Road passenger transport line passenger flow prediction method and system for special events

Abstract

The invention discloses a passenger transport hub passenger flow prediction method and system for special events, and relates to the technical field of informatization. According to the method, by screening and integrating multi-source heterogeneous data such as event attribute data, historical passenger flow data and weather data, key features are systematically extracted, various factors influencing passenger flow changes are known more comprehensively, and richer input information is provided for a prediction model. Secondly, a deep learning algorithm is utilized to automatically learn a complex relationship between different data sources, different prediction models do not need to be manually designed for various hubs, and non-linear changes and time-space characteristics of passenger flow of various hubs are captured end to end. And finally, through multi-source data fusion, a more robust and reliable prediction model is constructed, the prediction accuracy and stability are improved, and more scientific and convenient decision support is provided for a traffic management department.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method and system for predicting passenger flow on a road passenger transport route for special events. Background Art

[0002] Special events are defined as mass activities that occur within a specific time frame and designated location, cause large crowds to gather, disperse or reoccur, cause an abnormal increase in traffic flow, are different from daily traffic patterns, disrupt traffic order, and require the formulation of passenger flow organization and evacuation plans. They include sports events, cultural performances, exhibitions and other events.

[0003] Data dimensionality reduction is usually divided into two methods: feature extraction and feature selection. Feature extraction refers to mapping the original feature space to a new feature space through a certain transformation method. The new feature space usually has a lower dimension, but still retains the important information of the original data. Common feature extraction methods include principal component analysis, independent component analysis, linear discriminant analysis, etc. Feature selection is to select the most representative, most relevant or most predictive feature subset from the original feature set to reduce the data dimension and retain the most important information without changing the original structure of the data. Feature selection is usually divided into three methods: filtering, wrapping and hybrid.

[0004] A dataset consisting of N features can theoretically generate 2N feature subsets. Feature selection is to select features that are meaningful to a specific task from the original dataset to form a feature subset. The selected feature subset should not only have as few features as possible, but also make the model performance better. The feature selection process is roughly divided into four steps, namely, generating candidate subsets, evaluating subsets, termination conditions, and result verification.

[0005] Traditional single feature selection methods cannot fully consider the impact of feature subsets on specific learning tasks. Although the filtering feature selection method is computationally efficient, can perform feature selection independently of the model, and has no assumptions about the given learning algorithm, it does not consider the correlation and redundancy between features when selecting feature subsets, and cannot obtain the globally optimal feature subset. At the same time, the wrapping feature selection method is superior to the filtering feature selection method in terms of prediction performance, and directly uses the learning algorithm as the evaluation criterion, so it can select the globally optimal feature subset. However, since it needs to iteratively search the entire feature space and generate the optimal feature subset, it has a higher computational overhead than the filtering feature selection method, and there is also a risk of overfitting.

[0006] In special events or emergencies, the accuracy of the passenger flow prediction model applicable to normal scenarios will be greatly reduced because it is impossible to observe the passenger flow change pattern based on historical data. The passenger flow at a station under special events is different from the normal passenger flow, which is reflected in the surge in passenger flow and sudden changes in the trend. The existing technology lacks the accuracy and real-time performance corresponding to special times;

[0007] Therefore, a new solution to the above problems needs to be proposed. Summary of the invention

[0008] The purpose of the present invention is to provide a passenger flow prediction method and system for passenger transport hubs under special events. Through the idea of ​​deep reinforcement learning, multi-source heterogeneous data such as network information data, weather data, and historical passenger flow data are integrated to improve the accuracy and real-time performance of passenger flow prediction at road passenger transport hubs under special events, so as to solve the technical problems raised in the background technology.

