CN117217445A - Method and device for predicting freight demand of multiple railway stations - Google Patents
Method and device for predicting freight demand of multiple railway stations Download PDFInfo
- Publication number
- CN117217445A CN117217445A CN202311067340.1A CN202311067340A CN117217445A CN 117217445 A CN117217445 A CN 117217445A CN 202311067340 A CN202311067340 A CN 202311067340A CN 117217445 A CN117217445 A CN 117217445A
- Authority
- CN
- China
- Prior art keywords
- time
- space
- freight
- diagram
- railway
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000010586 diagram Methods 0.000 claims abstract description 108
- 239000011159 matrix material Substances 0.000 claims abstract description 58
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 37
- 230000005540 biological transmission Effects 0.000 claims description 12
- 230000001364 causal effect Effects 0.000 claims description 6
- 238000012384 transportation and delivery Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 239000000470 constituent Substances 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000012549 training Methods 0.000 description 10
- 230000006870 function Effects 0.000 description 9
- 238000013528 artificial neural network Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 3
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000007906 compression Methods 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000000547 structure data Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 241001536358 Fraxinus Species 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method and a device for predicting the freight demands of multiple railway stations, wherein the predicting method comprises the following steps: s1, acquiring a railway freight network topological graph, and constructing a railway freight network time-space diagram based on the railway freight network topological graph; the space-time diagram comprises a time sequence and a space matrix, wherein the time sequence is a freight characteristic quantity sequence of all stations in the railway freight network topological diagram in a preset historical time step, and the space matrix is a matrix used for representing the space characteristics of the railway freight network; s2, inputting the space-time diagram into a pre-trained space-time diagram convolutional neural network to obtain freight characteristic quantities of all stations in a topological diagram of a future preset time step. The technical problem that the influence of space-time characteristics of a railway topology network on prediction precision is ignored in the traditional railway freight demand is solved.
Description
Technical Field
The invention relates to the technical field of railway freight demand prediction, in particular to a method and a device for predicting railway multi-station freight demand.
Background
The railway freight has the advantages of large freight volume, high speed, energy saving and the like, occupies a great proportion in a transportation system, has a low 2022 year ago, has an operating mileage of 15.5 km, and has a total delivery capacity of 49.84 hundred million tons in a national railway freight year. The scientific prediction of the development trend of the freight traffic of the railway has positive practical significance for the adjustment of the railway transportation structure, the optimization of the social resource allocation and the improvement of the railway freight management level.
However, railway freight is less active than passenger transport, the freight needs to be manually negotiated and ordered, and complex processes such as transportation, transfer and classification are required to be carried out in the transportation process, so that the transportation service cannot be directly and simply purchased and automatically arrived and departed like passengers, and the transportation process is greatly influenced by unreliability, which makes the establishment of railway freight schedules extremely difficult. In addition, the railway freight requirement has complex characteristics of multiple scattered and concentrated freight requirements, multiple-class freight classification and assembly, multiple-destination batch transportation and the like, so that the railway freight requirement overall shows nonlinear variation. The current freight demand prediction mainly comprises medium-long-term prediction based on a time sequence prediction method, wherein the prediction mainly comprises coarse granularity prediction taking quarterly and year as time units, and the fine granularity prediction on the short-term freight demand of a station is lacking; meanwhile, due to the internal coupling relation between freight station sites, the freight flow transportation is influenced by the space-time characteristics of different sites, so that the fusion of the space-time characteristics in freight flow data plays an important role in improving the prediction precision of freight demand.
For this reason, there is a need for an improved method of predicting railroad freight demand.
Disclosure of Invention
First, the technical problem to be solved
In view of the above-mentioned shortcomings and disadvantages of the prior art, the present invention provides a method and apparatus for predicting a railway multi-station freight demand, which solve the technical problem that the traditional railway freight demand ignores the influence of space-time characteristics of a railway topology network on prediction accuracy.
(II) technical scheme
In order to achieve the above purpose, the main technical scheme adopted by the invention comprises the following steps:
in a first aspect, the present invention provides a method for predicting freight demand for multiple railway stations, comprising:
s1, acquiring a railway freight network topological graph, and constructing a railway freight network time-space diagram based on the railway freight network topological graph; the space-time diagram comprises a time sequence and a space matrix, wherein the time sequence is a freight characteristic quantity sequence of all stations in the railway freight network topological diagram in a preset historical time step, and the space matrix is a matrix used for representing the space characteristics of the railway freight network;
s2, inputting the space-time diagram into a pre-trained space-time diagram convolutional neural network to obtain freight characteristic quantities of all stations in a topological diagram of a future preset time step;
the space-time diagram convolutional neural network comprises at least one space-time convolutional module and one output layer which are sequentially connected, wherein a time sequence is input into a first space-time convolutional module, a space matrix is respectively input into each space-time convolutional module, the space-time convolutional module outputs a convolutional time sequence as the input of the next space-time convolutional module connected with the space-time convolutional module, and the output layer maps the convolutional time sequence output by the last space-time convolutional module to freight characteristic values of all stations in a space-time diagram of a future preset time step;
the space-time convolution module comprises a first gating time convolution layer, a space diagram convolution layer and a second gating time convolution layer which are sequentially connected, and the space matrix is input into the space diagram convolution layer.
