Logistics demand prediction method and device based on space-time diagram convolutional network

Abstract

The invention relates to a logistics demand prediction method and device based on a space-time diagram convolutional network, electronic equipment and a storage medium, and relates to the field of logistics demand prediction, and the method comprises the following steps: obtaining logistics information of a logistics network in a target area; based on the logistics information, clustering the logistics network nodes to obtain N types of logistics network nodes; dividing the logistics information according to a time sequence period and constructing hyperedges of similar logistics network nodes between adjacent periods to obtain a space-time hypergraph; inputting the space-time hypergraph into a trained logistics demand prediction model, and obtaining the logistics quantity of each logistics network node; therefore, the logistics quantity of each logistics network node in the target area can be predicted more accurately.

Description

Technical Field

[0001] The present application relates to the field of neural network technology, and specifically to a logistics demand prediction method and device based on a spatiotemporal graph convolutional network. Background Art

[0002] With the rapid development of e-commerce, the logistics industry has become an indispensable component of the modern economy. The complexity and dynamic nature of logistics networks require precise demand forecasting to optimize resource allocation and efficient scheduling. The accuracy of logistics demand forecasting directly impacts logistics companies' operational efficiency, cost control, and customer satisfaction.

[0003] Traditional logistics demand forecasting methods include qualitative forecasting, time series models, and regression analysis. Qualitative forecasting methods, such as the Delphi method, use expert consultation and opinions to predict logistics demand. This method relies on expert experience and judgment and is suitable for situations where data is insufficient or difficult to quantify. Time series models, such as exponential smoothing, assign different weights to data at different time points based on a moving average to adapt to data trends. However, the selection of weights often relies on experience. Regression analysis analyzes the relationship between various factors affecting logistics demand and logistics demand, establishing a regression model for forecasting. However, the accuracy of this method is affected by data quality and variable selection. In summary, traditional logistics demand forecasting methods are highly subjective and easily influenced by subjective factors, resulting in unstable forecast results. They are also highly data-dependent and require a large amount of historical data as support. However, in practice, complete data is often difficult to obtain.

[0004] Furthermore, with the development of machine learning technology, many studies have attempted to use machine learning methods to solve spatiotemporal data prediction problems. Existing work includes the Support Vector Machine (SVM) model. Furthermore, drawing on the ideas of the SARIMA (Seasonal Autoregressive Integrated Moving Average) model, existing work has also considered the cyclical characteristics of the data and proposed two novel SVR models, demonstrating the importance of data characteristics to prediction accuracy. Existing work also includes the use of a multi-layer structure optimization strategy based on an advanced genetic algorithm and combining it with an ANN (Artificial Neural Network) to improve prediction accuracy. This work also uses an improved KNN (K-Nearest Neighbor) model to predict short-term traffic data. Although machine learning-based methods have improved the accuracy of spatiotemporal data prediction tasks, they still cannot capture the highly complex spatiotemporal correlations in real spatiotemporal data.

[0005] While these methods can capture the spatiotemporal correlations in spatiotemporal data and improve the accuracy of spatiotemporal data predictions, they still have two shortcomings. First, existing logistics demand forecasting models extract temporal and spatial features from traffic data separately and combine them to represent spatiotemporal characteristics. The spatiotemporal characteristics captured by this method are indirect and partial, failing to accurately reflect the true spatiotemporal correlations in spatiotemporal data. Second, existing graph-based neural network models for logistics forecasting, where each edge connects two nodes, are difficult to apply to the individual nodes of a logistics network. Because connections between nodes in a logistics network carry more information than point-to-point connections, graph-based neural network models only use a regular graph to represent the logistics interaction graph, ignoring the higher-level information implicit in each node, thus affecting prediction accuracy. Summary of the Invention

[0006] The present application provides a logistics demand prediction method, device, electronic device and storage medium based on a spatiotemporal graph convolutional network to accurately obtain the logistics interaction volume of each logistics network node.

[0007] To achieve the above objectives, this application provides the following solutions.