[0009] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a passenger flow prediction system for a passenger transport hub for special events, comprising a multi-source data acquisition module, a multi-source data preprocessing module and a passenger flow prediction model based on deep reinforcement learning;

[0010] The multi-source data acquisition module is used to build an activity feature framework based on at least event features, time features, previous passenger flow difference, previous passenger flow ratio and social network features, so as to achieve the purpose of event information extraction;

[0011] The multi-source data preprocessing module is a multi-source data preprocessing module selected based on TG, and the multi-source data preprocessing module is used to preprocess all data to meet the data requirements before entering the prediction model;

[0012] The passenger flow prediction model based on deep reinforcement learning is a DQN model. By adopting deep learning technology, it can effectively extract key features from large-scale, high-dimensional data. The key features include at least historical passenger flow data, weather conditions, holiday information, and relevant data of special events, so that the DQN model can more accurately capture the dynamic changes of hub passenger flow.

[0013] Furthermore, the event characteristic is the large passenger flow impact faced by the hub passenger flow operation network under special events, which is specifically referred to in the following formula:

[0014]

[0015] Where:

[0016] A s.d (t) is used to define whether a special event occurs at hub s on day d.

[0017] Furthermore, the time feature is used to reflect that the active passenger flow of the hub outbound passenger flow and the track inbound passenger flow after the event changes regularly with the time of the distance event, and based on the time propagation characteristics of the station passenger flow under the special event, the time difference between the predicted period and the start period of the special event is used as one of the event information, specifically refer to the following formula:

[0018]

[0019] Where:

[0020] T s.d (t) represents the temporal characteristics of the activity.

[0021] Furthermore, the difference of the previous passenger flow indicates that the hub passenger flow has a continuous correlation in the time dimension. The passenger flow in period t evolves from the previous hub passenger flow in period t-1. The hub passenger flow can be decomposed into stable passenger flow and fluctuating passenger flow. The change of station passenger flow caused by special events is called fluctuating passenger flow. The fluctuation degree of the hub station passenger flow directly reflects the impact of the event. That is, the higher the impact of the event, the greater the volatility of the passenger flow at the target station. Since the prediction period t corresponds to I s.d (t) and O s.d (t) is unknown, so the passenger flow deviation of the station in the t-1 period adjacent to the t period is used as the input variable, where M s.d (t) is used to represent the difference between the actual passenger flow of the research site s at the t-day period and the previous period without special events, that is, the passenger flow at the normal period, refer to the following formula:

[0022]

[0023] Where: n represents the number of cycles with day as the preceding sequence; I' s.d-n (t) and O' s.d-n( t) represent the normal inbound and outbound passenger flow when there is no special event in the previous sequence. Here, M is selected s.d (t-1) is the passenger flow difference characteristic of the research site s at the dth day and the tth period;

[0024] The selection basis of the previous passenger flow ratio and the previous passenger flow difference is similar, R s.d (t) is used to represent the ratio of the actual passenger flow of the research site s at the t-day period to the passenger flow of the previous period without special events, that is, to the passenger flow of the normal period, see the following formula:

[0025]

[0026] Where: variable I' s.d-n (t) and O' s.d-n The definition of (t) is the same as the previous passenger flow difference formula, and R is selected here. s.d (t-1) is used as the passenger flow ratio characteristic of the research site s at the dth day and the tth period.

[0027] Furthermore, the social network features analyze the correlation of traffic passenger flow through the number of posts and sentiment in the social network. Here, N s.d (t) is used to represent the number of posts in the t-th period on the d-th day, S s.d (t) is used to represent the average sentiment score of the text in the t-th period of the d-th day, see the following formula:

[0028]

[0029] Where: t is a continuous time segment, k is a specific time node, k∈t means that time k is within time segment t; p i,k That is, the positive sentiment probability value corresponding to the i-th text, and the time when the text was published corresponds to time k, p i,k The value range is [0,1]; m is the total number of texts in the time segment t, meaning N s.d (t), that is, the total number of posts in the t-period on the d-th day.

[0030] Furthermore, the multi-source data preprocessing module includes data cleaning, data fusion, data screening, TG prediction and selection, and feature selection.

[0031] The passenger flow prediction method for a passenger transport hub for special events includes at least the following steps:

[0032] S1: A multi-source data acquisition module is used to extract event information. Combined with the characteristics of various data sources in passenger transport hubs, multi-source data are divided into four categories: historical flow, spatial information, station type, and special. The historical passenger flow information of each station, line, and road network is obtained through ticket data. Meteorological indicators such as wind speed, air pressure, relative humidity, and weather phenomena in each time period are obtained from the website of the Meteorological Bureau through the Internet. The time, location, scale, and other information of special events are obtained from various websites, social media, and forums. Holiday information is obtained from government websites, and point of interest data is obtained from map websites.

[0033] S2: The design of data preprocessing module is adopted to facilitate TG selection to preprocess the collected multi-source data to meet the data requirements before entering the prediction model. The preprocessing includes five steps: data cleaning, data fusion, data screening, TG prediction and selection, and feature selection;

[0034] S3: A deep reinforcement learning prediction model is used, and the fused data is used as the state input of the model. The input includes historical passenger flow data, current environmental status, and possible future prediction factors. Through the interaction between the agent and the environment, the parameters of the policy network are continuously updated to minimize the prediction error or maximize the cumulative reward. By continuously iterating and adjusting the model parameters, the prediction accuracy and generalization ability of the model under special events are improved.

[0035] Furthermore, the data cleaning in S2 includes historical passenger flow data, weather data, network information data and special event data;

[0036] The data fusion in S2 is to integrate special events, network information, weather, and historical passenger flow data sets;

[0037] The data screening in S2 is to screen the special event type, event scope, weather conditions, hub sites and passenger flow trends, and further screen the time granularity to determine whether the prediction accuracy and granularity are met;

[0038] The TG evaluation and selection in S2 adopts a TG evaluation model, and the TG evaluation model includes XGBoost, LSTM and SVR.

[0039] Furthermore, the feature selection in S2 at least includes the following steps: specifying a fused data set of granularity; and extracting the selected data set using the RF-RFE model.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The present invention systematically extracts key features by screening and integrating multi-source heterogeneous data such as event attribute data, historical passenger flow data, and weather data, so as to more comprehensively understand the various factors that affect passenger flow changes and provide richer input information for the prediction model. Secondly, by using deep learning algorithms, the complex relationships between different data sources are automatically learned, and there is no need to manually design different prediction models for various hubs. The nonlinear changes and spatiotemporal characteristics of passenger flows at various hubs are captured end-to-end. Finally, through the fusion of multi-source data, a more robust and reliable prediction model is constructed to improve the accuracy and stability of the prediction, providing more scientific and convenient decision-making support for transportation management departments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0043] Figure 1 It is a multi-source data preprocessing module of the present invention;

[0044] Figure 2 This is a structural diagram of the artificial neural network of the present invention;

[0045] Figure 3 This is a schematic diagram of the DQN algorithm structure of the present invention;

[0046] Figure 4 This is a diagram of the experience playback mechanism of the present invention;

[0047] Figure 5 The figure is a flowchart of the whole process. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0049] Embodiment 1:

[0050] The passenger flow prediction system for passenger transport hubs for special events includes a multi-source data acquisition module, a multi-source data preprocessing module, and a passenger flow prediction model based on deep reinforcement learning;

[0051] The multi-source data collection module is used to build an activity feature framework based on at least event features, time features, previous passenger flow difference, previous passenger flow ratio and social network features, so as to achieve the purpose of event information extraction;

[0052] Multi-source data preprocessing module is based on the multi-source data preprocessing module selected by TG. The multi-source data preprocessing module is used to preprocess all data to meet the data requirements before entering the prediction model;

[0053] The passenger flow prediction model based on deep reinforcement learning is the DQN model. By adopting deep learning technology, it can effectively extract key features from large-scale, high-dimensional data. The key features include at least historical passenger flow data, weather conditions, holiday information, and relevant data of special events, enabling the DQN model to more accurately capture the dynamic changes of hub passenger flow.

[0054] The event characteristics are the large passenger flow impact faced by the hub passenger flow operation network under special events. For details, please refer to the following formula:

[0055]

[0056] Where:

[0057] A s.d (t) is used to define whether a special event occurs at hub s on day d.

