A traffic flow mixed spatiotemporal prediction method, system, device and medium
By constructing a control unit graph in a sparse road network and using a hybrid model of graph convolutional neural network and recurrent neural network, processing traffic flow timing data, the problem of difficult-to-describe impact between nodes in a sparse road network is solved, and the sparse road is achieved The traffic flow prediction of the network has improved the overall control capabilities of the road network.
Patent Information
- Application Number
- CN202310027858.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-01-09
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2043-01-09
AI Technical Summary
The existing traffic flow prediction methods are difficult to effectively describe the mutual influence between nodes in the road network, especially in the sparse road network, where a single device is difficult to predict the traffic state of the road section, resulting in insufficient control capabilities of the sparse road network.
By dividing management and control units, a road network diagram is constructed, a hybrid model of graph convolutional neural network and recurrent neural network is adopted, and a mobile window function is used to process traffic flow timing data, a spatiotemporal impact factor matrix is established, a traffic flow prediction model is trained, and a sparse road network is realized. short-term and medium-term traffic flow forecasts.
Improve the traffic flow prediction capability of sparse road networks, effectively reflect the influence factors of road network operation in three aspects, including people, vehicles, roads, etc., solve the problem that a single device is difficult to predict traffic status on the road section, and takes into account both the demand from the macro and micro levels, improving the road The overall management and control capabilities of the network.
Smart Images

Figure CN116071920B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent transportation, and more specifically, to a traffic flow mixed spatiotemporal prediction method and system, device and medium. Background Art
[0002] Highway traffic flow prediction is to use historical traffic data to predict the traffic flow of the target road in the future, which plays a vital role in highway road control. Traffic flow data has the characteristics of nonlinearity, randomness, spatiotemporal variability, etc., making traffic flow prediction one of the research hotspots and difficulties in the field of transportation. Existing prediction algorithm models mainly focus on learning the spatiotemporal characteristics of traffic flow vehicle detectors, while taking into account the strong correlation between adjacent vehicle detector nodes in the road network that directly affects each other.
[0003] Traditional methods based on mathematical statistics and classical machine learning methods often model, analyze and predict the time series data of a certain observation point in the road network. With the development of deep neural networks, end-to-end convolutional neural networks (CNN), recurrent neural networks (RNN) and their variants, long short-term memory (LSTM) networks, and gated recurrent units (GRU) and other network models are widely used in time series prediction problems.
[0004] The above methods can capture the nonlinear characteristics of time series data, but it is difficult to describe the mutual influence between nodes for time series prediction of a single node. The traffic flow prediction problem that depends on the road network structure obviously cannot meet the needs of analyzing a single node one by one. Summary of the invention
[0005] The object of the present invention is to provide a method and system, device and medium based on mixed spatiotemporal prediction of traffic flow to solve at least one of the problems existing in the related art.
[0006] In order to achieve the above object, the present invention adopts the following technical solutions:
[0007] The first aspect of the present invention provides a traffic flow mixed spatiotemporal prediction method, comprising:
[0008] Divide the control units according to the road network structure of the target expressway, and construct a road network map with the control units as the core;
[0009] Establish a traffic flow time series data set of the control unit through the data collected by each sensor in the control unit;
[0010] A moving window function is set based on a traffic flow time series data period, and the traffic flow time series data set is processed based on the moving window function to obtain a training set and a verification set of traffic flow time series data with spatiotemporal characteristics;
[0011] Build a hybrid model of graph convolutional neural network and recurrent neural network;
[0012] Based on the periodic statistical values of traffic flow time series data, a traffic flow periodic statistical feature sequence of the graph convolutional neural network node feature input X is established;
[0013] Establish a sequence of periodic spatiotemporal influence factor adjacency matrices of edge feature input A of graph convolutional neural network;
[0014] Input the adjacency matrix sequence of the periodic spatiotemporal impact factors and the traffic flow periodic statistical feature sequence into the graph convolutional neural network of the hybrid model, and input the traffic flow time series data training set with spatiotemporal characteristics into the recurrent neural network of the hybrid model for training to obtain a traffic flow prediction model;
[0015] Traffic flow prediction is performed according to the traffic flow prediction model.
[0016] Optionally, the step of establishing a traffic flow periodic statistical feature sequence of a graph convolutional neural network node feature input X based on periodic statistical values of traffic flow time series data includes:
[0017] A graph convolutional neural network is constructed with the control unit as the node and the periodic statistical value of the traffic flow data as the node feature. Based on the periodic data trends and characteristics of the traffic flow represented by the graph convolutional neural network, a traffic flow periodic statistical feature sequence of the graph convolutional neural network node feature input X is established.
[0018] Optionally, the periodic statistical values of the traffic flow data include the mean, median, standard deviation, skewness and kurtosis values of the data within the moving window function period.
[0019] Optionally, the control unit includes a common unit and a special unit;
[0020] The common unit includes the road section between toll stations of the target highway;
[0021] The special units include bridges and tunnels, interchanges, provincial boundary toll stations and service areas with emergency rescue capabilities.
[0022] Optionally, the control unit is provided with at least one monitoring point, each of which is provided with a vehicle detector, a video detector, a microwave detector, a radar detector or a traffic volume surveyor;
[0023] The traffic parameters include the traffic volume, average vehicle speed and occupancy rate monitored by the control unit.
[0024] Optionally, the step of setting a moving window function based on a traffic flow time series data period, and processing the traffic flow time series data set based on the moving window function to obtain a training set and a validation set of traffic flow time series data with spatiotemporal characteristics comprises:
[0025] According to prior knowledge, determine the period T of traffic flow sampling time series data;
[0026] The moving window function is set as a one-dimensional or multi-dimensional input sequence with a period equal to the period of the traffic flow sampling time series data, and the one-dimensional or multi-dimensional output of the target period is predicted. The complete relevant data set is processed by the moving window function to convert the traffic flow time series data set into supervised learning data to obtain a traffic flow time series data training set and a verification set with spatiotemporal characteristics.
[0027] Optionally, the step of establishing a periodic spatiotemporal impact factor adjacency matrix sequence includes:
[0028] Determine the spatiotemporal factors that affect the traffic flows between the control units, and generate spatiotemporal influence factors related to the spatiotemporal factors as dynamic characteristics of the control units through a multiplication model;
[0029] An adjacency matrix is constructed based on the relationship between the traffic volume within the period between the control units and the spatiotemporal influencing factors, and an adjacency matrix sequence of the spatiotemporal influencing factors of the traffic flow period of the graph convolutional neural network edge feature input A is generated.
[0030] The second aspect of the present invention provides a hybrid spatiotemporal prediction system for traffic flow, which comprises a road network diagram construction module, a data acquisition module, a data processing module, a model training module and a prediction module, wherein:
[0031] The traffic flow network construction module is used to divide the control units according to the network structure of the target expressway and construct a road network map with the control units as the core;
[0032] The data acquisition module is used to establish a traffic flow time series data set of the control unit through the data collected by each sensor in the control unit;
[0033] The data processing module is used to set a moving window function based on the traffic flow time series data cycle, process the traffic flow time series data set based on the moving window function to obtain a traffic flow time series data training set and a verification set with spatiotemporal characteristics; establish a traffic flow cycle statistical feature sequence of the graph convolutional neural network node feature input X based on the traffic flow time series data cycle statistics; and establish a periodic spatiotemporal influence factor adjacency matrix sequence;
[0034] The model training module is used to establish a hybrid model of a graph convolutional neural network and a recurrent neural network; input the adjacency matrix sequence of the periodic spatiotemporal impact factor and the traffic flow periodic statistical feature sequence into the graph convolutional neural network of the hybrid model, and input the traffic flow time series data training set with spatiotemporal characteristics into the recurrent neural network of the hybrid model for training to obtain a traffic flow prediction model;
[0035] The prediction module is used to perform traffic flow prediction based on the traffic flow prediction model.
[0036] The third aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method provided in the first aspect of the present invention when executing the program.
[0037] A fourth aspect of the present invention provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided by the first aspect of the present invention.
