CN111260121B - Urban-range pedestrian flow prediction method based on deep bottleneck residual error network - Google Patents
Urban-range pedestrian flow prediction method based on deep bottleneck residual error network Download PDFInfo
- Publication number
- CN111260121B CN111260121B CN202010028983.5A CN202010028983A CN111260121B CN 111260121 B CN111260121 B CN 111260121B CN 202010028983 A CN202010028983 A CN 202010028983A CN 111260121 B CN111260121 B CN 111260121B
- Authority
- CN
- China
- Prior art keywords
- data
- prediction
- brbm
- network
- layer
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Probability & Statistics with Applications (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a city-wide pedestrian flow prediction method based on a deep bottleneck residual error network, which comprises the following steps: 1) acquiring original traffic flow data; 2) constructing a BRBM data reconstruction mechanism, inputting the acquired original traffic flow data into the BRBM data reconstruction mechanism, and performing dimensionality reduction and data reconstruction to obtain traffic flow data after BRBM reconstruction; 3) constructing a cooperative prediction mechanism, taking traffic flow data obtained after BRBM reconstruction as input data of the cooperative prediction mechanism, and obtaining a prediction result after the cooperative prediction mechanismX R (ii) a 4) Constructing an auxiliary prediction mechanism, and processing external factors influencing the pedestrian flow by adopting the auxiliary prediction mechanism to obtain a prediction resultX E (ii) a 5) The prediction result to be obtainedX R AndX E and fusing to obtain a final people flow prediction result. The method not only greatly reduces the computational complexity of the people flow prediction model and the time of model training, but also improves the people flow prediction precision.
Description
Technical Field
The invention relates to the technical field of urban pedestrian flow prediction, in particular to an urban range pedestrian flow prediction method based on a deep bottleneck residual error network.
Background
In recent years, with the rapid growth of population and the development of socioeconomic performance, population density has been increasing and traffic flow has been rapidly increasing. A crowd pedaling event and a series of safety accidents due to crowding often occur. Therefore, the method for predicting the rapid crowd gathering caused by various public events and emergencies in advance plays an important role in city planning management and city public safety. With the continuous enrichment of urban data, deep learning has become an effective people flow prediction method. Compared with the traditional prediction method based on machine learning, the deep learning method can be used for mining nonlinear characteristics from traffic flow data and improving the prediction precision. Currently, some research efforts utilize deep learning techniques and traffic flow data to predict human traffic. However, the flow of people is affected not only by temporal and spatial factors, but also by many external factors, such as weather, holidays, activities, and the like. Therefore, prediction methods based on multi-source data (e.g., traffic data, weather data, holiday data) are more competitive. In addition, traffic flow data of a city is usually high-dimensional, which causes more computing resources and time to be spent on training the network model, and the network model is difficult to obtain an optimal solution along with a large amount of high-dimensional input data.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a city-wide people flow prediction method based on a deep bottleneck residual error network.
The technical scheme for realizing the purpose of the invention is as follows:
a city-wide pedestrian flow prediction method based on a deep bottleneck residual error network comprises the following steps:
1) acquiring original traffic flow data;
2) constructing a BRBM data reconstruction mechanism, inputting the original traffic flow data acquired in the step 1) into the BRBM data reconstruction mechanism, and performing dimension reduction and data reconstruction to obtain traffic flow data after BRBM reconstruction;
3) constructing a cooperative prediction mechanism, taking the traffic flow data reconstructed by the BRBM obtained in the step 2) as input data of the cooperative prediction mechanism, and obtaining a prediction result X after passing through the cooperative prediction mechanismR;
4) Constructing an auxiliary prediction mechanism, processing external factors influencing the pedestrian flow by adopting the auxiliary prediction mechanism,obtaining a predicted result XE;
5) The prediction result X obtained in the step 3) is usedRWith X obtained in step 4)EAnd fusing to obtain a final people flow prediction result.
In the step 1), the original traffic flow data is obtained by using urban taxi GPS data, meteorological data and bicycle track data.
