A method and system for predicting the sharing rate of multiple transportation modes in a comprehensive transportation hub

Abstract

The present invention relates to the field of transportation technology, and in particular to a method and system for predicting the share rate of multiple transportation modes in a comprehensive transportation hub; the method comprises collecting historical total passenger flow data entering and leaving the comprehensive transportation hub, passenger data transported by various transportation modes, capacity data of various transportation modes and flight information data, calculating the port arrival and departure share rates of various transportation modes, forming a training set and a verification set, constructing a SCINet time series prediction model, inputting the port arrival and departure share rates of various transportation modes in the past period of time, the time series consisting of the capacity data of various transportation modes and flight information data of various transportation modes into the SCINet time series prediction model, and obtaining the predicted values ​​of the port arrival and departure share rates of various transportation modes in the future period of time; the present invention takes into account the influencing factors of the transportation mode share rate, and takes into account flight delays or cancellations, so as to realize the accurate prediction of the share rate of each transportation mode, so as to arrange and adjust each transportation mode in advance.

Description

Technical Field

[0001] The present invention relates to the field of transportation technology, and in particular to a method and system for predicting the sharing rate of multiple transportation modes in a comprehensive transportation hub. Background Art

[0002] As an important part of the comprehensive transportation system and a key node of the comprehensive transportation network, the comprehensive transportation hub is the main carrier for the efficient connection and integrated organization of various modes of transportation. It plays an important role in improving the efficiency of the comprehensive transportation network, optimizing the transportation structure, enhancing the level of multimodal transportation development, and accelerating the transformation and development of transportation. Passengers often have to wait for a long time when leaving the comprehensive transportation hub. An important reason for this situation is that the comprehensive transportation hub cannot accurately predict the proportion of passengers leaving the transportation they take, and thus cannot reasonably provide the appropriate amount of transportation capacity in advance. In this case, how to accurately predict the share of transportation modes for passengers entering and leaving the transportation hub has become an urgent problem to be solved.

[0003] Among the current methods for calculating the share rate, it is common to use traditional machine learning algorithms or deep learning methods such as long short-term memory networks (LSTM) to calculate the share rates of taxis, buses and other modes of transportation. The above methods usually do not consider the data of multiple modes of transportation together, do not calculate the impact of changes in the capacity of a certain mode of transportation on the share rate of other modes of transportation, and do not consider the impact of emergencies such as flight delays. At the same time, with the development of smart chips and deep learning technology, traditional methods based on statistics or probability and methods based on recurrent neural networks such as LSTM are gradually being replaced by more effective algorithms due to their limitations such as insufficient fitting ability and low accuracy. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention aims to provide a method and system for predicting the share rate of multiple transportation modes in a comprehensive transportation hub, so as to solve the problem of inaccurate transportation share rate in the existing prediction methods, and to achieve the arrangement and adjustment of each transportation mode in advance by accurately predicting the share rate of each transportation mode.

[0005] In order to solve the above problems, the present invention adopts the following technical solutions:

[0006] A method for predicting the sharing rate of multiple transportation modes in a comprehensive transportation hub, comprising:

[0007] Collect historical total passenger flow data in and out of the integrated transportation hub, passenger data transported by various modes of transportation, capacity data of various modes of transportation, and flight information data;

[0008] Calculate the port entry and exit share of various modes of transportation;

[0009] The time series is constructed by using the inbound and outbound share rates of various modes of transportation within a certain period of time, and the capacity data and flight information data of various modes of transportation are used as covariates of the time series to form training sets and validation sets;

[0010] The training set and the validation set are used to train and validate the SCINet time series prediction model in a self-supervised learning manner to construct the SCINet time series prediction model;

[0011] The time series consisting of the inbound and outbound share rates of various modes of transportation in the past period, the capacity data of various modes of transportation and the flight information data are input into the SCINet time series prediction model to obtain the predicted values ​​of the inbound and outbound share rates of various modes of transportation in the future period.

[0012] As an implementable embodiment, the covariate of the time series further includes work / holiday information and weather information.

[0013] As an implementable method, the inbound and outbound share rate of the various modes of transportation is the ratio of the passenger data transported by the various modes of transportation to the total passenger flow data entering and leaving the comprehensive transportation hub.

