Traffic flow prediction method and device based on adaptive hypergraph convolutional neural network

Abstract

The application discloses a traffic flow prediction method and device based on an adaptive hypergraph convolutional neural network, which captures dynamic node and hyperedge relations in a traffic flow structure that cannot be captured by a predefined graph or hypergraph through adaptive hypergraph learning; then uses a hypergraph convolution method to capture the spatial characteristics of the traffic flow in the adaptive hypergraph relations, uses the obtained spatial characteristics in a recurrent neural network structure to capture time characteristics, and finally obtains predicted traffic flow. The implementation method of the application is relatively simple, the network structure is simple, the model can be obtained by training through collected data, and manual modeling of graph network relations is not needed; the adaptive hypergraph learning spatial relations have strong robustness and strong interference ability for missing node position data; compared with other same type models, the application has better prediction effect, is beneficial to improving prediction accuracy, and has good model performance.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent transportation systems in computer network technology, and particularly relates to a traffic flow prediction method and device based on an adaptive hypergraph convolutional neural network. BACKGROUND

[0002] With the continuous improvement of urbanization process and social economic development level, a large number of vehicles begin to flow into the road, and the road congestion situation is getting worse. The intelligent transportation system uses advanced sensing technology to continuously collect vehicle data on the road, and modeling and analyzing the road vehicles have become the mainstream of relieving road congestion at home and abroad. As an important part of the road, it is of great significance to accurately model and predict the space-time of the road traffic flow for relieving the road congestion.

[0003] The current traffic flow prediction method mainly uses static graph structure or dynamic graph structure to simulate the relationship of road network nodes, but there are the following problems: 1. The static graph structure needs to obtain the road network data in advance and can only describe the road network relationship statically, but the traffic flow is in a dynamic process at all times, and obviously the simple static graph structure cannot accurately capture the dynamic relationship in the traffic flow. 2. The ordinary dynamic graph obtains the dynamic abstract relationship in the traffic flow through the node and the node relationship, and ignores the abstract relationship between the node and the edge. The above situations make the system have the problems of low prediction accuracy and poor model performance in the space-time modeling of the traffic flow. It is a problem to be solved to further think and solve this problem. SUMMARY

[0004] The present application proposes a traffic flow prediction method and device based on an adaptive hypergraph convolutional neural network to solve the problems of the prior art.

[0005] The purpose of the present application is realized by the following technical scheme: the first aspect of the embodiment of the present application provides a traffic flow prediction method based on an adaptive hypergraph convolutional neural network, including the following steps:

[0006] (1) Obtain traffic data by laying detectors in advance to construct multi-dimensional structured feature data; wherein the traffic data includes traffic flow, speed, density, and vehicle proportion data;

[0007] (2) Determine the network structure of the model, the network structure includes a data input layer, a space-time convolution layer and an output prediction layer; wherein the data input layer includes a space-time convolution module, the space-time convolution layer includes a plurality of space-time convolution modules, and the output prediction layer adopts a two-dimensional convolution operator;

[0008] (3) constructing a spatio-temporal convolution module to extract spatio-temporal features of the input information; the spatio-temporal convolution module mainly includes two parts: a spatial convolution network and a time convolution network; wherein the spatial convolution network uses adaptive hypergraph convolution as a spatial feature extractor, captures the spatial relationship between dynamic nodes through adaptive hypergraph, and performs convolution operation on the spatial relationship and traffic volume to complete spatial feature extraction of the spatial convolution network; the time convolution network uses a recurrent network to obtain time features from the spatial data obtained by the spatial convolution network;

[0009] (4) constructing a data input layer using the spatio-temporal convolution module obtained in the step (3);

[0010] (5) constructing a spatio-temporal convolution layer by stacking multiple spatio-temporal convolution modules obtained in the step (3) through a network;

[0011] (6) inputting the features obtained through the data input layer and the spatio-temporal convolution layer into an output prediction layer to obtain a prediction result of the traffic volume, obtaining a prediction error according to a loss function, and iterating model parameters through an optimization algorithm until the model converges to obtain an optimal model;

[0012] (7) inputting the processed historical data into the optimal model obtained in the step (6) to predict the traffic volume.

