A method for predicting passenger flow status of urban rail transit based on spatiotemporal characteristics

Abstract

The present invention discloses a method for predicting the passenger flow state of urban rail transit based on spatiotemporal characteristics. First, the AFC data is preprocessed to obtain the cross-sectional passenger flow; the urban rail line network diagram is converted into an adjacency topology diagram, and the spatial features are extracted using a graph convolutional neural network; the data is then input into a long short-term memory network to construct an urban rail passenger flow prediction model to obtain the passenger flow of the next time granularity; a full-parameter prediction model based on risk duration is further established for this passenger flow to obtain the passenger flow duration; a passenger flow time series is constructed based on all data, and a line network passenger flow state sequence is formed using the Fisher ordered sample clustering algorithm to obtain the predicted value and duration of the passenger flow state. The present invention not only provides the predicted value of the passenger flow state in the next stage, but also provides the state duration through the urban rail transit passenger flow state prediction method based on spatiotemporal characteristics, which provides a more accurate basis for the operation organization of the urban rail transit.

Description

Technical Field

[0001] The invention relates to a method for predicting the passenger flow state of urban rail transit, and in particular to a method for predicting the passenger flow state of urban rail transit based on time-space characteristics. Background Art

[0002] Whether or not a city has urban rail transit has become an important factor in measuring the level of comprehensive development of a city. At present, the networked operation pattern of urban rail transit in various countries has basically taken shape. The future development of urban rail transit will tend to be characterized by multiple lines, large passenger flows, and diversified system formats.

[0003] In order to improve the operational efficiency and overall service level of the entire rail transit, the urban rail transit system uses the automatic ticket vending and checking system (AFC) to count passenger flow data. Using the historical passenger flow data of AFC to make real-time predictions of the future passenger flow of the station has very important application prospects. The urban rail transit train operation diagram is a specific manifestation of the full-day driving plan. By analyzing the changing trend of the urban rail transit passenger flow transfer law, it provides data support for the driving plan arrangement. More importantly, the prediction of passenger flow can provide real-time dynamic adjustment measures for the urban rail decision-making level and formulate appropriate operation plans. Since the passenger flow state transfer is a nonlinear random fluctuation process, the simple time-sharing calculation and prediction method often produces poor results. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method for predicting passenger flow status of urban rail transit based on spatiotemporal characteristics, which ensures the reliability of train scheduling and provides a more accurate basis for the safety management and operation organization decision-making of urban rail transit.

[0005] Technical solution: The urban rail transit passenger flow state prediction method based on spatiotemporal characteristics of the present invention comprises the following steps:

[0006] (1) Obtain the historical AFC passenger flow as input data, perform preprocessing on the urban rail passenger flow data, and obtain the OD distribution probability matrix and section passenger flow of the urban rail network;

[0007] (2) The node edge information of the urban rail network graph is extracted to obtain the line adjacency topology graph, and the node adjacency matrix is ​​obtained based on the topology graph. The urban rail network section passenger flow is used as the input of the graph convolutional neural network (GCN) to extract the spatial characteristics of the urban rail network passenger flow data;

[0008] (3) The passenger flow output by GCN is used as the input of the long short-term memory network (LSTM) to build a rail transit OD passenger flow prediction model based on time and space factors to obtain the passenger flow of the urban rail network at the next time granularity;

[0009] (4) Based on the predicted passenger flow of the urban rail network, a risk-based duration full-parameter model in survival analysis statistics is established to explore the distribution pattern of the next duration of the predicted passenger flow data;

[0010] (5) Based on the predicted results of the subsequent time distribution law of passenger flow data and the historical passenger flow data, the section passenger flow time series is constructed, and the Fisher ordered sample clustering algorithm is used to optimally segment the ordered data stream to form the best classification of the section passenger flow rate of the urban rail network interval;

[0011] (6) Based on the optimal segmentation statistics of the section passenger flow data, the section full load rate is obtained, and the passenger flow state sequence of the urban rail network is formed according to the full load rate threshold interval, and the predicted state and duration of the urban rail network are obtained.

[0012] Furthermore, in step (1), the preprocessing includes the following steps: data cleaning of outliers and abnormal points in passenger flow data, and then noise denoising.

