High-speed railway train delay prediction method and device based on graph convolutional neural network, and storage medium

Abstract

The invention relates to the technical field of intelligent transportation, in particular to a high-speed railway train delay prediction method and device based on a graph convolutional neural network and a storage medium. The method comprises the steps that first operation data of a train at front stations are obtained, the front stations are n stations located in front of a to-be-predicted target station on a running route of the train, n is a positive integer, and the first operation data are used for indicating the delay condition and the operation plan of the train at the front stations; according to the first operation data, a trained delay prediction model is called for delay prediction, a delay prediction result is output and obtained, the delay prediction result is used for indicating the predicted delay condition of the train at the target station, and the delay prediction model is a model based on a graph convolutional neural network. According to the embodiment of the invention, through the delay prediction model based on the graph convolutional neural network, the delay condition of the train at the target station is predicted by using the delay condition and the operation plan of the train at the previous station, and the train delay prediction effect is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of intelligent transportation technology, and in particular to a method, device and storage medium for predicting high-speed railway train delays based on graph convolutional neural networks (GCN). Background Art

[0002] With the rapid development of high-speed railway networks, train delays have received increasing attention. Delays not only affect passengers' travel experience, but may also have a negative impact on railway transportation efficiency and safety. Therefore, accurate prediction of train delays is crucial to improving the quality of railway transportation services.

[0003] Traditional prediction methods may rely on simple statistical analysis or models based on specific assumptions. These methods may have limitations when dealing with complex train operation data and cannot accurately reflect the dynamic changes and propagation characteristics of train delays.

[0004] Regarding predicting train delays, the relevant technology has not yet provided a reasonable and effective delay prediction method. Summary of the invention

[0005] In view of this, the present invention proposes a method, device and storage medium for predicting high-speed railway train delays based on graph convolutional neural network.

[0006] According to one aspect of the present disclosure, a method for predicting high-speed railway train delays based on a graph convolutional neural network is provided, the method comprising:

[0007] Acquire first operation data of the train at a preceding station, where the preceding station is n stations located before the target station to be predicted on the travel route of the train, where n is a positive integer, and the first operation data is used to indicate the delay situation and operation plan of the train at the preceding station;

[0008] According to the first operating data, the trained delay prediction model is called to perform delay prediction, and a delay prediction result is output. The delay prediction result is used to indicate the predicted delay of the train at the target station. The delay prediction model is a model based on a graph convolutional neural network.

[0009] In a possible implementation, the first operation data includes first delay data of the train at the preceding station and first operation plan data, the first delay data including the difference between the arrival time of the train at the preceding station and the planned arrival time, and the first operation plan data including the planned stop time of the train at the preceding station and the planned travel time between stations at the preceding station.

[0010] In another possible implementation, before calling a trained delay prediction model to perform a delay prediction based on the first operation data and outputting a delay prediction result, the method further includes:

[0011] Acquire second operation data of an adjacent train at the preceding station, the second operation data being used to indicate a delay situation and an operation plan of the adjacent train at the preceding station, the adjacent train being m trains adjacent to the departure time interval and / or departure sequence of the train on the travel route of the train, where m is a positive integer;

[0012] The step of calling the trained delay prediction model to perform delay prediction according to the first operation data and outputting a delay prediction result includes:

[0013] The trained delay prediction model is called to perform delay prediction according to the first operation data and the second operation data, and the delay prediction result is outputted.

[0014] In another possible implementation, the delay prediction model is used to extract a first delay feature and a second delay feature based on the first operating data and the second operating data, and determine the delay prediction result based on the first delay feature and the second delay feature, the first delay feature including features of the mutual influence between the train and the adjacent trains, and the second delay feature including features of the train itself in a time series.

[0015] In another possible implementation, the delay prediction model includes a graph convolution module, a time series processing module and a fully connected layer, and calling the trained delay prediction model to perform delay prediction according to the first operation data and the second operation data, and outputting the delay prediction result includes:

[0016] According to the first operation data and the second operation data, calling the graph convolution module output to obtain the first delay feature;

[0017] According to the first operation data, calling the time series processing module output to obtain the second delay feature;

[0018] The first delay feature and the second delay feature are concatenated and input into the fully connected layer, and the delay prediction result is outputted.

[0019] In another possible implementation, the first operation data includes first delay data and first operation plan data of the train at the preceding station, the second operation data includes second delay data and second operation plan data of the adjacent train at the preceding station, and the calling of the graph convolution module output to obtain the first delay feature according to the first operation data and the second operation data includes:

[0020] Determining an adjacency matrix according to the first operation plan data and the second operation plan data, wherein the adjacency matrix is ​​used to indicate the degree of mutual influence between the train and the adjacent trains;

[0021] According to the first delay data, the second delay data and the adjacency matrix, the graph convolution module is called to output the first delay feature.

[0022] In another possible implementation, the first operation plan data includes the planned departure time of the train at the previous station, the second operation plan data includes the planned departure time of the adjacent train at the previous station, and determining the adjacency matrix according to the first operation plan data and the second operation plan data includes:

[0023] Determining the adjacency matrix according to the scheduled departure times of the train and a plurality of adjacent trains at the previous station;

[0024] The adjacency matrix includes the influence weight between the train and each of the adjacent trains, and the influence weight is negatively correlated with the planned departure time interval between the corresponding train and the adjacent train.

[0025] In another possible implementation, the time series processing module is a time convolution module based on a spatial-temporal graph convolutional network (STGCN), and the time series processing module includes a plurality of one-dimensional convolution layers and an activation layer. The second delay feature is obtained by calling the output of the time series processing module according to the first running data, including:

[0026] Inputting the first operation data into the multiple one-dimensional convolutional layers of the time series processing module, and outputting intermediate data;

[0027] The intermediate data is input into the activation layer of the time series processing module, and the second delay feature is obtained as output.

[0028] According to another aspect of the present disclosure, a method and device for predicting high-speed railway train delays based on a graph convolutional neural network is provided, the device comprising:

[0029] an acquisition unit, configured to acquire first operation data of a train at a preceding station, wherein the preceding station is n stations located before a target station to be predicted on a travel route of the train, wherein n is a positive integer, and the first operation data is used to indicate a delay situation and an operation plan of the train at the preceding station;

[0030] A prediction unit is used to call a trained delay prediction model to perform delay prediction according to the first operation data, and output a delay prediction result, wherein the delay prediction result is used to indicate the predicted delay of the train at the target station, and the delay prediction model is a model based on a graph convolutional neural network.

[0031] In a possible implementation, the first operation data includes first delay data of the train at the preceding station and first operation plan data, the first delay data including the difference between the arrival time of the train at the preceding station and the planned arrival time, and the first operation plan data including the planned stop time of the train at the preceding station and the planned travel time between stations at the preceding station.

[0032] In another possible implementation, the acquiring unit is further configured to:

[0033] Acquire second operation data of an adjacent train at the preceding station, the second operation data being used to indicate a delay situation and an operation plan of the adjacent train at the preceding station, the adjacent train being m trains adjacent to the departure time interval and / or departure sequence of the train on the travel route of the train, where m is a positive integer;

[0034] The prediction unit is further used for:

[0035] The trained delay prediction model is called to perform delay prediction according to the first operation data and the second operation data, and the delay prediction result is outputted.

[0036] In another possible implementation, the delay prediction model is used to extract a first delay feature and a second delay feature based on the first operating data and the second operating data, and determine the delay prediction result based on the first delay feature and the second delay feature, the first delay feature including features of the mutual influence between the train and the adjacent trains, and the second delay feature including features of the train itself in a time series.

[0037] In another possible implementation, the delay prediction model includes a graph convolution module, a time series processing module and a fully connected layer, and the prediction unit is further used to:

[0038] According to the first operation data and the second operation data, calling the graph convolution module output to obtain the first delay feature;

[0039] According to the first operation data, calling the time series processing module output to obtain the second delay feature;

[0040] The first delay feature and the second delay feature are concatenated and input into the fully connected layer, and the delay prediction result is outputted.

[0041] In another possible implementation, the first operation data includes first delay data and first operation plan data of the train at the preceding station, the second operation data includes second delay data and second operation plan data of the adjacent train at the preceding station, and the prediction unit is further configured to:

[0042] Determining an adjacency matrix according to the first operation plan data and the second operation plan data, wherein the adjacency matrix is ​​used to indicate the degree of mutual influence between the train and the adjacent trains;

[0043] According to the first delay data, the second delay data and the adjacency matrix, the graph convolution module is called to output the first delay feature.