[0013] Further, the step (1) includes the following sub-steps:

[0014] (1.1) laying N sensors in a designated traffic network;

[0015] (1.2) sampling the historical traffic volume data of Y time periods before T time of the N sensors according to a sampling time interval Δt to obtain traffic volume data;

[0016] (1.3) sampling other historical traffic data of Y time periods before T time of the N sensors according to a sampling time interval Δt, and obtaining traffic data of K time periods of T+t1, T+t2, …, T+t k according to the sampled historical traffic data to obtain traffic situation data;

[0017] (1.4) constructing multi-dimensional structured feature data according to the traffic volume data obtained in the step (1.2) and the traffic situation data obtained in the step (1.3).

[0018] Further, the data input layer is a feature input encoder for realizing encoding conversion from spatio-temporal data to spatio-temporal feature data; the output prediction layer is used to convert the spatio-temporal feature data into a final prediction result.

[0019] Further, the step (3) comprises the following sub-steps:

[0020] (3.1) An adaptive hypergraph is calculated by an adaptive hypergraph generator, the Euclidean distance between the input data is calculated according to the distance calculation formula, a distance matrix is obtained, and a new adaptive hypergraph is obtained by multiplying the adaptive hypergraph calculated before the input;

[0021] (3.2) The spatial convolution network obtains spatial features through the hypergraph convolution network;

[0022] (3.3) The time convolution network adopts a gated recurrent unit (GRU) as a time convolution module, inputs the obtained spatial feature data into the gated recurrent unit, and filters long and short time feature data by using a gating mechanism. After the weight parameter calculation in the time convolution network forward propagation formula, the final time sequence feature output is obtained.

[0023] Further, the spatial convolution network obtains spatial features through the hypergraph convolution network, specifically: the degree matrix and edge matrix of the adaptive hypergraph are calculated by the spatial convolution formula, and the input signal is convolved in the adaptive hypergraph convolution manner to obtain the spatial features.

[0024] Further, the time convolution network forward propagation formula is:

[0025] H t =ReLU{dis(x)·H t-1}

[0026] z t =σ(W z [X t ,h t-1 ]+b z )

[0027] r t =σ(W r [X t ,h t-1 ]+b r )

[0028]

[0029]

[0030] Wherein, x is input data, X t represents input data at time t, dis(·) is a distance equation, ReLU is an activation function of a function, H t-1 is the previous adaptive hypergraph, H t is the updated adaptive hypergraph, z t , rt and h t are the update gate, reset gate and hidden gate of GRU respectively, b z , b r are the bias of the update gate and reset gate respectively, W z , W r are the weights of the update gate and reset gate respectively, and sigma (·) is an activation function sigmoid function.

[0031] Further, the expression of the loss function is:

[0032]

[0033] wherein y n is a true value, is a predicted value, and n represents the number of nodes.

[0034] Further, the step (6) of iteratively optimizing the model parameters by using an optimization algorithm is specifically: using an adam optimizer as a model parameter iterative optimizer to iteratively optimize the model parameters.

[0035] The second aspect of the embodiment of the application provides a vehicle flow prediction device based on an adaptive hypergraph convolutional neural network, comprising one or more processors, which are used to implement the vehicle flow prediction method based on the adaptive hypergraph convolutional neural network.

[0036] The third aspect of the embodiment of the application provides a computer readable storage medium, which stores a program, and the program is executed by a processor to implement the vehicle flow prediction method based on the adaptive hypergraph convolutional neural network.

[0037] The application has the advantages that the application can effectively solve the problem that the node and hyperedge relationship cannot be further expressed and modeled in other graph network prediction models based on the relationship between nodes and nodes, the implementation method of the application is relatively simple, the network structure is simple, the model can be obtained by training the collected data directly, and the graph network relationship does not need to be manually modeled, the adaptive hypergraph learning space relationship has strong robustness and strong interference ability for node position data loss, compared with other same type models, the application has better prediction effect, is beneficial to improving the prediction precision, and the performance of the model is good. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 is an adaptive hypergraph convolutional neural network algorithm structure;

[0039] Figure 2 is a network model training process schematic diagram;

[0040] Figure 3A structural schematic diagram of a vehicle flow prediction device based on an adaptive hypergraph convolutional neural network. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without creative work fall within the protection scope of the present application.