[0013] In step (1), the statistical granularity of the urban rail network is 30-60 minutes. The passenger boarding and alighting volume of the network is calculated through the OD distribution profile matrix table, and the section passenger flow is obtained by combining the OD distribution information.

[0014] In step (2), the graph convolutional neural network GCN is a graph structure with signal nodes, and the data is analyzed by convolution operation using the spectral graph method. The specific topological graph Laplacian matrix L is in the canonical form:

[0015]

[0016] where A∈R N*N The adjacency matrix is ​​used to represent the edge relationship between nodes, D represents the degree matrix, and I N is the difference between the adjacency matrix with self-connection and the node adjacency matrix;

[0017] Perform Fourier transform on the image signal, and the result can be expressed as Use the convolution kernel g θ Convolution prediction for graph G;

[0018] in, is the graph Fourier transform of x, U is the eigenvector matrix of the normalized graph Laplace, and x is the node information value.

[0019] In step (2), the specific process of constructing the graph convolutional neural network GCN is as follows:

[0020] (a) Convert the line network into a graph model in the spectral domain, where the line relationships are connections and the sites are edges;

[0021] (b) perform spectral filtering, fast filtering, convolution operations and fast approximation;

[0022] (c) After the fully connected layer reverse learning adjusts the parameters, the predicted value is outputted through multiple iterations.

[0023] In step (2), the input features designed using the graph convolutional neural network (GCN) structure include historical features, time features, road section features, and traffic features. The network polynomial parameters are more than 5 layers. The Relu activation function is added, and cross-validation is used to regularize the parameters to construct a multi-layer graph convolutional neural network.

[0024] In step (3), the LSTM is divided into an input gate, a forget gate, and an output gate inside the neuron. The forget gate reads the hidden layer state h at the previous moment. t-1 , current input state x t , and then output the result to C t Assignment, C t The value indicates whether the neural unit information is retained or discarded. The specific algorithm is as follows: t =σ(W f ·[h t-1 ,x t ]+b f )

[0025] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0026] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0027] h t =o t tanh(C t )

[0028] Where: f t is the output value of the forget gate, i t is the input value, o t is the output value, b i ,b f ,b o are all bias constants, h t is the final output value of the LSTM neural unit, σ is the sigmoid activation function, W f ,W i ,W o is the weight matrix.

[0029] The steps of the LSTM training process in step (3) are as follows:

[0030] (a) Initialize all neuron parameters;

[0031] (b) Forward calculation to obtain the output value of LSTM neuron;

[0032] (c) Reversely calculate the neuron error value and perform reverse iteration to obtain the weight parameter gradient;

[0033] (d) Use the stochastic gradient descent method to continuously iterate and update the parameters to the set error range.

[0034] Construct LSTM input data and divide it into training set and test set for normalization. Construct input layer, hidden layer and output layer. Use full connection form and add Dropout layer to complete the prediction of rail transit passenger flow through training.

[0035] In step (4), a risk-based duration model is used to perform regression analysis on the data through the logistic survival analysis distribution function, and the evaluation is performed based on the minimum Akaike information criterion. The risk function and survival function expressions of the logistic accelerated failure model are:

[0036]

[0037] in, is the characteristic factor variable, is a parameter, γ represents the shape parameter of the benchmark function, and the duration value of the final predicted state is obtained according to the above formula, where t is the duration value.

[0038] In step (5), Fisher optimal segmentation is used to process the time series of urban rail passenger flow data, and passenger flow data with close correlation is classified and counted to form a cross-sectional passenger flow classification.

[0039] Using Fisher's optimal segmentation includes the following steps:

[0040] (a) Calculate the class diameter matrix of sample data to determine the intra-class differences and similarities;

[0041] (b) Minimum classification loss function to achieve optimal data segmentation;

[0042] (c) Dynamic recursive optimization is used to obtain the classification results of cross-sectional passenger flow data.

[0043] In step (6), the passenger flow is divided into six traffic flow states based on the passenger load factor threshold interval: extremely congested, moderately congested, slightly congested, basically unobstructed, unobstructed, and completely unobstructed.