In step 2), the BRBM data reconstruction mechanism comprises a BRBM network visible layer and a hidden layer; the node of the hidden layer is used as a nonlinear feature detector for reducing dimension and reconstructing input data, and extracting high-level feature information from original traffic flow data, wherein the data reconstruction process comprises the following steps:
2-1) carrying out normalization processing on input original traffic flow data by adopting a data normalization method, and mapping traffic flow data values of each region of a city between [0 and 1] to obtain inflow and outflow flow data of the city region;
2-2) respectively inputting the normalized urban area inflow and outflow flow data into the BRBM network for dimensionality reduction and reconstruction;
2-2-1) the BRBM network is an energy-based probability distribution model, and for a given state vector h and v, the current energy function of the BRBM network is expressed as:
wi,jrepresenting the weight between the ith neuron of the visible layer and the jth neuron of the hidden layer, aiRepresenting the bias value of the ith neuron in the visible layer, bjA bias value representing the jth neuron in the hidden layer;
2-2-2) based on the energy function obtained in step 2-2-1), then the state of the BRBM network is defined as a given v, and the joint probability distribution of h is:
wherein Z is a normalization factor and the expression is:
2-2-3) obtaining the marginal probability distribution of the pedestrian flow data of the visible layer according to the joint probability distribution formula in the step 2-2-2):
2-2-4) the process of training the BRBM network is to solve a set of parameters θ ═ { w, a, b }, so that the final output of the visual layer has an approximate probability distribution with the input traffic flow data, and to achieve this goal, based on the marginal probability distribution formula of the pedestrian flow data in step 2-2-3), the following loss function is defined:
solving a group of optimal solutions theta, v by minimizing a loss functioniThe people flow data of the ith area of a city, and s is the number of divided areas of the city;
2-2-5) minimizing the loss function of the step 2-2-4), wherein the conditional probability distribution of the input traffic flow data in a visible layer and a hidden layer in the BRBM network is respectively as follows:
wi,jrepresenting the weight between the ith neuron of the visible layer and the jth neuron of the hidden layer, aiRepresenting the bias of the ith neuron in the visible layerValue, bjRepresenting the bias value of the jth neuron in the hidden layer, wherein sigmoid is an activation function, and the expression of the activation function is as follows:
2-2-6) repeating step 2-2-5) k times to represent the result of the k-th sampling, thereby obtaining a probability distribution approximate to the input dataCarrying out forward and reverse training on input data between a visible layer and a hidden layer of the BRBM network, obtaining probability distribution similar to the input data by updating a weight, and completing the training of the BRBM network through k iterations;
2-2-7) extracting important characteristic information from input traffic flow data by the trained BRBM network to obtain traffic flow data reconstructed by the BRBM;
in step 3), the collaborative prediction mechanism is composed of prediction oriented to spatio-temporal data, and the prediction method of the collaborative prediction mechanism comprises the following steps:
3-1) dividing the area of a city into a plurality of P × Q grid maps according to longitude and latitude, each grid representing an area, and setting ri,jIs one of the P × Q grids, for one grid (P)1,q1) E (P, Q) which indicates that the lattice is located at the P-th1Line and q1Row ofEach grid comprises inflow and outflow two types of people flows, and the inflow and outflow of people in a city within a period of time are respectively converted into corresponding two-dimensional people flow matrixes;
3-2) selecting the data of different timestamps to connect together according to the time attribute of the people flow data by the two-dimensional people flow matrix obtained by conversion in the step 3-1), and respectively modeling three time dimensions: one hour, one day, one week, and then inputting corresponding time stamp data into the space-time orientedThe predicted three separate time dimensions of the data. The prediction facing the space-time data is composed of three time dimensions of one hour, one day and one week, network structures of the three time dimensions are all composed of a deep bottleneck residual error network, based on traffic flow data reconstructed by BRBM, a two-dimensional people flow matrix corresponding to each time dimension is input into the three separated time dimensions to model three time attributes of the people flow data, and the space dependency and the time dependency between the people flow data are processed through the deep bottleneck residual error network structure; three time dimensions: the output results of one hour, one day and one week are respectively defined as
3-3) adopting a fusion method based on a parameter matrix to fuse the output results of the three time dimension networks, wherein the fused result is defined as XRSpecifically, Hadamard product is adopted to calculate fusion result X of three time dimensionsR,XRThe calculation formula of (a) is as follows:
using trainable parameters W in the formulac、Wd、WwIndicating the degree to which the flow of the population in different regions is affected at different time periods, by pair Wc、Wd、WwThe values are adjusted to obtain the degree of influence of different regions on three time dimensions of one hour, one day, one week and the like, wherein L represents that L bottleneck residual error units are adopted, and "o" represents a Hadamard product.