[0058] The time feature is used to reflect that the active passenger flow of the hub outbound passenger flow and the track inbound passenger flow after the end of the event changes regularly with the time of the distance event. Based on the time propagation characteristics of the station passenger flow under special events, the time difference between the predicted period and the start period of the special event is used as one of the event information. For details, please refer to the following formula:

[0059]

[0060] Where:

[0061] T s.d (t) represents the temporal characteristics of the activity.

[0062] The difference of previous passenger flow indicates that the hub passenger flow has a continuous correlation in the time dimension. The passenger flow in period t evolves from the previous hub passenger flow in period t-1. The hub passenger flow can be decomposed into stable passenger flow and fluctuating passenger flow. The change of station passenger flow caused by special events is called fluctuating passenger flow. The fluctuation degree of hub station passenger flow directly reflects the impact of the event. That is, the higher the impact of the event, the greater the volatility of the passenger flow at the target station. Since the prediction period t corresponds to I s.d (t) and O s.d (t) is unknown, so the passenger flow deviation of the station in the t-1 period adjacent to the t period is used as the input variable, where M s.d (t) is used to represent the difference between the actual passenger flow of the research site s at the t-day period and the previous period without special events, that is, the passenger flow at the normal period, refer to the following formula:

[0063]

[0064] Where: n represents the number of cycles with the day as the preceding sequence; I′ s.d-n (t) and O′ s，d-n (t) represent the normal passenger flow in and out of the station when there is no special event in the previous sequence. Here, M is selected s.d (t-1) is the passenger flow difference characteristic of the research site s at the dth day and the tth period;

[0065] The selection criteria of the previous passenger flow ratio and the previous passenger flow difference are similar. s.d (t) is used to represent the ratio of the actual passenger flow of the research site s at the t-day period to the passenger flow of the previous period without special events, that is, to the passenger flow of the normal period, see the following formula:

[0066]

[0067] Where: variable I′ s.d-n (t) and O′ s.d-n The definition of (t) is the same as the previous passenger flow difference formula, and R is selected here. s.d (t-1) is used as the passenger flow ratio characteristic of the research site s at the dth day and the tth period.

[0068] Social network features analyze the correlation of traffic passenger flow by the number of posts and sentiment in social networks. Here, N is introduced s.d (t) is used to represent the number of posts in the t-th period on the d-th day, S s.d(t) is used to represent the average sentiment score of the text in the t-th period of the d-th day, see the following formula:

[0069]

[0070] Where: t is a continuous time segment, k is a specific time node, k∈t means that time k is within time segment t; p i，k That is, the positive sentiment probability value corresponding to the i-th text, and the time when the text was published corresponds to time k, p i，k The value range is [0, 1]; m is the total number of texts in the time segment t, meaning N s.d (t), that is, the total number of posts in the t-period on the d-th day.

[0071] See also Figure 1 ,The multi-source data preprocessing module includes data cleaning, data fusion, data screening, TG prediction and selection, and feature selection.

[0072] Passenger flow prediction model based on deep reinforcement learning:

[0073] Deep learning technology, especially convolutional neural networks (CNN) and recurrent neural networks (RNN), can effectively extract key features from large-scale, high-dimensional data, including historical passenger flow data, weather conditions, holiday information, and data related to special events (such as large-scale events, natural disasters, etc.). This ability enables the DQN model to more accurately capture the dynamic changes in hub passenger flow. The fully connected neural network of deep learning includes an input layer, a hidden layer, and an output layer. The input layer is determined by the dimension of the state variable, and the output layer is determined by the number of action choices in each interval. The specific structure is as follows Figure 2 As shown in the figure. Reinforcement learning can make optimal decisions in complex environments through continuous trial and error and optimization strategies. In passenger flow prediction, this means that the DQN model can dynamically adjust the prediction strategy based on real-time data and prediction results to cope with the uncertainty caused by special events.

[0074] Deep reinforcement learning DQN fits the state value function or state-action value function through the fitting ability of deep neural network. At present, deep reinforcement learning theory has a variety of model frameworks and is widely used in infinite continuous state space. Deep reinforcement network and policy gradient are two typical model frameworks. Deep reinforcement network is the most widely used value function-based algorithm in deep reinforcement learning. DQN is a structure that applies neural network to reinforcement learning. The block diagram of DQN learning algorithm is as follows: Figure 3 shown.