[0008] In a first aspect, the present application provides a logistics demand forecasting method based on a spatiotemporal graph convolutional network, the method comprising the following steps: Obtain the logistics interaction volume between various types of logistics network nodes in the logistics network within the target area; Acquire the first logistics interaction in the first time series period according to the logistics interaction information amount and the second logistics interaction amount in the second time series period; Constructing a first spatiotemporal hypergraph based on a first logistics interaction amount within the first time series period, and constructing a second spatiotemporal hypergraph based on a second logistics interaction amount within the first time series period; The first spatiotemporal hypergraph and the second spatiotemporal hypergraph are input into a trained logistics demand forecasting model to obtain the logistics interaction volume of each type of logistics network node.

[0009] Furthermore, the method of obtaining the logistics interaction volume between various types of logistics network nodes in the logistics network within the target area previously includes: obtaining the historical logistics interaction volume of each logistics network node within the target area; clustering the logistics network nodes according to the changing characteristics of the historical logistics interaction volume to obtain K categories of logistics network nodes.

[0010] Furthermore, the logistics demand forecasting model includes two spatiotemporal modules and a fully connected layer; wherein each of the spatiotemporal modules is composed of a layer of time domain convolution module and a spatiotemporal graph convolution module in sequence.

[0011] Furthermore, the time domain convolution module includes a one-dimensional convolutional neural network layer CNN and a gated linear unit GLU; wherein, CNN is used to capture the time information in the spatiotemporal data, and GLU is used to select the retained time information; the spatiotemporal graph convolution module includes a graph convolutional neural network GCN.

[0012] Furthermore, based on the historical logistics information, the logistics network nodes are clustered to obtain N categories of logistics network nodes, including: parsing the number of fixed-interval time points of logistics volume, the number of logistics network nodes and the entry and exit directions of logistics network nodes from the logistics information; using a clustering algorithm to classify the flow characteristics of the logistics network nodes; recursively processing the logistics network nodes to obtain the number of categories of the logistics network nodes.

[0013] Furthermore, the clustering algorithm is used to classify the traffic characteristics of the logistics network nodes, including the following steps: starting from any unvisited target logistics network node, calculating the characteristic similarity between the target logistics network node and other logistics network nodes; All logistics network nodes are processed recursively to obtain the number of categories of the logistics network nodes.

[0014] Furthermore, the calculation of the feature similarity between the target logistics network node and other logistics network nodes is specifically performed using Euclidean distance: Among them, x and y represent two different logistics network nodes in n-dimensional space; n represents the number of time points of logistics information, x i 、y i Represents the flow rate of node i in the logistics network at time.

[0015] In a second aspect, the present application provides a logistics demand forecasting device based on a spatiotemporal graph convolutional network, the device comprising: Logistics information acquisition module, which is used to obtain logistics information of the logistics network in the target area; A classification module, which is used to cluster the logistics network nodes based on the logistics information to obtain N categories of logistics network nodes; A spatiotemporal hypergraph acquisition module, which is used to divide the logistics information according to time series periods and construct hyperedges of similar logistics network nodes between adjacent periods to obtain a spatiotemporal hypergraph; The logistics volume acquisition module is used to input the spatiotemporal hypergraph into the trained logistics demand forecasting model to obtain the logistics volume of each logistics network node.

[0016] This application creates a logistics demand prediction model, and predicts the flow of the current node based on the historical logistics interaction volume of various types of logistics network nodes. It can obtain the different impacts of changes in logistics interaction volume in different time and space on the current logistics network node, including the impact of different logistics interaction volumes in time on the logistics volume of the current logistics network node, the impact of other sites in space on the flow of the current site, and the impact of other sites at different time points on the current logistics network node, so as to more accurately predict the logistics interaction volume of the current area. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0018] Figure 1 This is a flowchart of the steps of the logistics demand forecasting method based on the spatiotemporal graph convolutional network provided in an embodiment of the present application; Figure 2 This is a schematic diagram of the logistics demand forecasting model provided in the embodiments of this application; Figure 3 This is a schematic diagram of the time period division provided in the embodiments of the present application; Figure 4 This is a step flow chart of the classification module provided in the embodiments of the present application. DETAILED DESCRIPTION

[0019] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0020] The embodiments of the present application are further described in detail below with reference to the accompanying drawings.