Optionally, the constructed railway freight network time space diagram isWherein (1)>Representing freight characteristic quantity of n stations at time t in a railway freight network topology map, epsilon representing connectivity between stations in the railway freight network topology map, and +.>Representation->Is a weighted adjacency matrix,/>Representing a real number;
the weighted adjacency matrix W is obtained according to interval distances among n stations in the railway freight network topological graph, and the expression is as follows:
wherein d ij Is the distance between sites i and j; sigma (sigma) 2 And λ is the distribution threshold and sparsity threshold of W, respectively; omega ij Is an element in W, represented by d ij The weight of the edge determined.
Optionally, the shipment characteristic includes shipment delivery and truck loading rate.
Optionally, between S1 and S2 further comprises:
s11, preprocessing a time sequence; the pretreatment step comprises the following steps:
a1, replacing a value with zero freight transmission capacity in the time sequence by adopting an average value of freight transmission capacity in the time sequence, and replacing a value with zero freight transmission capacity in the time sequence by adopting an average value of freight full rate in the time sequence;
a2, carrying out normalization processing on the independent variable of the time sequence;
s2, inputting the preprocessed time sequence and the preprocessed space matrix into a pre-trained space-time diagram convolutional neural network.
Optionally, the time-gated convolutional layer is composed of a one-dimensional causal convolutional kernel and a nonlinear gated linear unit.
Optionally, the expression of the spatial convolution layer is:
wherein Θ is a graph convolution kernel; x is the input signal of the spatial convolution layer;vector representing polynomial coefficients of chebyshev polynomial limiting Θ to Λ,/->I n Is a unit matrix; />Is an angle matrix, wherein D ij ＝∑ j W ij ；/>Meaning that x has C i A signal of the channel; c (C) 0 An output channel which is a space convolution layer; y is j Having C as a spatial convolution layer j An output of the channel;
the gating time convolution layer has the expression:
wherein,is a convolution kernel; />Is represented as having C i The input signal of the channel, M is the length of the input sequence of the space-time convolution module; p, Q are the inputs of gates in the gated linear units, P being related to the constituent structure and dynamic variance in the time series, σ (Q) controlling the corresponding input values of the current state; the product of Hadamard of the element is indicated.
Optionally, the expression of the spatio-temporal convolution module is:
wherein,representing the input of a space-time convolution module I, M is the length of an input sequence of the space-time convolution module, n is the number of stations in a railway freight network topological graph, and C l An input channel of the space-time convolution module l;representing the output, K, of the spatio-temporal convolution module l t Is the width of a one-dimensional causal convolution kernel, C l+1 The output channel of the space-time convolution module l is also the input channel of the real-time space-time convolution module l+1; />Is a first time convolution kernel within the space-time convolution module l; />Is a second time convolution kernel within the spatio-temporal convolution module l; theta (theta) l The method comprises the steps of (1) performing graph convolution kernel for a space-time convolution module; reLU (·) is a linear activation function.
Optionally, the mapping, by the output layer, the convolution time sequence output by the last space-time convolution module to freight feature quantities of all stations in a space-time diagram of a preset time step in the future includes:
wherein,representing the output of the last spatio-temporal convolution module; />Is a weight vector; b is the deviation.
Optionally, the space-time convolutional neural network comprises two space-time convolutional modules and one output layer which are sequentially connected.
Optionally, before S1, the method further includes: receiving a railway freight network time space diagram for training, and generating a training sample; based on the loss function, the training samples are input into a space-time diagram convolutional neural network for model training.
In a second aspect, the present invention provides a railway multi-station freight demand prediction apparatus comprising:
the acquisition module is used for acquiring a railway freight network topological graph and constructing a railway freight network time-space diagram based on the railway freight network topological graph; the space-time diagram comprises a time sequence and a space matrix, wherein the time sequence is a freight characteristic quantity sequence of all stations in the railway freight network topological diagram in a preset historical time step, and the space matrix is a matrix used for representing the space characteristics of the railway freight network;
the prediction module is used for inputting the time sequence and the space matrix into a pre-trained space-time diagram convolutional neural network to obtain freight characteristic quantities of all stations in a space-time diagram of a preset time step in the future;
the space-time diagram convolutional neural network comprises at least one space-time convolutional module and one output layer which are sequentially connected, wherein a time sequence is input into the first space-time convolutional module, a space matrix is respectively input into each space-time convolutional module, the space-time convolutional module outputs a convolutional time sequence as the input of the next space-time convolutional module connected with the space-time convolutional module, and the output layer maps the convolutional time sequence output by the last space-time convolutional module to freight characteristic quantities of all stations in a space-time diagram of a future preset time step; the space-time convolution module comprises a first gating time convolution layer, a space diagram convolution layer and a second gating time convolution layer which are sequentially connected, and the space matrix is input into the space diagram convolution layer.
(III) beneficial effects
The beneficial effects of the invention are as follows:
according to the railway multi-station freight demand prediction method, the railway freight network time-space diagram is constructed based on the railway freight network topological diagram, so that the time-space characteristics of the railway freight network topology can be extracted, the time-space diagram is input into the time-space diagram convolutional neural network to predict freight demands of stations in the topological diagram, the influence of the railway topological network time-space characteristics on station freight demand prediction is considered, and the prediction accuracy is higher.