[0044] In another possible implementation, the first operation plan data includes the planned departure time of the train at the previous station, the second operation plan data includes the planned departure time of the adjacent train at the previous station, and the prediction unit is further used to:

[0045] Determining the adjacency matrix according to the scheduled departure times of the train and a plurality of adjacent trains at the previous station;

[0046] The adjacency matrix includes the influence weight between the train and each of the adjacent trains, and the influence weight is negatively correlated with the planned departure time interval between the corresponding train and the adjacent train.

[0047] In another possible implementation, the time series processing module is a time convolution module based on a spatiotemporal graph convolutional network, the time series processing module includes a plurality of one-dimensional convolutional layers and an activation layer, and the prediction unit is further used to:

[0048] Inputting the first operation data into the multiple one-dimensional convolutional layers of the time series processing module, and outputting intermediate data;

[0049] The intermediate data is input into the activation layer of the time series processing module, and the second delay feature is obtained as output.

[0050] According to another aspect of the present disclosure, a method and device for predicting high-speed railway train delays based on a graph convolutional neural network is provided, the device comprising:

[0051] processor;

[0052] a memory for storing processor-executable instructions;

[0053] Wherein, the processor is configured to implement the above method when executing the instructions stored in the memory.

[0054] According to another aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, and the computer program instructions implement the above method when executed by a processor.

[0055] According to another aspect of the present disclosure, a computer program product is provided, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of a computing device, the processor in the computing device executes the above method.

[0056] The disclosed embodiment provides a method for predicting high-speed railway train delays based on a graph convolutional neural network, which can obtain the first operation data of the train at the previous station, and the first operation data is used to indicate the delay situation and operation plan of the train at the previous station; according to the first operation data, the trained delay prediction model is called to perform delay prediction, and the delay prediction result is output to obtain the delay prediction result, and the delay prediction result is used to indicate the predicted delay situation of the train at the target station. The delay prediction model is a model based on a graph convolutional neural network; that is, a delay prediction model is constructed using a graph convolutional neural network, and the model can better capture the complex relationships and patterns in the train operation data, thereby improving the prediction accuracy of the train delay situation. The model not only takes into account the operation plan of the train, but also takes into account the delay situation of the train at the previous station, and realizes the prediction of the train delay situation at the target station by using the delay situation of the train at the previous station and the operation plan, which helps the model understand the continuity and propagation of train delays, and greatly improves the train delay prediction effect.

[0057] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0059] Figure 1A flowchart of a high-speed railway train delay prediction method based on a graph convolutional neural network provided by an exemplary embodiment of the present disclosure is shown.

[0060] Figure 2 A schematic diagram of train delay prediction logic provided by an exemplary embodiment of the present disclosure is shown.

[0061] Figure 3 A schematic diagram of the structure of a delay prediction model provided by an exemplary embodiment of the present disclosure is shown.

[0062] Figure 4 A flowchart of a high-speed railway train delay prediction method based on a graph convolutional neural network provided by another exemplary embodiment of the present disclosure is shown.

[0063] Figure 5 A schematic diagram of a calculation function of an adjacency matrix provided by an exemplary embodiment of the present disclosure is shown.

[0064] Figure 6 A schematic diagram of the structure of a graph convolution module provided by an exemplary embodiment of the present disclosure is shown.

[0065] Figure 7 A schematic diagram of the structure of a time series processing module provided by an exemplary embodiment of the present disclosure is shown.

[0066] Figure 8 A schematic diagram showing the variation of model loss with the number of training times during the model training process provided by an exemplary embodiment of the present disclosure is shown.

[0067] Fig. 9 A bar chart showing the distribution of differences between predicted values ​​and actual values ​​of train delays provided by an exemplary embodiment of the present disclosure is shown.

[0068] Fig.10 A statistical chart showing predicted values ​​and actual values ​​of train delays provided by an exemplary embodiment of the present disclosure is shown.

[0069] Fig.11 A schematic diagram of the structure of a high-speed railway train delay prediction device based on a graph convolutional neural network provided by an exemplary embodiment of the present disclosure is shown.

[0070] Fig.12 It is a block diagram of a device for executing a deep learning training method according to an exemplary embodiment. DETAILED DESCRIPTION

[0071] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0072] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0073] In the embodiments of the present disclosure, the terms "first", "second", etc. are used to distinguish different elements in terms of title, but do not represent the spatial arrangement or temporal order of these elements, etc., and these elements should not be limited by these terms. The term "and / or" includes any one and all combinations of one or more of the associated listed terms. The terms "comprising", "including", "having", etc. refer to the presence of the stated features, elements, components or components, but do not exclude the presence or addition of one or more other features, elements, components or components. The terms "pixel" and "voxel" are used interchangeably.

[0074] In the embodiments of the present disclosure, the singular forms "a", "the", etc. include plural forms and should be broadly understood as "a kind" or "a type" rather than being limited to the meaning of "one"; in addition, the term "said" should be understood to include both singular and plural forms, unless the context clearly indicates otherwise. In addition, the term "according to" should be understood as "at least partially based on...", and the term "based on" should be understood as "at least partially based on...", unless the context clearly indicates otherwise. The special word "exemplary" used here means "used as an example, embodiment, or illustrative". Any embodiment described here as "exemplary" is not necessarily interpreted as being superior or better than other embodiments.

[0075] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0076] With the rapid development of my country's railways, especially the rapid development of high-speed railways, energy-saving, comfortable and fast riding experience have made railways the first choice for more and more passengers. High-speed railway trains need to run through established lines and timetables to avoid conflicts between trains and achieve transportation purposes efficiently. However, due to weather reasons, human factors and possible failures, trains may not be able to run according to the established timetable during the travel section and at the stop station, which will cause delays. Delays in train operation may have an adverse impact on all aspects. From the perspective of passengers, train delays will cause inconvenience to passengers' travel. From the perspective of freight, the delay in transportation time may cause loss of cargo property. From the perspective of the operation of the entire railway system, the failure of trains to run according to the established timetable will affect the operation of adjacent trains, further leading to the spread, accumulation and expansion of delays, which is a loss of public transportation resources.

[0077] On the one hand, the study of delay prediction can provide passengers with timely itinerary information and help passengers reasonably adjust their travel plans according to the train delay status. On the other hand, accurate delay prediction can lay the foundation for train operation adjustment. Delay prediction models can usually be implemented using simulation models or data-driven methods. The evolution of train delays is affected by multi-source heterogeneous factors, such as extreme weather, peak hours, and train failures. When constructing a simulation model for delay prediction, it is difficult to use some quantitative variables to describe the impact of these influencing factors on the evolution of delays. In addition, when constructing a simulation model, it is usually necessary to make certain assumptions and simplifications on the delay evolution process based on human experience (prior knowledge), which may omit certain influencing factors and cause some factors to be not considered, thereby affecting the stability and accuracy of the simulation results. However, the prediction model based on data-driven treats the delay prediction problem as a black box, and uses historical data to explore the influence relationship between various factors and actual train delays, without artificially adding, quantifying and simulating the impact of various factors on delays. Therefore, some studies have found that the prediction model constructed based on the data-driven method can better capture the impact of some random factors that are difficult to predict in train operation on the evolution of train delays, thereby improving the accuracy of delay prediction. In addition, when predicting the corresponding delay based on the simulation model, it is necessary to fully simulate the delay of all trains. When the scale of the problem gradually expands, it will take too long to predict, thereby reducing the timeliness of the corresponding delay value. The data-driven method usually uses the offline training-online prediction method, which can achieve delay prediction in a shorter time and can better guarantee the real-time nature of the prediction. Furthermore, the train delay prediction model built based on the data-driven approach only needs to be trained based on different data for different railways (or trained using methods such as transfer learning), and has high portability. The simulation model only models and simulates specific situations. When the scenario to be simulated changes, large-scale adjustments may be required, resulting in more unnecessary consumption. Therefore, the delay prediction model built based on the data-driven approach is more applicable to different scenarios.

[0078] In view of the good applicability of data-driven methods to train delay prediction problems, the embodiment of the present disclosure uses a graph convolutional neural network to build a delay prediction model based on train operation data. Below, several exemplary embodiments are used to introduce the high-speed railway train delay prediction method based on a graph convolutional neural network provided by the embodiment of the present disclosure.

[0079] Please refer to Figure 1 , which shows a flowchart of a high-speed railway train delay prediction method based on a graph convolutional neural network provided by an exemplary embodiment of the present disclosure, and this embodiment is illustrated by using the method in a computing device. The method includes the following steps.

[0080] Step 101, obtaining the first operation data of the train at the preceding station, where the preceding station is n stations located before the target station to be predicted on the train's route, n is a positive integer, and the first operation data is used to indicate the delay situation and operation plan of the train at the preceding station.

[0081] A train is a train that runs on a high-speed railway, and the trains operate according to a set schedule and track route. In a high-speed railway system, the trains usually run at high speeds and stop at various stations along the line to pick up and drop off passengers.