[0042] The vehicle flow prediction method based on the adaptive hypergraph convolutional neural network in the present application, as shown in the figure, comprises the following steps: Figure 1

[0043] (1) Obtain traffic data by laying detectors in advance, and construct multi-dimensional structured feature data. The traffic data includes but is not limited to vehicle flow, speed, density and vehicle proportion data.

[0044] (1.1) Lay N sensors in advance in a designated traffic network.

[0045] (1.2) Sample the historical vehicle flow data of Y time periods before T time of the N sensors according to a sampling time interval of Δt, and obtain vehicle flow data. The number of sampling data of the vehicle flow data is Y / Δt.

[0046] (1.3) Sample other historical traffic data of Y time periods before T time of the N sensors according to a sampling time interval of Δt, and obtain traffic data on K time periods of T+t1, T+t2, … T+t k according to the sampled historical traffic data, so as to obtain traffic situation data. The number of sampling data of the traffic situation data is Y / Δt.

[0047] It should be understood that the traffic situation data is the historical data of Y time periods before T time and the traffic situation data on K time periods of T+t1, T+t2, … T+t k at the N sensor laying position; the data included therein include speed, density and vehicle proportion data.

[0048] (1.4) Construct multi-dimensional structured feature data according to the vehicle flow data obtained in step (1.2) and the traffic situation data obtained in step (1.3).

[0049] ​For example, the PEMS-D4 dataset is used, which records the data collected by sensors on the highways in the San Francisco Bay Area, USA, and specifically covers data collected by 307 sensors from January 1, 2018 to February 28, 2018, with data collected every five minutes. In this case, the traffic flow data is mainly used. Through data analysis and processing tools such as pandas, the sensor grid data is sorted to obtain graph network data.

[0050] Among them, each hour is taken as a time window (12 consecutive five minutes), and the traffic flow data at each time is taken as historical data, and the data of the next hour is taken as label data. The final training data dimension is (10173, 307, 12, 1), and the prediction data dimension is (10173, 307, 12).

[0051] (2) Determine the network structure of the model, which mainly includes three main parts: data input layer, space-time convolution layer and output prediction layer. Among them, the data input layer includes a space-time convolution module, the space-time convolution layer includes multiple space-time convolution modules, and the output prediction layer uses a two-dimensional convolution operator.

[0052] In this embodiment, the space-time convolution module is used as the data input layer, which is essentially a feature input encoder, mainly used to realize the encoding conversion from space-time data to space-time feature data. The space-time convolution layer is composed of multiple space-time convolution modules. The data input layer can extract space-time features by inputting data into the space-time convolution layer. The output prediction layer uses a two-dimensional convolution operator as the output of the last layer. Through two-dimensional convolution, the space-time feature data is converted into the final prediction result.

[0053] (3) Construct a space-time convolution module, which is mainly divided into two parts: a spatial convolution network and a temporal convolution network. Among them, the spatial convolution network uses adaptive hypergraph convolution as a spatial feature extractor, which captures the spatial relationship between dynamic nodes through adaptive hypergraph, and performs convolution operation on the spatial relationship and traffic volume to complete the spatial feature extraction of the spatial convolution network. The time convolution network uses a recurrent network to obtain the time features of the spatial data obtained by the spatial convolution network.

[0054] Through the space-time convolution module, the space-time features of the input information can be extracted.

[0055] (3.1) Calculate the adaptive hypergraph through the adaptive hypergraph generator (AHC), calculate the Euclidean distance between the input data according to the distance calculation formula dis(·), obtain the distance matrix, and multiply the adaptive hypergraph calculated before the input to obtain a new adaptive hypergraph. The expression of the updated adaptive hypergraph is:

[0056] Ht = ReLU {dis(x) · H t-1}

[0057] where x is input data, dis(·) is a distance calculation formula, ReLU is an activation function of a function, H t-1 is a previous adaptive hypergraph, H t is an updated adaptive hypergraph.

