherein, w is the first abnormal weight, and 1-w is the second abnormal weight.

[0078] According to an embodiment of the present application, the weighted fusion mechanism of the abnormal traffic data is determined according to the comparison result of the abnormal weighted value and the preset weighted value, the confidence of different abnormal detection methods is comprehensively considered, and the accuracy and robustness of the determination of the abnormal traffic data are effectively improved, which avoids the limitation of a single method and ensures the rigor of the abnormal confirmation through the preset weighted threshold.

[0079] According to an embodiment of the present application, the target traffic data sequence is subjected to feature extraction based on the time information of the target traffic data sequence and the traffic correlation relationship between the plurality of traffic intersections, to obtain the time feature and the space feature of the target traffic intersection, including: processing a plurality of target traffic data in the target traffic data sequence based on the time information of the target traffic data sequence to obtain the time feature of the target traffic intersection; inputting the target traffic data sequence and the traffic correlation relationship into an attention graph convolution network to obtain the space feature of the target traffic intersection and a traffic network graph, the traffic network graph including a plurality of nodes and edges between the plurality of nodes, the plurality of nodes representing a plurality of traffic intersections at the target position, and the edges representing the traffic correlation relationship.

[0080] In the embodiments of this application, the methods for determining the time characteristics of traffic intersections may include extracting periodic features such as seconds, minutes, and hours, statistically analyzing traffic flow by time period, calculating the statistical features of a sliding time window, extracting time series trend features, using Fourier transform to extract periodic features, and calculating the distance to key time points.

[0081] According to embodiments of this application, spatial features of traffic intersections are obtained through attention map convolutional networks, effectively integrating the spatiotemporal characteristics of traffic data. This captures the dynamic patterns of traffic flow changes over time and models the spatial dependencies between different intersections in the road network. The attention mechanism can adaptively learn the correlation strength between nodes, accurately identify key traffic impact paths, and thus construct a more precise traffic state representation network. This collaborative extraction of spatiotemporal features provides a more comprehensive and accurate data foundation for traffic prediction and anomaly detection.

[0082] According to an embodiment of this application, processing multiple target traffic data in a target traffic data sequence based on the time information of the target traffic data sequence to obtain the time features of a target traffic intersection includes: using a forward long short-term memory network to process multiple target traffic data in the target traffic data sequence according to the time order of the target traffic data sequence to obtain the forward hidden features of the target traffic intersection; using a reverse long short-term memory network to process the target traffic data sequence in reverse time order to obtain the reverse hidden features of the target traffic intersection; and fusing the forward hidden features and the reverse hidden features to obtain the time features of the target traffic intersection.

[0083] In the embodiments of this application, the temporal order of the target traffic data sequence is arranged from left to right based on the principle of time from earliest to latest. The temporal reverse order of the target traffic data sequence is arranged from right to left based on the principle of time from earliest to latest. Feature fusion of forward and reverse hidden features is performed by concatenation.

[0084] In one embodiment, a bidirectional long short-term memory network model is constructed to capture the temporal dependence of traffic flow as a temporal feature of traffic intersections.

[0085] A gate mechanism is introduced to capture short-term and long-term sequence dependencies. Three gate structures can be used (input gate...). Forgot the door Output gate Information exchange. The model structure of each gate is calculated as follows:

[0086] (3).

[0087] in: is the previous network state output, is the current network state input, and are the weight matrix and bias vector of each gate, respectively.

[0088] The candidate memory state is calculated. The bidirectional long short-term memory network uses the candidate memory state to control whether the input gate is accepted, meaning the potential current new information. The forget gate and input gate functions are combined to form the current memory state, and the candidate memory is obtained according to the memory cell state update based on the current memory state. The hidden state is combined with the output gate to form the current output, which is calculated as follows:

[0089] (4).

[0090] wherein: is the candidate memory, and are the weight matrix and corresponding bias vector for generating the candidate memory, is the previous memory cell state, is the current memory cell state, is the current hidden state output.

[0091] The bidirectional long short-term memory network model structure is designed. Based on the correlation of time series data, a bidirectional long short-term memory network composed of two general long short-term memory network modules is proposed, and the forward and backward training data are used to learn the past and future information, and improve the modeling ability of time series data.

[0092] (5).

[0093] wherein: are the forward and backward network hidden state outputs of formula (5) , respectively, is the joint output (forward / backward splicing) of the bidirectional long short-term memory network model LSTM, that is, the time feature of the traffic intersection.