[0021] The embodiments of the present application provide a logistics demand prediction method, device, electronic device and storage medium based on a spatiotemporal graph convolutional network, thereby achieving accurate prediction of the logistics volume of each logistics network node in a target area.

[0022] To achieve the above technical effects, the overall idea of ​​this application is as follows: like Figure 1 As shown, a logistics demand forecasting method based on spatiotemporal graph convolutional network includes the following steps: S1. Obtain the logistics interaction volume between various types of logistics network nodes in the logistics network within the target area; Among them, the logistics information includes the logistics demand, timestamp, geographical location, processing capacity and logistics flow of each logistics network node.

[0023] S2. Obtaining a first logistics interaction volume within a first time series period and a second logistics interaction volume within a second time series period according to the logistics interaction information; It should be noted that if Figure 3 As shown, the time series cycle includes daily cycle and weekly cycle time series; T daily cycle is a time series of the same time period as the forecast period in the past few days. d represents the logistics interaction volume divided by the daily cycle, t represents the prediction time point, d represents the dth day of the daily cycle time series, and D represents the number of days included in the logistics interaction volume; the weekly cycle is a time series of the same time period as the prediction period in the past few weeks, T w represents the logistics interaction volume divided by weekly cycles, w represents the wth week of the weekly time series, t represents the prediction time point, and W represents the number of weeks included in the logistics interaction volume. After division, the logistics interaction volume T of the two division methods is obtained d and T w .

[0024] Specifically, the number of fixed-interval time points of logistics volume, the number of logistics network nodes, and the entry and exit directions of logistics network nodes are parsed from the logistics information in step S1; and the flow characteristics of the logistics network nodes are classified using a clustering algorithm.

[0025] S3. Constructing a first spatiotemporal hypergraph based on the first logistics interaction volume within the first time series period, and constructing a second spatiotemporal hypergraph based on the second logistics interaction volume within the first time series period; Specifically, the statistical logistics information is divided according to the time series period; after the logistics information is divided according to the period, according to the classification of the logistics network nodes in step S2, a hyperedge is connected between the logistics network nodes belonging to the same category at adjacent period time points to construct a hypergraph across time and space.

[0026] It should be noted that the logistics network nodes of the same category in adjacent cycle time points are connected to form a hyperedge, and a hypergraph and adjacency matrix across time and space are constructed; After the traffic data is divided into cycles, according to the classification of subway stations in step S12, a hyperedge is connected between subway stations belonging to the same category in adjacent cycle time points to construct a cross-time and space hypergraph, which is expressed as G = (V, E). The node set V includes not only the nodes at the current moment, but also the nodes at adjacent moments; E represents the set of hyperedges. In order to facilitate subsequent calculations, the hypergraph is represented in the form of an adjacency matrix. The size of the adjacency matrix of the spatiotemporal hypergraph is (2N, K), where N represents the number of nodes, that is, the number of traffic stations. Since the spatiotemporal hypergraph connects the two previous and next time points, the number of nodes is twice the number of nodes at one moment, that is, 2N. And K represents the number of hyperedges, that is, the number of categories classified in step S12. The element hij in the adjacency matrix of the hypergraph is expressed as: The rows and columns of the hypergraph adjacency matrix represent nodes and hyperedges, respectively (i, j represent the order of the rows and columns, i.e. ). If a node belongs to this hyperedge, the position is marked as 1, otherwise it is marked as 0. For example: V = {a, b, c, d, a', b', c', d'}, where {a, b, c, d} are the nodes at the previous cycle time, and {a', b', c', d'} are the corresponding nodes at the next cycle time; E = {e0, e1, e2}, where e0 connects four vertices e0 = {a, b, a', b'}, e1 connects six vertices e1 = {b, c, d, b', c', d'}, and e2 connects four vertices e2 = {c, d, c', d'}.

[0027] S4. Input the spatiotemporal hypergraph into the trained logistics demand forecasting model to obtain the logistics volume of each logistics network node.

[0028] Among them, the logistics demand forecasting model is the STHGCN model; The spatiotemporal hypergraph convolutional network model consists of two identical spatiotemporal modules and a fully connected layer. Each spatiotemporal module is composed of a time domain convolution module and a hypergraph convolution module in sequence. The traffic flow data obtained by processing steps S1 and S2 and the first spatiotemporal hypergraph and the second spatiotemporal hypergraph are input into the STHGCN model respectively, and the results of the two periodic spatiotemporal hypergraphs are input into the fully connected layer to obtain the final prediction results. Through the above steps, the prediction results of the logistics interaction volume of each type of logistics network node can be obtained, which can be expressed as X = {X0, X1, ... X N}, where N represents N transportation stations. N Represents the future traffic flow forecast for station N, where the dimension of XN is the number of predicted time points.

[0029] It should be noted that the logistics demand forecasting model includes two spatiotemporal modules and one fully connected layer; among them, each spatiotemporal module is composed of a layer of time domain convolution module and a spatiotemporal graph convolution module in sequence.

[0030] This application creates a logistics demand prediction model and predicts the flow of the current site based on the historical flow data of other sites. It can obtain the different impacts of changes in logistics volume at different times and spaces on the current site, including the impact of different logistics volumes in time on the logistics volume of the current logistics network node, the impact of other sites in space on the flow of the current site, and the impact of other sites at different time points on the current logistics network node, so as to more accurately predict the logistics volume of each logistics network node in the target area.

[0031] In one embodiment of the application, Figure 2 As shown, the logistics demand forecasting model includes two spatiotemporal modules and a fully connected layer; wherein each of the spatiotemporal modules is composed of a layer of time domain convolution module and a spatiotemporal graph convolution module in sequence.

[0032] Specifically, the spatiotemporal hypergraph convolutional network model consists of two identical spatiotemporal modules and a fully connected layer. Each spatiotemporal module is composed of a time domain convolution module and a hypergraph convolution module. The spatiotemporal hypergraph is input into the trained logistics demand forecasting model to obtain the final forecast results.

[0033] Among them, the time domain convolution module includes a one-dimensional convolutional neural network layer CNN and a gated linear unit GLU; among them, CNN is used to capture the time information in the spatiotemporal data, and GLU is used to select the retained time information; the spatiotemporal graph convolution module includes a graph convolutional neural network GCN.

[0034] In an embodiment of the present application, a hypergraph convolution module is constructed to perform convolution on the constructed spatiotemporal hypergraph, and the spatiotemporal features contained in the spatiotemporal hypergraph are learned. Specifically, the spatiotemporal hypergraph of the present invention is first converted into an ordinary graph and then convolution is performed using the graph convolution method of a single layer of a graph convolutional neural network GCN: in the constructed hypergraph convolution module, for each hyperedge, since the hyperedge corresponds to many nodes, the present invention only selects the two vertices with the farthest Euclidean distance on the same edge, and uses the edge between these two nodes to represent the hyperedge. Then connect the remaining intermediate nodes to the selected nodes. After the above steps, the hypergraph is converted into an ordinary graph, and then convolution is performed using the graph convolution method of a single layer of a graph convolutional neural network GCN.

[0035] In one embodiment of the application, Figure 4 As shown, clustering the logistics network nodes based on the historical logistics information to obtain N categories of logistics network nodes includes: S11, parsing the logistics information to obtain the number of fixed-interval time points of logistics volume, the number of logistics network nodes, and the entry and exit directions of the logistics network nodes; S12, use clustering algorithm to classify the flow characteristics of logistics network nodes.

[0036] Specifically, starting from any unvisited target logistics network node, the feature similarity between the target logistics network node and other logistics network nodes is calculated; all logistics network nodes are recursively processed to obtain the number of categories of logistics network nodes.

[0037] Among them, the feature similarity between the target logistics network node and other logistics network nodes is calculated, specifically using the Euclidean distance: Among them, x and y represent two different logistics network nodes in n-dimensional space; n represents the number of time points of logistics information, x i 、y i represents the flow rate at time instant i in the logistics network. After calculating the Euclidean distance, find all nodes within a Euclidean distance R of the node. The value of R is set according to different data, for example: R = 1, 2, 3, etc. If the Euclidean distance between two sites is ≤ R, the two sites are considered to be of the same type, and the counted sites are set as visited. Then, recursively process all unvisited sites in the data using the same method.