[0075] In the reinforcement learning algorithm, the Q value is updated by iteratively approximating the Bellman equation. The specific process is as follows:

[0076] Q(s t , a t )←Q(s t , a t )+α[r t+1 +γmax Q(s t+1 , a t+1 )-Q(s t , a t )]

[0077] Where α is the learning rate and its value is (0, 1), and γ is the discount factor and its value is [0, 1].

[0078] In DQN, the design of the experience replay pool mechanism means that the agent's environment exploration samples at each step will be stored in the experience replay pool D. DQN is a kind of of-policy offline learning method, which can learn both the current sample data set and the past sample data set. The sample pool includes the current state s t 、Action a t , Reward t and the next state s t+1 When the network is updated, DQN randomly extracts sample sets from the experience pool, which is the memory bank of previous learning. This means that DQN can randomly extract some previous sample data for learning each time. Random batch extraction of sample sets can not only eliminate the correlation of training data, but also make the samples meet the requirements of independent distribution, thereby avoiding the non-convergence state that may occur during learning. At the same time, the experience storage pool has an upper capacity limit. Therefore, during the training process, if the sample data set exceeds the upper capacity limit of the experience storage pool, the new sample data will replace the old sample data, thereby improving the full coverage of the learning process. The experience replay mechanism process is as follows: Figure 4 shown.

[0079] DQN introduces two networks with the same structure but different parameters to improve the reliability of the algorithm, namely the target value Q(s, a; θ) network and the predicted value Q(s, a; θ) network. - ) network. The role of the target value Q network is also to disrupt the correlation. The predicted value Q network is called the current value network, which is used to predict the estimated value of the current state-action pair, and the target value network is used to predict the actual value y of Q. The mean square error of the predicted value Q network and the target value Q network is calculated as the error loss function:

[0080] y=r+γmax Q(s,a′;θ - )

[0081] L(θ)=E[yQ(s,a;θ)]

[0082] Where, θ is the predicted value Q network parameter; θ- is the target value Q network parameter

[0083] The parameters θ of the prediction Q network are updated and optimized using the gradient descent method, and are kept updated as the loss function is iterated. The parameters of the target value Q network are copied from the prediction value Q network at regular intervals, which reduces the correlation between the Q values ​​obtained by the prediction value Q network and the Q values ​​obtained by the target value Q network, thereby improving the stability and accuracy of the algorithm. Generally speaking, the larger the target network parameter update time interval, the higher the algorithm stability and the slower the algorithm convergence speed.

[0084] Deep reinforcement learning has strong perception and decision-making capabilities, efficient processing of time series data, flexible response to complex scenarios, real-time and accuracy, adaptability and scalability, etc., which makes it have broad application prospects and important practical value in the prediction of passenger flow at road passenger transport hubs under special events.

[0085] Embodiment 2:

[0086] Based on the above embodiment 1, a passenger flow prediction method for a passenger transport hub for special events is proposed, which at least includes the following steps:

[0087] S1: A multi-source data acquisition module is used to extract event information. Combined with the characteristics of various data sources in passenger transport hubs, multi-source data are divided into four categories: historical flow, spatial information, station type, and special. The historical passenger flow information of each station, line, and road network is obtained through ticket data. Meteorological indicators such as wind speed, air pressure, relative humidity, and weather phenomena in each time period are obtained from the website of the Meteorological Bureau through the Internet. The time, location, scale, and other information of special events are obtained from various websites, social media, and forums. Holiday information is obtained from government websites, and point of interest data is obtained from map websites.

[0088] S2: The design of data preprocessing module is adopted to facilitate TG to select and preprocess the collected multi-source data to meet the data requirements before entering the prediction model. The preprocessing includes five steps: data cleaning, data fusion, data screening, TG prediction and selection, and feature selection;

[0089] S3: A deep reinforcement learning prediction model is used, and the fused data is used as the state input of the model. The input includes historical passenger flow data, current environmental status (such as weather, time, special events, etc.) and possible future prediction factors. Through the interaction between the agent and the environment (i.e., multi-source data), the parameters of the policy network are continuously updated to minimize the prediction error or maximize the cumulative reward. By continuously iterating and adjusting the model parameters, the prediction accuracy and generalization ability of the model under special events are improved.