[0082] A route is a track that trains follow on a specific track path on the railway network. A route contains multiple stations, and trains move between these stations.

[0083] The previous station is located on the train's route and before the target station to be predicted. The previous station includes the n consecutive stations before the target station to be predicted. n is a positive integer and can be preset according to the needs of the delay prediction model. For example, n can be set to 2, which means considering the data of the two stations before the target station.

[0084] The first operation data refers to the data of the train's operation status and operation plan at the previous station. These data are crucial for predicting the train's delay at subsequent stations. The first operation data includes the first delay data and the first operation plan data of the train at the previous station. The first delay data is used to indicate the train's delay at the previous station, and the first operation plan data is used to indicate the train's operation plan at the previous station.

[0085] The first delay data may include the delay time of each train at the first n stations. The delay time is also called the delay value. The delay value is the difference between the arrival time of the train and the planned arrival time. This difference can be used to quantify the degree of delay of the train. If the arrival time of the train is later than the planned arrival time, it means that the train is delayed. The delay time of each train at the first n stations may be the actual delay time, or the predicted delay time, or part of it may be the actual delay time and the other part may be the predicted delay time.

[0086] The first operation plan data may include the planned stop time of the train at each of the first n stations and the planned travel time between stations. The planned stop time of the train at the station is the length of time the train stops at the station. That is, the planned stop time of the train at the station is the difference between the planned departure time and the planned arrival time of the train at the station. The station section of a station is the section from the previous station to the current station, and the planned travel time of the station section of the train at the station is the planned travel time of the train from the previous station to the current station. That is, the planned travel time of the station section of the train at the station is the difference between the planned departure time of the train at the previous station and the planned arrival time of the current station. The first operation plan data may include the planned stop time of the train and the planned travel time between stations arranged alternately.

[0087] The first operation plan data may also include the planned arrival time of the train at each of the first n stations (ie, the time when the train arrives at the station) and / or the planned departure time (ie, the time when the train departs from the station).

[0088] The first operation plan data may be a default setting, that is, a plan pre-set according to the train schedule; or may be a dynamically updated data, that is, a plan adjusted according to real-time operation conditions. Dynamically updated data may improve the accuracy of prediction.

[0089] Schematically, taking n as 2 and the preceding stations being the two stations before the target station to be predicted on the train's route as an example, the first operating data of the train at the preceding station obtained includes: the delay data, planned stop time, and planned travel time between stations of the first station before the target station, and the delay data, planned stop time, and planned travel time between stations of the second station before the target station.

[0090] Step 102: Based on the first operation data, the trained delay prediction model is called to perform delay prediction, and a delay prediction result is output. The delay prediction result is used to indicate the predicted delay of the train at the target station. The delay prediction model is a model based on a graph convolutional neural network.

[0091] Optionally, based on the departure time of the train at the previous station, the first delay data at the previous station and the first operation plan data, the trained delay prediction model is called to perform delay prediction, and the delay prediction result is output. The previous station is the previous station of the target station. The departure time of the train at the previous station refers to the time of departure from the previous station, also known as the time when the train enters the station section of the target station. The departure time can be the actual departure time or the predicted departure time. For example, the delay time of the train at the previous station is predicted, and the departure time of the train at the previous station can be determined based on the predicted delay time and the planned departure time of the train at the previous station.

[0092] The delay prediction model is a model trained using historical operation data to predict the possible delays of trains at target stations. The delay prediction model predicts the delay results of trains by analyzing the input data.

[0093] The delay prediction result is the output of the delay prediction model, and the delay prediction result may include at least one of the delay time, delay probability and delay duration distribution. The delay time is also called the delay value, and the delay value is the difference between the actual arrival time of the train at the target station and the planned arrival time, usually in minutes. The delay probability is the possibility of the train being delayed at the target station, which can be expressed as a percentage. The delay duration distribution is the possible duration range of the train delay, for example, it may be delayed by 0-5 minutes, 5-10 minutes, more than 10 minutes, etc. For the convenience of explanation, the following only takes the delay prediction result including the delay value as an example for introduction. The delay prediction model can perform delay prediction when the train is not delayed or delayed. The delay value of not delayed or early arrival is defined as 0. The delay value is output when the train is delayed, which can assist the dispatching and command of the train group.

[0094] The delay prediction model is based on a graph convolutional neural network. A graph convolutional network is a deep learning network based on graph structured data. Its core idea is to propagate and aggregate information on a graph through convolution operations. Unlike traditional convolutional neural networks, graph convolutional networks take into account the connection relationship between nodes, so that the mutual influence between nodes can be characterized. The nodes in a graph convolutional neural network usually refer to the vertices in the graph. Each node has its own feature vector, and the nodes are connected to each other through edges to form the structure of the entire graph.

[0095] In theory, graph convolutional neural networks can also be applied to the case of considering only a single train. That is, the different states or attributes of the train can be regarded as nodes in the graph, and the mutual influence between these states or attributes can be regarded as edges. In this way, graph convolutional neural networks can capture the dynamic changes and mutual relationships of a single train under different states or attributes. For example, the nodes in the graph can include factors such as the current speed of the train, the scheduled arrival time of the train, the track usage in front of the train, and weather conditions. These factors are connected by edges because they all affect the arrival time of the train. Graph convolutional neural networks can capture the complex relationships between these nodes and predict train delays.

[0096] The evolution of train delays has obvious correlation and accumulation. The delay of a train is not only related to its own operation, but also has a mutual influence relationship with the operation of adjacent trains. For the case of considering trains and adjacent trains, graph convolutional neural networks can also be applied. Each train can be regarded as a node, in which case the node is also called a train node, and the train delay data is regarded as the attribute characteristics of the train node. The delay data generated by the train at the next station depends not only on the node characteristics of the train itself, but also on the influence of the adjacent nodes. The weight of the mutual influence between nodes is the edge weight between train nodes, which can be determined according to the departure interval of the train. In the graph convolutional neural network, the processing of node features in the graph not only considers the characteristics of the node itself, but also adds the characteristics of the adjacent nodes. This provides a good way to express the mutual influence relationship between train delays, so it is very effective to use the graph convolution method to characterize the mutual influence relationship between trains. In some embodiments, while obtaining the first operating data of the train at the preceding station, the computing device also obtains the second operating data of the adjacent train at the preceding station, the second operating data being used to indicate the delay situation and operation plan of the adjacent train at the preceding station, the adjacent trains being m trains adjacent to the departure time interval and / or departure sequence of the train on the train's route, where m is a positive integer; thereby, based on the first operating data and the second operating data, the trained delay prediction model can be called to perform delay prediction, and the delay prediction result can be output.

[0097] In addition, the continuous accumulation and propagation of delays between stations is also a manifestation of correlation. Train delays are actually the cumulative effect of local delays. The process from one station to the next station is also a cumulative adjustment process of delays. Therefore, the accumulation of delays at a certain station is largely affected by the delays at the previous station. Therefore, by extracting the time series features in the train operation data, the accumulation of train delays in line operation can be described and predicted. Optionally, the delay prediction model is used to extract the first delay feature and the second delay feature based on the first operation data and the second operation data, and determine the delay prediction result based on the first delay feature and the second delay feature. The first delay feature includes the characteristics of the mutual influence between the train and the adjacent train, and the second delay feature includes the characteristics of the train itself in the time series.

[0098] It should be noted that the details of calling the trained delay prediction model to perform delay prediction and outputting the delay prediction result can be referred to the relevant description in the following embodiment, which will not be introduced here.

[0099] The delay prediction model provided by the disclosed embodiment is intended to provide a practical delay reference tool. The model does not pursue extreme prediction speed, but can quickly compare the effects of different operation plans on delay adjustment on the basis of ensuring a certain degree of accuracy. Compared with traditional simulation models, this model is more adaptable to various interference factors encountered in actual operation and shows higher stability and adaptability.

[0100] The core goal of the delay prediction model provided by the embodiment of the present disclosure is to predict the delay of the train at each station on the entire route by analyzing the delay of the train and its operation plan on the route. The embodiment of the present disclosure adopts a step-by-step approach, that is, first using the delay data of the train at the first n stations (for example, n is 2) to predict its delay at the next station, and then and so on, so as to obtain the delay of the train at each station on the entire route.