[0094] According to the embodiments of the present application, the bidirectional LSTM structure can comprehensively capture the time dependence of traffic data, considering both the influence of historical trends on the current state and the feedback of future changes on the current mode. By fusing the forward and backward hidden features, the complex time dynamic characteristics of the traffic intersection can be more accurately represented, and the representation of the time feature is enhanced, providing more rich time context information for subsequent traffic prediction and analysis.

[0095] According to embodiments of this application, the target traffic data sequence and traffic association relationships are input into an attention graph convolutional network to obtain spatial features of traffic intersections. This includes: determining the neighborhood datasets of multiple nodes based on the edges between multiple nodes in the traffic network graph. The neighborhood datasets include at least one neighborhood data subset, which is the target traffic data sequence of neighboring nodes connected by a node through l edges, where l is an integer greater than or equal to 1; weighting and aggregating the multiple neighborhood data subsets based on the learnable projection matrix and adaptive weighting factor corresponding to each node to obtain aggregated features; and processing the aggregated features using a nonlinear activation function to obtain spatial features.

[0096] In the embodiments of this application, attention map convolution is used to mine spatial features. A multi-level neighborhood information set for each node is defined. The traffic network can be represented as a graph. ,in It is a set of traffic nodes. These are the edges between these nodes. Represents a node and Direct path between, neighborhood Represents a node The neighborhood contains all directly connected to The node, the node of The set of neighborhood information is defined as follows:

[0097] (6).

[0098] Specifically, nodes of A collection of neighborhood information, a neighborhood dataset. Indicates from node Set off, cross The set of all nodes reachable by an edge. Typically, (itself); (First-order neighbors, all nodes) Directly connected nodes), nodes of -1 order neighborhood information set neighborhood dataset Indicates from node Set off, cross The set of all nodes reachable by an edge of -1. ;right Calculate using the above recurrence relation. , Indicates from node Set off, cross -2 sets of all nodes that can be reached by an edge.

[0099] The spatial features of the expanded attention map convolution output of the neighborhood information are designed. Each adjacent node is weighted according to the relative importance of the adjacent node in the graph, and the weight value is dynamically determined according to the similarity or correlation between nodes, and the adjacent nodes with strong spatial correlation are given greater weight values:

[0100] (7)。

[0101] wherein: is an adaptive weighting factor between node and node , is a learnable projection matrix, is the remaining connection of node . is the aggregation (including residual) of the single-order neighborhood, includes multiple single-order neighborhoods; any single-order neighborhood can represent a subset of neighborhood data.

[0102] In the method for obtaining the time feature of the traffic intersection and in the method for obtaining the spatial feature of the traffic intersection are related, and are different ways of obtaining the time feature and the spatial feature of the same "spatiotemporal feature tensor".

[0103] A high-order expanded attention map convolution is constructed to capture more information in a larger field of view. By introducing a larger scale of node neighborhood, information transmission is performed in a larger receptive field, and the complex relationship between distant nodes is learned.

[0104] (8)。

[0105] wherein, indicates the learnable coefficient of the th expanded order, which is used to adjust the contribution size of each order neighborhood to the output when multi-order information is aggregated. is the aggregation of the multi-order neighborhood according to .

[0106] Step S304, a nonlinear activation function is established to process the output.

[0107] (9)。

[0108] wherein, is the final output obtained after the multi-order aggregation and the nonlinear activation .

[0109] According to the embodiments of the present application, the correlation strength between traffic intersections is dynamically learned through the attention mechanism, which can adaptively capture the complex spatial dependence in the road network, thereby more accurately extracting the spatial features reflecting the traffic flow propagation pattern.

[0110] According to the embodiments of the present application, the traffic state of the target traffic intersection is determined based on the time feature and the space feature, including: inputting the time feature, the space feature and the traffic network graph into a hierarchical progressive attention module in a cascaded progressive attention network to obtain a space-time feature; inputting the space-time feature into a feedforward network in the cascaded progressive attention network to obtain a traffic prediction result of the target traffic intersection, the traffic prediction result including the traffic state.

[0111] In the embodiments of the present application, the cascaded progressive attention includes a plurality of hierarchical progressive attention blocks and a feedforward network. After the space-time feature passes through all the hierarchical progressive attention modules, the final time feature representation is processed using the feedforward network and converted into a prediction result.