[0090] See also Figure 1 :

[0091] Data cleaning in S2 includes historical passenger flow data, weather data, network information data, and special event data;

[0092] The data fusion in S2 is to integrate special events, network information, weather, and historical passenger flow data sets;

[0093] The data screening in S2 is to screen the special event type, event scope, weather conditions, hub stations and passenger flow trends, and further screen the time granularity to determine whether the prediction accuracy and granularity are met;

[0094] The TG evaluation and selection in S2 adopts the TG evaluation model, which includes XGBoost, LSTM and SVR.

[0095] The feature selection in S2 includes at least the following steps: specifying a fused data set of granularity; and extracting the selected data set using the RF-RFE model.

[0096] The development of the Internet in recent years has provided a new channel for mining information related to urban transportation. The Internet, including various websites, social media, and forums, is full of a large amount of information released by users, companies, and institutions. This information may include traffic conditions, accidents, climate, road construction information, and special event news. Through the information on the Internet, we can know the information of special events that have occurred, are occurring, or are about to occur. The present invention regards the passenger flow of special events as the superposition of normal passenger flow and activity passenger flow, and constructs a characteristic variable framework for passenger flow prediction of passenger hubs under special events from the two aspects of normal characteristics and activity characteristics.

[0097] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A passenger flow prediction system for road passenger transport routes for special events, characterized by: It includes multi-source data acquisition module, multi-source data preprocessing module and passenger flow prediction model based on deep reinforcement learning; The multi-source data acquisition module is used to build an activity feature framework based on at least event features, time features, previous passenger flow difference, previous passenger flow ratio and social network features, so as to achieve the purpose of event information extraction; The multi-source data preprocessing module is a multi-source data preprocessing module selected based on TG, and the multi-source data preprocessing module is used to preprocess all data to meet the data requirements before entering the prediction model; The passenger flow prediction model based on deep reinforcement learning is a DQN model. By adopting deep learning technology, it can effectively extract key features from large-scale, high-dimensional data. The key features include at least historical passenger flow data, weather conditions, holiday information, and relevant data of special events, so that the DQN model can more accurately capture the dynamic changes of hub passenger flow.

2. The passenger flow prediction system for road passenger transport routes for special events according to claim 1 is characterized by: The event characteristics are the large passenger flow impact faced by the hub passenger flow operation network under special events, please refer to the following formula for details: Where: A s.d (t) is used to define whether a special event occurs at hub s on day d.

3. The method and system for predicting passenger flow of road passenger transport routes for special events according to claim 2, characterized in that: The time feature is used to reflect that the active passenger flow of the hub outbound passenger flow and the track inbound passenger flow after the end of the event changes regularly with the time of the distance event, and based on the time propagation characteristics of the station passenger flow under special events, the time difference between the predicted period and the start period of the special event is used as one of the event information, specifically refer to the following formula: Where: T s.d (t) represents the temporal characteristics of the activity.