[0101] Consider a train route consisting of N+1 stations denoted as {S0,S1…S N}, N is a positive integer, where the distance between stations is fixed. At the same time, for each train, it has a fixed operation plan, which contains two parts. One is the planned stop time of the train at the corresponding station, recorded as {T s1 ,T s2 …,T sN The other part includes the speed of the train in the station section. The fixed spacing between stations can be used to convert the train speed into the planned travel time between stations, thereby unifying the train operation plan unit, which is recorded as {T r1 ,T r2 ,…,T rN In addition, there are departure times of different trains at the same station, denoted as {ST0, ST1, …, ST K}(the total number of selected trains is K+1), and the delay of the train at each station. Taking the delay time as an example, the delay time of the train at each station is recorded as {D0,D1,…,D K}, D j ∈R N+1 , D j ={D j0 ,D j1 ,…,D jN , R is the planned travel time of the train in the station section, D j is the delay time of train j at each station, D jiis the delay time of train j at station i, j is an integer, j ranges from 0 to K, i is an integer, i ranges from 0 to N. The prediction part is to predict the delay time of the train in future operation under the premise of knowing the train operation plan and part of the delay time. Optionally, the delay of the train at the target station is predicted using the delay of the train at the previous station and the operation plan, and then the next station of the target station is determined as the target station, and the above-mentioned prediction of the delay of the target station using the delay of the train at the previous station and the operation plan is performed again (at this time, the delay of the train at the previous station includes the predicted delay), and finally the prediction of the delay of each station on the entire route of the train is gradually completed.

[0102] In an illustrative example, taking the use of data of the first two stations of a train at a target station to predict the delay of a target station as an example, a schematic diagram of a train delay prediction logic provided by an exemplary embodiment of the present disclosure is as follows: Figure 2 As shown. Among them, si represents the train delay data and operation plan data at station i, s0 is the initial station, there is no delay data and operation plan data, s1 represents the train delay data and operation plan data at station 1, s2 represents the train delay data and operation plan data at station 2, and so on. The operation plan data of station i includes the planned stop time of station i and the planned travel time of the station section. The station section of station i is the section from station i-1 to station i. The entry section time of station i (i.e. the time when the train enters the station section of station i) refers to the time of departure from station i-1, and the planned travel time of the station section of station i refers to the time required from station i-1 to station i. The delay situation and operation plan of the train at station i and station i+1 are used to predict the delay situation of station i+2, and then the delay situation of station i+3 is predicted by the delay situation and operation plan of station i+1 and the predicted delay situation and operation plan of station i+2, and finally the prediction of the train delay situation on the entire train operation line is gradually completed. That is to say, s1 and s2 can be used to predict the delay situation in s3, and then s2 and s3 can be used to predict the delay situation in s4, and finally the prediction of train delay situation on the entire train operation line can be gradually completed.

[0103] In some embodiments, the problem is divided, and three consecutive stations are considered first. The train delays at the first two stations, the operation plan, and the time to enter the station section of the third station (i.e., the target station) are known. The train delays at the target station are predicted, and a graph G = {V, E, A} is constructed. V represents a set of nodes, where each train entering the station section is regarded as a node, and its nodes contain the characteristics of the train's operation plan, delays, and time to enter the section. Other features can also be included according to demand. E represents a set of edges connected between nodes. Here, train pairs with possible delays that have a greater mutual influence are selected to establish edge connections. A represents an adjacency matrix to describe the mutual influence between train operations. It should be noted that the determination method of the adjacency matrix can refer to the relevant description in the following embodiment, which will not be introduced here. Since the predicted time range is large, the influence that may exist between trains with closer departure times is greater, so the weight of the edges between nodes is considered. Finally, the train delays at the target station are predicted. Therefore, in one possible implementation, the input of the delay prediction model is: the departure time of the train at the previous station, the first delay data of the train at the first n stations and the first operation plan data, for example, the value of n is 2; the output of the delay prediction model is: the delay prediction result of the train at the target station.

[0104] Optionally, the train may be one or more trains to be predicted. When the train is multiple trains to be predicted, the multiple trains may be multiple trains on the same route. In this case, the first operation data of the train at the previous station includes the first operation data of the multiple trains at the previous station, and the delay prediction result includes the delay prediction results of the multiple trains at the target station. That is, in another possible implementation, the input of the delay prediction model is: the departure time of multiple trains at the previous station, the first delay data of multiple trains at the first n stations, and the first operation plan data of multiple trains at the first n stations (or the planned operation diagram of the entire train group), for example, the value of n is 2; the output of the delay prediction model is: the delay prediction results of all trains, that is, the delay prediction results of each of the multiple trains at the target station.

[0105] In summary, the high-speed railway train delay prediction method based on graph convolutional neural network provided by the embodiment of the present disclosure can obtain the first operation data of the train at the previous station, the previous station is the n stations located before the target station to be predicted on the train's travel route, n is a positive integer, and the first operation data is used to indicate the train's delay and operation plan at the previous station; according to the first operation data, the trained delay prediction model is called to perform delay prediction, and the delay prediction result is output, and the delay prediction result is used to indicate the predicted train delay at the target station. The delay prediction model is a model based on graph convolutional neural network; that is, the spatiotemporal characteristics of train operation data are modeled using graph convolutional neural network, which can more accurately capture the dynamic spatiotemporal correlation in train operation, thereby improving the accuracy of delay prediction. This method not only considers the delay of a single train, but also comprehensively considers the temporal and spatial correlation between multiple trains and multiple lines, so that the prediction result is more comprehensive.

[0106] It should be noted that train delays can be roughly divided into two types: initial delays and accumulated delays. Initial delays often come from some accidental situations, such as extreme weather and train failures. Due to the randomness of these delays, predictions of them are not very accurate. The prediction of train delays in the disclosed embodiment mainly considers the accumulated delays of trains, that is, the situation of secondary delays.

[0107] The accumulation of delays is mainly caused by two methods: one is the existing delay at the previous station. If the train still stops according to the operation plan before the delay, the delay value of the train at the next station will not be less than the delay value of the train at the previous station. The other is the mutual influence between trains. When a train is delayed, it will cause the time for the train to enter and leave to be delayed. For example, when the delay occurs at a train stop, the departure time of the train needs to be postponed, but because the subsequent trains need to use the same arrival and departure line resources, the two trains cannot leave the station at the same time. At this time, the train needs to stop and wait, and this waiting time belongs to the delay caused by the mutual influence between trains.

[0108] The first reason is that the train is late at the previous station. If the actual operation is not adjusted, the delay will gradually accumulate to the subsequent stations. Therefore, the delay of the train at one station will be directly affected by the delay of the train at the previous station. Therefore, in order to predict the delay of the train at the subsequent station, the first delay data of the train at the previous station needs to be used as input.

[0109] Another reason is the mutual influence between trains. According to the above analysis, there will be mutual influence between train delay values. Therefore, when predicting the delay of a certain train, the second operation data of other trains on the same route need to be used as input. However, due to the increase in the time span, the number of train objects considered for prediction will gradually increase, so it is impossible to use all train operation conditions as input. There will also be a relationship between the influence of trains. For example, the departure time interval between train A and train B is 25 minutes, and the departure time interval between train A and train C is 40 minutes. Compared with train B, when train C is delayed, the impact on the operation of train A will be smaller, because on the one hand, in the existing departure time interval, the train may adjust the delay back to the original planned operation by accelerating, reducing the stop time, etc., so it will not affect the subsequent train operation. On the other hand, due to the large departure time interval, train A and train C will not meet at the station to conflict, and such influence between trains is often indirectly realized through other trains between train A and train C. Therefore, the impact of train C's delay on the operation of train A may be manifested as an indirect impact. In this part, it can be considered that the smaller the departure time interval between trains and / or the closer the departure order between trains, the greater the mutual influence between their operations. Therefore, when predicting the delay of a certain train, the second operation data of several adjacent trains can be used as input. In order to further improve the accuracy of the delay prediction model, the adjustment of the dispatching system to the train operation is taken into consideration. For the station interval on the train route, the train originally has {S1, R1, S2, R2…, S i ,R i …}, where S i is the planned stop time of the train at station i, R i is the planned travel time of the train at station i. It can be seen that the delay of the train at station i and the subsequent S i ,R i ,S i+1 The operation plan before the station where the train is delayed is not related to the operation plan, so it is necessary to take the operation plan before the target station to be predicted as input. However, due to the long line of the train, taking all factors into consideration will result in too large input. The operation plan of the previous station is directly related to the delay of the train at the target station. Therefore, only the first operation plan data of the train at the previous station can be considered as input. At the same time, since the train dispatching system mainly adjusts the operation of the train by adjusting the planned stop time of the train and the planned travel time of the station interval, taking this part as input can make the prediction model work better with the simulation model and the subsequent adjustment model.

[0110] Through the above analysis, the main influencing factors for the train delay at the target station are the train delay at the previous station, the running speed of the previous section of the train (which can be converted into the planned travel time of the station section through the distance) and the planned stop time at the previous station. In addition, the operation of adjacent trains also needs to be considered.