[0112] (10)。

[0113] wherein, is a node-by-node and multi-step prediction of the future traffic state (flow). represents the final time feature after the space-time feature passes through all the hierarchical progressive attention modules, and is because After passing through times of progressive attention fusion, the space-time dependence is most completely encoded, which is suitable as the input of the FFN (Feed-Forward Network, feedforward network) for task mapping. And is the intermediate layer feature, which is used for layer-by-layer accumulation and amplification of effective space-time information.

[0114] According to the predicted traffic flow and congestion condition, the signal light duration is dynamically adjusted to give priority to the evacuation of congested sections; traffic control is implemented in specific areas to guide the reasonable distribution of vehicles and ensure the smoothness of the road and the efficiency of the traffic operation.

[0115] According to the embodiments of the present application, the time feature, the space feature and the traffic network graph are input into the hierarchical progressive attention module in the cascaded progressive attention network to obtain the space-time feature, including: extracting the spatial dependence relationship between a plurality of nodes from the traffic network graph, the spatial dependence relationship representing the topological relationship between the plurality of nodes; based on a learnable coefficient, the space feature is fused into the time feature to obtain an initial fusion feature, the learnable coefficient being used to control the fusion strength between the space feature and the time feature; the initial fusion feature and the spatial dependence relationship are input into the hierarchical progressive attention module to obtain the space-time feature.

[0116] According to an embodiment of the present application, each hierarchical progressive attention block gradually integrates temporal features and spatial knowledge into spatio-temporal dependencies, and the hierarchical progressive attention block is determined as:

[0117] (11).

[0118] wherein, and is the output of the i-1th and the ith hierarchical progressive attention module, is a low-dimensional space of graph embedding. is a module that performs attention aggregation on a multi-order expanded neighborhood and injects spatial knowledge into temporal representation layer by layer and step by step, and the output of the module is which will be used by the next layer to continue to integrate spatio-temporal dependencies hierarchically and progressively.

[0119] According to an embodiment of the present application, the hierarchical progressive attention module realizes multi-scale fusion of spatio-temporal features, and can adaptively capture local and global traffic pattern dependencies. The cascading attention mechanism effectively balances the importance of features in different spatio-temporal dimensions, enhances the representation ability of the model for complex traffic dynamics, and thus significantly improves the accuracy and robustness of multi-intersection collaborative prediction.

[0120] Figure 3 A flowchart of a traffic state determination method based on multi-source heterogeneous data according to another embodiment of the present application is shown.

[0121] As shown in Figure 3 , the traffic state determination method based on multi-source heterogeneous data according to the embodiment includes operations S10-S40.

[0122] In operation S10, anomaly detection: a feature extraction method is introduced to detect and remove abnormal nodes.

[0123] In an embodiment, the intelligent transportation system, which is a typical complex system, can be represented as a multi-level traffic data flow system. It is considered that the system has high-efficiency real-time data processing capability, and the system is a city or regional traffic flow monitoring area. The input of the intelligent transportation system includes , C, , and . Based on the support vector data description algorithm, the final out-of-bound strength score of the i-th sample and the corresponding spatio-temporal node area of each traffic flow are obtained according to formula (1). At the same time, an anomaly detection module based on the isolation forest algorithm ​, the abnormal data detection model is constructed by using the module, and then the outlier elimination strategy is designed, and the path abnormal score is obtained according to formula (2) and are fused, and S i characterization and , using the weighted average method, filtering out abnormal samples, obtaining new samples , which improves the detection accuracy and accuracy of abnormal points.

[0124] In operation S20, time modeling: a bidirectional long short-term memory network model is constructed to capture the time dependence of traffic flow.

[0125] According to the change of traffic flow data in different areas, a dynamic adjustment strategy can be designed, which takes a smooth change between 0.5 and 1.5, and this strategy is responsible for real-time adjustment of the flexibility of data cleaning algorithm.

[0126] Among them, for the bidirectional long short-term memory network model with gate mechanism , the network is used to build the short-term and long-term time series dependence modeling of traffic flow. The former gives the short-term prediction result of traffic flow, and the latter gives the long-term trend prediction result. As shown in formulas (3) to (5), based on the candidate memory state , the state update and the forgetting control are calculated, and the control strategy has the functions of optimizing space-time characteristics and accurate prediction. After obtaining the bidirectional long short-term memory network model structure containing gate mechanism and memory unit state, it is used for traffic flow prediction to obtain the prediction results at multiple times, that is, the bidirectional long short-term memory network is used to process the sequence data of new samples in multiple time (t-τ, ……t), so as to obtain the prediction results corresponding to each time .

[0127] In operation S30, space modeling: attention map convolution is used to mine spatial features.