[0044] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: after preprocessing the historical passenger flow data, the spatial and temporal characteristics of the passenger flow are extracted by combining the graph convolutional neural network GCN and the long short-term memory network LSTM, and then a risk-based duration full-parameter prediction model is established for the passenger flow, a state sequence is constructed, and the predicted state value and duration of the urban rail network are obtained, which improves the accuracy of passenger flow prediction, ensures the reliability of driving scheduling, and provides a more accurate basis for the safety management and operation organization decision-making of urban rail transit. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is the structure diagram of the graph convolutional neural network GCN;

[0046] Figure 2 This is the structure diagram of the long short-term memory network LSTM;

[0047] Figure 3 It is an operation flow chart of the present invention. DETAILED DESCRIPTION

[0048] The technical solution of the present invention is further described below in conjunction with embodiments.

[0049] Urban rail transit passenger flow state prediction method based on spatiotemporal characteristics, such as Figure 1-3 As shown. The following steps are included:

[0050] Step 1: Obtain AFC historical passenger flow as input data, pre-process the urban rail network passenger flow data, mainly clean up data outliers and abnormal points, and then perform noise denoising; take 30 minutes as the statistical granularity, calculate the number of passengers getting on and off the line network through the OD distribution profile matrix table to obtain the urban rail network OD distribution probability matrix and section passenger flow;

[0051] Step 2: Extract node edge information from the urban rail network graph to obtain a line adjacency topology graph, obtain a node adjacency matrix based on the topology graph, and use the passenger flow of the urban rail network section as the input of the graph convolutional neural network GCN;

[0052] Step 2-1: Convert the line network into a graph model in the spectral domain, where the line relationships are connections and the sites are edges;

[0053] Step 2-2: perform spectral filtering, fast filtering, convolution operation and fast approximation;

[0054] Step 2-3: After reverse learning of the fully connected layer, adjust the parameters, iterate multiple times to output the predicted values ​​and obtain the spatial features.

[0055] Step 3: Use the historical passenger flow with extracted spatial information as the input of the long short-term memory network LSTM to build an urban rail transit OD passenger flow prediction model based on multiple factors of time and space;

[0056] Step 3-1: Initialize all neuron parameters;

[0057] Step 3-2: Forward calculation to obtain the LSTM neuron output value;

[0058] Step 3-3: Reversely calculate the neuron error value and perform reverse iteration to obtain the weight parameter gradient;

[0059] Step 3-4: Use the stochastic gradient descent method to continuously iterate and update the parameters to the set error range to obtain the passenger flow of the urban rail network at the next time granularity.

[0060] Step 4: Based on the predicted urban rail transit passenger flow, a risk-based duration full-parameter model in survival analysis statistics is further established. The data is regressed through the Logistic survival analysis distribution function and evaluated based on the minimum Akaike information criterion. The risk function and survival function expressions of the Logistic accelerated failure model are:

[0061]

[0062] in, is the characteristic factor variable, is a parameter, γ represents the shape parameter of the benchmark function, and the duration value of the final predicted state is obtained according to the above formula.

[0063] Step 5: Construct the section passenger flow time series based on the prediction results and historical data, using the Fisher ordered sample clustering algorithm;

[0064] Step 5-1: Calculate the class diameter matrix of the sample data to determine the intra-class differences and similarities;

[0065] Step 5-2: Minimum classification loss function to achieve optimal data segmentation;

[0066] Step 5-3: Use dynamic recursive optimization to obtain the classification results of section passenger flow data and form the best classification of section passenger flow rates in urban rail network intervals.

[0067] Step 6: Based on the optimal segmentation statistics of the section passenger flow data, the section full load rate is obtained, and the passenger flow state sequence of the urban rail network is formed according to the full load rate threshold interval, which is mainly divided into six traffic flow states: extremely crowded, moderately crowded, slightly crowded, basically unobstructed, unobstructed and completely unobstructed. Finally, the predicted state and duration of the urban rail network are obtained.