[0111] The embodiments of the present disclosure define these influencing factors as follows: Delay represents the delay data of the train at the station, Stoptime represents the planned stop time of the train at the station, Runtime represents the planned travel time of the train in the station section, and Starttime represents the departure time of the train at the station.

[0112] Regarding the processing method of the above-mentioned influencing factors, the accumulation of delays is the accumulation generated when the train runs on the route, and the input is the train delay data and operation plan data at different stations, both of which have obvious time series characteristics. Since the definition of delay is based on the time when the train arrives at the station and the planned arrival time, the delay of the train at each station can be regarded as a sequential process. The delay of the train at the corresponding station is the state of the train. Due to different operation plans and different execution of operation plans, the train will reach another delayed state based on the original state, and these two delayed states are strongly correlated. Considering the sequential occurrence mode, the embodiment of the present disclosure chooses to use the time series processing module to extract the time series characteristics of the data, so as to characterize the accumulation of train delays over time.

[0113] The time series processing module can be a time convolution module based on the space-time graph convolution network. The space-time graph convolution network is a deep learning framework that can effectively characterize the space-time dependencies in the data, and has a good effect in traffic prediction. Among them, the time convolution module can well extract the time series features in the data. Compared with other network structures that process time series features, the time convolution module has the characteristics of fast training speed, simple structure, and no constraints from the previous steps. On the time axis, the entire convolution structure is used to capture the time dynamic behavior in the data, and a layer is formed through a multi-layer convolution structure. Using this structure to process the delay features of trains that appear one after another on the line can effectively extract the time series features existing in the train delay data.

[0114] The mutual influence between train delays is determined by the train's own delays and the delays of adjacent trains, and can therefore be seen as a connection between train delays. This connection does not change due to the different stations where the trains are located. Therefore, the disclosed embodiment uses a graph convolution module to capture the mutual influence between train delays.

[0115] In addition, in the transportation network, the pattern of train delays often shows similar fluctuation characteristics across platforms and sections. Graph convolution operations are particularly suitable for identifying and extracting such similar fluctuation patterns. Based on this, the embodiment of the present disclosure chooses to use a graph convolution module to extract key delay features from the transportation network. The graph convolution module is selected instead of other graph models because other models may lack the ability to effectively extract these similar fluctuation features.

[0116] Therefore, the above-mentioned graph convolution module can be used to determine the first delay feature according to the first operation data and the second operation data, and the above-mentioned time series processing module can be used to determine the second delay feature according to the first operation data. In this way, the delay prediction model provided by the embodiment of the present disclosure can more deeply understand the mutual influence between train delays and the time series characteristics of train delays, thereby providing a more accurate delay prediction result according to the first delay feature and the second delay feature.

[0117] In some embodiments, a schematic diagram of a delay prediction model provided by an exemplary embodiment of the present disclosure is as follows: Figure 3 As shown. The delay prediction model may include a preprocessing module, a graph convolution module, a time series processing module and a fully connected layer. According to the first running data and the second running data, data preprocessing is performed by the preprocessing module to obtain the preprocessed first running data and the second running data (as the characteristics of each train node), and the adjacency matrix is ​​determined, and then the characteristics of each train node and the calculated adjacency matrix are processed by the graph convolution module and the time series processing module to obtain respective delay features, and the delay features output by both are then processed by the fully connected layer to obtain the final output. It should be noted that the embodiment of the present disclosure can discretize each time to be processed, such as discretizing the delay time of the delay prediction model and the planned stop time and planned travel time in the train operation diagram into integers, and using the rounding up method, the unit 1 of the input time part represents the actual 30 seconds. For example, if the train is delayed by 1 minute and 30 seconds, then in the delay prediction model, this delay time will be represented as 3, because 1 minute and 30 seconds is equal to 90 seconds, and 90 seconds divided by 30 seconds is 3.

[0118] The preprocessing module can be used to calculate the influence weights between train nodes through the departure time of the train (i.e., Starttime), and finally store the influence weights of the train nodes in the form of a sparse matrix, so as to characterize the mutual influence between train operations for subsequent module calls. The graph convolution module can be used to process the train delay data, input the delay data of the train and the adjacent train (i.e., Delay), and obtain a delay feature after passing through the graph convolution module. The time convolution module can be used to process the train delay data, the planned stop time, and the planned travel time in the station section (i.e., Stoptime and Runtime) through three one-dimensional convolution modules to obtain a second delay feature. Finally, the two parts of the delay features are connected and input into the fully connected network to obtain the corresponding delay prediction result, and then the delay prediction result is rounded up. The loss function set by the model can be a mean squared error loss function (Mean Squared Error loss, MSEloss), and the optimizer can be an adaptive moment estimation optimizer (Adaptive Moment Estimation, Adam). This is not limited in the embodiment of the present disclosure.

[0119] In one possible implementation, see Figure 4 , which shows a flowchart of a high-speed railway train delay prediction method based on a graph convolutional neural network provided by another exemplary embodiment of the present disclosure, and this embodiment is illustrated by using the method in a computing device. The method includes the following steps.

[0120] Step 401, obtain the first operation data of the train at the preceding station and the second operation data of the adjacent train at the preceding station, the first operation data including the first delay data and the first operation plan data of the train at the preceding station, and the second operation data including the second delay data and the second operation plan data of the adjacent train at the preceding station.

[0121] The adjacent train has the same route as the train, and the adjacent train is the m trains that are adjacent to the train in the route of the train in terms of departure time interval and / or departure order. m is a positive integer and can be preset according to the needs of the delay prediction model. For example, if m is set to 6, then when predicting the delay of a certain train, the operation data of the 6 trains before and after it will be considered.

[0122] In a possible implementation, adjacent trains may be defined as trains that are adjacent to the train to be predicted in the train departure sequence table, regardless of the departure time interval. For example, if the train to be predicted is train 4, and the departure sequence on the route is train 0, train 1, train 2, train 3, train 4, train 5, train 6, train 7 and train 8, then the adjacent trains of train 4 may include train 1, train 2, train 3 and train 5, train 6, train 7.

[0123] In another possible implementation, adjacent trains can also be determined based on the proximity of departure times. In this case, a train with a shorter time interval with the train to be predicted will be considered as an adjacent train, regardless of the departure sequence.

[0124] In another possible implementation, the departure sequence and departure time interval are combined to determine adjacent trains. Adjacent trains can also be defined as trains that are adjacent to the train to be predicted in the departure sequence and whose departure time interval is within a preset threshold. In this way, not only the proximity relationship on the physical line can be taken into account, but also the proximity in time can be taken into account, thereby more comprehensively capturing the mutual influence between trains.

[0125] The second operation data includes the second delay data and the second operation plan data of the adjacent train at the preceding station, the second delay data is used to indicate the delay of the adjacent train at the preceding station, and the second operation plan data is used to indicate the operation plan of the adjacent train at the preceding station. The second delay data may include the delay time of the adjacent train at the first n stations, and the second operation plan data may include the planned stop time of the adjacent train at the first n stations and the planned travel time between stations. The second operation plan data may also include the planned arrival time and / or planned departure time of the adjacent train at the first n stations. It should be noted that the relevant details of the second operation data can be referred to by analogy with the relevant description of the first operation data, and will not be repeated here.

[0126] Step 402, preprocessing the first operation data and the second operation data through a preprocessing module, and determining an adjacency matrix according to the first operation plan data and the second operation plan data, where the adjacency matrix is ​​used to indicate the degree of mutual influence between a train and adjacent trains.

[0127] The first operation data and the second operation data are preprocessed by the preprocessing module to obtain the preprocessed first operation data and the second operation data. The data preprocessing includes converting the first operation data and the second operation data into a format that can be processed by the delay prediction model. The data preprocessing also includes deleting a portion of data that is incomplete or has obvious errors. Subsequent steps are performed based on the preprocessed first operation data and the second operation data. After preprocessing, the adjacency matrix between the train nodes in the graph convolutional network is determined based on the first operation plan data and the second operation plan data.

[0128] Optionally, the first operation plan data includes the planned departure time of the train at the previous station, and the second operation plan data includes the planned departure time of an adjacent train at the previous station, and an adjacency matrix is ​​determined based on the planned departure times of the train and multiple adjacent trains at the previous station; wherein the adjacency matrix includes influence weights between the train and each adjacent train, and the influence weights are negatively correlated with the planned departure time intervals between the corresponding train and the adjacent trains.

[0129] Indicatively, the scheduled departure time of the train at the previous station is used as the basis for calculation, and only the three trains before and after the train enters are taken as nodes with edge connections to calculate the edge weights. The calculation method of the influence weight between two trains is as follows:

[0130] ST a ,ST b |ab|<4

[0131] A a,b =A b,a =tanh(|ST a -ST b | / 20)

[0132] Inf a,b =exp(-A a,b )

[0133] Among them, ST a is the scheduled departure time of train a at the previous station, ST b is the scheduled departure time of train b at the previous station, A a,b , A b,a is an intermediate variable, Inf a,b is the influence weight between train a and train b, where a and b are two different values.