Drawings
FIG. 1 is a flow chart of a method for predicting freight demand for multiple railway stations according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a space-time convolutional neural network according to an embodiment of the present invention;
FIG. 3 is a railway freight network topology according to an embodiment of the present invention;
FIG. 4 is a adjacency matrix heatmap of a railway network topology;
FIG. 5 is a schematic diagram of a shipping volume sequence for the east well set station of FIG. 3 over a historical time step;
FIG. 6 is a schematic diagram of a truck loading rate sequence for the east well set station of FIG. 3 over a historical time step.
Detailed Description
The invention will be better explained by the following detailed description of the embodiments with reference to the drawings.
According to the railway multi-station freight demand prediction method provided by the embodiment of the invention, the railway freight network time-space diagram is constructed based on the railway freight network topological diagram, so that the time-space characteristics of the railway freight network topology can be extracted, the time-space diagram is input into the time-space diagram convolutional neural network to predict the freight demands of stations in the topological diagram, the influence of the railway topological network time-space characteristics on the station freight demand prediction is considered, and the prediction accuracy is higher.
In order that the above-described aspects may be better understood, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
Fig. 1 is a flow chart of a method for predicting the freight demand of multiple railway stations according to the present embodiment. As shown in fig. 1, the method for predicting the freight demand of multiple railway stations according to the present embodiment includes the following steps:
step S1, acquiring a railway freight network topological graph, and constructing a railway freight network time-space diagram based on the railway freight network topological graph; the space-time diagram comprises a time sequence and a space matrix, wherein the time sequence is a freight characteristic quantity sequence of all stations in the railway freight network topological diagram in a preset historical time step, and the space matrix is a matrix used for representing the space characteristics of the railway freight network.
The railway freight network topology graph is a network graph constructed by taking stations as nodes and taking intervals as edges, as shown in fig. 3.
Preferably, the time-space diagram of the railway freight network is constructed as followsWherein (1)>Is a limited vertex set and represents freight characteristic quantity of n stations at time t in a railway freight network topological graph; epsilon represents connectivity between sites in the railway freight network topology; />Representation->Is a weighted adjacency matrix,/>Representing a real number. Thus, the structured railway freight network time space diagram focuses on the structured freight time sequence, v t Are not independent but are connected by pairs of edges in the topology.
Further, the weighted adjacency matrix W is obtained according to the interval distance among n stations in the railway freight network topological graph, and the expression is:
wherein d ij Is the distance between sites i and j; sigma (sigma) 2 And λ is the distribution threshold and sparsity threshold of W, respectively; omega ij Is an element in W, represented by d ij The weight of the edge determined. Further preferably, sigma 2 10 and lambda of 0.5.
Preferably, the shipment characteristic amount includes a shipment transmission amount and a truck loading rate. I.e.Each node of (a) has two shipping characteristics: freight delivery and truck loading rate. Wherein the "freight delivery amount" corresponds to the total weight of the freight delivered by the station in a specific time, generally in tons, and the "freight rate" corresponds to the ratio of the actual weight of the loaded freight in the railway freight car delivered by the station in a specific time to the maximum freight weight that the freight car can accommodate, expressed as a percentage. In this way, in this embodiment, the freight transmission amount and the truck loading rate in the actual physical topological graph are used as node characteristics, so that space-time correlation between different nodes (stations) is sought, and further the freight transmission amount is predicted.
Further preferably, the "freight delivery amount" corresponds to the total weight of the freight delivered by the station in one day, and the "freight full rate" corresponds to the ratio of the actual weight of the loaded freight in the railway freight car delivered by the station in one day to the maximum freight weight that the freight car can accommodate. Therefore, the problem of coarse granularity of the traditional prediction space-time is solved by taking the fine granularity time sequence as the data base of prediction, and the short-term demand of the station freight can be predicted.
Step S11, preprocessing the time sequence.
Preferably, the step of pre-treating comprises:
a1, replacing a value with zero freight transmission quantity in the time sequence by using an average value of freight transmission quantity in the time sequence, and replacing a value with zero freight transmission quantity in the time sequence by using an average value of freight full rate in the time sequence. Thus, the problem of excessive zero values in the historical data is solved.
A2, carrying out normalization processing on the independent variable of the time sequence. In this way, the influence of different dimensions on the prediction accuracy can be eliminated.
Further preferably, the independent variable of the time series is Z-Score normalized.
And S2, inputting the space-time diagram into a pre-trained space-time diagram convolutional neural network to obtain freight characteristic quantities of all stations in a topological diagram of a future preset time step.
The space-time diagram convolutional neural network comprises at least one space-time convolutional module and one output layer which are sequentially connected, wherein a time sequence is input into the first space-time convolutional module, a space matrix is respectively input into each space-time convolutional module, the space-time convolutional module outputs the convolved time sequence as the input of the next space-time convolutional module connected with the space-time convolutional module, and the output layer maps the convolved time sequence output by the last space-time convolutional module to freight characteristic quantities of all stations in a space-time diagram of a preset time step in the future; the space-time convolution module comprises a first gating time convolution layer, a space diagram convolution layer and a second gating time convolution layer which are sequentially connected, and the space matrix is input into the space diagram convolution layer.