[0128] The time step in the process of bidirectional long short-term memory network can be optimized on the basis of the above model, which can effectively ensure the accuracy of traffic flow prediction. In the implementation of traffic flow data, the effective acquisition of spatial features is an important link to ensure the effect of traffic flow prediction, but it is not practical to let all nodes enter global calculation in actual situation, therefore, the set of node multi-order neighborhood information is defined. The neighborhood of the node is the traffic flow characteristics around the node, and the extended attention map convolution based on neighborhood information is established according to the traffic flow data, and the spatial feature ​In the data stream processing process, a high-order expansion attention graph convolution structure based on neighborhood information is established, a nonlinear activation function is used to process the output, and effective extraction of spatial features is realized. According to formulas (6) to (9), based on attention graph convolution, the sample The corresponding new sample is modeled as a graph structure G(V, E), where V represents the nodes of the graph structure G(V, E), E represents the edges of the graph structure G(V, E), and the set of multi-order neighborhood information of the nodes is combined to output spatial features and a low-dimensional space GE of graph embedding.

[0129] In operation S40, spatio-temporal fusion and control: the spatio-temporal dependence relationship is integrated by using the cascaded progressive attention module, and the traffic flow is controlled in real time.

[0130] The expansion attention graph convolution method can use graph convolutional neural networks, extended attention mechanisms, etc., but the benchmark must be multi-level processing of graph structure data. The output of the hierarchical progressive attention module is determined by using the cascaded attention method . The dependence relationship rules between nodes are determined by using graph embedding information. Then, after all the hierarchical progressive attention modules, a feedforward network is used to process the final temporal feature representation and convert it into a prediction result process, realizing real-time optimization and control of traffic flow; specifically, according to formulas (10) and (11), through the cascaded progressive attention module, the temporal features and spatial graph features GE are fused, and the final temporal features are obtained step by step and iteratively, and a prediction result is generated. Finally, based on the prediction result , the control action U is output, completing the closed loop from feature fusion to decision control.

[0131] According to operations S10 to S40, a determination model of traffic state is constructed; when training the determination model of traffic state, the traffic flow data of February, May, August and December can be used as historical traffic data sets, the historical traffic data sets are used as training samples, and the traffic flow data of December is selected as a test sample of the historical traffic data set. The training sample is applied to train the determination model of traffic state, the test sample is applied to test the trained determination model of traffic state, and the accuracy and effectiveness of the training sample of the determination model of traffic state are evaluated.

[0132] The training samples record daily average traffic flow from various sensors, revealing trends of increasing, decreasing, or stable traffic volume (for a single road segment, weekday traffic flow is generally similar, but there are significant differences between weekday and weekend traffic flow). Traffic volume statistics and distribution patterns are analyzed after filtering out abnormal data from each sensor, obtaining the median location from different sensors to assess traffic conditions in different spatial areas. By integrating spatiotemporal dependencies, the predicted traffic flow data achieves a high degree of consistency with actual traffic data, ensuring consistency between actual and predicted traffic data at future time steps.

[0133] Figure 4 This is a graph of abnormal data detected by a sensor according to one embodiment. Figure 5 This is a graph of abnormal data detected by another sensor according to one embodiment. Figure 6 This is a comparison chart of predicted traffic flow and actual traffic flow according to one embodiment. Figure 7 This is a scatter plot of predicted and actual traffic data according to one embodiment.

[0134] like Figures 4-7 As shown, this study considers two periods: abnormal weather and peak holiday periods, used to detect abnormal data. Generally, around midnight, people are asleep, and traffic flow is close to zero. Figure 4 One of the sensors detected an unusual surge in traffic flow at midnight, which is clearly illogical. Figure 5 Another sensor recorded a traffic flow spectrum that was too sparse, with a large amount of data loss, which is also an anomaly. Figure 6 The displayed prediction results show that the predicted traffic flow curve is highly consistent with the actual traffic flow curve, indicating that this application can accurately capture traffic flow trends. Furthermore, Figure 7 The scatter plot further illustrates the consistency between the actual traffic flow data and the predicted traffic flow data at future time steps.

[0135] Therefore, this application can not only ensure high data accuracy, but also predict future traffic volume and perform intelligent optimization scheduling to improve traffic congestion.

[0136] Based on the above-mentioned traffic state determination method based on multi-source heterogeneous data, this application also provides a traffic state determination device based on multi-source heterogeneous data. The following will be combined with... Figure 8 The device is described in detail.

[0137] Figure 8 This is a structural block diagram of a traffic state determination device based on multi-source heterogeneous data, according to an embodiment of this application.

[0138] like Figure 8As shown, the traffic state determination apparatus 800 based on multi-source heterogeneous data of the embodiment includes an anomaly detection module 810, a denoising module 820, a feature extraction module 830, and a determination module 840.

[0139] The anomaly detection module 810 is configured to perform anomaly detection on the plurality of initial traffic data according to distances between the plurality of initial traffic data arranged according to time information in the initial traffic data sequence and preset data, and abnormal values of the plurality of initial traffic data, to obtain abnormal traffic data. The initial traffic data represents a traffic state of a target traffic intersection at a target location, the preset data is determined according to a Lagrange multiplier obtained by inputting a historical traffic data set of a plurality of traffic intersections at the target location into a quadratic programming solver, and the abnormal value is obtained by inputting the traffic data sequence into an isolation forest algorithm. In an embodiment, the anomaly detection module 810 can be configured to perform the operation S210 described above, and details are not described herein again.

[0140] The denoising module 820 is configured to perform denoising on the initial traffic data sequence according to the abnormal traffic data, to obtain a target traffic data sequence. In an embodiment, the denoising module 820 can be configured to perform the operation S220 described above, and details are not described herein again.

[0141] The feature extraction module 830 is configured to perform feature extraction on the target traffic data sequence based on time information of the target traffic data sequence and a traffic correlation relationship between the plurality of traffic intersections, to obtain a time feature and a space feature of the target traffic intersection. In an embodiment, the feature extraction module 830 can be configured to perform the operation S230 described above, and details are not described herein again.

[0142] The determination module 840 is configured to determine a traffic state of the target traffic intersection based on the time feature and the space feature. In an embodiment, the determination module 840 can be configured to perform the operation S240 described above, and details are not described herein again.

[0143] According to an embodiment of the present application, the anomaly detection module 810 includes a first determination sub-module, a second determination sub-module, and a third determination sub-module. The first determination sub-module is configured to determine a first abnormal data set including at least one first abnormal data from the plurality of initial traffic data according to distances between the plurality of initial traffic data and preset data. The second determination sub-module is configured to determine a second abnormal data set including at least one second abnormal data from the plurality of initial traffic data according to abnormal values of the plurality of traffic data. The third determination sub-module is configured to determine abnormal traffic data according to repeated data between the first abnormal data set and the second abnormal data set.

[0144] According to an embodiment of the present application, the first determining sub-module comprises a first determining unit. The first determining unit is configured to determine the initial traffic data as the first abnormal data when a distance between the initial traffic data and the preset data is greater than a preset distance, and the preset distance is determined according to a Lagrange multiplier obtained by inputting the historical traffic data set into a quadratic programming solver.

[0145] According to an embodiment of the present application, the second determining sub-module comprises a third determining unit. The third determining unit is configured to determine the initial traffic data as the second abnormal data when the abnormal value is less than a preset abnormal value.

[0146] According to an embodiment of the present application, the third determining sub-module comprises a weighted sum unit and a fourth determining unit. The weighted sum unit is configured to perform weighted sum on the repeated data according to the first abnormal weight and the second abnormal weight to obtain an abnormal weighted value of the repeated data; and the fourth determining unit is configured to determine the repeated data as the abnormal traffic data when the abnormal weighted value is greater than a preset weighted value.

[0147] According to an embodiment of the present application, the feature extraction module 830 comprises a first obtaining sub-module and a second obtaining sub-module. The first obtaining sub-module is configured to process a plurality of target traffic data in the target traffic data sequence based on time information of the target traffic data sequence to obtain a time feature of the target traffic intersection; and the second obtaining sub-module is configured to input the target traffic data sequence and the traffic correlation into an attention graph convolution network to obtain a spatial feature of the target traffic intersection and a traffic network graph, the traffic network graph comprising a plurality of nodes and edges between the plurality of nodes, the plurality of nodes representing a plurality of traffic intersections at the target location, and the edges representing the traffic correlation.

[0148] According to an embodiment of the present application, the first obtaining sub-module comprises a first processing unit, a second processing unit and a feature fusion unit. The first processing unit is configured to process the plurality of target traffic data in the target traffic data sequence according to a time sequence of the target traffic data sequence by using a forward long short-term memory network to obtain a forward hidden feature of the target traffic intersection; the second processing unit is configured to process the target traffic data sequence according to a reverse time sequence of the target traffic data sequence by using a reverse long short-term memory network to obtain a reverse hidden feature of the target traffic intersection; and the feature fusion unit is configured to perform feature fusion on the forward hidden feature and the reverse hidden feature to obtain the time feature of the target traffic intersection.