[0134] In an illustrative example, the schematic diagram of the calculation function of the adjacency matrix is ​​as follows Figure 5As shown in . Among them, the horizontal axis represents the departure time difference between the two trains at the previous station, and the unit 1 represents the actual time of 30 seconds. The vertical axis represents the influence weight between the two trains, that is, the edge weight between the nodes, and the unit is 1. It can be observed from the figure that when the departure time of the two trains differs from 0 to ±50 time units, that is, when the departure time of the two trains differs by 25 minutes, the weight of their mutual influence is relatively large. Within this time difference range, the closer the planned departure time between the trains, the greater the weight of their mutual influence, which indicates that when delays occur, the mutual influence between trains with closely adjacent departure times is more significant. This weight setting ensures that the influence weight is negatively correlated with the planned departure time interval between the corresponding train and the adjacent train, that is, as the planned departure time interval between the trains increases, the influence weight between them will decrease accordingly, and the mutual influence between them will decrease accordingly, thereby reasonably reflecting the regulatory effect of the time interval in train operation on the degree of mutual influence. After completing this weight calculation, the adjacency matrix can be converted into a sparse matrix format for storage to optimize storage efficiency and facilitate subsequent graph convolutional network processing.

[0135] The adjacency matrix can include only the influence weights between a train and each adjacent train, or it can include the influence weights between any two trains among all considered trains. The influence weights between any two trains among all considered trains are stored in a common adjacency matrix. When a certain train is considered later, the corresponding rows and columns will be proposed and input into the subsequent graph convolution module.

[0136] Step 403: Based on the first delay data, the second delay data and the adjacency matrix, the graph convolution module is called to output the first delay feature.

[0137] After the preprocessing is completed, the first delay data, the second delay data and the adjacency matrix are input into the graph convolution module, and the corresponding first delay feature is output. The first delay feature can be in vector form, and the first delay feature includes the feature of the mutual influence between the train and the adjacent train. Optionally, the first delay data includes the delay value of the train at the previous station, and the second delay data includes the delay value of the adjacent train at the previous station. The delay values ​​of the train and the adjacent train at the previous station are used as the features of the corresponding nodes, and are input into the graph convolution module together with the adjacency matrix, and the corresponding first delay feature is output.

[0138] In a possible implementation, a schematic diagram of the structure of a graph convolution module provided by an exemplary embodiment of the present disclosure is as follows: Figure 6As shown. The graph convolution module includes two graph convolution layers. The input data (for example, the delay value and adjacency matrix of the train and the adjacent train at the previous station) first undergoes the first graph convolution layer, and then is converted into intermediate features through the action of the activation function (such as the ReLU function). These intermediate features are then input into the second graph convolution layer, and after further processing, they are finally converted into hidden features, namely the first delay features. Among them, the function of each graph convolution layer is to combine the node features of the adjacent trains with the weights of the adjacency matrix to achieve feature aggregation. In this process, the node features of the adjacent trains are multiplied by the adjacency matrix, and the multiplication result is added to the node features of the train itself, and the result of the addition is added to the bias term to output the output features of the layer. Schematically, the basic settings of the two graph convolution layers are as follows: the first graph convolution layer expands the input channels from 7 to 140, while the second graph convolution layer compresses the feature dimension from 140 to 10, that is, the input and output channels of the two graph convolution layers are one: (7, 140) and two: (140, 10). Through this structure, the graph convolution module not only takes into account the delay of the train itself, but also takes into account the influence of neighboring trains through the edge weights between nodes, thereby revealing the mutual influence relationship between train delays and extracting the hidden features of the mutual influence between trains, namely the first delay feature.

[0139] In the graph convolutional neural network, nodes will take the features of neighboring nodes into account in their own features, and this will enable nodes to fully consider the impact of neighboring nodes on themselves. At the same time, because the features of neighboring nodes are directly added to the features of the target node when constructing the graph convolutional network, when the position of the train node changes, the corresponding graph structure and the feature value of the node will not change due to the change of the node position, that is, the corresponding graph topology structure has not changed. For the prediction of train delays, there is a similar situation in terms of the mutual influence between trains. The delay of a train will have an impact on neighboring trains, and this impact will not change due to the different stations or travel sections where a train is located at a certain time. Therefore, using a graph convolutional neural network to characterize the mutual influence between trains can effectively extract the characteristics of the mutual influence between trains.

[0140] Step 404: Based on the first operation data, call the time series processing module output to obtain the second delay feature.

[0141] After the preprocessing is completed, according to the first delay data of the train at the previous station and the first operation plan data, the time series processing module is called to output the corresponding second delay feature, which can be in vector form and includes the characteristics of the train itself in the time series. It should be noted that step 403 and step 404 can be executed in parallel, or step 403 can be executed first and then step 404, or step 404 can be executed first and then step 403, and the embodiment of the present disclosure is not limited to this.

[0142] The time series processing module can be a time convolution module based on the spatiotemporal graph convolution network, which can capture the time series features between data through three one-dimensional convolution layers. For the time series processing module, its input data is the first delay data of the train at the previous station and the first operation plan data. The corresponding hidden feature, namely the second delay feature, is obtained through the input data. In a possible implementation, the structural diagram of the time series processing module is as follows Figure 7 The time series processing module includes multiple one-dimensional convolution layers (only three one-dimensional convolution layers are schematically shown in the figure) and an activation layer. For the processing of time series data, the first operation data (for example, the delay value of the train at the previous station, the departure time of the train at the previous station, and the operation plan data of the train at the previous station) are input into the multiple one-dimensional convolution layers of the time series processing module, and the intermediate data are output; the intermediate data are input into the activation layer of the time series processing module (for example, including Sigmoid function and ReLU function), and the corresponding time series feature, i.e., the second delay feature, is output.

[0143] Since the spatiotemporal graph convolutional neural network is a deep learning network specifically used to process spatiotemporal data, it can effectively extract features from data in the dimensions of time and space. Here, the temporal feature extraction of data is realized through the convolutional neural network method, because compared with the processing of time series data by recurrent neural networks, convolutional neural networks do not rely too much on the memory information in the previous step, so they can better capture the jitter changes of data, which is the jitter of train delays. In addition, convolutional neural networks do not require complex iterative training and gating mechanisms, so using convolutional neural networks to process time series data has better results.

[0144] Step 405, concatenate the first delay feature and the second delay feature and input them into the fully connected layer, and output the delay prediction result.

[0145] There are two types of factors that affect train delays: delay features affected by adjacent trains (i.e., the first delay feature) and the time series features of the train’s own delay (i.e., the second delay feature). These two features are concatenated through vectors and input into the fully connected layer to output the delay prediction result.

[0146] In the deep learning architecture, the extracted first delay feature and second delay feature are concatenated to form a comprehensive target delay feature. This concatenation process, i.e. feature concatenation, ensures that the network can capture the complex pattern of train delays from multiple angles and levels by concatenating the two features into vectors.

[0147] The concatenated target delay features are then fed into the fully connected layer, which is the key part of the model responsible for the final prediction. The fully connected layer is used to indicate the correspondence between the target delay features and the delay prediction results. The fully connected layer learns the correspondence between the target delay features and the delay prediction results through its dense weight matrix, thereby outputting a quantitative prediction of the train delay. In order to make the delay prediction results more in line with the actual application scenario, the output delay prediction results are rounded up. This step ensures that the delay prediction results are considered according to the worst case, thereby providing a conservative and safe estimate for train scheduling and passenger notification.

[0148] In an illustrative example, the data used in the disclosed embodiment is the high-speed railway line running section from Guangzhou South Station to Chibi North Station in the Wuhan-Guangzhou high-speed railway line, and its path includes 15 stations on the route (Guangzhou South, Guangzhou North, Qingyuan, Yingde West, Shaoguan, Lechang East, Binzhou West, Leiyang West, Hengyang East, Hengshan West, Zhuzhou West, Changsha South Wuhan-Guangzhou, Miluo East, Yueyang East, Chibi North) and 14 sections of station intervals (the travel distance of the train station interval can be obtained according to the query). The data includes the train number of each train, the departure date of the train, the name of the station to which the train arrives, the time when the train arrives at each station, the time when the train is scheduled to arrive at each station, the time when the train departs from each station, the scheduled departure time of the train at each station, the track number when the train enters the station, and the delay time (the delay here refers to the difference between the actual arrival time of the train at the station and the planned arrival time).

[0149] Except for some larger stations on this line, it is generally less affected by cross-line trains. The simulation is of a single line, so only trains departing from Guangzhou South Station to Chibi North Station are considered. Trains entering this line in the middle and trains leaving through a station are not taken into consideration.