[0014] As an implementable method, the construction of the SCINet time series prediction model includes:

[0015] Construct SCI-Block and complete downsampling and interactive learning operations in one SCI-Block: downsample the input time series, separate the odd and even elements into odd subsequences and even subsequences respectively; use different convolution filters to process odd subsequences and even subsequences respectively, and mine and extract temporal feature information in each subsequence; and perform interactive learning operations between odd subsequences and even subsequences to realize information interaction between the two subsequences and discover the temporal dependency between different sequences;

[0016] The SCINet architecture is constructed by arranging multiple SCI-Blocks in layers to obtain a binary tree structure framework; the input time series is processed through multiple layers of downsampling, convolution, and interaction to mine its feature information at different time resolutions and gradually accumulate it to obtain several feature subsequences; after passing through the last layer of the binary tree structure, the obtained feature subsequences are spliced ​​and arranged to form a new sequence, which is added to the original input time series through residual connection to generate a new feature sequence with enhanced perceptibility;

[0017] The new feature sequence is processed using a fully connected layer and a Softmax function to output the share of various current transportation modes. The cross entropy loss is then calculated to iteratively optimize the various parameters of the model until the SCINet time series prediction model with the best performance is obtained, completing the model training.

[0018] As an implementable method, the flight information data includes the planned take-off and landing time of the flight, the actual take-off and landing time and the number of passengers.

[0019] A multi-transportation mode share rate prediction system for a comprehensive transportation hub, comprising a data collection module, a calculation module, a data set construction module, a model construction module and a share rate prediction module;

[0020] The data collection module is used to collect historical total passenger flow data in and out of the comprehensive transportation hub, passenger data transported by various modes of transportation, capacity data of various modes of transportation, and flight information data;

[0021] The calculation module is used to calculate the port entry and exit sharing rates of various modes of transportation;

[0022] The data set construction module is used to construct a time series using the inbound and outbound share rates of various modes of transportation within a certain period of time, and use the capacity data and flight information data of various modes of transportation as covariates of the time series to form a training set and a validation set;

[0023] The model building module is used to train and verify the SCINet time series prediction model by using the training set and the verification set in a self-supervised learning manner to build the SCINet time series prediction model;

[0024] The share rate prediction module is used to input the time series consisting of the inbound and outbound share rates of various modes of transportation in the past period of time, the capacity data of various modes of transportation and flight information data into the SCINet time series prediction model to obtain the predicted values ​​of the inbound and outbound share rates of various modes of transportation in the future period of time.

[0025] As an implementable embodiment, the covariate of the time series further includes work / holiday information and weather information.

[0026] As an implementable method, the inbound and outbound share rate of the various modes of transportation is the ratio of the passenger data transported by the various modes of transportation to the total passenger flow data entering and leaving the comprehensive transportation hub.

[0027] As an implementable method, the construction of the SCINet time series prediction model includes:

[0028] Construct SCI-Block and complete downsampling and interactive learning operations in one SCI-Block: downsample the input time series, separate the odd and even elements into odd subsequences and even subsequences respectively; use different convolution filters to process odd subsequences and even subsequences respectively, and mine and extract temporal feature information in each subsequence; and perform interactive learning operations between odd subsequences and even subsequences to realize information interaction between the two subsequences and discover the temporal dependency between different sequences;

[0029] The SCINet architecture is constructed by arranging multiple SCI-Blocks in layers to obtain a binary tree structure framework; the input time series is processed through multiple layers of downsampling, convolution, and interaction to mine its feature information at different time resolutions and gradually accumulate it to obtain several feature subsequences; after passing through the last layer of the binary tree structure, the obtained feature subsequences are spliced ​​and arranged to form a new sequence, which is added to the original input time series through residual connection to generate a new feature sequence with enhanced perceptibility;

[0030] The new feature sequence is processed using a fully connected layer and a Softmax function to output the share of various current transportation modes. The cross entropy loss is then calculated to iteratively optimize the various parameters of the model until the SCINet time series prediction model with the best performance is obtained, completing the model training.

[0031] As an implementable method, the flight information data includes the planned take-off and landing time of the flight, the actual take-off and landing time and the number of passengers.

[0032] The beneficial effects of the present invention are:

[0033] 1. When predicting the share of one or several modes of transportation in a comprehensive transportation hub, use the historical share and capacity information of all modes of transportation as input. Because the scenes in a comprehensive transportation hub are complex and there are many emergencies, the increase or decrease in the share of other modes of transportation and the sudden change of the capacity of other modes of transportation due to various reasons will affect the share of the mode of transportation that needs to be predicted.