[0149] According to an embodiment of the present application, the second obtaining submodule includes a fourth determining unit, a weighted aggregation unit and a processing unit. The fourth determining unit is configured to determine a plurality of node respective neighborhood data sets based on edges between the plurality of nodes in the traffic network graph, the neighborhood data set including at least one neighborhood data subset, the neighborhood data subset being a target traffic data sequence of neighborhood nodes connected to the node through l edges, l being an integer greater than or equal to 1; the weighted aggregation unit is configured to perform weighted aggregation on the plurality of neighborhood data subsets based on the plurality of node respective corresponding learnable projection matrices and adaptive weighting factors, to obtain aggregated features; and the processing unit is configured to process the aggregated features using a nonlinear activation function to obtain spatial features.

[0150] According to an embodiment of the present application, the determining module 840 includes a first input submodule and a second input submodule. The first input submodule is configured to input the time features, the spatial features and the traffic network graph into a hierarchical progressive attention module in the cascaded progressive attention network to obtain spatiotemporal features; and the second input submodule is configured to input the spatiotemporal features into a feedforward network in the cascaded progressive attention network to obtain a traffic prediction result of the target traffic intersection, the traffic prediction result including a traffic state.

[0151] According to an embodiment of the present application, the first input submodule includes an extraction unit, a fusion unit and an input unit. The first input submodule is configured to extract spatial dependency relationships between the plurality of nodes from the traffic network graph, the spatial dependency relationships representing topological relationships between the plurality of nodes; the fusion unit is configured to fuse the spatial features into the time features based on a learnable coefficient to obtain initial fusion features, the learnable coefficient being used to control a fusion strength between the spatial features and the time features; and the input unit is configured to input the initial fusion features and the spatial dependency relationships into the hierarchical progressive attention module to obtain the spatiotemporal features.

[0152] According to an embodiment of the present application, any of the modules of the anomaly detection module 810, the de-noising module 820, the feature extraction module 830 and the determination module 840 can be combined in one module, or any of the modules can be split into multiple modules. Alternatively, at least part of the function of one or more of the modules can be combined with at least part of the function of the other modules, and implemented in one module. According to an embodiment of the present application, at least one of the anomaly detection module 810, the de-noising module 820, the feature extraction module 830 and the determination module 840 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on chip, a system on board, a system on package, an application specific integrated circuit (ASIC), or any other reasonable manner of integrating or packaging a circuit, etc. in hardware or firmware, or implemented in any one of the three implementation manners of software, hardware and firmware, or in a proper combination of any of the three. Alternatively, at least one of the anomaly detection module 810, the de-noising module 820, the feature extraction module 830 and the determination module 840 can be at least partially implemented as a computer program module which, when executed, can perform the corresponding function.

[0153] Figure 9 A block diagram of an electronic device suitable for implementing the traffic state determination method based on multi-source heterogeneous data according to an embodiment of the present application is shown.

[0154] As shown in Figure 9 The electronic device 900 according to an embodiment of the present application includes a processor 901 which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 to a random access memory (RAM) 903. The processor 901 can include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset, and / or a special-purpose microprocessor (e.g., an application specific integrated circuit (ASIC)), etc. The processor 901 can also include an on-board memory for cache use. The processor 901 can include a single processing unit or multiple processing units for performing different actions of the method processes according to embodiments of the present application.

[0155] In the RAM 903, various programs and data required for the operation of the electronic device 900 are stored. The processor 901, the ROM 902, and the RAM 903 are connected to each other via the bus 904. The processor 901 performs various operations of the method flow according to the embodiments of the present application by executing the programs in the ROM 902 and / or the RAM 903. It should be noted that the programs can also be stored in one or more memories other than the ROM 902 and the RAM 903. The processor 901 can also perform various operations of the method flow according to the embodiments of the present application by executing the programs stored in the one or more memories.

[0156] According to the embodiments of the present application, the electronic device 900 can further include an input / output (I / O) interface 905, which is also connected to the bus 904. The electronic device 900 can further include one or more of the following components connected to the input / output (I / O) interface 905: an input part 906 including a keyboard, a mouse, and the like; an output part 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), and the like, and a speaker, and the like; a storage part 908 including a hard disk, and the like; and a communication part 909 including a network interface card such as a LAN card, a modem, and the like. The communication part 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as necessary. A removable medium 911 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, and the like is mounted on the drive 910 as necessary, so that a computer program read therefrom is installed in the storage part 908 as necessary.

[0157] The present application also provides a computer readable storage medium, which can be included in the device / apparatus / system described in the above embodiments; or can exist separately without being assembled into the device / apparatus / system. The above computer readable storage medium carries one or more programs, when the one or more programs are executed, the method according to the embodiments of the present application is implemented.

[0158] According to an embodiment of the present application, the computer readable storage medium can be a non-transitory computer readable storage medium, for example, can include but not limited to: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In this application, a computer readable storage medium can be any tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present application, the computer readable storage medium can include one or more of the above-described ROM 902 and / or RAM 903 and / or a memory other than the ROM 902 and the RAM 903.

[0159] Embodiments of the present application also include a computer program product, which includes a computer program containing program codes for executing the methods shown in the flowcharts. When the computer program product is run in a computer system, the program codes are used to make the computer system implement the traffic state determination method based on multi-source heterogeneous data provided by embodiments of the present application.

[0160] The above-described functions defined in the system / device of the embodiments of the present application are performed when the computer program is executed by the processor 901. According to an embodiment of the present application, the above-described system, device, module, unit, etc. can be implemented by computer program modules.

[0161] In one embodiment, the computer program can rely on tangible storage media such as optical storage media, magnetic storage media, etc. In another embodiment, the computer program can also be transmitted, distributed, and downloaded in the form of a signal on a network medium, and be downloaded and installed through the communication part 909, and / or installed from the detachable medium 911. The program codes contained in the computer program can be transmitted by any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.

[0162] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 909, and / or installed from the detachable medium 911. When the computer program is executed by the processor 901, the above-described functions defined in the system of the embodiments of the present application are performed. According to an embodiment of the present application, the above-described system, device, apparatus, module, unit, etc. can be implemented by computer program modules.

[0163] According to embodiments of the present application, program code for implementing the computer programs provided by embodiments of the present application can be written in any combination of one or more programming languages, and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. The programming language includes, but is not limited to, such languages as Java, C++, python, "C" language, or the like. The program code can execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the remote computing device, or entirely on the remote computing device or server. In the latter scenario, the remote computing device can be connected to the user's computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computing device, such as through the Internet using an Internet Service Provider.

[0164] The computer program instructions can also be loaded onto a computer or other programmable information processing apparatus to cause a series of operations to be performed on the computer or other programmable information processing apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable information processing apparatus implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0165] Those skilled in the art will understand that features recited in the various embodiments of the present application can be combined and / or integrated in various ways, even if such combinations or integrations are not expressly noted in the present application. In particular, features recited in the various embodiments of the present application can be combined and / or integrated in ways that do not depart from the spirit and teachings of the present application. All such combinations and / or integrations are within the scope of the present application.

[0166] The embodiments of the present application have been described above. However, these embodiments are merely for the purpose of illustration, and are not intended to limit the scope of the present application. Although the embodiments are described separately above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Those skilled in the art can make various substitutions and modifications without departing from the scope of the present application, and these substitutions and modifications shall fall within the scope of the present application.

Claims

1. A traffic state determination method based on multi-source heterogeneous data, characterized in that, The traffic state determination method comprises: According to the distance between the multiple initial traffic data arranged according to the time information in the initial traffic data sequence and the preset data, and the abnormal value of each of the multiple initial traffic data, performing abnormal detection on the multiple initial traffic data to obtain abnormal traffic data, comprising: According to the distance between the multiple initial traffic data and the preset data, determining a first abnormal data set including at least one first abnormal data from the multiple initial traffic data; According to the abnormal value of each of the multiple initial traffic data, determining a second abnormal data set including at least one second abnormal data from the multiple initial traffic data; According to the repeated data between the first abnormal data set and the second abnormal data set, determining the abnormal traffic data; wherein the initial traffic data represents the traffic state of the target traffic intersection at the target position, the preset data is determined according to the Lagrange multiplier obtained by inputting the historical traffic data set of multiple traffic intersections at the target position into a quadratic programming solver, and the abnormal value is obtained by inputting the initial traffic data sequence into an isolation forest algorithm; According to the abnormal traffic data, denoising the initial traffic data sequence to obtain a target traffic data sequence; Based on the time information of the target traffic data sequence and the traffic correlation relationship between the multiple traffic intersections, feature extraction is performed on the target traffic data sequence to obtain the time feature and the space feature of the target traffic intersection; Based on the time feature and the space feature, determining the traffic state of the target traffic intersection.