In the space-time diagram convolutional neural network provided by the embodiment, each space-time convolution module comprises a three-layer structure, namely an upper layer and a lower layer are gating time convolution layers for extracting time features, and a middle layer is a space diagram convolution layer for extracting space features, so that a sandwich structure is formed. The space layer is connected with the two time layers, so that the space state from graph convolution to time convolution can be rapidly propagated; the sandwich structure is beneficial to the neural network to fully apply bottleneck strategies, and the channels are downscaled and upscaled through the graph convolution layer, so that the scale compression and the feature compression are realized. Therefore, the space-time diagram is input into the space-time diagram convolutional neural network, useful space-time information is extracted from the space-time diagram by applying the convolutional theorem and the map theory, the freight demand of the station in the topological diagram is predicted, the influence of the space-time characteristics of the railway topological network on the freight demand prediction of the station is considered, and the prediction accuracy is higher.
Specifically, layer normalization is used in each spatio-temporal convolution module to prevent overfitting.
Preferably, in the space-time convolution module, the expression of the spatial convolution layer is:
wherein Θ is a graph convolution kernel; x is the input signal of the spatial convolution layer;vector representing polynomial coefficients of chebyshev polynomial limiting Θ to Λ,/->I n Is a unit matrix; />Is an angle matrix, wherein D ij ＝∑ j W ij ；/>Meaning that x has C i A signal of the channel; c (C) 0 An output channel which is a space convolution layer; y is j Having C as a spatial convolution layer j The output of the channel.
Specifically, the derivation process of the spatial convolution layer expression is as follows.
Graph convolution is used as a signal by introducing a concept of a graph convolution operator' based on a concept of spectrogram convolution (utilizing a graph theory)Product of the product with the convolution kernel Θ:
Θ*x＝Θ(L)x＝Θ(UΛU T )x＝UΘ(Λ)U T x (1)；
wherein the method comprises the steps ofI n Is an identity matrix>Is an angle matrix, D ij ＝∑ j W ij ；/>Is a eigenvector matrix of the normalized graph Laplace matrix L; u (U) T Is the transposition of U; />Is the diagonal matrix of eigenvalues of L; Θ (Λ) is the angle matrix.
The graph signal x is filtered by the kernel Θ and fourier transformed at Θ and graph U according to the definition of equation (1) T Multiplying x, calculating the graph convolution kernel theta by the formula (1), wherein the complexity of the method is O (n) 2 ) Too time consuming, two approximation strategies are applied to overcome this problem, namely Chebyshev polynomial approximation and first order approximation.
The basic idea of Chebyshev polynomial approximation is to locate the filter and reduce the number of parameters, limiting the graph convolution kernel Θ to polynomials of Λ, such asWherein (1)>Is a vector of polynomial coefficients, K is the kernel size of the graph convolution, determining the maximum radius from the center node to the convolution. Conventionally, chebyshev polynomials T k (x) The kernel can be approximated as a truncated expansion of the K-1 order, e.g. +.>Rescaling->λ max Representing the maximum eigenvalue of L. Thus, applying the Chebyshev polynomial approximation strategy, equation (1) can be transformed into:
wherein,is in scaling Laplace operator +.>The K-th order chebyshev polynomial calculated below.
By recursively computing the K partial convolution with polynomial approximation, the complexity of equation (2) can be reduced. Due to scaling and normalization in neural networks, λ can be further assumed max Approximately 2, equation (2) can be reduced to:
wherein θ 0 、θ 1 Is two shared parameters of the graph convolution kernel.
To constrain the parameters and stabilize the numerical performance, θ 0 And theta 1 Replaced by a single parameter θ, let θ=θ 0 ＝-θ 1 The method comprises the steps of carrying out a first treatment on the surface of the W and D are respectively passed throughAnd->And (5) renormalizing. The graph convolution can be expressed as:
the information of the (K-1) order neighborhood from the central node is utilized to realize the graph convolution stack with the first order approximation, and the effect similar to the K local convolution level is realized. In this case, K is the number of continuous filtering operations or convolution layers in the model. Furthermore, hierarchical linear structures are parametrically economical and efficient for large scale graphs, as the approximate order is limited to 1. Is defined inThe above volume integrator "x" can be extended to multidimensional tensors, for having C i Signal of channel->The graph convolution can be summarized as:
wherein,with C i ×C o Vector, C i ×C o Input of feature map respectivelyAnd the magnitude of the output, and,
preferably, the time-gated convolutional layer consists of a one-dimensional causal convolutional kernel and a nonlinear gated linear unit.
Preferably, in the space-time convolution module, the gating time convolution layer has the expression:
wherein,for convolution kernel, it is intended to map the input Y to a single output element +.>P, Q is divided into two halves, and the sizes are the same; />Is represented as having C i The input signal of the channel, M is the length of the input sequence of the space-time convolution module; p, Q are the inputs of gates in the gated linear units, P being related to the constituent structure and dynamic variance in the time series, σ (Q) controlling the corresponding input values of the current state; the product of Hadamard of the element is indicated.