[0150] First, subtract the departure time and arrival time of the train at a station to get the train's stop time at different stations, then subtract the departure time of the train at a station from the arrival time of the train at the next station to get the train's travel time between stations, and alternately arrange the train's stop time and travel time between stations to form the train's operation plan data. Subtract the departure time of all trains from the departure time of the first train to get the departure time relative to the first train. Subtract the time when the train arrives at the station from the time when the train is scheduled to arrive at the station to get the train's delay value at the corresponding station. All time variables are in the same unit, and each unit time represents 30 seconds of actual time.

[0151] After removing the data with wrong values, the data set contains 35194 sets of data, which are divided into training set, validation set and test set in the ratio of 6:2:2. 80% of the data are used to train and validate the model, 75% of the data are used as training set, 25% of the data are used as validation set, and the remaining 20% ​​of the data are used as test set.

[0152] In the train delay prediction part, a step-by-step prediction method is adopted: the train delay value at station i+2 is predicted based on the train delay values ​​at station i and station i+1 and the train operation plan data at these two stations. In the prediction of the entire route, the train delay value at the station will be predicted step by step using the above method. For example, the train delay value at station i+2 is first predicted based on the train data at station i and station i+1, and then the train delay value at station i+3 is predicted based on the train data at station i+1 and station i+2, and so on, to predict the delay values ​​of subsequent trains on the route.

[0153] Schematically, the training results of the train delay prediction model are as follows: Figures 8 to 10 shown. Figure 8 A schematic diagram showing the change of model loss with the number of training times during the model training process provided by an exemplary embodiment of the present disclosure is shown. The model loss includes training loss (English: Training Loss) and validation loss (English: Validation Loss). The training loss refers to the performance of the delay prediction model on the training data set, that is, the difference between the model prediction value and the actual value. The validation loss refers to the performance of the delay prediction model on the validation data set, which is usually used to adjust the model parameters to avoid overfitting. The validation data set is a data set that the delay prediction model has not seen during the training process.

[0154] Fig. 9A bar chart showing the distribution of the difference between the predicted value and the actual value of train delay provided by an exemplary embodiment of the present disclosure is shown. The horizontal axis represents the difference between the predicted value and the actual value of the train delay, and the difference per unit time is measured in units of 30 seconds, and the vertical axis represents the quantity statistics.

[0155] Fig.10 A statistical chart of predicted values ​​and actual values ​​of train delays provided by an exemplary embodiment of the present disclosure is shown. The horizontal axis represents the predicted value of train delays, expressed in minutes, and the vertical axis represents the actual value of train delays, also expressed in minutes.

[0156] It can be seen that the model loss gradually decreases with the increase of the number of training times until it reaches a stable value, which will be different for different stations. For the delay prediction results, the predicted value of the train delay is rounded up, and the accuracy of the prediction can be measured by the accuracy. The accuracy is defined as the degree of match between the delay prediction result and the actual delay result. Taking the delay prediction result as the predicted value of the train delay and the actual delay result as the actual value of the train delay as an example, several key statistical variables are defined to evaluate the accuracy of the prediction:

[0157] acc_diff_1=count[delay_t-2, delay_t]

[0158] acc_diff_2=count[delay_t, delay_t+2]

[0159] acc_diff_3=count[delay_t-2, delay_t+2]

[0160] Among them, acc_diff_1: counts the number of predicted values ​​of train delays within the first 2 minutes of the actual value, that is, [delay_t-2, delay_t]. acc_diff_2: counts the number of predicted values ​​of train delays within the last 2 minutes of the actual value, that is, [delay_t, delay_t+2]. acc_diff_3: counts the number of predicted values ​​of train delays within the 2 minutes before and after the actual value, that is, [delay_t-2, delay_t+2]. Through these statistical variables, it can be observed that the vast majority of predicted values ​​of train delays are within ±2 minutes of the actual value of the train delay, that is, within ±60 seconds of the actual time, which indicates that the delay prediction model provided by the embodiment of the present disclosure has a high accuracy. Indicatively, the statistical variables and distribution ratios listed in Table 1 provide a quantitative evaluation of the performance of the train delay prediction model.

[0161] Table 1

[0162] Statistical variables Distribution ratio Accuracy 0.5510 acc_diff_1 0.6029 acc_diff_2 0.3861 acc_diff_3 0.9890

[0163] Among them, accuracy (English: Accuracy): 0.5510 means that the accuracy of the model prediction is 55.10%. This means that among all the predictions, 55.10% of the predicted values ​​are consistent with the actual values ​​of the train delay. acc_diff_1: 0.6029 means that the accuracy of the predicted value of the train delay within the first 2 minutes of the actual value is 60.29%. acc_diff_2: 0.3861 means that the accuracy of the predicted value of the train delay within the last 2 minutes of the actual value is 38.61%. acc_diff_3: 0.9890 means that the accuracy of the predicted value of the train delay within the range of 2 minutes before and after the actual value is 98.90%, which covers almost all the predicted values, further confirming the prediction accuracy of the delay prediction model.

[0164] Schematically, the comparison results between the predicted value of train delay generated by the delay prediction model and the actual value of train delay are shown in Table II.

[0165] Table 2

[0166]

[0167]

[0168] In the process of developing the train delay prediction model, we went through a series of attempts and optimizations. Initially, the model used the operation plan of the entire line and the initial delay situation as input, trying to predict the train delay situation at each station. However, the prediction effect of this method was not ideal. After in-depth analysis, it was found that the problem was that the time and space span of the prediction was too large, and there was a lack of direct causal relationship between the initial delay situation and the delays of subsequent stations.

[0169] To address this problem, the model was optimized to use the delays at the previous station and the operation plan between that station and the current station as input to predict the delays at the next station. Although this modification brought some improvement, the prediction results were still unsatisfactory. Further analysis showed that insufficient input information may be the main reason for the poor prediction results. Based on this finding, a more sophisticated prediction method was adopted: using the train delays at the first two stations and the train operation plan between them to predict the train delays at the next station. This method significantly improved the prediction results.

[0170] In addition, during the data preprocessing stage, an initial attempt was made to convert the train's travel time in the station interval into the train's interval running speed through the running distance, but the prediction results obtained by this method had a low accuracy rate, and the loss function did not show a downward trend during the training process. After analysis, this may be due to the inconsistency of the units, which resulted in too much clutter in the input information, affecting the model's ability to learn data features. Therefore, the train's interval running speed was reconverted into the travel time in the station interval. This change significantly improved the model's prediction ability. Through this series of attempts and optimizations, the delay prediction model provided by the embodiment of the present disclosure has been significantly improved in terms of accuracy and reliability, providing strong data support for train operation management.

[0171] In summary, the embodiment of the present disclosure uses a graph convolutional neural network to construct a delay prediction model, thereby realizing accurate prediction of train delays at the target station. This model makes full use of the train delays and operation plans at the previous station, and carefully processes and extracts features of multiple factors affecting train delays. The core of the model lies in the application of two key technologies: graph convolution module and time series processing module. The graph convolution module deeply explores the mutual influence between trains and captures the complex network dynamic characteristics. The time series processing module focuses on the timing characteristics of trains in line operation, ensuring the timeliness and continuity of the prediction. Through this innovative combination, the delay prediction model not only demonstrates a strong delay prediction capability at each station, but also realizes a comprehensive prediction of delays at the level of the entire transportation network. More importantly, the delay prediction model adopts an experience-based time series processing method when processing the collaborative characteristics of train delay propagation. This method not only improves the interpretability of the model, but also enhances its credibility. Compared with those methods that rely solely on deep learning models, the delay prediction model provided by the embodiment of the present disclosure is closer to the actual operation situation and has higher practical value.

[0172] The following is an apparatus embodiment of the present disclosure. For parts not described in detail in the apparatus embodiment, reference may be made to the technical details disclosed in the above method embodiment.

[0173] Please refer to Fig.11 , which shows a schematic diagram of the structure of a high-speed railway train delay prediction device based on a graph convolutional neural network provided by an exemplary embodiment of the present disclosure. The device can be implemented in whole or in part by software, hardware, or a combination of the two, which is called a computing device. The device includes: an acquisition unit 1110 and a prediction unit 1120.

[0174] The acquisition unit 1110 is used to acquire first operation data of the train at the preceding station, where the preceding station is n stations located before the target station to be predicted on the travel route of the train, where n is a positive integer, and the first operation data is used to indicate the delay situation and operation plan of the train at the preceding station;

[0175] The prediction unit 1120 is used to call the trained delay prediction model to perform delay prediction according to the first operation data, and output a delay prediction result, which is used to indicate the predicted delay of the train at the target station. The delay prediction model is a model based on a graph convolutional neural network.