In step 4), the auxiliary prediction mechanism is a component of a prediction method of the pedestrian flow of the deep bottleneck residual error network, and the prediction method comprises the following steps:
4-1) manually extracting data characteristics from external factor data (such as meteorological data, holidays and activity events), converting the extracted data characteristics of the external factors into binary vectors, carrying out standardization processing, and mapping the data between [0, 1 ];
4-2) inputting the extracted external factor data characteristics into a two-layer fully-connected network for training, wherein the first layer fully-connected network is regarded as an embedded layer of each sub-factor, the output result of the embedded layer is input into an activation function for nonlinear transformation, and the second layer fully-connected network is used for mapping the output result of the first layer from a low dimension to XtThe same high dimension;
4-3) obtaining an output result X of an auxiliary prediction mechanism through the processing of a two-layer full-connection networkE,XEAnd XtWith the same data dimension.
In step 5), the prediction results of two components of the people flow prediction model based on the deep bottleneck residual error network are fused, and the instant empty prediction result X is obtainedRAnd auxiliary prediction result XEThe fusion method comprises the following steps:
5-1) Total output X of three time dimensions consisting of a deep bottleneck residual networkRAnd output of auxiliary prediction XEAre directly combined into XFusion,XFusionThe expression of (a) is as follows:
XFusion=XR+XE
5-2) fusing the final result X by Tanh activation functionFusionMapping at [ -1, 1 [)]And the predicted value of the flow rate of people in the t-th time interval is expressed asNamely, it isFor the final result of the prediction of the traffic,is defined as follows:
the urban-range pedestrian flow prediction method based on the deep bottleneck residual error network provided by the invention constructs a data reconstruction mechanism based on a Bernoulli limited Boltzmann machine (BRBM) so as to reduce the dimension and reconstruct sample data. And cooperatively predicting the human flow by adopting the space-time data prediction based on the bottleneck residual error network and the auxiliary prediction based on the full-connection network. Compared with other existing methods, the method provided by the invention not only greatly reduces the computational complexity of the pedestrian flow prediction model and the time for model training, but also improves the prediction precision of the pedestrian flow.
Drawings
FIG. 1 is a diagram of a deep space-time bottleneck residual error network pedestrian flow prediction method architecture;
FIG. 2 is a schematic diagram of a BRBM network structure;
FIG. 3 is a flow chart of a data reconstruction process;
FIG. 4 is a flow chart of a data reconstruction training algorithm for BRBM;
FIG. 5 is a diagram of a bottleneck residual block;
FIG. 6 is a diagram of an in/out meta-model of a city population;
fig. 7 is a flowchart of a training algorithm of the prediction method of the deep bottleneck residual error network.
Detailed Description
The invention will be further elucidated with reference to the drawings and examples, without however being limited thereto.
Example (b):
as shown in fig. 1, a city-wide pedestrian volume prediction method based on a deep bottleneck residual error network includes the following steps:
1) acquiring original traffic flow data;
2) constructing a BRBM data reconstruction mechanism, inputting the original traffic flow data acquired in the step 1) into the BRBM data reconstruction mechanism, and performing dimension reduction and data reconstruction to obtain traffic flow data after BRBM reconstruction;
3) constructing a cooperative prediction mechanism, taking the traffic flow data obtained after BRBM reconstruction in the step 2) as input data of the cooperative prediction mechanism, and obtaining a prediction result after the cooperative prediction mechanismFruit XR;
4) Constructing an auxiliary prediction mechanism, and processing external factors influencing the pedestrian flow by adopting the auxiliary prediction mechanism to obtain a prediction result XE;
5) The prediction result X obtained in the step 3) is usedRWith X obtained in step 4)EAnd fusing to obtain a final people flow prediction result.