[0034] 2. When predicting the share of one or several modes of transportation in a comprehensive transportation hub, the flight schedule information over the past period of time should be considered to avoid the difficulty in predicting the share of each mode of transportation due to emergencies such as flight delays and flight cancellations.

[0035] 3. Use the sample convolution and interaction network (SCINet) algorithm to predict the traffic hub share rate. Compared with traditional machine learning algorithms and LSTM algorithms, the SCINet algorithm has stronger fitting ability, more parameters, larger receptive field and better parallel computing ability. It can better solve the gradient vanishing and forgetting problems, and supports longer input sequences. It can learn a lot of time series rules in the training data. Applying SCINet in the prediction of traffic hub share rate can achieve better accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic flow chart of a method for predicting the sharing rate of multiple transportation modes in a comprehensive transportation hub according to an embodiment of the present invention.

[0037] Figure 2 Schematic diagram of a system for predicting the sharing rate of multiple transportation modes in a comprehensive transportation hub in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The present invention is further described in detail below in conjunction with specific embodiments.

[0039] It should be noted that these embodiments are only used to illustrate the present invention rather than to limit the present invention. Simple improvements to the method based on the concept of the present invention all fall within the scope of protection claimed by the present invention.

[0040] See also Figure 1 , which is a method for predicting the sharing rate of multiple transportation modes in a comprehensive transportation hub, including:

[0041] S100: Collect historical total passenger flow data in and out of the integrated transportation hub, passenger data transported by various modes of transportation, capacity data of various modes of transportation, and flight information data, wherein the flight information data includes the planned take-off and landing time of the flight, the actual take-off and landing time, and the number of passengers.

[0042] S200, calculating the port-entry and port-departure sharing rate of various modes of transportation. The ratio of the passenger data transported by various modes of transportation at the corresponding time to the total passenger flow data entering and leaving the comprehensive transportation hub is the port-entry and port-departure sharing rate of various modes of transportation. .

[0043] S300, port entry and exit share of various modes of transportation within a certain period of time Constructing a time series [ Sr 0 , Sr 1 , Sr 2 …] , and the capacity data of various modes of transportation and flight information data As a covariate of the time series, the resulting time series is [ Cap 0 , FlightInfo 0 , Cap 1 , FlightInfo 1 , Cap 2 , FlightInfo 2 … ] , forming a training set and a validation set. Among them, the covariates of the time series also include work / holiday information and weather information.

[0044] S400, using the training set and the validation set to train and validate the SCINet time series prediction model in a self-supervised learning manner, and constructing the SCINet time series prediction model.

[0045] S500, input the port inbound and outbound share of various modes of transportation over the past period of time [ Sr t-n , Sr t-n+1 , Sr t-n+2 ,…, Sr t-1 , Sr t ] , capacity data and flight information data of various modes of transportation [ Cap t-n , FlightInfo t-n , Cap t-n+1 , FlightInfo t-n+1 ,…, Cap t , FlightInfo t ] The constructed time series is applied to the SCINet time series prediction model to obtain the predicted values ​​of the port entry and exit share of various modes of transportation in the future. [ Sr t+1 , Sr t+2 ,…, Sr t+m ] .

[0046] As an implementable method, the construction of the SCINet time series prediction model includes:

[0047] Construct SCI-Block and complete downsampling and interactive learning operations in one SCI-Block: downsample the input time series, separate the odd and even elements into odd subsequences and even subsequences respectively; use different convolution filters to process odd subsequences and even subsequences respectively, and mine and extract temporal feature information in each subsequence; and perform interactive learning operations between odd subsequences and even subsequences to realize information interaction between the two subsequences and discover the temporal dependency between different sequences;

[0048] The SCINet architecture is constructed by arranging multiple SCI-Blocks in layers to obtain a binary tree structure framework; the input time series is processed through multiple layers of downsampling, convolution, and interaction to mine its feature information at different time resolutions and gradually accumulate it to obtain several feature subsequences; after passing through the last layer of the binary tree structure, the obtained feature subsequences are spliced ​​and arranged to form a new sequence, which is added to the original input time series through residual connection to generate a new feature sequence with enhanced perceptibility;

[0049] The new feature sequence is processed using a fully connected layer and a Softmax function to output the share of various current transportation modes. The cross entropy loss is then calculated to iteratively optimize the various parameters of the model until the SCINet time series prediction model with the best performance is obtained, completing the model training.