Preferably, the expression of the spatio-temporal convolution module is:
wherein the input and output of the space-time convolution module are 3-D tensors;representing the input of the space-time convolution module, M is the length of the input sequence of the space-time convolution module, and n is the railway freightNumber of stations in network topology, C l The number of input channels is the space-time convolution module l; />Representing the output, K, of the spatio-temporal convolution module l t Is the width of a one-dimensional causal convolution kernel, C l+1 The output channel of the space-time convolution module l is also the input channel of the real-time space-time convolution module l+1; />Is a first time convolution kernel within the space-time convolution module l; />Is a second time convolution kernel within the spatio-temporal convolution module l; theta (theta) l The method comprises the steps of (1) performing graph convolution kernel for a space-time convolution module; reLU (·) is a linear activation function.
As one example, the expression for the linear activation function is:
where x is the input signal.
Preferably, channel c=16 of the spatial convolution layer and channel c=64 of the gating time convolution layer. In this way, the space-time diagram convolutional neural network prediction is made more accurate.
Preferably, the space-time convolutional neural network comprises two space-time convolutional modules and one output layer which are connected in sequence. In this way, the space-time diagram convolutional neural network prediction is made more accurate.
Preferably, the mapping, by the output layer, the convolution time sequence output by the last space-time convolution module to freight feature quantities of all stations in a space-time diagram of a preset time step in the future includes:
wherein,representing the output of the last spatio-temporal convolution module; />Is a weight vector; b is the deviation.
Preferably, before step S1, the method further comprises: receiving a railway freight network time space diagram for training, and generating a training sample; based on the loss function, the training samples are input into a space-time diagram convolutional neural network for model training.
Specifically, training a time space graph convolutional neural network model according to preset iteration step number, batch size, learning rate, learner and other parameters, and manually adjusting model parameters according to loss results to enable loss values to be converged and at optimal values.
Preferably, the expression of the loss function is:
wherein W is θ Is all trainable parameters in the model; v t+1 Is a true value;is a predictive value of the model.
According to the railway multi-station freight demand prediction method provided by the embodiment, after the railway freight demand is severely increased, the freight flow to be increased in a specific area and a station is predicted in time, trucks are allocated in time and grouped in time, and an order of an employer is completed according to the points in time; when the railway freight demand is reduced, the idle trucks are allocated in time to participate in other transportation orders, so that the utilization rate of railway transportation resources is maximized; for railway freight dispatchers, the change condition of future freight flows can be predicted in advance, errors caused by manual work under the condition of hasty handling are avoided, the formulation of freight grouping strategies is more reasonable and intelligent, and the high intelligence of railway freight management level is improved. The auxiliary dispatching department timely senses the fluctuation of future freight demands, timely adjusts the freight train grouping plan to meet new freight demands, improves the capacity of freight organizations, maximizes the utilization rate of train resources and promotes the development of the railway system for adjusting the traffic flow to process the traffic flow from passive reaction to active decision.
To measure and evaluate the performance of the rail multi-station freight demand prediction method proposed in this embodiment, L is selected 2 The loss function was used as the loss function of the model and the space-time diagram convolutional neural network model proposed in this example was compared with the following reference model:
(1) Attention network (Graph attention network, GAT)
Importance among nodes of the graph structure data is allocated by using an attention mechanism, and the graph structure data belongs to graph neural network representation based on airspace.
(2) Convolutional neural network of figure (Graph Convolutional Neural Network, GCN)
The graph convolution neural network is a graph neural network representation based on a spectral domain, after graph embedding representation learning is carried out on graph data, the problem of high complexity caused by direct application of a convolution idea in the traditional CNN to the graph embedding representation data is solved, and the graph convolution neural network is a further version of a chebnet model.
(3) Chebnet model
The model is an improvement to the convolution kernel using chebyshev polynomials (ChebShev Polynomial) and demonstrates that computational complexity can be reduced.
Specifically, as the index for measuring and evaluating the model performance, average absolute error (Mean Absolute Error, MAE), average absolute percent error (Mean Absolute Percentage Error, MAPE), and root mean square error (Root Mean Square Error, RMSE) are used, and the formula is as follows:
wherein y is i Representing the actual value;representing the predicted value; n represents the number of predictions; the smaller the RMSE, MAPE and MAE values, the smaller the error between the predicted and actual values; note that y cannot be 0.
Specifically, the model performance is evaluated, when a comparison test is performed, parameters such as iteration step number, batch size, learning rate, learner and the like are required to be set uniformly, and then a model result is obtained by analyzing MAE, MAPE, RMSE and time consuming procedures and analyzing a comparison result.
The model performance will be evaluated by comparative experiments in conjunction with railway specific networks. 20 interconnected stations are selected to construct a railway freight network topology as shown in fig. 3, and a weighted adjacency matrix of the railway freight network topology is shown in fig. 4. 20 station shipment data from day-to-day records of 2021, 1-1, 12-31, 2022 were used and named "daqin dataset". The main freight characteristics considered in this data are freight volume and full rate. To ensure data integrity and interpretability, an Dongjin set station in the Dafraxinus dataset was chosen as a representative example, as shown in FIGS. 5 and 6.