[0038] It can be understood that the clustering algorithm can be a DBSCAN clustering algorithm, which classifies logistics network nodes according to the characteristics of the logistics volume of each logistics network node changing over time, for example, logistics network nodes such as logistics terminals and regional Cainiao stations.

[0039] In one embodiment of the application, the spatiotemporal hypergraph convolutional network model consists of two identical spatiotemporal modules and a fully connected layer. A convolutional layer and an activation layer may be added after the fully connected layer, which can be expressed as: in, f represents the tanh activation function, w res Represents the learnable parameters of the convolutional layer.

[0040] It's worth noting that different regions are affected by local spatiotemporal, daily, and weekly periodicities, but the extent of these influences may vary. For some regions, long-term dependencies are often particularly significant. Therefore, when fusing the two components, the differences in their contributions should be fully considered and each should be multiplied by a different weight. This weighting parameter can be automatically learned from historical data.

[0041] Based on the same inventive concept as the real-time example of the method, an embodiment of the present application provides a logistics demand forecasting device based on a spatiotemporal graph convolutional network, which includes: The logistics interaction volume acquisition module is used to obtain the logistics interaction volume between various types of logistics network nodes in the logistics network in the target area; a period arrangement module, configured to obtain a first logistics interaction volume within a first time series period and a second logistics interaction volume within a second time series period according to the logistics interaction information; a spatiotemporal hypergraph acquisition module, configured to construct a first spatiotemporal hypergraph based on a first logistics interaction amount within the first time series period, and to construct a second spatiotemporal hypergraph based on a second logistics interaction amount within the first time series period; The prediction module is used to input the first spatiotemporal hypergraph and the second spatiotemporal hypergraph into the trained logistics demand prediction model to obtain the logistics interaction volume of each type of logistics network node.

[0042] This application creates a logistics demand prediction model and predicts the flow of the current site based on the historical flow data of other sites. It can obtain the different impacts of changes in logistics volume at different times and spaces on the current site, including the impact of different logistics volumes in time on the logistics volume of the current logistics network node, the impact of other sites in space on the flow of the current site, and the impact of other sites at different time points on the current logistics network node, so as to more accurately predict the logistics volume of each logistics network node in the target area.

[0043] In one embodiment of the application, the logistics interaction volume acquisition module includes: an interaction volume acquisition sub-module, which is used to obtain the logistics interaction volume between various types of logistics network nodes in the logistics network within the target area; a clustering sub-module, which is used to cluster the logistics network nodes according to the change characteristics of the historical logistics interaction volume to obtain K categories of logistics network nodes.

[0044] In one embodiment of the application, the logistics demand forecasting model includes two spatiotemporal modules and a fully connected layer; each spatiotemporal module is composed of a temporal convolution module and a spatiotemporal graph convolution module. The temporal convolution module includes a one-dimensional convolutional neural network layer (CNN) and a gated linear unit (GLU); the CNN is used to capture temporal information in spatiotemporal data, and the GLU is used to select the retained temporal information; the spatiotemporal graph convolution module includes a graph convolutional neural network (GCN).

[0045] In one embodiment of the application, the clustering submodule includes a parsing unit, which is used to parse the number of fixed-interval time points of logistics volume, the number of logistics network nodes and the entry and exit directions of logistics network nodes from the logistics information; and a classification unit, which is used to use a clustering algorithm to classify the flow characteristics of the logistics network nodes.

[0046] In one embodiment of the application, the classification unit includes a calculation subunit, which is used to start from any unvisited target logistics network node and calculate the feature similarity between the target logistics network node and other logistics network nodes; a recursive subunit, which is used to recursively process all logistics network nodes to obtain the number of categories of the logistics network nodes.

[0047] In one embodiment of the application, the feature similarity between the target logistics network node and other logistics network nodes is calculated by using the Euclidean distance: Among them, x and y represent two different logistics network nodes in n-dimensional space; n represents the number of time points of the logistics information, x i 、y i Represents the flow rate of node i in the logistics network at time.

[0048] It should be noted that the logistics demand forecasting device based on the spatiotemporal graph convolutional network provided in the embodiment of the present application, and its corresponding technical problems, technical means and technical effects are similar to the principles of the logistics demand forecasting method based on the spatiotemporal graph convolutional network at the principle level.