[0050] See also Figure 2 , which is a comprehensive transportation hub multi-transportation mode share rate prediction system, including a data collection module 100, a calculation module 200, a data set construction module 300, a model construction module 400 and a share rate prediction module 500;

[0051] The data collection module 100 is used to collect historical total passenger flow data in and out of the integrated transportation hub, passenger data transported by various modes of transportation, capacity data of various modes of transportation, and flight information data;

[0052] The calculation module 200 is used to calculate the port entry and exit sharing rates of various modes of transportation;

[0053] The data set construction module 300 is used to construct a time series using the inbound and outbound share rates of various modes of transportation within a certain period of time, and use the capacity data and flight information data of various modes of transportation as covariates of the time series to form a training set and a validation set;

[0054] The model building module 400 is used to train and verify the SCINet time series prediction model by using the training set and the verification set in a self-supervised learning manner to build the SCINet time series prediction model;

[0055] The share rate prediction module 500 is used to input the time series consisting of the inbound and outbound share rates of various modes of transportation in the past period of time, the capacity data of various modes of transportation and flight information data into the SCINet time series prediction model, and obtain the predicted values ​​of the inbound and outbound share rates of various modes of transportation in the future period of time.

[0056] The covariates of the time series also include work / holiday information and weather information.

[0057] Among them, the entry and exit share rate of the various modes of transportation is the ratio of the passenger data transported by the various modes of transportation to the total passenger flow data entering and leaving the comprehensive transportation hub.

[0058] Wherein, the construction of the SCINet time series prediction model includes:

[0059] Construct SCI-Block and complete downsampling and interactive learning operations in one SCI-Block: downsample the input time series, separate the odd and even elements into odd subsequences and even subsequences respectively; use different convolution filters to process odd subsequences and even subsequences respectively, and mine and extract temporal feature information in each subsequence; and perform interactive learning operations between odd subsequences and even subsequences to realize information interaction between the two subsequences and discover the temporal dependency between different sequences;

[0060] The SCINet architecture is constructed by arranging multiple SCI-Blocks in layers to obtain a binary tree structure framework; the input time series is processed through multiple layers of downsampling, convolution, and interaction to mine its feature information at different time resolutions and gradually accumulate it to obtain several feature subsequences; after passing through the last layer of the binary tree structure, the obtained feature subsequences are spliced ​​and arranged to form a new sequence, which is added to the original input time series through residual connection to generate a new feature sequence with enhanced perceptibility;

[0061] The new feature sequence is processed using a fully connected layer and a Softmax function to output the share of various current transportation modes. The cross entropy loss is then calculated to iteratively optimize the various parameters of the model until the SCINet time series prediction model with the best performance is obtained, completing the model training.

[0062] The flight information data includes the planned take-off and landing time, actual take-off and landing time and the number of passengers of the flight.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described with reference to the preferred embodiments of the present invention, it should be understood by those skilled in the art that various changes may be made in form and details without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A method for predicting the sharing rate of multiple transportation modes in a comprehensive transportation hub, characterized in that: include: Collect historical total passenger flow data in and out of the integrated transportation hub, passenger data transported by various modes of transportation, capacity data of various modes of transportation, and flight information data; Calculate the port entry and exit share of various modes of transportation; The time series is constructed by using the inbound and outbound share rates of various modes of transportation within a certain period of time, and the capacity data and flight information data of various modes of transportation are used as covariates of the time series to form training sets and validation sets; The training set and the validation set are used to train and validate the SCINet time series prediction model in a self-supervised learning manner to construct the SCINet time series prediction model; Input the time series consisting of the inbound and outbound share rates of various modes of transportation in the past period of time, the capacity data of various modes of transportation, and the flight information data to the SCINet time series prediction model to obtain the predicted value of the inbound and outbound share rates of various modes of transportation in the future period of time; The construction of the SCINet time series prediction model includes: Construct SCI-Block and complete downsampling and interactive learning operations in one SCI-Block: downsample the input time series, separate the odd and even elements into odd subsequences and even subsequences respectively; use different convolution filters to process odd subsequences and even subsequences respectively, and mine and extract temporal feature information in each subsequence; and perform interactive learning operations between odd subsequences and even subsequences to realize information interaction between the two subsequences and discover the temporal dependency between different sequences; The SCINet architecture is constructed by arranging multiple SCI-Blocks in layers to obtain a binary tree structure framework; the input time series is processed through multiple layers of downsampling, convolution, and interaction to mine its feature information at different time resolutions and gradually accumulate it to obtain several feature subsequences; after passing through the last layer of the binary tree structure, the obtained feature subsequences are spliced ​​and arranged to form a new sequence, which is added to the original input time series through residual connection to generate a new feature sequence with enhanced perceptibility; The new feature sequence is processed using a fully connected layer and a Softmax function, the share of various current transportation modes is output, and then the cross entropy loss is calculated, so as to iteratively optimize the various parameters of the model until the SCINet time series prediction model with the best performance is obtained, and the model training is completed; The flight information data includes the planned take-off and landing time of the flight, the actual take-off and landing time and the number of passengers.