It should be noted that due to the limitations of the acquisition system, the shipping data may not be recorded every day, resulting in many zero values in the raw data. In this embodiment, this problem is solved by replacing the zero value in 759 instances of the large ash dataset with the average of the two years of collected data. During the performance of the simulation experiments of the present invention, the data set of fraxinus was divided into a training set (containing 80% of the data) and a test set (containing 20% of the data). Experiments were performed on a Lennovo computer equipped with Intel (R) Core (TM) i5-6500 CPU operating at 3.20GHz with 4 CPUs and 16GB of memory. For comparison experiments, GAT, GCN and Chebnet were chosen as reference models. In order to reduce the computational complexity, the model parameters of the STGCN model and the comparison model in all simulation experiments are required to be consistent. Specifically, the freight volume of the past 12 time steps (i.e., h=12) was predicted for 1 time step in the future, and experimental analysis and evaluation were performed on the space-time convolutional neural network, GAT, GCN, and Chebnet based on the actual freight data set, and the results are shown in table 1.
Table 1 comparison of freight demand and forecast results for "daqin" actual railway network
As can be seen from table 1, MAE, RMSE, RMAPE of the space-time diagram convolutional neural network model is 0.059, 0.09 and 0.01 respectively, and simulation results show that compared with other reference models, the space-time diagram convolutional neural network model has the highest prediction precision and the lowest time consumption. The space-time diagram convolutional neural network model can be used for capturing the inherent complex space-time relationship in the railway freight demand data, and can provide more accurate information for railway freight demand prediction.
The embodiment also provides a railway multi-station freight demand prediction device, which comprises: the acquisition module is used for acquiring a railway freight network topological graph and constructing a railway freight network time-space diagram based on the railway freight network topological graph; the space-time diagram comprises a time sequence and a space matrix, wherein the time sequence is a freight characteristic quantity sequence of all stations in the railway freight network topological diagram in a preset historical time step, and the space matrix is a matrix used for representing the space characteristics of the railway freight network; the prediction module is used for inputting the time sequence and the space matrix into a pre-trained space-time diagram convolutional neural network to obtain freight characteristic quantities of all stations in a space-time diagram of a preset time step in the future. The space-time diagram convolutional neural network comprises at least one space-time convolutional module and one output layer which are sequentially connected, wherein a time sequence is input into the first space-time convolutional module, a space matrix is respectively input into each space-time convolutional module, the space-time convolutional module outputs a convolutional time sequence as the input of the next space-time convolutional module connected with the space-time convolutional module, and the output layer maps the convolutional time sequence output by the last space-time convolutional module to freight characteristic quantities of all stations in a space-time diagram of a future preset time step; the space-time convolution module comprises a first gating time convolution layer, a space diagram convolution layer and a second gating time convolution layer which are sequentially connected, and the space matrix is input into the space diagram convolution layer.
It should be noted that, specific functions of each module in the railway multi-station freight demand prediction device and the railway multi-station freight demand prediction flow provided in this embodiment may refer to the detailed description of the railway multi-station freight demand prediction method provided in the foregoing embodiment, and will not be repeated herein.
In summary, the railway multi-station freight demand prediction device provided by the embodiment of the invention constructs the railway freight network time-space diagram based on the railway freight network topological diagram, can extract the time-space characteristics of the railway freight network topology, inputs the time-space diagram into the time-space diagram convolutional neural network to predict the freight demands of stations in the topological diagram, considers the influence of the time-space characteristics of the railway topological network on the prediction of the freight demands of the stations, and ensures higher prediction accuracy.
In the description of the present invention, it should be understood that the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
In the present invention, unless explicitly specified and limited otherwise, the terms "mounted," "connected," "secured," and the like are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally formed; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium; may be a communication between two elements or an interaction between two elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
In the present invention, unless expressly stated or limited otherwise, a first feature is "on" or "under" a second feature, which may be in direct contact with the first and second features, or in indirect contact with the first and second features via an intervening medium. Moreover, a first feature "above," "over" and "on" a second feature may be a first feature directly above or obliquely above the second feature, or simply indicate that the first feature is higher in level than the second feature. The first feature being "under", "below" and "beneath" the second feature may be the first feature being directly under or obliquely below the second feature, or simply indicating that the first feature is level lower than the second feature.
In the description of the present specification, the terms "one embodiment," "some embodiments," "examples," "particular examples," or "some examples," etc., refer to particular features, structures, materials, or characteristics described in connection with the embodiment or example as being included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
While embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that alterations, modifications, substitutions and variations may be made in the above embodiments by those skilled in the art within the scope of the invention.
Claims (10)
1. A method for predicting freight demand for multiple railway stations, comprising:
s1, acquiring a railway freight network topological graph, and constructing a railway freight network time-space diagram based on the railway freight network topological graph; the space-time diagram comprises a time sequence and a space matrix, wherein the time sequence is a freight characteristic quantity sequence of all stations in the railway freight network topological diagram in a preset historical time step, and the space matrix is a matrix used for representing the space characteristics of the railway freight network;
s2, inputting the space-time diagram into a pre-trained space-time diagram convolutional neural network to obtain freight characteristic quantities of all stations in a topological diagram of a future preset time step;
the space-time diagram convolutional neural network comprises at least one space-time convolutional module and one output layer which are sequentially connected, wherein a time sequence is input into a first space-time convolutional module, a space matrix is respectively input into each space-time convolutional module, the space-time convolutional module outputs a convolutional time sequence as the input of the next space-time convolutional module connected with the space-time convolutional module, and the output layer maps the convolutional time sequence output by the last space-time convolutional module to freight characteristic values of all stations in a space-time diagram of a future preset time step;
the space-time convolution module comprises a first gating time convolution layer, a space diagram convolution layer and a second gating time convolution layer which are sequentially connected, and the space matrix is input into the space diagram convolution layer.