[0038] The beneficial effects of the present invention are as follows:
[0039] The mixed spatiotemporal prediction method for traffic flow provided by the present invention reflects the influencing factors of three aspects of road network operation, namely, people, vehicles and roads, through the designed spatiotemporal influencing factors of the control unit, so as to solve the problem that a single device is difficult to predict the traffic status of a road section due to the small number of external monitoring facilities in a sparse road network; according to the control principle combining generality with particularity, taking into account both macro and micro needs, the control capability of a sparse road network is improved from the overall road network level. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings.
[0041] Figure 1 A flow chart of a traffic flow mixed spatiotemporal prediction method proposed by an embodiment of the present invention is shown.
[0042] Figure 2 A schematic diagram showing a hybrid model proposed in one embodiment of the present invention.
[0043] Figure 3 A schematic diagram of a traffic flow hybrid spatiotemporal prediction system proposed in one embodiment of the present invention is shown.
[0044] Figure 4 A schematic diagram showing the structure of a computer device proposed in one embodiment of the present application is shown. DETAILED DESCRIPTION
[0045] In order to explain the present invention more clearly, the present invention is further described below in conjunction with embodiments and drawings. Similar components in the drawings are represented by the same reference numerals. It should be understood by those skilled in the art that the content specifically described below is illustrative rather than restrictive, and the protection scope of the present invention should not be limited thereto.
[0046] To solve at least one of the above problems, the present invention proposes a traffic flow mixed spatiotemporal prediction method and system, device and medium. The traffic flow mixed spatiotemporal prediction method and system, device and medium are described below through several specific embodiments.
[0047] The inventors found that although some traditional methods consider the connection properties between observation points in the traffic topology, they have not been fully utilized. As a discrete irregularly arranged structure that is easily affected by distance, the road network is naturally applied to traffic prediction problems using graph representation and graph convolutional networks (GCNs) and has achieved excellent results. However, these research objects still remain at the spatiotemporal correlation of a certain detector. There are few reports on traffic flow prediction research that focuses on the essence of traffic management, that is, traffic control capabilities, especially at the regional road network level.
[0048] Sparse road network usually refers to a road network with a small total amount of transportation, a low level of transportation infrastructure construction and large node spacing. In view of the characteristics of sparse road network, such as low level of transportation infrastructure construction and large node spacing, large road network control area, few external control facilities, and low traffic volume on most sections, we have established an abstract control unit layer on the road network equipment perception layer to meet the growing traffic control needs now and in the future. The abstraction refers to the unified abstract representation of the underlying perception device data at the control layer. Regardless of the specific type and number of perception devices, the control unit defines unified data attributes.
[0049] Therefore, from the perspective of sparse road network control, we define adaptive control units and construct a road network graph structure with a set of road network control units as the core. Through the spatiotemporal data set of road traffic flow perceived by each control unit, we apply the graph convolutional neural network GCN and recurrent neural network LSTM hybrid model algorithm to realize data-driven short-term / medium- and long-term traffic flow prediction at the sparse regional road network level. Among them, short-term traffic flow prediction is the traffic flow prediction after 5-30 minutes, and medium- and long-term traffic flow prediction is the traffic flow prediction after 30 minutes to several hours.
[0050] An embodiment of the present invention provides a mixed spatiotemporal prediction method for traffic flow, such as Figure 1 As shown, including,
[0051] Divide the control units according to the road network structure of the target expressway, and construct a road network map with the control units as the core;
[0052] Establish a traffic flow time series data set of the control unit through the data collected by each sensor in the control unit;
[0053] A moving window function is set based on a traffic flow time series data period, and the traffic flow time series data set is processed based on the moving window function to obtain a training set and a verification set of traffic flow time series data with spatiotemporal characteristics;
[0054] Build a hybrid model of graph convolutional neural network and recurrent neural network;
[0055] Based on the periodic statistical values of traffic flow time series data, a traffic flow periodic statistical feature sequence of the graph convolutional neural network node feature input X is established;
[0056] Establish a sequence of periodic spatiotemporal influence factor adjacency matrices of edge feature input A of graph convolutional neural network;
[0057] Input the adjacency matrix sequence of the periodic spatiotemporal impact factors and the traffic flow periodic statistical feature sequence into the graph convolutional neural network of the hybrid model, and input the traffic flow time series data training set with spatiotemporal characteristics into the recurrent neural network of the hybrid model for training to obtain a traffic flow prediction model;
[0058] like Figure 2 As shown, it is a schematic diagram of the input and output of the prediction model, the adjacency matrix sequence of the periodic spatiotemporal impact factor and the traffic flow periodic statistical feature sequence are input into the graph convolutional neural network of the hybrid model, and the traffic flow time series data training set with spatiotemporal characteristics is input into the recurrent neural network of the hybrid model for training to obtain the traffic flow prediction model;
[0059] Predictions are made based on the traffic flow prediction model.
[0060] This embodiment is aimed at sparse road networks. It divides the control units according to the road network structure of the target expressway, that is, establishes an abstract control unit layer on the road network equipment perception layer, constructs a road network graph structure with a set of road network control units as the core, and uses the spatiotemporal data set of road section traffic flow composed of perceptions of various control units to solve the problem of small total transportation volume and large node spacing in sparse road networks; through the designed spatiotemporal influencing factors of the control units, it reflects the influencing factors of road network operation in three aspects, namely, people, vehicles and roads, and solves the problem that a single device is difficult to predict the traffic status of a road section due to the small number of external monitoring facilities in sparse road networks; in accordance with the control principle of combining generality with particularity, it takes into account both macro and micro needs, and improves the control capability of sparse road networks from the overall road network level.
[0061] The traffic flow time series data set is processed by a moving window function to obtain a training set and a verification set of traffic flow time series data with spatiotemporal characteristics, and the data is secondary extracted by a moving window function to segment the traffic flow time series data set.
[0062] In a possible implementation, the control unit refers to the smallest functional unit in the highway control system that has basic control capabilities, a single direction, and satisfies certain control strategies; the control unit includes a common unit and a special unit;
[0063] The common unit includes a road section between toll stations on the target expressway, such as a road section between a first toll station and a second toll station;
[0064] The special units include bridges and tunnels, interchanges, provincial boundary toll stations, and service areas with emergency rescue capabilities. The comprehensive status of the evaluation control area is the comprehensive status of multiple monitoring points within its coverage area.
[0065] In a possible implementation, a traffic flow time series data set of the control unit is established by summarizing, analyzing, judging and optimizing the data collected by various sensors in the control unit. The data structure of the traffic flow time series data set is shown in Table 1:
[0066] Table 1: Data structure of traffic flow time series dataset
[0067]
[0068] The control unit is provided with at least one monitoring point, each of which is provided with a vehicle detector, a video detector, a microwave detector, a radar detector or a traffic volume surveyor; the sampling period of each sensor is 5 minutes; the traffic parameters include the traffic volume monitored by the control unit, the average speed of the vehicle, the occupancy rate and whether there are traffic incidents in the unit.
[0069] In a possible implementation, the step of setting a moving window function based on a traffic flow time series data period, and processing the traffic flow time series data set based on the moving window function to obtain a training set and a validation set of traffic flow time series data with spatiotemporal characteristics includes:
[0070] According to prior knowledge, the traffic flow time series data period T of the 5-minute sampling time series data of the traffic flow is determined, where T is 60 minutes;
[0071] 80% of the traffic flow time series data are selected as training samples, and 20% are used for training verification;
[0072] The general demand in engineering applications is to predict traffic volume and average speed. A moving window function is designed according to the determined traffic flow time series data period. The period of the moving window function is set to be equal to the one-dimensional or multi-dimensional input sequence xTS of the traffic flow time series data period T. The moving window is a one-dimensional or multi-dimensional output label yTS of the predicted target period yT, where yT=1 / 3 / 6 / 12, that is, the prediction period is 5 minutes, 15 minutes, 30 minutes or one hour; xTS and yTS are both processed by standard normalization, the time series data set is converted into supervised learning data, and a training sample sequence is generated. The traffic flow time series data set is converted into supervised learning data to obtain a training set and a verification set of traffic flow time series data with spatiotemporal characteristics.