2. The method for predicting the multi-transportation mode share rate of a comprehensive transportation hub according to claim 1 is characterized in that: The covariates of the time series also include work / holiday information and weather information.

3. The method for predicting the multi-transportation mode share rate of a comprehensive transportation hub according to claim 1 is characterized in that: The port entry and exit sharing rate of the various modes of transportation is the ratio of the passenger data transported by the various modes of transportation to the total passenger flow data entering and leaving the comprehensive transportation hub.

4. A comprehensive transportation hub multi-transportation mode share prediction system, characterized in that: It includes data collection module, calculation module, data set construction module, model construction module and sharing rate prediction module; The data collection module is used to collect historical total passenger flow data in and out of the comprehensive transportation hub, passenger data transported by various modes of transportation, capacity data of various modes of transportation, and flight information data; The calculation module is used to calculate the port entry and exit sharing rates of various modes of transportation; The data set construction module is used to construct a time series using the inbound and outbound share rates of various modes of transportation within a certain period of time, and use the capacity data and flight information data of various modes of transportation as covariates of the time series to form a training set and a validation set; The model building module is used to train and verify the SCINet time series prediction model by using the training set and the verification set in a self-supervised learning manner to build the SCINet time series prediction model; The share rate prediction module is used to input the time series consisting of the inbound and outbound share rates of various modes of transportation in the past period of time, the capacity data of various modes of transportation, and the flight information data into the SCINet time series prediction model to obtain the predicted values ​​of the inbound and outbound share rates of various modes of transportation in the future period of time; The construction of the SCINet time series prediction model includes: Construct SCI-Block and complete downsampling and interactive learning operations in one SCI-Block: downsample the input time series, separate the odd and even elements into odd subsequences and even subsequences respectively; use different convolution filters to process odd subsequences and even subsequences respectively, and mine and extract temporal feature information in each subsequence; and perform interactive learning operations between odd subsequences and even subsequences to realize information interaction between the two subsequences and discover the temporal dependency between different sequences; The SCINet architecture is constructed by arranging multiple SCI-Blocks in layers to obtain a binary tree structure framework; the input time series is processed through multiple layers of downsampling, convolution, and interaction to mine its feature information at different time resolutions and gradually accumulate it to obtain several feature subsequences; after passing through the last layer of the binary tree structure, the obtained feature subsequences are spliced ​​and arranged to form a new sequence, which is added to the original input time series through residual connection to generate a new feature sequence with enhanced perceptibility; The new feature sequence is processed using a fully connected layer and a Softmax function, the share of various current transportation modes is output, and then the cross entropy loss is calculated, so as to iteratively optimize the various parameters of the model until the SCINet time series prediction model with the best performance is obtained, and the model training is completed; The flight information data includes the planned take-off and landing time of the flight, the actual take-off and landing time and the number of passengers.

5. The comprehensive transportation hub multi-transportation mode share prediction system according to claim 4 is characterized in that: The covariates of the time series also include work / holiday information and weather information.

6. The comprehensive transportation hub multi-transportation mode share prediction system according to claim 4 is characterized in that: The port entry and exit sharing rate of the various modes of transportation is the ratio of the passenger data transported by the various modes of transportation to the total passenger flow data entering and leaving the comprehensive transportation hub.

Images