4. The road passenger transport route passenger flow prediction system for special events according to claim 3 is characterized by: The difference of the previous passenger flow indicates that the hub passenger flow has a continuous correlation in the time dimension. The passenger flow in period t evolves from the previous hub passenger flow in period t-1. The hub passenger flow can be decomposed into stable passenger flow and fluctuating passenger flow. The change of station passenger flow caused by special events is called fluctuating passenger flow. The fluctuation degree of the hub station passenger flow directly reflects the impact of the event. That is, the higher the impact of the event, the greater the volatility of the passenger flow at the target station. Since the prediction period t corresponds to I s.d (t) and O s.d (t) is unknown, so the passenger flow deviation of the station in the t-1 period adjacent to the t period is used as the input variable, where M s.d (t) is used to represent the difference between the actual passenger flow of the research site s at the t-day period and the previous period without special events, that is, the passenger flow at the normal period, refer to the following formula: Where: n represents the number of cycles with day as the preceding sequence; I' s.d-n (t) and O' s.d-n (t) represent the normal passenger flow in and out of the station when there is no special event in the previous sequence. Here, M is selected s.d (t-1) is the passenger flow difference characteristic of the research site s at the dth day and the tth period; The selection basis of the previous passenger flow ratio and the previous passenger flow difference is similar, R s.d (t) is used to represent the ratio of the actual passenger flow of the research site s at the t-day period to the passenger flow of the previous period without special events, that is, to the passenger flow of the normal period, see the following formula: Where: variable I' s.d-n (t) and O' s.d-n The definition of (t) is the same as the previous passenger flow difference formula, and R is selected here. s.d (t-1) is used as the passenger flow ratio characteristic of the research site s at the dth day and the tth period.

5. The passenger flow prediction system for passenger transport hubs facing special events according to claim 4 is characterized by: The social network features analyze the correlation of traffic passenger flow by the number of posts and sentiment in the social network. Here, N s.d (t) is used to represent the number of posts in the t-th period on the d-th day, S s.d (t) is used to represent the average sentiment score of the text in the t-th period of the d-th day, see the following formula: Where: t is a continuous time segment, k is a specific time node, k∈t means that time k is within time segment t; p i,k That is, the positive sentiment probability value corresponding to the i-th text, and the time when the text was published corresponds to time k, p i,k The value range is [0,1]; m is the total number of texts in the time segment t, meaning N s.d (t), that is, the total number of posts in the t-period on the d-th day.

6. The passenger flow prediction system for passenger transport hubs facing special events according to claim 1, characterized in that: The multi-source data preprocessing module includes data cleaning, data fusion, data screening, TG prediction and selection, and feature selection.

7. A passenger flow prediction method for a passenger transport hub for special events, used in a passenger flow prediction system for a passenger transport hub for special events as described in any one of claims 1 to 6, characterized in that: At least the following steps are included: S1: A multi-source data acquisition module is used to extract event information. Combined with the characteristics of various data sources in passenger transport hubs, multi-source data are divided into four categories: historical flow, spatial information, station type, and special. The historical passenger flow information of each station, line, and road network is obtained through ticket data. Meteorological indicators such as wind speed, air pressure, relative humidity, and weather phenomena in each time period are obtained from the website of the Meteorological Bureau through the Internet. The time, location, scale, and other information of special events are obtained from various websites, social media, and forums. Holiday information is obtained from government websites, and point of interest data is obtained from map websites. S2: The design of data preprocessing module is adopted to facilitate TG selection to preprocess the collected multi-source data to meet the data requirements before entering the prediction model. The preprocessing includes five steps: data cleaning, data fusion, data screening, TG prediction and selection, and feature selection; S3: A deep reinforcement learning prediction model is used, and the fused data is used as the state input of the model. The input includes historical passenger flow data, current environmental status, and possible future prediction factors. Through the interaction between the agent and the environment, the parameters of the policy network are continuously updated to minimize the prediction error or maximize the cumulative reward. By continuously iterating and adjusting the model parameters, the prediction accuracy and generalization ability of the model under special events are improved.

8. The passenger flow prediction method for passenger transport hubs facing special events according to claim 7 is characterized by: The data cleaning in S2 includes historical passenger flow data, weather data, network information data and special event data; The data fusion in S2 is to integrate special events, network information, weather, and historical passenger flow data sets; The data screening in S2 is to screen the special event type, event scope, weather conditions, hub sites and passenger flow trends, and further screen the time granularity to determine whether the prediction accuracy and granularity are met; The TG evaluation and selection in S2 adopts a TG evaluation model, and the TG evaluation model includes XGBoost, LSTM and SVR.

9. The passenger flow prediction method for passenger transport hubs facing special events according to claim 8, characterized in that: The feature selection in S2 at least includes the following steps: specifying a fused data set of granularity; and extracting the selected data set using the RF-RFE model.

Images