[0058] Further, the expression of the distance calculation formula dis(x) is as follows:

[0059]

[0060] where x i and x j are the i-th data and the j-th data of the input data respectively, and n represents that the input data has n in total.

[0061] (3.2) The spatial convolution network obtains spatial features through a hypergraph convolution network (AHCN). Specifically, the degree matrix and the edge matrix of the adaptive hypergraph are calculated through the formula of the spatial convolution network, the input signal is convolved in the adaptive hypergraph convolution manner, and the output of the spatial convolution network is the spatial features.

[0062] where the spatial convolution formula is:

[0063]

[0064] where X is input data, H is a hypergraph, I N is a unit matrix, W is a diagonal matrix of hyperedges, D v is a degree matrix of the adaptive hypergraph, D e is an edge matrix of the adaptive hypergraph, E G is an embedding matrix, W G is a weight pool matrix, and b G is a bias pool matrix.

[0065] (3.3) The time convolution network adopts a gated recurrent unit (GRU) as a time convolution module, inputs the obtained spatial feature data into the gated recurrent unit, and uses the gating mechanism to filter to obtain long and short time feature data. After calculation by the weight parameters in the forward propagation formula of the time convolution network, the final time sequence feature output is obtained.

[0066] where the forward propagation formula of the time convolution network is:

[0067] H t = ReLU {dis(x) · H t-1}

[0068] zt = σ(W z [X t ,h t-1 ]+b z )

[0069] r t = σ(W r [X t ,h t-1 ]+b r )

[0070]

[0071]

[0072] wherein x is input data, X t represents input data at time t, dis(·) is a distance equation, ReLU is an activation function of a function, H t-1 is a previous adaptive hypergraph, H t is an updated adaptive hypergraph, z t , r t and h t are an update gate, a reset gate and a hidden gate of the GRU respectively, b z , b r are biases of the update gate and the reset gate respectively, W z , W r are weights of the update gate and the reset gate respectively, and σ(·) is an activation function sigmoid function. It should be understood that, is only an intermediate variable for easy understanding and has no actual meaning.

[0073] Exemplarily, specific network parameters of the spatio-temporal convolution module are shown in Table 1.

[0074] Table 1: Network parameters

[0075] Parameter name Parameter amount Hyperedge number 25 Number of spatio-temporal convolution modules in spatio-temporal convolution layer 64

[0076] (4) A data input layer is constructed by using the spatio-temporal convolution module obtained according to step (3).

[0077] (5) A spatio-temporal convolution layer is constructed by stacking multiple spatio-temporal convolution modules obtained according to step (3) in a network.

[0078] It should be understood that the network parameters of the spatio-temporal convolution module of the data input layer and the spatio-temporal convolution module of the spatio-temporal convolution layer are different.

[0079] In addition, the number of spatio-temporal convolution modules in the spatio-temporal convolution layer can be selected according to actual needs.

[0080] (6) The features obtained through the data input layer and the spatio-temporal convolution layer are input into the output prediction layer, and the prediction result of the final required traffic volume can be obtained. By calculating the loss function, the prediction error is obtained, and the model parameters are iterated using an optimization algorithm to achieve model convergence and obtain the optimal model.

[0081] The training process of the model is shown in Figure 2 The learning rate and loss weight are first initialized, that is, the prediction error is obtained according to the loss function, and the model parameters are obtained. Then, it is determined whether the model converges. If the model converges, the optimal model required in this embodiment is obtained. If the model does not converge, the model parameters are iterated and optimized, the adaptive hypergraph data is updated, the prediction is performed again, the prediction result is output, the loss function is calculated according to the new prediction result, and it is determined again whether the model converges until the model converges.

[0082] In this embodiment, two-dimensional convolution is used as the output prediction layer when outputting.

[0083] The specific data of the two-dimensional convolution layer is shown in Table 2.