In the step 1), the original traffic flow data is obtained by using urban taxi GPS data, meteorological data and bicycle track data.
In step 2), the BRBM data reconfiguration mechanism includes a BRBM network visible layer and a hidden layer, as shown in fig. 2; the nodes of the hidden layer are used as a nonlinear feature detector for reducing dimension and reconstructing input data, and extracting high-level feature information from original traffic flow data, wherein the data reconstruction process comprises the following steps, as shown in fig. 3:
2-1) carrying out normalization processing on input original traffic flow data by adopting a data normalization method, and mapping traffic flow data values of each region of a city between [0 and 1] to obtain inflow and outflow flow data of the city region;
2-2) respectively inputting the normalized urban area inflow and outflow flow data into the BRBM network for dimensionality reduction and reconstruction;
2-2-1) the BRBM network is an energy-based probability distribution model, and for a given state vector h and v, the current energy function of the BRBM network is expressed as:
wi,jrepresenting the weight between the ith neuron of the visible layer and the jth neuron of the hidden layer, aiRepresenting the bias value of the ith neuron in the visible layer, bjA bias value representing the jth neuron in the hidden layer;
2-2-2) based on the energy function obtained in step 2-2-1), then the state of the BRBM network is defined as a given v, and the joint probability distribution of h is:
wherein Z is a normalization factor and the expression is:
2-2-3) obtaining the marginal probability distribution of the pedestrian flow data of the visible layer according to the joint probability distribution formula in the step 2-2-2):
2-2-4) the process of training the BRBM network is to solve a set of parameters θ ═ { w, a, b }, so that the final output of the visual layer has an approximate probability distribution with the input traffic flow data, and to achieve this goal, based on the marginal probability distribution formula of the pedestrian flow data in step 2-2-3), the following loss function is defined:
solving a group of optimal solutions theta, v by minimizing a loss functioniThe people flow data of the ith area of a city, and s is the number of divided areas of the city;
2-2-5) minimizing the loss function of the step 2-2-4), wherein the conditional probability distribution of the input traffic flow data in a visible layer and a hidden layer in the BRBM network is respectively as follows:
wi,jrepresenting the weight between the ith neuron of the visible layer and the jth neuron of the hidden layer, aiRepresenting the bias value of the ith neuron in the visible layer, bjRepresenting the bias value of the jth neuron in the hidden layer, wherein sigmoid is an activation function, and the expression of the activation function is as follows:
2-2-6) repeating step 2-2-5) k times to represent the result of the k-th sampling, thereby obtaining a probability distribution approximate to the input dataCarrying out forward and reverse training on input data between a visible layer and a hidden layer of the BRBM network, obtaining probability distribution similar to the input data by updating a weight, and completing the training of the BRBM network through k iterations;
2-2-7) extracting important characteristic information from input traffic flow data by the trained BRBM network to obtain traffic flow data reconstructed by the BRBM;
a flow chart of a data reconstruction training algorithm for a BRBM network is shown in figure 4,
in step 3), the collaborative prediction mechanism is composed of prediction oriented to spatio-temporal data, and the prediction method of the collaborative prediction mechanism comprises the following steps:
3-1) dividing the area of a city into a plurality of P × Q grid maps each representing an area according to longitude and latitude, as shown in FIG. 6, ri,jIs one of the P × Q grids, and for a grid (P1, Q1) e (P, Q), it indicates that the grid is located at the pth1Line and q1Row ofEach grid comprises inflow and outflow two types of people flows, and the inflow and outflow of people in a city within a period of time are respectively converted into corresponding two-dimensional people flow matrixes;
3-2) selecting the data of different timestamps to connect together according to the time attribute of the people flow data by the two-dimensional people flow matrix obtained by conversion in the step 3-1), and respectively modeling three time dimensions: one hour, one day, one week, then corresponding timestamp data is entered into three separate time dimensions for prediction of spatio-temporal data. The prediction facing the space-time data is composed of three time dimensions of one hour, one day and one week, network structures of the three time dimensions are all composed of a deep bottleneck residual error network, based on traffic flow data reconstructed by BRBM, a two-dimensional people flow matrix corresponding to each time dimension is input into the three separated time dimensions to model three time attributes of the people flow data, the space dependency and the time dependency between the people flow data are processed through the deep bottleneck residual error network structure, and a bottleneck residual error block structure is shown in FIG. 5; three time dimensions: the output results of one hour, one day and one week are respectively defined as
3-3) adopting a fusion method based on a parameter matrix to fuse the output results of the three time dimension networks, wherein the fused result is defined as XRSpecifically, Hadamard product is adopted to calculate fusion result X of three time dimensionsR,XRThe calculation formula of (a) is as follows:
using trainable parameters W in the formulac、Wd、WwIndicating the degree to which the flow of the population in different regions is affected at different time periods, by pair Wc、Wd、WwThe values are adjusted to obtain the degree of influence of different regions on three time dimensions of one hour, one day, one week and the like, wherein L represents that L bottleneck residual error units are adopted, and "o" represents a Hadamard product.