Claims

1. A method for predicting the passenger flow state of urban rail transit based on spatiotemporal characteristics, characterized in that: The steps include: (1) Obtain the historical AFC passenger flow as input data, perform preprocessing on the urban rail passenger flow data, and obtain the OD distribution probability matrix and section passenger flow of the urban rail network; (2) The node edge information of the urban rail network graph is extracted to obtain the line adjacency topology graph, and the node adjacency matrix is ​​obtained based on the topology graph. The urban rail network section passenger flow is used as the input of the graph convolutional neural network (GCN) to extract the spatial characteristics of the urban rail network passenger flow data; (3) The passenger flow output by GCN is used as the input of the long short-term memory network (LSTM) to build a rail transit OD passenger flow prediction model based on time and space factors to obtain the passenger flow of the urban rail network at the next time granularity; (4) Based on the predicted passenger flow of the urban rail network, a risk-based duration full-parameter model in survival analysis statistics is established to explore the distribution pattern of the next duration of the predicted passenger flow data; The risk-based duration full parameter model uses the Logistic survival analysis distribution function to perform regression analysis on the data and uses the minimum Akaike information as the criterion for evaluation. The risk function and survival function expressions of the Logistic accelerated failure model are: in, is the characteristic factor variable, is a parameter, γ represents the shape parameter of the benchmark function, and the duration value of the final predicted state is obtained according to the above formula, where t is the duration value; (5) Based on the predicted results of the subsequent time distribution law of passenger flow data and the historical passenger flow data, the section passenger flow time series is constructed. The Fisher ordered sample clustering algorithm is used to optimally segment the ordered data stream, process the time series of urban rail passenger flow data, classify and count the passenger flow data with close correlation, and form the best classification of the section passenger flow rate of the urban rail network interval; The optimal segmentation comprises the following steps: (a) Calculate the class diameter matrix of sample data to determine the intra-class differences and similarities; (b) Minimum classification loss function to achieve optimal data segmentation; (c) Using dynamic recursive optimization, we can obtain the classification results of the cross-section passenger flow data; (6) Based on the optimal segmentation statistics of the section passenger flow data, the section full load rate is obtained, and the passenger flow state sequence of the urban rail network is formed according to the full load rate threshold interval, and the predicted state and duration of the urban rail network are obtained; The predicted status of the urban rail network is: based on the passenger load rate threshold interval, the passenger flow is divided into six traffic flow states: extremely congested, moderately congested, slightly congested, basically unobstructed, unobstructed and completely unobstructed.

2. The urban rail transit passenger flow state prediction method based on spatiotemporal characteristics according to claim 1, characterized in that: In step (1), the preprocessing includes the following steps: data cleaning of outliers and abnormal points in passenger flow data, and then noise denoising.

3. The urban rail transit passenger flow state prediction method based on spatiotemporal characteristics according to claim 1, characterized in that: In step (1), the statistical granularity of the urban rail network is 30-60 minutes, and the passenger boarding and alighting volume of the network is calculated through the OD distribution profile matrix table, and the cross-sectional passenger flow is obtained in combination with the OD distribution information.

4. The urban rail transit passenger flow state prediction method based on spatiotemporal characteristics according to claim 1, characterized in that: In step (2), the specific process of constructing the graph convolutional neural network GCN is as follows: (a) Convert the line network into a graph model in the spectral domain, where the line relationships are connections and the sites are edges; (b) performing spectral filtering, fast filtering, convolution operations, and fast approximation; (c) After the fully connected layer reverse learning adjusts the parameters, the predicted value is outputted through multiple iterations.

5. The urban rail transit passenger flow state prediction method based on spatiotemporal characteristics according to claim 1, characterized in that: In step (2), the input features of the graph convolutional neural network (GCN) structure design include historical features, time features, road section features and traffic features. The network polynomial parameters are more than 5 layers. The Relu activation function is added, and cross-validation is used to regularize the parameters to construct a multi-layer graph convolutional neural network.

6. The urban rail transit passenger flow state prediction method based on spatiotemporal characteristics according to claim 1, characterized in that: The steps of the LSTM training process in step (3) are as follows: (a) Initialize all neuron parameters; (b) Forward calculation to obtain the output value of LSTM neuron; (c) Reversely calculate the neuron error value and perform reverse iteration to obtain the weight parameter gradient; (d) Use the stochastic gradient descent method to continuously iterate and update the parameters to the set error range.

Images