[0073] The moving window function can be directly imported through a mature module, and its code is "fromtsmoothie.utils_func import create_windows"; the moving window function can also be created, and the specific code includes:
[0074]
[0075] return np.squeeze(window_data,1)
[0076] The description of each parameter includes:
[0077] data: 2D array (timetemps, pandasseries) of time series of the same size;
[0078] window_shape: integer, greater than 1, the size of the sliding window used to divide the input sequence;
[0079] step: integer, default = 1, used to generate the step of the sliding window. The overlap of two adjacent windows can be defined as (window_shape-step);
[0080] start_id: integer, default = None, the starting position of the sliding operation. It should be noted that if it is None, the window is generated from index 0;
[0081] end_id: integer, default = None, the end position of the sliding operation. It should be noted that if the end position is not defined, the window ends at the last available position.
[0082] Window_data: 3D array window_slices, window_shape, series. The input data is divided into windows of the same length.
[0083] y_prime is the label sequence, X is the observation sequence, alllabels is the label list, FeatureFunction is the feature function, and weights is the weight list;
[0084] The code for generating the input sequence and label column sequence is as follows:
[0085] Generate input sequence: xTS=create_windows(raw_df, window_shape=T, end_id=-yT), where raw_df is the original data of traffic flow prediction item.
[0086] Generate label column sequence: yTS = create_windows(raw_df, window_shape = T, start_id = yT) yTS = y_train.reshape(-1, yT*n_features)
[0087] Among them, raw_df is the original data of traffic flow prediction items, n_features is the number of prediction items of traffic flow data,
[0088] In a specific embodiment, the prediction items include traffic volume and average speed, and n_features=number of units*2.
[0089] In a possible implementation, the traffic flow periodic statistical feature sequence of the graph convolutional neural network node feature input X is established based on the periodic statistical value of the traffic flow time series data, including:
[0090] A graph convolutional neural network is constructed with the control unit as the node and the periodic statistical value of the traffic flow data as the node feature. Based on the periodic data trends and characteristics of the traffic flow represented by the graph convolutional neural network, a traffic flow periodic statistical feature sequence of the graph convolutional neural network node feature input X is established.
[0091] Specifically, the main code for obtaining the traffic flow periodic statistical feature sequence is as follows:
[0092] The main code is as follows:
[0093]
[0094]
[0095] In a possible implementation, the traffic flow data periodic statistical values include the mean, median, standard deviation, skewness and kurtosis values of the data within the moving window function period.
[0096] In a possible implementation, establishing a periodic spatiotemporal impact factor adjacency matrix sequence includes:
[0097] Determine the spatiotemporal factors that affect the traffic flows between the control units, and generate spatiotemporal influence factors related to the spatiotemporal factors as dynamic characteristics of the control units through a multiplication model;
[0098] An adjacency matrix is constructed based on the relationship between the traffic volume within the period between the control units and the spatiotemporal influencing factors, and an adjacency matrix sequence of the spatiotemporal influencing factors of the traffic flow period of the graph convolutional neural network edge feature input A is generated.
[0099] Specifically, for sparse highway networks, in addition to factors such as environmental weather and events, the main factors that affect the traffic flow between control units involve three spatiotemporal factors: traffic volume, average speed, and control unit length;
[0100] From the overall consideration of the sparse road network, there are explicit and implicit traffic correlations between the control units as nodes of the graph convolutional neural network. The edges of the graph convolutional neural network are the traffic correlations between nodes, including the spatiotemporal characteristics between nodes. Therefore, the adjacency matrix is constructed based on the mutual correlation between the traffic volume and spatiotemporal influencing factors within the cycle between the control units, and the adjacency matrix sequence of the traffic flow cycle spatiotemporal influencing factors of the graph convolutional neural network edge feature input A is established.