In step 4), the auxiliary prediction mechanism is a component of a prediction method of the pedestrian flow of the deep bottleneck residual error network, and the prediction method comprises the following steps:
4-1) manually extracting data characteristics from external factor data (such as meteorological data, holidays and activity events), converting the extracted data characteristics of the external factors into binary vectors, carrying out standardization processing, and mapping the data between [0, 1 ];
4-2) inputting the extracted external factor data characteristics into a two-layer fully-connected network for training, wherein the first layer fully-connected network is regarded as an embedded layer of each sub-factor, the output result of the embedded layer is input into an activation function for nonlinear transformation, and the second layer fully-connected network is used for mapping the output result of the first layer from a low dimension to XtThe same high dimension;
4-3) obtaining an output result X of an auxiliary prediction mechanism through the processing of a two-layer full-connection networkE,XEAnd XtWith the same data dimension.
In step 5), the prediction results of two components of the people flow prediction model based on the deep bottleneck residual error network are fused, and the instant empty prediction result X is obtainedRAnd auxiliary prediction result XEThe fusion method comprises the following steps:
5-1) Total output X of three time dimensions consisting of a deep bottleneck residual networkRAnd output of auxiliary prediction XEAre directly combined into XFusion,XFusionThe expression of (a) is as follows:
XFusion=XR+XE
5-2) fusing the final result X by Tanh activation functionFusionMapping at [ -1, 1 [)]And the predicted value of the flow rate of people in the t-th time interval is expressed asNamely, it isFor the final result of the prediction of the traffic,is defined as follows:
the processing flow of the people flow prediction method based on the deep bottleneck residual error network according to the steps (3) to (5) is shown in fig. 7.
Claims (5)
1. a city-wide pedestrian flow prediction method based on a deep bottleneck residual error network is characterized by comprising the following steps:
1) acquiring original traffic flow data;
2) constructing a BRBM data reconstruction mechanism, inputting the original traffic flow data acquired in the step 1) into the BRBM data reconstruction mechanism, and performing dimension reduction and data reconstruction to obtain traffic flow data after BRBM reconstruction;
3) constructing a cooperative prediction mechanism, taking the traffic flow data reconstructed by the BRBM obtained in the step 2) as input data of the cooperative prediction mechanism, and obtaining a prediction result X after passing through the cooperative prediction mechanismR;
The collaborative prediction mechanism consists of prediction oriented to space-time data, and the prediction method of the collaborative prediction mechanism comprises the following steps:
3-1) dividing the area of a city into a plurality of P × Q grid maps according to longitude and latitude, each grid representing an area, and setting ri,jIs one of the P × Q grids, and for a grid (P1, Q1) e (P, Q), it indicates that the grid is located at the pth1Line and q1Row ofEach grid comprises inflow and outflow two types of people flows, and the inflow and outflow of people in a city within a period of time are respectively converted into corresponding two-dimensional people flow matrixes;
3-2) selecting the data of different timestamps to connect together according to the time attribute of the people flow data by the two-dimensional people flow matrix obtained by conversion in the step 3-1), and respectively modeling three time dimensions: one hour, one day and one week, then inputting corresponding timestamp data into three separated time dimensions of prediction for space-time data, wherein the prediction for the space-time data comprises three time dimensions of one hour, one day and one week, network structures of the three time dimensions are all formed by a depth bottleneck residual error network, based on traffic flow data reconstructed by BRBM, a two-dimensional people flow matrix corresponding to each time dimension is input into the three separated time dimensions to model three time attributes of the people flow data, and the space dependency and the time dependency between the people flow data are processed through the depth bottleneck residual error network structure; three time dimensions: the output results of one hour, one day and one week are respectively defined as