[0084] Table 2: Convolution layer data

[0085] Convolution layer Number of convolution kernels Convolution kernel size Whether to bias Conv2D Prediction length + output dimension (1, hidden layer dimension) Yes

[0086] In this embodiment, when calculating the loss of the model, the mean absolute error suitable for the regression problem is used as the loss function of the model, and the expression is:

[0087]

[0088] where y n is the true value, is the predicted value, and n represents the number of nodes.

[0089] Further, when the model parameters are iterated, the adam optimizer is used as the model parameter iteration optimizer. It should be understood that the Adam optimizer can update the variables according to the shock of the historical gradient and the true historical gradient after filtering the shock.

[0090] (7) The processed historical data is input into the optimal model obtained in step (6) to predict the traffic volume.

[0091] It should be understood that the historical data of a certain road section, i.e., the corresponding sensor, is input into the optimal model to predict the traffic volume.

[0092] Corresponding to the foregoing embodiments of the traffic volume prediction method based on the adaptive hypergraph convolutional neural network, the present application also provides embodiments of a traffic volume prediction device based on the adaptive hypergraph convolutional neural network.

[0093] See Figure 3 The present invention provides a traffic flow prediction device based on an adaptive hypergraph convolutional neural network, comprising one or more processors for implementing the traffic flow prediction method based on the adaptive hypergraph convolutional neural network in the above embodiments.

[0094] The embodiments of the traffic flow prediction device based on adaptive hypergraph convolutional neural networks of the present invention can be applied to any device with data processing capabilities, such as a computer. The device embodiments can be implemented in software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of any data processing device loading the corresponding computer program instructions from non-volatile memory into memory for execution. From a hardware perspective, such as... Figure 3 The diagram shown is a hardware structure diagram of any data processing-capable device, including the traffic flow prediction device based on an adaptive hypergraph convolutional neural network according to the present invention. Except for... Figure 3 In addition to the processor, memory, network interface, and non-volatile memory shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0095] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0096] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0097] This invention also provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements the traffic flow prediction method based on an adaptive hypergraph convolutional neural network described in the above embodiments.

[0098] The computer readable storage medium can be an internal storage unit of any of the aforementioned devices with data processing capability, such as a hard disk or a memory. The computer readable storage medium can also be any of the aforementioned devices with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. Further, the computer readable storage medium can also include both an internal storage unit of any of the aforementioned devices with data processing capability and an external storage device. The computer readable storage medium is used to store the computer program and other programs and data required by the aforementioned devices with data processing capability, and can also be used to temporarily store data that has been output or will be output.

[0099] The above embodiments are only used to illustrate the technical solutions of the present application, rather than limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalent features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A traffic flow prediction method based on an adaptive hypergraph convolutional neural network, characterized in that, Includes the following steps: (1) Traffic data is acquired by pre-installing detectors and multi-dimensional structured feature data is constructed; wherein, the traffic data includes traffic flow, speed, density and vehicle ratio data. (2) Determine the network structure of the model, which includes a data input layer, a spatiotemporal convolutional layer and an output prediction layer; wherein, the data input layer includes a spatiotemporal convolutional module, the spatiotemporal convolutional layer includes multiple spatiotemporal convolutional modules, and the output prediction layer uses a two-dimensional convolutional layer; (3) Construct a spatiotemporal convolution module to extract spatiotemporal features of the input information; the spatiotemporal convolution module is mainly divided into two parts: a spatial convolutional network and a temporal convolutional network; wherein, the spatial convolutional network uses an adaptive hypergraph convolution as a spatial feature extractor, and captures the spatial relationships between dynamic nodes through the adaptive hypergraph, and performs convolution operations on the spatial relationships and traffic flow to complete the extraction of spatial features by the spatial convolutional network; the temporal convolutional network uses a recurrent network to obtain temporal features from the spatial data obtained by the spatial convolutional network through the recurrent network; Step (3) includes the following sub-steps: (3.1) The adaptive hypergraph is calculated by the adaptive hypergraph generator. According to the distance calculation formula, the Euclidean distance between the input data is calculated to obtain the distance matrix. The distance matrix is ​​then multiplied with the previously calculated adaptive hypergraph to obtain a new adaptive hypergraph. (3.2) Spatial convolutional networks obtain spatial features through hypergraph convolutional networks; (3.3) The temporal convolutional network uses a gated recurrent unit as the temporal convolution module. The obtained spatial feature data is input into the gated recurrent unit, and the long and short time feature data are obtained by using the gating mechanism. After the weight parameters in the forward propagation formula of the temporal convolutional network are calculated, the final temporal feature output is obtained. (4) Construct a data input layer using the spatiotemporal convolution module obtained in step (3); (5) Construct a spatiotemporal convolutional layer by stacking multiple spatiotemporal convolutional modules obtained in step (3); (6) Input the features from the data input layer and the spatiotemporal convolutional layer into the output prediction layer to obtain the traffic flow prediction result, and obtain the prediction error according to the loss function. Iterate the model parameters through the optimization algorithm until the model converges to obtain the optimal model. (7) The obtained historical data is processed and then input into the optimal model obtained in step (6) to predict traffic flow.