[0101] The main code for establishing the adjacency matrix sequence is as follows:
[0102]
[0103]
[0104] Among them, the spatiotemporal impact factor of the control unit TSL = T*L*S,
[0105] Among them, T is the average traffic volume in a 5-minute section, unit: vehicles / 5 minutes, S is the average speed in a 5-minute section, unit: kilometers / hours, L is the mileage of the control unit, unit: kilometers.
[0106] It should be noted that the spatiotemporal impact of traffic control units involves three factors: people, vehicles and roads. From the perspective of control, the greater the flow, the greater the speed, the longer the route involved in the control, the more disturbance factors there are, and thus the greater the impact on the entire sparse road network and other related control units.
[0107] An embodiment of the present invention provides a traffic flow hybrid spatiotemporal prediction system, such as Figure 3 As shown, the traffic flow hybrid spatiotemporal prediction system includes a road network diagram construction module 201, a data acquisition module 202, a data processing module 203, a model training module 204 and a prediction module 205, wherein:
[0108] The traffic flow network construction module 201 is used to divide the control units according to the network structure of the target expressway and construct a road network map with the control units as the core;
[0109] The data acquisition module 202 is used to establish a traffic flow time series data set of the control unit through the data collected by each sensor in the control unit;
[0110] The data processing module 203 is used to set a moving window function based on the traffic flow time series data cycle, process the traffic flow time series data set based on the moving window function to obtain a traffic flow time series data training set and a verification set with spatiotemporal characteristics; establish a traffic flow cycle statistical feature sequence of the graph convolutional neural network node feature input X based on the traffic flow time series data cycle statistics; and establish a periodic spatiotemporal impact factor adjacency matrix sequence;
[0111] The model training module 204 is used to establish a hybrid model of a graph convolutional neural network and a recurrent neural network; input the periodic spatiotemporal impact factor adjacency matrix sequence and the traffic flow periodic statistical feature sequence into the graph convolutional neural network of the hybrid model, and input the traffic flow time series data training set with spatiotemporal characteristics into the recurrent neural network of the hybrid model for training to obtain a traffic flow prediction model.
[0112] The prediction module 205 is used to perform traffic flow prediction according to the traffic flow prediction model.
[0113] It should be noted that the principles and workflow of the traffic flow hybrid spatiotemporal prediction system provided in this embodiment are similar to the principles and workflow of the above-mentioned traffic flow hybrid spatiotemporal prediction method. The relevant parts can be referred to the above description and will not be repeated here.
[0114] like Figure 4 As shown, a schematic diagram of the structure of a computer device provided by the third embodiment of the present invention. Suitable for implementing the traffic flow mixed spatiotemporal prediction method provided by the above embodiment, including a central processing module (CPU), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) or the program loaded from the storage part to the random access memory (RAM). In the RAM, various programs and data required for the operation of the computer device are also stored. The CPU, ROM and RAM are connected to each other through a bus. The input / output (I / O) interface is also connected to the bus.
[0115] The following components are connected to the I / O interface: an input section including a keyboard, a mouse, etc.; an output section including a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section including a hard disk, etc.; and a communication section including a network interface card such as a LAN card, a modem, etc. The communication section performs communication processing via a network such as the Internet. A drive is also connected to the I / O interface as needed. Removable media, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., are installed on the drive as needed so that a computer program read therefrom is installed into the storage section as needed.
[0116] In particular, according to this embodiment, the process described in the flowchart above can be implemented as a computer software program. For example, this embodiment includes a computer program product, which includes a computer program tangibly contained on a computer-readable medium, and the computer program includes program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part, and / or installed from a removable medium.
[0117] A fourth embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements:
[0118] Divide the control units according to the road network structure of the target expressway, and construct a road network map with the control units as the core;
[0119] Establish a traffic flow time series data set of the control unit through the data collected by each sensor in the control unit;
[0120] A moving window function is set based on a traffic flow time series data period, and the traffic flow time series data set is processed based on the moving window function to obtain a training set and a verification set of traffic flow time series data with spatiotemporal characteristics;
[0121] Based on the periodic statistical values of traffic flow time series data, a traffic flow periodic statistical feature sequence of the graph convolutional neural network node feature input X is established;
[0122] Establish the adjacency matrix sequence of periodic space-time impact factors;
[0123] Build a hybrid model of graph convolutional neural network and recurrent neural network;
[0124] Input the adjacency matrix sequence of the periodic spatiotemporal impact factors and the traffic flow periodic statistical feature sequence into the graph convolutional neural network of the hybrid model, and input the traffic flow time series data training set with spatiotemporal characteristics into the recurrent neural network of the hybrid model for training to obtain a traffic flow prediction model;
[0125] Predictions are made based on the traffic flow prediction model.
[0126] In practical applications, the computer-readable storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.
[0127] It should be noted that the flowcharts and schematic diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product of the present embodiment. In this regard, each box in the flowchart or schematic diagram can represent a module, a program segment or a part of a code, and a part of the above-mentioned module, program segment or code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the schematic diagram and / or flowchart, and the combination of the boxes in the schematic diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0128] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not limitations on the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the embodiments here. All obvious changes or modifications derived from the technical solution of the present invention are still within the protection scope of the present invention.
Claims
1. A traffic flow mixed spatiotemporal prediction method, characterized in that: It includes dividing the control units according to the road network structure of the target expressway and constructing a road network map with the control units as the core; Establish a traffic flow time series data set of the control unit through the data collected by each sensor in the control unit; A moving window function is set based on a traffic flow time series data period, and the traffic flow time series data set is processed based on the moving window function to obtain a training set and a verification set of traffic flow time series data with spatiotemporal characteristics; Build a hybrid model of graph convolutional neural network and recurrent neural network; Based on the periodic statistical values of traffic flow time series data, a traffic flow periodic statistical feature sequence of the graph convolutional neural network node feature input X is established; Establish a sequence of periodic spatiotemporal influence factor adjacency matrices of edge feature input A of graph convolutional neural network; Input the adjacency matrix sequence of the periodic spatiotemporal impact factors and the traffic flow periodic statistical feature sequence into the graph convolutional neural network of the hybrid model, and input the traffic flow time series data training set with spatiotemporal characteristics into the recurrent neural network of the hybrid model for training to obtain a traffic flow prediction model; Performing traffic flow prediction according to the traffic flow prediction model; The traffic flow periodic statistical feature sequence of establishing the graph convolutional neural network node feature input X based on the periodic statistical value of the traffic flow time series data includes: A graph convolutional neural network is constructed with the control unit as a node and the periodic statistical value of traffic flow data as the node feature. A traffic flow periodic statistical feature sequence of the graph convolutional neural network node feature input X is established based on the traffic flow periodic data trend and features represented by the graph convolutional neural network. The traffic flow data periodic statistical values include the mean, median, standard deviation, skewness and kurtosis values of the data within the moving window function period; The control unit includes ordinary units and special units; The common unit includes the road section between toll stations of the target highway; The special units include bridges and tunnels, interchanges, provincial boundary toll booths, and service areas with emergency rescue capabilities; The control unit is provided with at least one monitoring point, each of which is provided with a vehicle detector, a video detector, a microwave detector, a radar detector or a traffic volume survey device; The traffic parameters of the traffic flow time series dataset include the traffic volume, average vehicle speed and occupancy rate monitored by the control unit; The step of setting a moving window function based on the traffic flow time series data period, and processing the traffic flow time series data set based on the moving window function to obtain a training set and a verification set of traffic flow time series data with spatiotemporal characteristics comprises: According to prior knowledge, determine the period T of traffic flow sampling time series data; The moving window function is set as a one-dimensional or multi-dimensional input sequence with a period equal to the period of the traffic flow sampling time series data, and the one-dimensional or multi-dimensional output of the target period is predicted. The traffic flow time series data set is processed by the moving window function to convert the traffic flow time series data set into supervised learning data to obtain a training set and a verification set of traffic flow time series data with spatiotemporal characteristics; The process of establishing a periodic spatiotemporal influence factor adjacency matrix sequence of the edge feature input A of the graph convolutional neural network includes: Determine the spatiotemporal factors that affect the traffic flows between the control units, and generate spatiotemporal influence factors related to the spatiotemporal factors as dynamic characteristics of the control units through a multiplication model; An adjacency matrix is constructed based on the relationship between the traffic volume within the period between the control units and the spatiotemporal influencing factors, and an adjacency matrix sequence of the spatiotemporal influencing factors of the traffic flow period of the graph convolutional neural network edge feature input A is generated.