[0049] In a second aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the water pump control method mentioned in the first aspect is implemented.

[0050] In a third aspect, an embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program running on the processor, and when the processor executes the computer program, the water pump control method mentioned in the first aspect is implemented.

[0051] It should be noted that, in this application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises", "includes" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0052] The above are merely specific embodiments of the present application to enable those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but rather is intended to conform to the widest scope consistent with the principles and novel features of the present application.

Claims

1. A logistics demand forecasting method based on spatiotemporal graph convolutional network, characterized by: The method comprises the following steps: Obtain the logistics interaction volume between various types of logistics network nodes in the logistics network within the target area; Acquire the first logistics interaction in the first time series period according to the logistics interaction information amount and the second logistics interaction amount in the second time series period; Constructing a first spatiotemporal hypergraph based on a first logistics interaction amount within the first time series period, and constructing a second spatiotemporal hypergraph based on a second logistics interaction amount within the first time series period; The first spatiotemporal hypergraph and the second spatiotemporal hypergraph are input into a trained logistics demand forecasting model to obtain the logistics interaction volume of each type of logistics network node.

2. The logistics demand forecasting method based on spatiotemporal graph convolutional networks according to claim 1, wherein obtaining the logistics interaction volume between various types of logistics network nodes in the logistics network within the target area comprises: Obtaining historical logistics interaction volume of each logistics network node in the target area; According to the change characteristics of the historical logistics interaction volume, the logistics network nodes are clustered to obtain K categories of logistics network nodes.

3. The logistics demand forecasting method based on spatiotemporal graph convolutional network according to claim 1 is characterized in that: The logistics demand forecasting model includes two spatiotemporal modules and a fully connected layer; wherein each of the spatiotemporal modules is composed of a layer of time domain convolution module and a spatiotemporal graph convolution module in sequence.

4. The logistics demand forecasting method based on spatiotemporal graph convolutional network according to claim 3 is characterized in that: The time domain convolution module includes a one-dimensional convolutional neural network layer CNN and a gated linear unit GLU; wherein, CNN is used to capture the time information in the spatiotemporal data, and GLU is used to select the retained time information; the spatiotemporal graph convolution module includes a graph convolutional neural network GCN.

5. The logistics demand forecasting method based on spatiotemporal graph convolutional network according to claim 1, characterized in that: The clustering of the logistics network nodes based on the historical logistics information to obtain N categories of logistics network nodes includes: Parsing the logistics information to determine the number of fixed-interval time points of logistics volume, the number of logistics network nodes, and the entry and exit directions of the logistics network nodes; A clustering algorithm is used to classify the traffic characteristics of the logistics network nodes.

6. The logistics demand forecasting method based on spatiotemporal graph convolutional network according to claim 5 is characterized in that: The method of using a clustering algorithm to classify the flow characteristics of the logistics network nodes includes the following steps: Starting from any unvisited target logistics network node, calculate the feature similarity between the target logistics network node and other logistics network nodes; All logistics network nodes are processed recursively to obtain the number of categories of the logistics network nodes.

7. The logistics demand forecasting method based on spatiotemporal graph convolutional network according to claim 1, characterized in that: The feature similarity between the target logistics network node and other logistics network nodes is calculated by using the Euclidean distance: Among them, x and y represent two different logistics network nodes in n-dimensional space; n represents the number of time points of the logistics information, x i 、y i Represents the flow rate of the logistics network node at time i.

8. A device for logistics demand forecasting based on spatiotemporal graph convolutional network, characterized in that: The device comprises: The logistics interaction volume acquisition module is used to obtain the logistics interaction volume between various types of logistics network nodes in the logistics network in the target area; a period arrangement module, configured to obtain a first logistics interaction amount within a first time series period and a second logistics interaction amount within a second time series period according to the logistics interaction information; a spatiotemporal hypergraph acquisition module, configured to construct a first spatiotemporal hypergraph based on a first logistics interaction amount within the first time series period, and to construct a second spatiotemporal hypergraph based on a second logistics interaction amount within the first time series period; The prediction module is used to input the first spatiotemporal hypergraph and the second spatiotemporal hypergraph into the trained logistics demand prediction model to obtain the logistics interaction volume of each type of logistics network node.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images