2. The traffic state determination method according to claim 1, characterized by, According to the distance between the multiple initial traffic data and the preset data, determining a first abnormal data set including at least one first abnormal data from the multiple initial traffic data, comprising: In the case where the distance between the initial traffic data and the preset data is greater than the preset distance, the initial traffic data is determined as the first abnormal data, and the preset distance is determined according to the Lagrange multiplier obtained by inputting the historical traffic data set into the quadratic programming solver.

3. The traffic state determination method according to claim 1, characterized by, According to the abnormal value of each of the multiple initial traffic data, determining a second abnormal data set including at least one second abnormal data from the multiple initial traffic data, comprising: In the case where the abnormal value is less than the preset abnormal value, the initial traffic data is determined as the second abnormal data.

4. The traffic state determination method according to any one of claims 2 to 3, characterized by, According to the repeated data between the first abnormal data set and the second abnormal data set, determining the abnormal traffic data, comprising: According to the first abnormal weight and the second abnormal weight, performing weighted summation on the repeated data to obtain an abnormal weighted value of the repeated data; In the case where the abnormal weighted value is greater than the preset weighted value, the repeated data is determined as the abnormal traffic data.

5. The traffic state determination method according to claim 1, characterized by, According to the time information of the target traffic data sequence and the traffic correlation relationship between the multiple traffic intersections, feature extraction is performed on the target traffic data sequence to obtain the time feature and the space feature of the target traffic intersection, comprising: processing a plurality of the target traffic data in the target traffic data sequence based on time information of the target traffic data sequence, to obtain a time feature of the target traffic intersection; inputting the target traffic data sequence and the traffic correlation into an attention graph convolution network, to obtain a spatial feature of the target traffic intersection and a traffic network graph, the traffic network graph comprising a plurality of nodes and edges between the nodes, the nodes representing a plurality of the traffic intersections at the target location, and the edges representing the traffic correlation between the traffic intersections.

6. The traffic state determination method according to claim 5, characterized by, The processing a plurality of the target traffic data in the target traffic data sequence based on time information of the target traffic data sequence, to obtain a time feature of the target traffic intersection, comprises: processing a plurality of the target traffic data in the target traffic data sequence in a time sequence of the target traffic data sequence by a forward long short-term memory network, to obtain a forward hidden feature of the target traffic intersection; processing the target traffic data sequence in a reverse time sequence of the target traffic data sequence by a reverse long short-term memory network, to obtain a reverse hidden feature of the target traffic intersection; performing feature fusion on the forward hidden feature and the reverse hidden feature, to obtain the time feature of the target traffic intersection.

7. The traffic state determination method according to claim 5, characterized by, The inputting the target traffic data sequence and the traffic correlation into an attention graph convolution network, to obtain a spatial feature of the target traffic intersection, comprises: determining a neighborhood data set of each of the nodes based on the edges between the nodes in the traffic network graph, the neighborhood data set comprising at least one neighborhood data subset, the neighborhood data subset being a target traffic data sequence of a neighborhood node connected to the node by l edges, l being an integer greater than or equal to 1; performing weighted aggregation on a plurality of the neighborhood data subsets based on a learnable projection matrix and an adaptive weighting factor corresponding to each of the nodes, to obtain an aggregated feature; processing the aggregated feature by a nonlinear activation function, to obtain the spatial feature.

8. The traffic state determination method according to claim 5, characterized by, The determining a traffic state of the target traffic intersection based on the time feature and the spatial feature, comprises: inputting the time feature, the spatial feature and the traffic network graph into a hierarchical progressive attention module in a cascaded progressive attention network, to obtain a spatio-temporal feature; inputting the spatio-temporal feature into a feedforward network in the cascaded progressive attention network, to obtain a traffic prediction result of the target traffic intersection, the traffic prediction result comprising the traffic state.

9. The traffic state determination method according to claim 8, characterized by, The inputting the time feature, the spatial feature and the traffic network graph into a hierarchical progressive attention module in a cascaded progressive attention network, to obtain a spatio-temporal feature, comprises: extracting a spatial dependency between the nodes from the traffic network graph, the spatial dependency representing a topological relationship between the nodes; fusing the spatial feature into the time feature based on a learnable coefficient, to obtain an initial fusion feature, the learnable coefficient being used to control a fusion strength between the spatial feature and the time feature; The initial fusion feature and the spatial dependency are input into the hierarchical progressive attention module to obtain the spatiotemporal feature.

Images