2. A traffic flow hybrid spatiotemporal prediction system, characterized in that: The traffic flow hybrid spatiotemporal prediction system includes a road network diagram construction module, a data acquisition module, a data processing module, a model training module and a prediction module, wherein: The road network diagram construction module is used to divide the control units according to the road network structure of the target expressway and construct the road network diagram with the control units as the core; The data acquisition module is used to establish a traffic flow time series data set of the control unit through the data collected by each sensor in the control unit; The data processing module is used to set a moving window function based on the traffic flow time series data cycle, process the traffic flow time series data set based on the moving window function to obtain a traffic flow time series data training set and a verification set with spatiotemporal characteristics; establish a traffic flow cycle statistical feature sequence of the graph convolutional neural network node feature input X based on the traffic flow time series data cycle statistics; and establish a periodic spatiotemporal influence factor adjacency matrix sequence; The model training module is used to establish a hybrid model of a graph convolutional neural network and a recurrent neural network; input the adjacency matrix sequence of the periodic spatiotemporal impact factor and the traffic flow periodic statistical feature sequence into the graph convolutional neural network of the hybrid model, and input the traffic flow time series data training set with spatiotemporal characteristics into the recurrent neural network of the hybrid model for training to obtain a traffic flow prediction model; The prediction module is used to perform traffic flow prediction according to the traffic flow prediction model; The traffic flow periodic statistical feature sequence of establishing the graph convolutional neural network node feature input X based on the periodic statistical value of the traffic flow time series data includes: A graph convolutional neural network is constructed with the control unit as a node and the periodic statistical value of traffic flow data as the node feature. A traffic flow periodic statistical feature sequence of the graph convolutional neural network node feature input X is established based on the traffic flow periodic data trend and features represented by the graph convolutional neural network. The traffic flow data periodic statistical values include the mean, median, standard deviation, skewness and kurtosis values of the data within the moving window function period; The control unit includes ordinary units and special units; The common unit includes the road section between toll stations of the target highway; The special units include bridges and tunnels, interchanges, provincial boundary toll booths, and service areas with emergency rescue capabilities; The control unit is provided with at least one monitoring point, each of which is provided with a vehicle detector, a video detector, a microwave detector, a radar detector or a traffic volume survey device; The traffic parameters of the traffic flow time series dataset include the traffic volume, average vehicle speed and occupancy rate monitored by the control unit; The step of setting a moving window function based on the traffic flow time series data period, and processing the traffic flow time series data set based on the moving window function to obtain a training set and a verification set of traffic flow time series data with spatiotemporal characteristics comprises: According to prior knowledge, determine the period T of traffic flow sampling time series data; The moving window function is set as a one-dimensional or multi-dimensional input sequence with a period equal to the period of the traffic flow sampling time series data, and the one-dimensional or multi-dimensional output of the target period is predicted. The traffic flow time series data set is processed by the moving window function to convert the traffic flow time series data set into supervised learning data to obtain a training set and a verification set of traffic flow time series data with spatiotemporal characteristics; The process of establishing a periodic spatiotemporal influence factor adjacency matrix sequence of the edge feature input A of the graph convolutional neural network includes: Determine the spatiotemporal factors that affect the traffic flows between the control units, and generate spatiotemporal influence factors related to the spatiotemporal factors as dynamic characteristics of the control units through a multiplication model; An adjacency matrix is constructed based on the relationship between the traffic volume within the period between the control units and the spatiotemporal influencing factors, and an adjacency matrix sequence of the spatiotemporal influencing factors of the traffic flow period of the graph convolutional neural network edge feature input A is generated.
3. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method as claimed in claim 1 is implemented.
4. A computer storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method as claimed in claim 1 is implemented.