2. The method for predicting freight demand for multiple stations on a railway in accordance with claim 1, wherein,
the constructed railway freight network time-space diagram is thatWherein (1)>Represents freight characteristic quantity of n stations at time t in the railway freight network topology diagram, epsilon represents connectivity among stations in the railway freight network topology diagram,representation->Is a weighted adjacency matrix,/>Representing a real number;
the weighted adjacency matrix W is obtained according to interval distances among n stations in the railway freight network topological graph, and the expression is as follows:
wherein d ij Is the distance between sites i and j; sigma (sigma) 2 And λ is the distribution threshold and sparsity threshold of W, respectively; omega ij Is an element in W, represented by d ij The weight of the edge determined.
3. The railway multi-station freight demand prediction method according to claim 1 or 2, wherein the freight characteristic quantity includes freight delivery quantity and freight full rate.
4. A method of predicting a freight demand for a railway plural station as in claim 3, further comprising between S1 and S2:
s11, preprocessing a time sequence; the pretreatment step comprises the following steps:
a1, replacing a value with zero freight transmission capacity in the time sequence by adopting an average value of freight transmission capacity in the time sequence, and replacing a value with zero freight transmission capacity in the time sequence by adopting an average value of freight full rate in the time sequence;
a2, carrying out normalization processing on the independent variable of the time sequence;
s2, inputting the preprocessed time sequence and the preprocessed space matrix into a pre-trained space-time diagram convolutional neural network.
5. The method for predicting freight demand for multiple stations on a railway in accordance with claim 2, wherein,
the time-gated convolution layer consists of a one-dimensional causal convolution kernel and a nonlinear gated linear unit.
6. The method for predicting freight demand for multiple stations on a railway in accordance with claim 2, wherein,
the expression of the spatial convolution layer is:
wherein Θ is a graph convolution kernel; x is the input signal of the spatial convolution layer;vector representing polynomial coefficients of chebyshev polynomial limiting Θ to Λ,/->I n Is a unit matrix; />Is an angle matrix, wherein D ij ＝∑ j W ij ；/>Meaning that x has C i A signal of the channel; c (C) 0 An output channel which is a space convolution layer; y is j Having C as a spatial convolution layer j An output of the channel;
the gating time convolution layer has the expression:
wherein,is a convolution kernel; />Is represented as having C i The input signal of the channel, M is the length of the input sequence of the space-time convolution module; p, Q are the inputs of gates in the gated linear units, P being related to the constituent structure and dynamic variance in the time series, σ (Q) controlling the corresponding input values of the current state; the product of Hadamard of the element is indicated.
7. The method of claim 5, wherein the expression of the space-time convolution module is:
wherein,representing the input of a space-time convolution module I, M is the length of an input sequence of the space-time convolution module, n is the number of stations in a railway freight network topological graph, and C l The number of input channels is the space-time convolution module l;representing the output, K, of the spatio-temporal convolution module l t Is the width of a one-dimensional causal convolution kernel, C l+1 The output channel of the space-time convolution module l is also the input channel of the real-time space-time convolution module l+1; />Is a first time convolution kernel within the space-time convolution module l; />Is a second time convolution kernel within the spatio-temporal convolution module l; theta (theta) l The method comprises the steps of (1) performing graph convolution kernel for a space-time convolution module; reLU (·) is a linear activation function.
8. The method for predicting freight demand for multiple stations on a railway as set forth in claim 6 or 7, wherein the outputting layer maps the convolution time sequence outputted by the last space-time convolution module to freight feature values of all stations in a space-time diagram of a preset time step in the future, comprising:
wherein,representing the output of the last spatio-temporal convolution module; />Is a weight vector; b is the deviation.
9. The method of claim 1, wherein the space-time convolutional neural network comprises two space-time convolutional modules and one output layer connected in sequence.