2. The traffic flow prediction method based on adaptive hypergraph convolutional neural network according to claim 1, characterized in that, Step (1) includes the following sub-steps: (1.1) Install N sensors in the designated traffic network; (1.2) Sampling is performed on the historical traffic flow data of N sensors before time T in the Y period according to the sampling time interval Δt, so as to obtain the traffic flow data; (1.3) Sample other historical traffic data from N sensors before time T in the Y period according to the sampling time interval Δt, and obtain the corresponding prediction data T+t1, T+t2, ... T+t based on the sampled historical traffic data. k Traffic data over K time periods is used to obtain traffic situation data; (1.4) Construct multidimensional structured feature data based on the traffic flow data obtained in step (1.2) and the traffic situation data obtained in step (1.3).

3. The traffic flow prediction method based on adaptive hypergraph convolutional neural network according to claim 1, characterized in that, The data input layer is a feature input encoder, used to realize the encoding conversion from spatiotemporal data to spatiotemporal feature data; the output prediction layer is used to convert the spatiotemporal feature data into the final prediction result.

4. The traffic flow prediction method based on adaptive hypergraph convolutional neural network according to claim 1, characterized in that, The spatial convolutional network obtains spatial features through a hypergraph convolutional network by: calculating the degree matrix and edge matrix of the adaptive hypergraph using the spatial convolution formula; and then convolving the input signal using the adaptive hypergraph convolution method to obtain the spatial features.

5. The traffic flow prediction method based on adaptive hypergraph convolutional neural network according to claim 1, characterized in that, The forward propagation formula for the temporal convolutional network is: H t =ReLU{say(x)·H t-1 } z t =σ(W z [X t ,h t-1 ]+b z ) r t =σ(W r [X t ,h t-1 ]+b r ) Where x is the input data, X t Let H represent the input data at time t, dis(·) be the distance equation, ReLU be the activation function, and H be the input data at time t. t-1 For the previous adaptive hypergraph, H t For the updated adaptive hypergraph, z t r t and h t These are the GRU update gate, reset gate, and hidden gate, respectively. z b r W represents the offset of the updated door and the offset of the reset door, respectively. z W r σ(·) represents the weights of the update gate and the reset gate, respectively, and σ(·) is the sigmoid activation function.

6. The traffic flow prediction method based on adaptive hypergraph convolutional neural network according to claim 1, characterized in that, The expression for the loss function is: Among them, y n For the true value, The value is the predicted value, and n represents the number of nodes.

7. The traffic flow prediction method based on adaptive hypergraph convolutional neural network according to claim 1, characterized in that, In step (6), iterating the model parameters through optimization algorithms specifically involves using the Adam optimizer as the model parameter iterative optimizer to iteratively optimize the model parameters.

8. A traffic flow prediction device based on an adaptive hypergraph convolutional neural network, characterized in that, It includes one or more processors for implementing the traffic flow prediction method based on an adaptive hypergraph convolutional neural network as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, is used to implement the traffic flow prediction method based on an adaptive hypergraph convolutional neural network as described in any one of claims 1-7.

Images