[0176] In one possible implementation, the first operation data includes the first delay data of the train at the previous station and the first operation plan data, the first delay data includes the difference between the arrival time of the train at the previous station and the planned arrival time, and the first operation plan data includes the planned stop time of the train at the previous station and the planned travel time between the stations at the previous station.

[0177] In another possible implementation, the acquiring unit 1110 is further configured to:

[0178] Acquire second operation data of the adjacent train at the preceding station, the second operation data being used to indicate the delay situation and operation plan of the adjacent train at the preceding station, the adjacent train being m trains adjacent to the departure time interval and / or departure sequence of the train on the travel route of the train, where m is a positive integer;

[0179] The prediction unit 1120 is further configured to:

[0180] The trained delay prediction model is called according to the first operation data and the second operation data to perform delay prediction, and a delay prediction result is outputted.

[0181] In another possible implementation, the delay prediction model is used to extract a first delay feature and a second delay feature based on the first operating data and the second operating data, and determine a delay prediction result based on the first delay feature and the second delay feature, the first delay feature including features of mutual influence between the train and adjacent trains, and the second delay feature including features of the train itself in a time series.

[0182] In another possible implementation, the delay prediction model includes a graph convolution module, a time series processing module and a fully connected layer, and the prediction unit 1120 is further used to:

[0183] According to the first running data and the second running data, calling the graph convolution module output to obtain a first delay feature;

[0184] According to the first operation data, the time series processing module output is called to obtain the second delay feature;

[0185] The first delay feature and the second delay feature are concatenated and input into the fully connected layer, and the delay prediction result is output.

[0186] In another possible implementation, the first operation data includes first delay data and first operation plan data of the train at the preceding station, the second operation data includes second delay data and second operation plan data of the adjacent train at the preceding station, and the prediction unit 1120 is further configured to:

[0187] Determine an adjacency matrix according to the first operation plan data and the second operation plan data, the adjacency matrix being used to indicate the degree of mutual influence between the train and adjacent trains;

[0188] According to the first delay data, the second delay data and the adjacency matrix, the graph convolution module is called to output the first delay features.

[0189] In another possible implementation, the first operation plan data includes the planned departure time of the train at the previous station, the second operation plan data includes the planned departure time of the adjacent train at the previous station, and the prediction unit 1120 is further configured to:

[0190] Determine an adjacency matrix based on the scheduled departure times of the train and multiple adjacent trains at the previous station;

[0191] The adjacency matrix includes the influence weights between a train and each adjacent train, and the influence weights are negatively correlated with the planned departure time intervals between the corresponding train and the adjacent train.

[0192] In another possible implementation, the time series processing module is a time convolution module based on a spatiotemporal graph convolution network, and the time series processing module includes a plurality of one-dimensional convolution layers and an activation layer, and the prediction unit 1120 is further used to:

[0193] Inputting the first running data into a plurality of one-dimensional convolutional layers of a time series processing module, and outputting intermediate data;

[0194] The intermediate data is input into the activation layer of the time series processing module, and the second delay feature is output.

[0195] It should be noted that the device provided in the above embodiment only uses the division of the above-mentioned functional modules as an example to implement its functions. In practical applications, the above-mentioned functions can be assigned to different functional modules according to actual needs, that is, the content structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0196] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0197] An embodiment of the present disclosure also provides a high-speed railway train delay prediction device based on a graph convolutional neural network, the device comprising: a processor; a memory for storing processor executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0198] The embodiment of the present disclosure further provides a non-volatile computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described in the claim is implemented.

[0199] The embodiments of the present disclosure also provide a computer program product, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of a computing device, the processor in the computing device executes the above method.

[0200] Fig.12 1 is a block diagram of an apparatus 1900 for executing a deep learning training method according to an exemplary embodiment. For example, the apparatus 1900 may be provided as a server or a terminal device. Fig.12 , the apparatus 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions, such as an application, that can be executed by the processing component 1922. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above method.

[0201] The device 1900 may also include a power supply component 1926 configured to perform power management of the device 1900, a wired or wireless network interface 1950 configured to connect the device 1900 to a network, and an input / output interface 1958 (I / O interface). The device 1900 may operate based on an operating system stored in the memory 1932, such as Windows Server 2000. TM , MacOS X TM , Unix TM ,Linux TM , FreeBSD TM or similar.

[0202] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the device 1900 to perform the above method.

[0203] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0204] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: 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), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0205] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0206] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0207] Various aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0208] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0209] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0210] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0211] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used in the embodiments of the present disclosure is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed in the embodiments of the present disclosure.

Claims

1. A high-speed railway train delay prediction method based on graph convolutional neural network, characterized in that: The method comprises: Acquire first operation data of the train at a preceding station, where the preceding station is n stations located before the target station to be predicted on the travel route of the train, where n is a positive integer, and the first operation data is used to indicate the delay situation and operation plan of the train at the preceding station; According to the first operating data, the trained delay prediction model is called to perform delay prediction, and a delay prediction result is output. The delay prediction result is used to indicate the predicted delay of the train at the target station. The delay prediction model is a model based on a graph convolutional neural network.

2. The method according to claim 1, characterized in that The first operation data includes the first delay data of the train at the preceding station and the first operation plan data, the first delay data includes the difference between the arrival time of the train at the preceding station and the planned arrival time, and the first operation plan data includes the planned stop time of the train at the preceding station and the planned travel time between stations at the preceding station.

3. The method according to claim 1 or 2, characterized in that: Before calling the trained delay prediction model to perform delay prediction according to the first operation data and outputting the delay prediction result, the method further includes: Acquire second operation data of an adjacent train at the preceding station, the second operation data being used to indicate a delay situation and an operation plan of the adjacent train at the preceding station, the adjacent train being m trains adjacent to the departure time interval and / or departure sequence of the train on the travel route of the train, where m is a positive integer; The step of calling the trained delay prediction model to perform delay prediction according to the first operation data and outputting a delay prediction result includes: The trained delay prediction model is called to perform delay prediction according to the first operation data and the second operation data, and the delay prediction result is outputted.

4. The method according to claim 3, characterized in that The delay prediction model is used to extract a first delay feature and a second delay feature based on the first operating data and the second operating data, and determine the delay prediction result based on the first delay feature and the second delay feature, the first delay feature including the feature of the mutual influence between the train and the adjacent train, and the second delay feature including the feature of the train itself in a time series.

5. The method according to claim 4, characterized in that The delay prediction model includes a graph convolution module, a time series processing module and a fully connected layer. The delay prediction model that has been trained is called according to the first operation data and the second operation data to perform delay prediction, and the delay prediction result is output, including: According to the first operation data and the second operation data, calling the graph convolution module output to obtain the first delay feature; According to the first operation data, calling the time series processing module output to obtain the second delay feature; The first delay feature and the second delay feature are concatenated and input into the fully connected layer, and the delay prediction result is outputted.

6. The method according to claim 5, characterized in that The first operation data includes first delay data and first operation plan data of the train at the preceding station, the second operation data includes second delay data and second operation plan data of the adjacent train at the preceding station, and the first delay feature is obtained by calling the output of the graph convolution module according to the first operation data and the second operation data, including: Determining an adjacency matrix according to the first operation plan data and the second operation plan data, wherein the adjacency matrix is ​​used to indicate the degree of mutual influence between the train and the adjacent trains; According to the first delay data, the second delay data and the adjacency matrix, the graph convolution module is called to output the first delay feature.

7. The method according to claim 6, characterized in that The first operation plan data includes the planned departure time of the train at the previous station, the second operation plan data includes the planned departure time of the adjacent train at the previous station, and the determining of the adjacency matrix according to the first operation plan data and the second operation plan data includes: Determining the adjacency matrix according to the scheduled departure times of the train and a plurality of adjacent trains at the previous station; The adjacency matrix includes the influence weight between the train and each of the adjacent trains, and the influence weight is negatively correlated with the planned departure time interval between the corresponding train and the adjacent train.

8. The method according to claim 5, characterized in that The time series processing module is a time convolution module based on a spatiotemporal graph convolutional network, and the time series processing module includes a plurality of one-dimensional convolutional layers and an activation layer. The second delay feature is obtained by calling the output of the time series processing module according to the first operation data, including: Inputting the first operation data into the multiple one-dimensional convolutional layers of the time series processing module, and outputting intermediate data; The intermediate data is input into the activation layer of the time series processing module, and the second delay feature is obtained as output.

9. A high-speed railway train delay prediction method and device based on graph convolutional neural network, characterized in that: The device comprises: processor; a memory for storing processor-executable instructions; The processor is configured to implement the method described in any one of claims 1 to 8 when executing the instructions stored in the memory.

10. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images