3-3) adopting a fusion method based on a parameter matrix to fuse the output results of the three time dimension networks, wherein the fused result is defined as XRSpecifically, Hadamard product is adopted to calculate fusion result X of three time dimensionsR,XRThe calculation formula of (a) is as follows:
using trainable parameters W in the formulac、Wd、WwIndicating the degree to which the flow of the population in different regions is affected at different time periods, by pair Wc、Wd、WwAdjusting the values to obtain the degree of influence of three time dimensions of one hour, one day and one week on different areas, wherein L represents that L bottleneck residual error units are adopted,representing a Hadamard product;
4) constructing an auxiliary prediction mechanism, and processing external factors influencing the pedestrian flow by adopting the auxiliary prediction mechanism to obtain a prediction result XE;
5) The prediction result X obtained in the step 3) is usedRWith X obtained in step 4)EAnd fusing to obtain a final people flow prediction result.
2. The urban-wide pedestrian volume prediction method based on the deep bottleneck residual error network according to claim 1, wherein in the step 1), the original traffic flow data is urban taxi GPS data, meteorological data and bicycle track data.
3. The urban-wide people flow prediction method based on the deep bottleneck residual error network according to claim 1, wherein in the step 2), the BRBM data reconstruction mechanism comprises a BRBM network visible layer and a hidden layer; the node of the hidden layer is used as a nonlinear feature detector for reducing dimension and reconstructing input data, and extracting high-level feature information from original traffic flow data, wherein the data reconstruction process comprises the following steps:
2-1) carrying out normalization processing on input original traffic flow data by adopting a data normalization method, and mapping traffic flow data values of each region of a city between [0 and 1] to obtain inflow and outflow flow data of the city region;
2-2) respectively inputting the normalized urban area inflow and outflow flow data into the BRBM network for dimensionality reduction and reconstruction;
2-2-1) the BRBM network is an energy-based probability distribution model, and for a given state vector h and v, the current energy function of the BRBM network is expressed as:
wi,jrepresenting the weight between the ith neuron of the visible layer and the jth neuron of the hidden layer, aiRepresenting the bias value of the ith neuron in the visible layer, bjA bias value representing the jth neuron in the hidden layer;
2-2-2) based on the energy function obtained in step 2-2-1), then the state of the BRBM network is defined as a given v, and the joint probability distribution of h is:
wherein Z is a normalization factor and the expression is:
2-2-3) obtaining the marginal probability distribution of the pedestrian flow data of the visible layer according to the joint probability distribution formula in the step 2-2-2):
2-2-4) the process of training the BRBM network is to solve a set of parameters θ ═ { w, a, b }, so that the final output of the visual layer has an approximate probability distribution with the input traffic flow data, and to achieve this goal, based on the marginal probability distribution formula of the pedestrian flow data in step 2-2-3), the following loss function is defined:
solving a group of optimal solutions theta, v by minimizing a loss functioniThe people flow data of the ith area of a city, and s is the number of divided areas of the city;
2-2-5) minimizing the loss function of the step 2-2-4), wherein the conditional probability distribution of the input traffic flow data in a visible layer and a hidden layer in the BRBM network is respectively as follows:
wi,jrepresenting the weight between the ith neuron of the visible layer and the jth neuron of the hidden layer, aiRepresenting the bias value of the ith neuron in the visible layer, bjRepresenting the bias value of the jth neuron in the hidden layer, wherein sigmoid is an activation function, and the expression of the activation function is as follows:
2-2-6) repeating step 2-2-5) k times to represent the result of the k-th sampling, thereby obtaining a probability distribution approximate to the input dataCarrying out forward and reverse training on input data between a visible layer and a hidden layer of the BRBM network, obtaining probability distribution similar to the input data by updating a weight, and completing the training of the BRBM network through k iterations;
2-2-7) extracting important characteristic information from input traffic flow data by the trained BRBM network to obtain traffic flow data reconstructed by the BRBM.
4. The urban-wide people flow prediction method based on the deep bottleneck residual error network according to claim 1, wherein in the step 4), the auxiliary prediction mechanism is a component of the urban-wide people flow prediction method based on the deep bottleneck residual error network, and the prediction method comprises the following steps:
4-1) manually extracting data characteristics from external factor data, converting the extracted data characteristics of the external factors into binary vectors, carrying out standardization processing, and mapping the data between [0, 1 ];
4-2) inputting the extracted external factor data characteristics into a two-layer fully-connected network for training, wherein the first layer fully-connected network is regarded as an embedded layer of each sub-factor, the output result of the embedded layer is input into an activation function for nonlinear transformation, and the second layer fully-connected network is used for mapping the output result of the first layer from a low dimension to XtThe same high dimension;
4-3) obtaining an output result X of an auxiliary prediction mechanism through the processing of a two-layer full-connection networkE,XEAnd XtWith the same data dimension.
5. The urban-wide pedestrian volume prediction method based on the deep bottleneck residual error network according to claim 1, wherein in the step 5), the prediction results of two components of the urban-wide pedestrian volume prediction model based on the deep bottleneck residual error network are fused, and an instant empty prediction result X is obtainedRAnd auxiliary prediction result XEThe fusion method comprises the following steps:
5-1) Total output X of three time dimensions consisting of a deep bottleneck residual networkRAnd output of auxiliary prediction XEAre directly combined into XFusion,XFusionThe expression of (a) is as follows:
XFusion=XR+XE
5-2) fusing the final result X by Tanh activation functionFusionMapping at [ -1, 1 [)]And the predicted value of the flow rate of people in the t-th time interval is expressed asNamely, it isFor the final result of the prediction of the traffic,is defined as follows:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010028983.5A CN111260121B (en) | 2020-01-12 | 2020-01-12 | Urban-range pedestrian flow prediction method based on deep bottleneck residual error network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010028983.5A CN111260121B (en) | 2020-01-12 | 2020-01-12 | Urban-range pedestrian flow prediction method based on deep bottleneck residual error network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111260121A CN111260121A (en) | 2020-06-09 |
CN111260121B true CN111260121B (en) | 2022-04-29 |
Family
ID=70948744
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010028983.5A Active CN111260121B (en) | 2020-01-12 | 2020-01-12 | Urban-range pedestrian flow prediction method based on deep bottleneck residual error network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111260121B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112541852B (en) * | 2020-12-24 | 2024-04-12 | 南方科技大学 | Urban people stream monitoring method and device, electronic equipment and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106295874A (en) * | 2016-08-08 | 2017-01-04 | 上海交通大学 | Traffic flow parameter Forecasting Methodology based on deep belief network |
CN108446798A (en) * | 2018-03-08 | 2018-08-24 | 重庆邮电大学 | Urban population flow prediction method based on dual path space-time residual error network |
CN109376969A (en) * | 2018-12-14 | 2019-02-22 | 中南大学 | City fining population distribution dynamic prediction method and device based on deep learning |
CN110310474A (en) * | 2018-05-14 | 2019-10-08 | 桂林远望智能通信科技有限公司 | A kind of vehicle flowrate prediction technique and device based on space-time residual error network |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106033555A (en) * | 2015-03-13 | 2016-10-19 | 中国科学院声学研究所 | Big data processing method based on depth learning model satisfying K-dimensional sparsity constraint |
CN108288109A (en) * | 2018-01-11 | 2018-07-17 | 安徽优思天成智能科技有限公司 | Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks |
-
2020
- 2020-01-12 CN CN202010028983.5A patent/CN111260121B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106295874A (en) * | 2016-08-08 | 2017-01-04 | 上海交通大学 | Traffic flow parameter Forecasting Methodology based on deep belief network |
CN108446798A (en) * | 2018-03-08 | 2018-08-24 | 重庆邮电大学 | Urban population flow prediction method based on dual path space-time residual error network |
CN110310474A (en) * | 2018-05-14 | 2019-10-08 | 桂林远望智能通信科技有限公司 | A kind of vehicle flowrate prediction technique and device based on space-time residual error network |
CN109376969A (en) * | 2018-12-14 | 2019-02-22 | 中南大学 | City fining population distribution dynamic prediction method and device based on deep learning |
Non-Patent Citations (2)
Title |
---|
城市短时交通流预测仿真研究;陆琳等;《计算机仿真》;20120515;第第29卷卷(第05期);第334-336,415页 * |
基于卷积神经网络的人口流量预测;蔡乐等;《电脑与信息技术》;20191215(第06期);第5-7页 * |
Also Published As
Publication number | Publication date |
---|---|
CN111260121A (en) | 2020-06-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109887282B (en) | Road network traffic flow prediction method based on hierarchical timing diagram convolutional network | |
CN111612206B (en) | Neighborhood people stream prediction method and system based on space-time diagram convolution neural network | |
Li et al. | A deep learning method based on an attention mechanism for wireless network traffic prediction | |
CN112532439B (en) | Network flow prediction method based on attention multi-component space-time cross-domain neural network model | |
CN111612243B (en) | Traffic speed prediction method, system and storage medium | |
CN111223301B (en) | Traffic flow prediction method based on graph attention convolution network | |
CN109508360B (en) | Geographical multivariate stream data space-time autocorrelation analysis method based on cellular automaton | |
CN113268916A (en) | Traffic accident prediction method based on space-time graph convolutional network | |
CN111860951A (en) | Rail transit passenger flow prediction method based on dynamic hypergraph convolutional network | |
CN112257934A (en) | Urban people flow prediction method based on space-time dynamic neural network | |
Zhang et al. | A Traffic Prediction Method of Bicycle-sharing based on Long and Short term Memory Network. | |
CN113222218B (en) | Traffic accident risk prediction method based on convolution long-time and short-time memory neural network | |
CN110570035B (en) | People flow prediction system for simultaneously modeling space-time dependency and daily flow dependency | |
CN113905391A (en) | Ensemble learning network traffic prediction method, system, device, terminal, and medium | |
Lu et al. | Short-term demand forecasting for online car-hailing using ConvLSTM networks | |
CN112561058A (en) | Short-term photovoltaic power prediction method based on Stacking-ensemble learning | |
CN114202122A (en) | Urban traffic flow prediction method based on Markov cluster map attention network | |
CN112801340B (en) | Crowd density prediction method based on multi-level city information unit portraits | |
CN112766600A (en) | Urban area crowd flow prediction method and system | |
CN114692984A (en) | Traffic prediction method based on multi-step coupling graph convolution network | |
CN112446489A (en) | Dynamic network embedded link prediction method based on variational self-encoder | |
CN114662791A (en) | Long time sequence pm2.5 prediction method and system based on space-time attention | |
CN113971373A (en) | Traffic flow interpolation method based on video restoration technology | |
CN111260121B (en) | Urban-range pedestrian flow prediction method based on deep bottleneck residual error network | |
CN115376317A (en) | Traffic flow prediction method based on dynamic graph convolution and time sequence convolution network |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20200609 Assignee: GUILIN JINFAMING TECHNOLOGY DEVELOPMENT CO.,LTD. Assignor: GUILIN University OF ELECTRONIC TECHNOLOGY Contract record no.: X2022450000400 Denomination of invention: A prediction method of city wide passenger flow based on deep bottleneck residual network Granted publication date: 20220429 License type: Common License Record date: 20221226 |
|
EE01 | Entry into force of recordation of patent licensing contract |