10. A railway multi-station freight demand prediction apparatus, comprising:
the acquisition module is used for acquiring a railway freight network topological graph and constructing a railway freight network time-space diagram based on the railway freight network topological graph; the space-time diagram comprises a time sequence and a space matrix, wherein the time sequence is a freight characteristic quantity sequence of all stations in the railway freight network topological diagram in a preset historical time step, and the space matrix is a matrix used for representing the space characteristics of the railway freight network;
the prediction module is used for inputting the time sequence and the space matrix into a pre-trained space-time diagram convolutional neural network to obtain freight characteristic quantities of all stations in a space-time diagram of a preset time step in the future;
the space-time diagram convolutional neural network comprises at least one space-time convolutional module and one output layer which are sequentially connected, wherein a time sequence is input into the first space-time convolutional module, a space matrix is respectively input into each space-time convolutional module, the space-time convolutional module outputs a convolutional time sequence as the input of the next space-time convolutional module connected with the space-time convolutional module, and the output layer maps the convolutional time sequence output by the last space-time convolutional module to freight characteristic quantities of all stations in a space-time diagram of a future preset time step; the space-time convolution module comprises a first gating time convolution layer, a space diagram convolution layer and a second gating time convolution layer which are sequentially connected, and the space matrix is input into the space diagram convolution layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311067340.1A CN117217445A (en) | 2023-08-23 | 2023-08-23 | Method and device for predicting freight demand of multiple railway stations |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311067340.1A CN117217445A (en) | 2023-08-23 | 2023-08-23 | Method and device for predicting freight demand of multiple railway stations |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117217445A true CN117217445A (en) | 2023-12-12 |
Family
ID=89045238
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311067340.1A Pending CN117217445A (en) | 2023-08-23 | 2023-08-23 | Method and device for predicting freight demand of multiple railway stations |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117217445A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113487066A (en) * | 2021-06-18 | 2021-10-08 | 山西云时代智慧城市技术发展有限公司 | Long-time-sequence freight volume prediction method based on multi-attribute enhanced graph convolution-Informer model |
| CN114580617A (en) * | 2022-03-11 | 2022-06-03 | 北京航空航天大学 | National road transport capacity distribution prediction method based on space-time graph convolutional network |
| CN115017809A (en) * | 2022-06-07 | 2022-09-06 | 上海海事大学 | Ship arrival amount prediction system and method |
| CN115223402A (en) * | 2022-06-29 | 2022-10-21 | 北京航空航天大学 | Space domain sector complexity prediction method based on space-time graph convolutional network |
| CN115829158A (en) * | 2022-12-28 | 2023-03-21 | 中铁第一勘察设计院集团有限公司 | Urban rapid freight volume prediction method based on deep learning |
-
2023
- 2023-08-23 CN CN202311067340.1A patent/CN117217445A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113487066A (en) * | 2021-06-18 | 2021-10-08 | 山西云时代智慧城市技术发展有限公司 | Long-time-sequence freight volume prediction method based on multi-attribute enhanced graph convolution-Informer model |
| CN114580617A (en) * | 2022-03-11 | 2022-06-03 | 北京航空航天大学 | National road transport capacity distribution prediction method based on space-time graph convolutional network |
| CN115017809A (en) * | 2022-06-07 | 2022-09-06 | 上海海事大学 | Ship arrival amount prediction system and method |
| CN115223402A (en) * | 2022-06-29 | 2022-10-21 | 北京航空航天大学 | Space domain sector complexity prediction method based on space-time graph convolutional network |
| CN115829158A (en) * | 2022-12-28 | 2023-03-21 | 中铁第一勘察设计院集团有限公司 | Urban rapid freight volume prediction method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| BING YU 等: "Spatio-Temporal Graph Convolutional Networks: A Deep Learning Framework for Traffic Forecasting", pages 2 - 4, Retrieved from the Internet <URL:《https://arxiv.org/pdf/1709.04875.pdf》> * |
| 毛海军: "《物流系统规划与设计 第2版》", 31 January 2017, 南京：东南大学出版社, pages: 125 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wan et al. | CTS-LSTM: LSTM-based neural networks for correlatedtime series prediction | |
| Ping et al. | Impact of driver behavior on fuel consumption: Classification, evaluation and prediction using machine learning | |
| CN113077090B (en) | Passenger flow prediction method, system and computer readable storage medium | |
| Zhang et al. | Network‐wide traffic speed forecasting: 3D convolutional neural network with ensemble empirical mode decomposition | |
| CN111860951A (en) | Rail transit passenger flow prediction method based on dynamic hypergraph convolutional network | |
| Li et al. | Graph CNNs for urban traffic passenger flows prediction | |
| CN116187640B (en) | Power distribution network planning method and device based on grid multi-attribute image system | |
| Fan et al. | Predicting vacant parking space availability: a long short-term memory approach | |
| CN114943356B (en) | Short-time demand integrated prediction method for airport arrival passenger to take taxi | |
| Zhao et al. | A novel hybrid model for short-term high-speed railway passenger demand forecasting | |
| Wang et al. | Deep learning of spatiotemporal patterns for urban mobility prediction using big data | |
| CN114462623B (en) | Data analysis method, system and platform based on edge calculation | |
| CN106650959A (en) | Power distribution network repair ability assessment method based on improved grey clustering | |
| CN114066073A (en) | Power grid load prediction method | |
| CN112308298B (en) | Multi-scenario performance index prediction method and system for semiconductor production line | |
| CN111008726A (en) | Class image conversion method in power load prediction | |
| Jiang et al. | A new form of deep learning in smart logistics with IoT environment | |
| CN116824861A (en) | Method and system for scheduling sharing bicycle based on multidimensional data of urban brain platform | |
| Bao et al. | Forecasting network-wide multi-step metro ridership with an attention-weighted multi-view graph to sequence learning approach | |
| Wu et al. | Forecasting metro rail transit passenger flow with multiple-attention deep neural networks and surrounding vehicle detection devices | |
| Li et al. | Short-term traffic demand prediction using graph convolutional neural networks | |
| Zhu et al. | A novel hybrid deep learning model for taxi demand forecasting based on decomposition of time series and fusion of text data | |
| Rodríguez-Rueda et al. | Origin–Destination matrix estimation and prediction from socioeconomic variables using automatic feature selection procedure-based machine learning model | |
| Ye et al. | Demand forecasting of online car‐hailing by exhaustively capturing the temporal dependency with TCN and Attention approaches | |
| CN117217445A (en) | Method and device for predicting freight demand of multiple railway stations |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |