CN114036135A - Method and system for estimating urban mobile source pollution emission by using incomplete information - Google Patents

Method and system for estimating urban mobile source pollution emission by using incomplete information Download PDF

Info

Publication number
CN114036135A
CN114036135A CN202111192115.1A CN202111192115A CN114036135A CN 114036135 A CN114036135 A CN 114036135A CN 202111192115 A CN202111192115 A CN 202111192115A CN 114036135 A CN114036135 A CN 114036135A
Authority
CN
China
Prior art keywords
emission
road
information
data
road section
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
Application number
CN202111192115.1A
Other languages
Chinese (zh)
Inventor
康宇
赵振怡
曹洋
许镇义
裴丽红
吕文君
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Science and Technology of China USTC
Original Assignee
University of Science and Technology of China USTC
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by University of Science and Technology of China USTC filed Critical University of Science and Technology of China USTC
Priority to CN202111192115.1A priority Critical patent/CN114036135A/en
Publication of CN114036135A publication Critical patent/CN114036135A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/21—Design, administration or maintenance of databases
    • G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2474—Sequence data queries, e.g. querying versioned data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2477—Temporal data queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/29—Geographical information databases
    • 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
    • G06N3/045—Combinations of 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
    • 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
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing

Abstract

The invention relates to a method and a system for estimating urban mobile source pollution emission by using incomplete information, which comprises the following steps of dividing road sections into 2 types according to the data density degree obtained on the road sections: the information missing road section which loses traffic mode information due to sparse acquired data and the information complete road section which has dense enough data; constructing a continuous speed field and an inferred flow value for the information complete road section through the motor vehicle driving related data, and calculating a corresponding traffic emission value through a microscopic emission factor model according to the continuous speed field and the inferred flow value; and respectively extracting topological characteristics of a road network and the time sequence dependency of the motor vehicle emission sequence on the information complete road section by using a graph convolution unit and a gate control linear unit, thereby reconstructing the emission sequence on the information missing road section and filling the missing emission distribution information. The method utilizes less urban traffic data to dig out more traffic modes and emission distribution characteristics, and realizes a task of fine estimation of urban road section level traffic pollution emission.

Description

Method and system for estimating urban mobile source pollution emission by using incomplete information
Technical Field
The invention relates to the technical field of environment detection processing, in particular to a method and a system for estimating urban mobile source pollution emission by using incomplete information.
Background
At present, the atmospheric environment situation in China is severe, regional atmospheric environment problems taking fine particles as characteristic pollutants are increasingly prominent, wherein mobile pollution sources represented by motor vehicles have great influence on air quality, and especially in super-large cities such as Beijing, Shanghai and the like and in the population-dense areas of the east, the proportion of mobile source emission to air pollution is up to more than 20% to 50%. Although China continuously makes relevant policies to prevent and control the tail gas pollution of motor vehicles, such as the elimination of unqualified vehicles, the upgrading of oil quality and the like, the current situation of mobile source pollution is still severe. The pollution emission information in the urban traffic network can provide theoretical support for the realization of tasks such as urban high-pollution emission area identification and the like, and provide theoretical basis for the control and decision of relevant parts of subsequent governments.
The traffic emission model is a main means widely used for quantifying the emission of motor vehicles and calculating the emission factor of the motor vehicles in the traffic field. The emission model is classified into an average speed-based emission model and a driving condition-based emission model according to the division of the main application parameters. The former extracts a representative average speed value of the motor vehicle from the overall driving characteristics contained in the motor vehicle driving data of a specific vehicle type, and calculates a corresponding emission value according to the representative average speed value. Therefore, models are typically focused on macro level studies and are used to estimate or predict total traffic emissions in a particular area (typically a region or city) over a particular period of time (typically a quarter or year). Emission models based on driving conditions generally focus on microscopic level research. The model is mainly used for analyzing the emission condition of the motor vehicle through a complete driving process according to transient second-by-second data obtained when the motor vehicle runs, and is generally suitable for the tasks of analyzing the emission of a single vehicle or calculating the emission of specific roads. However, the two methods have respective defects in a refined emission estimation task in an urban area, and have the challenges of large errors or unsupportable data size and the like.
Disclosure of Invention
The method and the system for estimating the urban mobile source pollution emission by using the incomplete information can solve the problem that the error of the estimated pollution value is large under the condition of small data volume in the existing method.
In order to achieve the purpose, the invention adopts the following technical scheme:
a method for estimating urban mobile source pollution emission by using incomplete information comprises the following steps,
s10, acquiring motor vehicle driving related data in urban traffic and urban traffic network data preprocessing and constructing a data space;
according to the data density degree obtained on the road sections, the road sections are divided into 2 types: the information missing road section which loses traffic mode information due to sparse acquired data and the information complete road section which has dense enough data;
s20, constructing a continuous speed field and an inferred flow value for the information complete road section through the motor vehicle driving related data, and calculating a corresponding traffic emission value through a microscopic emission factor model according to the continuous speed field and the inferred flow value;
and S30, extracting the topological characteristics of the road network and the time sequence dependency of the motor vehicle emission sequence on the information complete road section respectively by using the graph convolution unit and the gate control linear unit, thereby reconstructing the emission sequence on the information missing road section and filling up the missing emission distribution information.
Further, the step S10 is to acquire the data related to the driving of the motor vehicle in the urban traffic and the urban traffic network data, preprocess and construct a data space; according to the data density degree obtained on the road sections, the road sections are divided into 2 types: the information missing road sections and the information complete road sections with sufficiently dense data, which lack the traffic pattern information due to sparse acquired data, specifically include:
s11: urban road network data is constructed into a directed graph
Figure BDA0003301602690000022
Given a road network, all road intersections are first set as the edges epsilon of the connecting road sections in the graph, and the top point between two edges
Figure BDA0003301602690000021
Representing a collection of N road segments, each r having its own attributes, including a lane maximum speed value r.vmLane number r.n and road segment length r.len; the adjacency matrix W is abstracted as adjacency dependencies between vertices, and the connectivity between segments is represented by W: if the road sections i, j are connected, Wi,j1, otherwise Wi,j＝0；
S12: the method comprises the following steps that vehicle driving related data, namely GPS tracks of a sampled vehicle, each GPS track P comprises a series of time sequence GPS points, wherein each GPS point P comprises a time stamp t and a geographic space coordinate g, namely P is (t, g); the instant speed of the GPS point is firstly calculated according to the change of the geographic position and the time: given two GPS points p1And p2Then point p1The speed values of (a) are:
Figure BDA0003301602690000031
where dist (·) is a function for calculating the distance between two points, in the present invention, the Manhattan distance, i.e., the first-order norm; matching each GPS point to a corresponding road section by using a map matching algorithm, so that the attribute of each track point is expanded into p ═ t, g, v, r;
s13: next, defining a conceptual sampling density; firstly, counting the total number q of GPS points recorded on a road section r within one hourrThen the sampling density d of the road sectionr：
dr＝qr/r.len
By analysis, the road sections with the sampling density less than 20 are regarded as the road sections with the missing information
Figure BDA0003301602690000033
Taking the road section with the sampling density of more than or equal to 20 as the information complete road section
Figure BDA0003301602690000034
Further, the step S20 of constructing a continuous speed field and an inferred flow value for the information-completed road section through the vehicle driving related data, and calculating corresponding traffic emission through the microscopic emission factor model according to the continuous speed field and the inferred flow value specifically includes:
s21: given road section
Figure BDA0003301602690000032
GPS Point set of (1) { p }1,p2,…,pi,…,pn|pi＝(ti,gi,viR) as input, firstly, reconstructing a continuous speed field V of the road section by a Gaussian adaptive sliding methodr(t)：
Figure BDA0003301602690000041
Wherein the smoothing kernel function phii(t) decays with increasing distance in time, i.e. the function will generate smaller weight coefficients when the data points are far from the target computation point; thus, the kernel function is of the form:
Figure BDA0003301602690000042
wherein τ is a time width, and the larger the value of τ is, the larger the influence of the data point on the target calculation point is; the normalization function Φ (t) is the sum of all kernel functions:
Figure BDA0003301602690000043
therefore, all road sections in the urban road network construct a continuous speed field according to the formula;
s22: continuous speed field V for a given route section rrCalculating continuous emission rate TP by using EMIT microcosmic emission factor modell(t); the EMIT model consists of two modules: the system comprises an engine exhaust module and a tail gas pipe exhaust module; firstly, the traction force P of the motor vehicle is definedtract：
Figure BDA0003301602690000044
Wherein each parameter is defined as follows:
a: a rolling resistance item; b: speed correction of the rolling resistance item; c: an air resistance term; m: an engine mass; g: a gravitational constant; θ: road grade; definition l is a general emission class (CO, HC, NO)x) Engine emission rate EO of emission llThe expression is as follows:
Figure BDA0003301602690000045
wherein the parameters take the following values:
CO HC NOx
α 0.0316 0.00916 -0.00391
β 0.0 0.0 0.000305
δ 1.09e-07 7.55e-09 2.27e-08
ζ 0.00883 0.00111 0.00307
α′ 0.0261 0.00528 0.00323
exhaust emission rate TP of different pollutants under exhaust pipe emission modulelThere are different calculation formulas:
Figure BDA0003301602690000051
Figure BDA0003301602690000052
Figure BDA0003301602690000053
wherein the parameters take the following values:
Figure BDA0003301602690000054
s23: given the continuous speed field of the road section r, estimating the corresponding traffic flow according to a relevant model of a traffic theory: traffic flow Q of road section rr(t) and density kr(t) and velocity VrThe relationship of (t) is:
Qr(t)＝kr(t)×Vr(t)
because the density is unknown, the density value is deduced through the speed, and then the traffic flow value is calculated through the formula; using the velocity-density model:
Figure BDA0003301602690000061
wherein v ismIs the maximum velocity, kcIs the optimum density; thus, density
Figure BDA0003301602690000062
From the above formula, one can obtain:
Figure BDA0003301602690000063
therefore, all road sections in the urban road network estimate the traffic flow according to the formula;
s24: continuous discharge rate TP for a given road section rl(t) traffic flow Qr(t) and road segment attributes, the vehicle emission change for the road segment is described by the following function, with an emission value of:
Figure BDA0003301602690000064
further, the S30 utilizes a graph convolution unit and a gate control linear unit to respectively extract the topological feature of the road network and the time sequence dependency of the vehicle emission sequence on the information-complete road segment, thereby reconstructing the emission sequence on the information-missing road segment and filling up the missing emission distribution information, and specifically includes:
s31: firstly, constructing a vertex feature matrix of a graph structure in the Tth time interval
Figure BDA0003301602690000065
Wherein the set of informative full segments is
Figure BDA0003301602690000066
Corresponding road section in the vertex feature matrix
Figure BDA0003301602690000067
Using the emission value E obtained in step S20r(t), calculating various pollutant emission values of the road section:
Figure BDA0003301602690000071
for the section with the corresponding information missing in the vertex feature matrix
Figure BDA0003301602690000072
Setting the characteristic value to be 0;
complete set of road segments from information
Figure BDA0003301602690000073
Randomly deleting 20% of the vertex eigenvalues, and setting the eigenvalues of the deleted vertices to 0, thereby obtaining a new eigen matrix
Figure BDA0003301602690000074
Wherein the set of vertices of the deleted feature values is
Figure BDA0003301602690000075
The set of vertices remaining are
Figure BDA0003301602690000076
Since the network task is to estimate the historical emissions of the motor vehicle over 6 time intervals, a network input feature matrix is constructed
Figure BDA0003301602690000077
With a target output matrix of
Figure BDA0003301602690000078
Figure BDA0003301602690000079
S32: under the condition of giving a topological structure and partial characteristic information of a traffic road network, designing a space-time data interpolation model based on a generated countermeasure network, and generating a real characteristic value, namely an emission value, for an information-missing road section in the road network; the confrontation model consists of two confrontation neural networks, namely a feature generator and a feature discriminator based on an autoencoder: the feature generator captures data distribution and generates data similar to a real sample as much as possible to confuse the discriminator; a discriminator for distinguishing the real sample from the sample generated by the generator; both networks participate in max-min gambling against learning to improve the authenticity of the generated graph features; after network parameters are finely adjusted through training, a generator provides complete and real node characteristics as output according to input partial characteristic data;
s33: in an auto-encoder based feature generator: firstly, input graph characteristic matrix of network is expressed
Figure BDA00033016026900000710
Normalized to [0,1 ]]A range; the generator is formed by connecting 3 continuous space-time coding blocks in series and is used for feature learning; each space-time coding block is composed of a gate control linear unit and a graph convolution unit which are connected in series: wherein, the gate control linear unit is a convolution layer output without nonlinearity multiplied by a convolution layer output after nonlinear activation; thus, given an input matrix
Figure BDA0003301602690000081
Gated linear cell operation
Figure BDA0003301602690000082
Output of (2)
Figure BDA0003301602690000083
Comprises the following steps:
Figure BDA0003301602690000084
wherein
Figure BDA0003301602690000085
Is a convolution kernel parameter;
meanwhile, the graph convolution unit adopts the connectivity of the graph, namely an adjacent matrix, as a convolution filter of neighborhood mixing; concept based on spectrogram convolution, graph convolution Unit operation
Figure BDA0003301602690000086
Is shown for graph signals
Figure BDA0003301602690000087
Of the non-linear function of, the output of
Figure BDA0003301602690000088
Is equal to
Figure BDA0003301602690000089
Wherein, the Fourier base of the graph
Figure BDA00033016026900000810
Is a feature vector matrix of the normalized graph laplacian matrix L; laplace matrix
Figure BDA00033016026900000811
Wherein
Figure BDA00033016026900000812
Is a matrix of cells, which is,
Figure BDA00033016026900000813
is a diagonal matrix whose diagonal element values dii＝∑jWij；
Figure BDA00033016026900000814
Is the eigenvalue diagonal matrix of matrix L, so Θ (Λ) is also the diagonal matrix, and Θ is the graph convolution kernel parameter;
given the above-described gated linear unit and graph convolution unit operations, a spatio-temporal coding block is defined as
Figure BDA00033016026900000815
Operation, i.e. giving the input matrix of the jth space-time coding block
Figure BDA00033016026900000816
And nuclear parameters xij＝[ΓjΘj]Output of space-time coded blocks
Figure BDA00033016026900000817
Comprises the following steps:
Figure BDA00033016026900000818
wherein σ1(·),σ2(. is an activation function; therefore, the output after the three space-time coding blocks is:
Figure BDA00033016026900000819
Figure BDA00033016026900000820
Figure BDA00033016026900000821
wherein all activation functions are elu activation functions; after three space-time coding blocks, converting the input partial graph characteristics into an N multiplied by 512 middle characteristic matrix; the matrix is then decoded by three space-time decoding blocks; the structure of the space-time decoding block is the same as that of the space-time coding block, and the network structure is expressed by a formula as follows:
Figure BDA0003301602690000091
Figure BDA0003301602690000092
all of the activation functions described above are elu functions,
Figure BDA0003301602690000093
the activation function σ in this step2Activating a function for sigmoid; reconstructing graph features through three space-time decoding blocks;
s34: the graph feature discriminator is used for distinguishing the graph feature matrix generated by the feature generator
Figure BDA0003301602690000098
And a true graph feature matrix X; the network starts from a space-time coding block, the output of the space-time coding block is expanded into a one-dimensional matrix and then fed into three full-connection layers, namely, the space-time coding block is used for extracting hidden graph features from input data, and then completely communicated neurons in all the full-connection layers cooperate to classify the extracted features so as to judge whether the input data is real or not; finally, a sigmoid function is used for generating a probability between 0 and 1, which indicates the real probability of the input feature discrimination of the discriminator;
each operation and each full concatenation in space-time coding blockA LayerNorm layer is added after the layers are connected, and the LayerNorm layer is used for inputting
Figure BDA0003301602690000094
Or X, the output of the space-time coding block is as follows:
Figure BDA0003301602690000095
wherein LN (-) represents a LayerNorm layer, will
Figure BDA0003301602690000096
Spread into a one-dimensional matrix
Figure BDA0003301602690000097
Taking the obtained data as the input of three subsequent fully-connected layers, wherein the activation function of each layer is elu activation function and adding a LayerNorm operation, the activation function of the last layer is sigmoid function, and the output of the final discriminator is
Figure BDA0003301602690000101
For input
Figure BDA0003301602690000102
The output of the discriminator is
Figure BDA0003301602690000103
For input X, the output of the discriminator is
Figure BDA0003301602690000104
S35: in the generation of confrontation network training in confrontation learning, the training target of the feature generator is expressed as:
Figure BDA0003301602690000105
the training targets for the feature discriminators are represented as:
Figure BDA0003301602690000106
further, in the step S35, a gradient penalty term is added to the loss function of the feature discriminator in the training process, so that the loss function L of the feature discriminatorDComprises the following steps:
Figure BDA0003301602690000107
where λ is the tuning parameter.
Further, λ in step S35 is an adjustment parameter, and its value is 10.
Further, in step S35, an estimated mean square error term is added to the loss function of the generator, so that the loss function L of the feature generatorGComprises the following steps:
Figure BDA0003301602690000108
the feature generator and feature arbiter are alternately trained using an Adam optimizer until the two loss functions converge to a stationary state.
In another aspect, the present invention also discloses a system for estimating urban mobile source pollutant discharge by using incomplete information, comprising the following units,
the data acquisition and processing unit is used for acquiring motor vehicle driving related data in urban traffic and urban traffic network data preprocessing and constructing a data space;
and the traffic emission value calculating unit is used for dividing the road sections into 2 types according to the data density degree obtained on the road sections: the information missing road section which loses traffic mode information due to sparse acquired data and the information complete road section which has dense enough data; constructing a continuous speed field and an inferred flow value for the information complete road section through the motor vehicle driving related data, and calculating a corresponding traffic emission value through a microscopic emission factor model according to the continuous speed field and the inferred flow value;
and the discharge distribution information filling unit is used for respectively extracting the topological characteristics of the road network and the time sequence dependency of the motor vehicle discharge sequence on the information complete road section by utilizing the graph convolution unit and the gate control linear unit, thereby reconstructing the discharge sequence on the information missing road section and filling the missing discharge distribution information.
In a third aspect, the present invention also discloses a computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of the method as described above.
According to the technical scheme, the method and the system for estimating urban mobile source pollution emission by using incomplete information aim to utilize less urban traffic data to dig out more traffic modes and emission distribution characteristics so as to overcome the defects of the existing method and realize a refined estimation task of urban road section-level traffic pollution emission.
In the embodiment, the road sections are divided into two types according to the completeness of the obtained motor vehicle driving information, and different ideas are designed for calculating the motor vehicle emission of the two types of road sections. For the road section with complete information, a more accurate continuous speed field and flow change can be reconstructed based on the obtained data, and the motor vehicle emission value of the road section can be calculated as accurately as possible by utilizing a microscopic emission factor calculation model; for the information-missing road sections, by means of a generation countermeasure network composed of a graph convolution unit and a gate control linear unit, the emission sequence of the road sections is estimated by means of the spatial topological characteristics of the road network and the time sequence dependence of motor vehicle emission in the information-complete road sections, and therefore the traffic emission sequence of all the road sections in the urban road network within hours of history is calculated.
Drawings
FIGS. 1 and 2 are flows of a method for urban mobile source emission estimation;
FIG. 3 is a graph convolution and gated linear cell based generative confrontation network.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention.
As shown in fig. 1 and 2, the method for estimating urban mobile source pollutant emission by using incomplete information according to the embodiment includes the following steps:
step 1: and preprocessing the motor vehicle driving related data in the urban traffic and urban traffic network data and constructing a data space. According to the data density degree obtained on the road sections, the road sections are divided into 2 types: the information missing road section which lacks traffic mode information due to sparse collected data and the information complete road section which has dense enough data.
Step 2: and constructing a continuous speed field and an inferred flow value for the information complete road section through the motor vehicle driving related data, and calculating corresponding traffic emission through a microscopic emission factor model according to the continuous speed field and the inferred flow value.
And step 3: in order to reconstruct the emission distribution of the information missing road section with the history of hours, by utilizing the thought of confrontation generation learning, the topological characteristics of a road network and the time sequence dependency of the motor vehicle emission sequence on the information complete road section are respectively extracted by utilizing a graph convolution unit and a gate control linear unit, so that the emission sequence on the information missing road section is reconstructed, and the missing emission distribution information is filled.
The following is a detailed description:
further, the above step S1: and preprocessing and constructing data space for motor vehicle driving related data in urban traffic and urban traffic network data. According to the data density degree obtained on the road sections, the road sections are divided into 2 types: the method specifically comprises the following subdivision steps S11 to S13 that the acquired data are sparse and the information missing road sections of the traffic pattern information and the information full road sections of which the data points are dense enough are missing:
s11: the city road network data can be constructed into a directed graph
Figure BDA0003301602690000121
Given a road network, all road intersections are first set as the edges epsilon of the connecting road sections in the graph, and the top point between two edges
Figure BDA0003301602690000131
Representing a collection of N road segments, each r having its own attributes, including a lane maximum speed value r.vmNumber of lanes r.n and link length r.len. The adjacency matrix W can be abstracted as adjacency dependencies between vertices, and in the present invention, we denote the connectivity between segments by W: if the road sections i, j are connected, Wi,j1, otherwise Wi,j＝0。
S12: the vehicle driving related data is GPS tracks of a sampled vehicle, each GPS track P comprises a series of time sequence GPS points, wherein each GPS point P comprises a time stamp t and a geographic space coordinate g, namely P is (t, g). The instant speed of the GPS point is firstly calculated according to the change of the geographic position and the time: given two GPS points p1And p2Then point p1The speed values of (a) are:
Figure BDA0003301602690000132
where dist (·) is a function that calculates the distance between two points, in the present invention the Manhattan distance, i.e., the first-order norm. Each GPS point is matched to a corresponding road segment using a map matching algorithm, so that the attribute of each track point can be extended to p ═ t, g, v, r.
S13: next, a concept is defined: the density of the samples. Firstly, counting the total number q of GPS points recorded on a road section r within one hourrThen the sampling density d of the road sectionr：
dr＝qr/r.len
By analysis, the road sections with the sampling density less than 20 are regarded as the road sections with the missing information
Figure BDA0003301602690000135
Taking the road section with the sampling density of more than or equal to 20 as the information complete road section
Figure BDA0003301602690000134
The above step S2: and constructing a continuous speed field and an inferred flow value for the information complete road section through the motor vehicle driving related data, and calculating corresponding traffic emission through a microscopic emission factor model according to the continuous speed field and the inferred flow value. The method specifically comprises the following subdivision steps S21-S24:
s21: given road section
Figure BDA0003301602690000133
GPS Point set of (1) { p }1,p2,…,pi,…,pn|pi＝(ti,gi,viR) as input, firstly, reconstructing a continuous speed field V of the road section by a Gaussian adaptive sliding methodr(t)：
Figure BDA0003301602690000141
Wherein the smoothing kernel function phiiThe value of (t) decays with increasing distance in time, i.e. the function will generate smaller weight coefficients when the data points are far from the target calculation point. Thus, the kernel function is of the form:
Figure BDA0003301602690000142
where τ is the time width, the larger the value of τ, the greater the influence of the data point on the target computation point.
The normalization function Φ (t) is the sum of all kernel functions:
Figure BDA0003301602690000143
therefore, all road segments in the urban road network can construct a continuous speed field according to the formula.
S22: continuous speed field V for a given route section rrCalculating continuous emission rate TP by using EMIT microcosmic emission factor modell(t) of (d). The EMIT model consists of two modules: the device comprises an engine exhaust module and a tail gas pipe exhaust module. Firstly, the traction force P of the motor vehicle is definedtract：
Figure BDA0003301602690000144
Wherein each parameter is defined as follows:
a: a rolling resistance item; b: speed correction of the rolling resistance item; c: an air resistance term; m: an engine mass; g: a gravitational constant; θ: road grade; definition l is a general emission class (CO, HC, NO)x) Engine emission rate EO of emission llThe expression is as follows:
Figure BDA0003301602690000151
wherein the parameters take the following values:
CO HC NOx
α 0.0316 0.00916 -0.00391
β 0.0 0.0 0.000305
δ 1.09e-07 7.55e-09 2.27e-08
ζ 0.00883 0.00111 0.00307
α′ 0.0261 0.00528 0.00323
exhaust emission rate TP of different pollutants under exhaust pipe emission modulelThere are different calculation formulas:
Figure BDA0003301602690000152
Figure BDA0003301602690000153
Figure BDA0003301602690000154
wherein the parameters take the following values:
Figure BDA0003301602690000155
Figure BDA0003301602690000161
s23: given a continuous speed field of a road section r, the corresponding traffic flow can be estimated according to a relevant model of traffic theory: traffic flow Q of road section rr(t) and density kr(t) and velocity Vr(the relationship of t is:
Qr(t)＝kr(t)×Vr(t)
since the density is unknown, the density value needs to be first deduced by velocity and then the traffic flow value is calculated by the above formula. In the present invention, a velocity-density model is used:
Figure BDA0003301602690000162
wherein v ismIs the maximum velocity, kcIs the optimum density. Thus, density
Figure BDA0003301602690000163
From the above formula, one can obtain:
Figure BDA0003301602690000164
therefore, all the road segments in the urban road network can be used for estimating the traffic flow according to the formula.
S24: continuous discharge rate TP for a given road section rl(t) traffic flow Qr(t) and road segment attributes, the vehicle emission change for the road segment is described by the following function, with an emission value of:
Figure BDA0003301602690000165
further, the above step S3: in order to reconstruct the emission distribution of the information missing road section with the history of hours, by utilizing the thought of confrontation generation learning, the topological characteristics of a road network and the time sequence dependency of the motor vehicle emission sequence on the information complete road section are respectively extracted by utilizing a graph convolution unit and a gate control linear unit, so that the emission sequence on the information missing road section is reconstructed, and the missing emission distribution information is filled. The method specifically comprises the following subdivision steps S31-S35:
s31: first, a vertex feature matrix of a graph structure is constructed at the Tth period (each period is 1 hour)
Figure BDA0003301602690000171
Wherein the set of informative full segments is
Figure BDA0003301602690000172
Corresponding road section in the vertex feature matrix
Figure BDA0003301602690000173
Using the emission value E obtained in step 2r(t), calculating various pollutant emission values of the road section:
Figure BDA0003301602690000174
for the section with the corresponding information missing in the vertex feature matrix
Figure BDA0003301602690000175
The characteristic value thereof is set to 0.
To train a network, a complete set of segments is derived from information
Figure BDA0003301602690000176
Randomly deleting 20% of the vertex eigenvalues, and setting the eigenvalues of the deleted vertices to 0, thereby obtaining a new eigen matrix
Figure BDA0003301602690000177
Wherein the set of vertices of the deleted feature values is
Figure BDA0003301602690000178
The set of vertices remaining are
Figure BDA0003301602690000179
Due to network controlThe business is to estimate the historical emissions of the motor vehicle for 6 time intervals, thus constructing a network input feature matrix
Figure BDA00033016026900001710
With a target output matrix of
Figure BDA00033016026900001711
Figure BDA00033016026900001712
S32: under the condition of giving the topological structure and partial characteristic information of a traffic network, a space-time data interpolation model based on a generated countermeasure network is designed to generate a real characteristic value (emission value) for an information-missing road section in the traffic network. The confrontation model consists of two confrontation neural networks, namely a feature generator and a feature discriminator based on an autoencoder: the feature generator captures data distribution and generates data similar to a real sample as much as possible to confuse the discriminator; the discriminator distinguishes real samples from samples generated by the generator. Both networks participate in max-min gambling against learning to improve the authenticity of the generated graph features. After the network parameters are finely adjusted through training, the generator provides complete and real node characteristics as output according to the input partial characteristic data.
S33: in an auto-encoder based feature generator: firstly, input graph characteristic matrix of network is expressed
Figure BDA0003301602690000181
Normalized to [0,1 ]]And (3) a range. The generator is formed by connecting 3 continuous space-time coding blocks in series for feature learning. Each space-time coding block is composed of a gate control linear unit and a graph convolution unit which are connected in series: (1) the principle of a gated linear cell is that the output of a convolutional layer without non-linearization is multiplied by the output of a convolutional layer that has undergone non-linear activation. Thus, given an input matrix
Figure BDA0003301602690000182
Gated linear cell operation
Figure BDA0003301602690000183
Output of (2)
Figure BDA0003301602690000184
Comprises the following steps:
Figure BDA0003301602690000185
wherein
Figure BDA0003301602690000186
Is a convolution kernel parameter; (2) the graph convolution unit adopts the connectivity of the graph, namely the adjacency matrix, as a convolution filter of the neighborhood mixture. Concept based on spectrogram convolution, graph convolution Unit operation
Figure BDA0003301602690000187
Can be expressed as a graph signal
Figure BDA0003301602690000188
Of the non-linear function of, the output of
Figure BDA0003301602690000189
Is equal to
Figure BDA00033016026900001810
Wherein, the Fourier base of the graph
Figure BDA00033016026900001811
Is the eigenvector matrix of the normalized graph laplacian matrix L. Laplace matrix
Figure BDA00033016026900001812
Wherein
Figure BDA00033016026900001813
Is a matrix of cells, which is,
Figure BDA00033016026900001814
is a diagonal matrix whose diagonal element values dii＝∑jWij；
Figure BDA00033016026900001815
Is the eigenvalue diagonal matrix of matrix L, so Θ (Λ) is also the diagonal matrix, and Θ is the graph convolution kernel parameter. In the graph convolution operation, I ═ M is common.
Given the above-described gated linear cell and graph convolution cell operations, a spatio-temporal coding block can be defined as
Figure BDA00033016026900001816
Operation, i.e. giving the input matrix of the jth space-time coding block
Figure BDA00033016026900001817
And nuclear parameters xij＝[ΓjΘj]Output of space-time coded blocks
Figure BDA00033016026900001818
Comprises the following steps:
Figure BDA00033016026900001819
wherein σ1(·),σ2(. cndot.) is an activation function. Therefore, the output after the three space-time coding blocks is:
Figure BDA00033016026900001820
Figure BDA0003301602690000191
Figure BDA0003301602690000192
all of the activation functions are elu activation functions. After three space-time coding blocks, the input partial graph characteristics are converted into an N multiplied by 512 middle characteristic matrix. The matrix is then decoded by three spatio-temporal decoding blocks. The structure of the space-time decoding block is the same as that of the space-time coding block, and the network structure is expressed by a formula as follows:
Figure BDA0003301602690000193
Figure BDA0003301602690000194
all of the activation functions described above are elu functions,
Figure BDA0003301602690000195
the activation function σ in this step2The function is activated for sigmoid. The graph features are reconstructed by three spatio-temporal decoding blocks. Essentially, the feature generator consisting of three space-time coding blocks and three space-time decoding blocks is a self-encoder that encodes part of the input data into high-dimensional feature maps and then decodes them into the complete graph feature matrix
Figure BDA0003301602690000196
S34: the purpose of the graph feature discriminator is to distinguish the graph feature matrix generated by the feature generator
Figure BDA0003301602690000197
And a true graph feature matrix X. The network starts from a space-time coding block, the output of the space-time coding block is expanded into a one-dimensional matrix and then fed into three full-connection layers, namely, the space-time coding block is used for extracting hidden graph features from input data, and then completely-communicated neurons in all the full-connection layers cooperate to classify the extracted features so as to judge whether the input data is real or not.Finally, a sigmoid function is used to generate a probability between 0 and 1, indicating the true probability of the input feature being discriminated by the discriminator.
A LayerNorm layer is added after each operation and each full connection layer in the space-time coding block so as to solve the problems of gradient loss and parameter dual polarization in the counterstudy. For input
Figure BDA0003301602690000201
Or X, the output of the space-time coding block is as follows:
Figure BDA0003301602690000202
wherein LN (-) represents a LayerNorm layer. Will be provided with
Figure BDA0003301602690000203
Spread into a one-dimensional matrix
Figure BDA0003301602690000204
Taking the obtained data as the input of three subsequent fully-connected layers, wherein the activation function of each layer is elu activation function and adding a LayerNorm operation, the activation function of the last layer is sigmoid function, and the output of the final discriminator is
Figure BDA0003301602690000205
For input
Figure BDA0003301602690000206
The output of the discriminator is
Figure BDA0003301602690000207
For input X, the output of the discriminator is
Figure BDA0003301602690000208
S35: in generating confrontation network training in confrontation learning, the training target of the feature generator is generally expressed as:
Figure BDA0003301602690000209
the training targets for the feature discriminators are typically expressed as:
Figure BDA00033016026900002010
in the actual training process, in order to overcome the problem of gradient disappearance, a gradient penalty term is added into the loss function of the feature discriminator, so that the loss function L of the feature discriminatorDComprises the following steps:
Figure BDA00033016026900002011
where λ is the tuning parameter, which is 10 in this patent. Furthermore, to improve the quality of the samples generated by the feature generator, an estimated mean square error term is added to the loss function of the generator, so the loss function L of the feature generatorGComprises the following steps:
Figure BDA00033016026900002012
the feature generator and feature arbiter are alternately trained using an Adam optimizer until the two loss functions converge to a stationary state.
The embodiment of the invention has the following main characteristics:
1. in order to solve the problem of spatial imbalance of traffic state information, road sections in a road network are divided into two road sections according to the collected GPS density: the information-full section and the information-missing section, and for the two sections, different ideas are used to estimate the traffic emission values thereof, respectively.
2. For an information full road segment: in order to overcome the noise such as information discontinuity and error on time sequence, a Gaussian adaptive smoothing model is applied to construct a continuous speed domain through discrete GPS points, and the speed of each moment is obtained through time sequence correlation correction of other sampling points and the speed value of the moment, so that the error noise is counteracted. Then, the road section traffic flow is estimated according to the speed domain, so that the corresponding traffic emission sequence is calculated by using the microscopic emission model.
3. For information missing links: due to the fact that the urban road network has connectivity, the traffic states of the two road sections have interdependence relation, and the traffic mode and the emission distribution have similar rules. Therefore, a generation countermeasure network based on space-time self-coding is provided, a space-time coding block and a space-time decoding block which are composed of a graph convolution layer and a gate control linear unit are designed, and therefore the time sequence dependence characteristics of the traffic emission sequence of the complete information section and the space adjacent characteristic space of the urban road network are constructed, and the estimation task of the traffic emission value on the incomplete information section is achieved through the powerful characteristic learning capacity of the countermeasure generation network.
In summary, in the embodiment, the road sections are divided into two types according to the completeness of the obtained vehicle driving information, and different ideas are designed for calculating the vehicle emission of the two types of road sections. For the road section with complete information, a more accurate continuous speed field and flow change can be reconstructed based on the obtained data, and the motor vehicle emission value of the road section can be calculated as accurately as possible by utilizing a microscopic emission factor calculation model; for the information-missing road sections, by means of a generation countermeasure network composed of a graph convolution unit and a gate control linear unit, the emission sequence of the road sections is estimated by means of the spatial topological characteristics of the road network and the time sequence dependence of motor vehicle emission in the information-complete road sections, and therefore the traffic emission sequence of all the road sections in the urban road network within hours of history is calculated.
In another aspect, the present invention also discloses a system for estimating urban mobile source pollutant discharge by using incomplete information, comprising the following units,
the data acquisition and processing unit is used for acquiring motor vehicle driving related data in urban traffic and urban traffic network data preprocessing and constructing a data space;
and the traffic emission value calculating unit is used for dividing the road sections into 2 types according to the data density degree obtained on the road sections: the information missing road section which loses traffic mode information due to sparse acquired data and the information complete road section which has dense enough data; constructing a continuous speed field and an inferred flow value for the information complete road section through the motor vehicle driving related data, and calculating a corresponding traffic emission value through a microscopic emission factor model according to the continuous speed field and the inferred flow value;
and the discharge distribution information filling unit is used for respectively extracting the topological characteristics of the road network and the time sequence dependency of the motor vehicle discharge sequence on the information complete road section by utilizing the graph convolution unit and the gate control linear unit, thereby reconstructing the discharge sequence on the information missing road section and filling the missing discharge distribution information.
In a third aspect, the present invention also discloses a computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of the method as described above.
It is understood that the system provided by the embodiment of the present invention corresponds to the method provided by the embodiment of the present invention, and the explanation, the example and the beneficial effects of the related contents can refer to the corresponding parts in the method.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (8)

1. A method for estimating urban mobile source pollution emission by using incomplete information is characterized by comprising the following steps,
s10, acquiring motor vehicle driving related data in urban traffic and urban traffic network data preprocessing and constructing a data space;
according to the data density degree obtained on the road sections, the road sections are divided into 2 types: the information missing road section which loses traffic mode information due to sparse acquired data and the information complete road section which has dense enough data;
s20, constructing a continuous speed field and an inferred flow value for the information complete road section through the motor vehicle driving related data, and calculating a corresponding traffic emission value through a microscopic emission factor model according to the continuous speed field and the inferred flow value;
and S30, extracting the topological characteristics of the road network and the time sequence dependency of the motor vehicle emission sequence on the information complete road section respectively by using the graph convolution unit and the gate control linear unit, thereby reconstructing the emission sequence on the information missing road section and filling up the missing emission distribution information.
2. The method for estimating urban mobile source pollutant emission according to claim 1, characterized in that: the S10 acquires the motor vehicle driving related data in the urban traffic and the urban traffic network data preprocessing and data space construction; according to the data density degree obtained on the road sections, the road sections are divided into 2 types: the information missing road sections and the information complete road sections with sufficiently dense data, which lack the traffic pattern information due to sparse acquired data, specifically include:
s11: urban road network data is constructed into a directed graph
Figure FDA0003301602680000011
Given a road network, all road intersections are first set as the edges epsilon of the connecting road sections in the graph, and the top point between two edges
Figure FDA0003301602680000012
Figure FDA0003301602680000013
Representing a collection of N road segments, each r having an own attribute, including a lane maximum speed value r.vmLane number r.n and road segment length r.len; the adjacency matrix W is abstracted as adjacency dependencies between vertices, and the connectivity between segments is represented by W: if the road sections i, j are connected, Wi，j1, otherwise Wi，j＝0；
S12: maneuveringVehicle driving related data, namely GPS tracks of a sampled motor vehicle, wherein each GPS track P comprises a series of time sequence GPS points, wherein each GPS point P comprises a time stamp t and a geographic space coordinate g, namely P is (t, g); the instant speed of the GPS point is firstly calculated according to the change of the geographic position and the time: given two GPS points p1And p2Then point p1The speed values of (a) are:
Figure FDA0003301602680000021
where dist (·) is a function for calculating the distance between two points, in the present invention, the Manhattan distance, i.e., the first-order norm; matching each GPS point to a corresponding road section by using a map matching algorithm, so that the attribute of each track point is expanded into p ═ t, g, v, r;
s13: next, defining a conceptual sampling density; firstly, counting the total number q of GPS points recorded on a road section r within one hourrThen the sampling density d of the road sectionr：
dr＝qr/r.len
By analysis, the road sections with the sampling density less than 20 are regarded as the road sections with the missing information
Figure FDA0003301602680000024
Taking the road section with the sampling density of more than or equal to 20 as the information complete road section
Figure FDA0003301602680000025
3. The method for estimating urban mobile source pollutant emission according to claim 2, characterized in that: the S20, constructing a continuous speed field and an inferred flow value for the information complete road section through the vehicle driving related data, and calculating corresponding traffic emission through the microscopic emission factor model according to the continuous speed field and the inferred flow value specifically includes:
s21: given road section
Figure FDA0003301602680000026
GPS Point set of (1) { p }1，p2，...，pi，...，pn|pi＝(ti，gi，viR) as input, firstly, reconstructing a continuous speed field V of the road section by a Gaussian adaptive sliding methodr(t)：
Figure FDA0003301602680000022
Wherein the smoothing kernel function phii(t) decays with increasing distance in time, i.e. the function will generate smaller weight coefficients when the data points are far from the target computation point; thus, the kernel function is of the form:
Figure FDA0003301602680000023
wherein τ is a time width, and the larger the value of τ is, the larger the influence of the data point on the target calculation point is; the normalization function Φ (t) is the sum of all kernel functions:
Figure FDA0003301602680000031
therefore, all road sections in the urban road network construct a continuous speed field according to the formula;
s22: continuous speed field V for a given route section rrCalculating continuous emission rate TP by using EMIT microcosmic emission factor modell(t); the EMIT model consists of two modules: the system comprises an engine exhaust module and a tail gas pipe exhaust module; firstly, the traction force P of the motor vehicle is definedtract：
Figure FDA0003301602680000032
Wherein each parameter is defined as follows:
a: a rolling resistance item; b: speed correction of the rolling resistance item; c: an air resistance term; m: an engine mass; g: a gravitational constant; θ: road grade; definition l is a general emission class (CO, HC, NO)x) Engine emission rate EO of emission llThe expression is as follows:
Figure FDA0003301602680000033
wherein the parameters take the following values:
CO HC NOx α 0.0316 0.00916 -0.00391 β 0.0 0.0 0.000305 δ 1.09e-07 7.55e-09 2.27e-08 ζ 0.00883 0.00111 0.00307 α′ 0.0261 0.00528 0.00323
exhaust emission rate TP of different pollutants under exhaust pipe emission modulelThere are different calculation formulas:
Figure FDA0003301602680000041
Figure FDA0003301602680000042
Figure FDA0003301602680000043
wherein the parameters take the following values:
Figure FDA0003301602680000044
s23: given the continuous speed field of the road section r, estimating the corresponding traffic flow according to a relevant model of a traffic theory: traffic flow Q of road section rr(t) and density kr(t) and velocity VrThe relationship of (t) is:
Qr(t)＝kr(t)×Vr(t)
because the density is unknown, the density value is deduced through the speed, and then the traffic flow value is calculated through the formula; using the velocity-density model:
Figure FDA0003301602680000045
wherein v ismIs the maximum velocity, kcIs the optimum density; thus, density
Figure FDA0003301602680000046
From the above formula, one can obtain:
Figure FDA0003301602680000047
therefore, all road sections in the urban road network estimate the traffic flow according to the formula;
s24: continuous discharge rate TP for a given road section rl(t) traffic flow Qr(t) and road segment attributes, the vehicle emission change for the road segment is described by the following function, with an emission value of:
Figure FDA0003301602680000051
4. the method for estimating urban mobile source pollutant emission according to claim 3, characterized in that: the S30 utilizes the graph convolution unit and the gate control linear unit to extract the topological feature of the road network and the time sequence dependency of the vehicle emission sequence on the information-complete road segment, thereby reconstructing the emission sequence on the information-missing road segment and filling up the missing emission distribution information, and specifically includes:
s31: firstly, construct the vertex character of graph structure in the Tth periodSign matrix
Figure FDA0003301602680000052
Wherein the set of informative full segments is
Figure FDA0003301602680000053
Corresponding road section in the vertex feature matrix
Figure FDA0003301602680000054
Using the emission value E obtained in step S20r(t), calculating various pollutant emission values of the road section:
Figure FDA0003301602680000055
for the section with the corresponding information missing in the vertex feature matrix
Figure FDA0003301602680000056
Setting the characteristic value to be 0;
complete set of road segments from information
Figure FDA0003301602680000057
Randomly deleting 20% of the vertex eigenvalues, and setting the eigenvalues of the deleted vertices to 0, thereby obtaining a new eigen matrix
Figure FDA0003301602680000058
Wherein the set of vertices of the deleted feature values is
Figure FDA0003301602680000059
Figure FDA00033016026800000510
The set of vertices remaining are
Figure FDA00033016026800000511
Since the network task is to estimate the historical emissions of the motor vehicle over 6 time intervals, a network input feature matrix is constructed
Figure FDA00033016026800000512
With a target output matrix of
Figure FDA00033016026800000513
S32: under the condition of giving a topological structure and partial characteristic information of a traffic road network, designing a space-time data interpolation model based on a generated countermeasure network, and generating a real characteristic value, namely an emission value, for an information-missing road section in the road network; the confrontation model consists of two confrontation neural networks, namely a feature generator and a feature discriminator based on an autoencoder: the feature generator captures data distribution and generates data similar to a real sample as much as possible to confuse the discriminator; a discriminator for distinguishing the real sample from the sample generated by the generator; both networks participate in max-min gambling against learning to improve the authenticity of the generated graph features; after network parameters are finely adjusted through training, a generator provides complete and real node characteristics as output according to input partial characteristic data;
s33: in an auto-encoder based feature generator: firstly, input graph characteristic matrix of network is expressed
Figure FDA0003301602680000061
Figure FDA0003301602680000062
Normalized to [0,1 ]]A range; the generator is formed by connecting 3 continuous space-time coding blocks in series and is used for feature learning; each space-time coding block is composed of a gate control linear unit and a graph convolution unit which are connected in series: wherein, the gate control linear unit is a convolution layer output without nonlinearity multiplied by a convolution layer output after nonlinear activation; thus, given an input matrix
Figure FDA0003301602680000063
Gated linear cell operation
Figure FDA0003301602680000064
Output of (2)
Figure FDA0003301602680000065
Comprises the following steps:
Figure FDA0003301602680000066
wherein
Figure FDA0003301602680000067
Is a convolution kernel parameter;
meanwhile, the graph convolution unit adopts the connectivity of the graph, namely an adjacent matrix, as a convolution filter of neighborhood mixing; concept based on spectrogram convolution, graph convolution Unit operation
Figure FDA0003301602680000068
Is shown for graph signals
Figure FDA0003301602680000069
Of the non-linear function of, the output of
Figure FDA00033016026800000610
Is equal to
Figure FDA00033016026800000611
Wherein, the Fourier base of the graph
Figure FDA00033016026800000612
Is a feature vector matrix of the normalized graph laplacian matrix L; laplace matrix
Figure FDA00033016026800000613
Wherein
Figure FDA00033016026800000614
Is a matrix of cells, which is,
Figure FDA00033016026800000615
Figure FDA00033016026800000616
is a diagonal matrix whose diagonal element values dii＝∑jWij；
Figure FDA00033016026800000617
Is the eigenvalue diagonal matrix of matrix L, so Θ (Λ) is also the diagonal matrix, and Θ is the graph convolution kernel parameter;
given the above-described gated linear unit and graph convolution unit operations, a spatio-temporal coding block is defined as
Figure FDA00033016026800000618
Operation, i.e. giving the input matrix of the jth space-time coding block
Figure FDA00033016026800000619
And nuclear parameters xij＝[ΓjΘj]Output of space-time coded blocks
Figure FDA00033016026800000620
Comprises the following steps:
Figure FDA0003301602680000071
wherein σ1(·)，σ2(. is an activation function; therefore, the output after the three space-time coding blocks is:
Figure FDA0003301602680000072
Figure FDA0003301602680000073
Figure FDA0003301602680000074
wherein all activation functions are elu activation functions; after three space-time coding blocks, converting the input partial graph characteristics into an N multiplied by 512 middle characteristic matrix; the matrix is then decoded by three space-time decoding blocks; the structure of the space-time decoding block is the same as that of the space-time coding block, and the network structure is expressed by a formula as follows:
Figure FDA0003301602680000075
Figure FDA0003301602680000076
all of the activation functions described above are elu functions,
Figure FDA0003301602680000077
the activation function σ in this step2Activating a function for sigmoid; reconstructing graph features through three space-time decoding blocks;
s34: the graph feature discriminator is used for distinguishing the graph feature matrix generated by the feature generator
Figure FDA0003301602680000078
And a true graph feature matrix X; the network starts with a space-time coding block whose output is expanded into a one-dimensional matrixThen feeding in three full-connection layers, namely extracting hidden graph features from input data by utilizing a space-time coding block, then classifying the extracted features by fully-communicated neuron cooperation in each full-connection layer, and judging whether the input data is real or not; finally, a sigmoid function is used for generating a probability between 0 and 1, which indicates the real probability of the input feature discrimination of the discriminator;
adding a LayerNorm layer after each operation and each full connection layer in the space-time coding block, and regarding the input
Figure FDA0003301602680000079
Or X, the output of the space-time coding block is as follows:
Figure FDA00033016026800000710
wherein LN (-) represents a LayerNorm layer, will
Figure FDA00033016026800000711
Spread into a one-dimensional matrix
Figure FDA00033016026800000712
Taking the obtained data as the input of three subsequent fully-connected layers, wherein the activation function of each layer is elu activation function and adding a LayerNorm operation, the activation function of the last layer is sigmoid function, and the output of the final discriminator is
Figure FDA0003301602680000081
For input
Figure FDA0003301602680000082
The output of the discriminator is
Figure FDA0003301602680000083
For input X, the output of the discriminator is
Figure FDA0003301602680000084
S35: in the generation of confrontation network training in confrontation learning, the training target of the feature generator is expressed as:
Figure FDA0003301602680000085
the training targets for the feature discriminators are represented as:
Figure FDA0003301602680000086
5. the method for estimating urban mobile source pollutant emission according to claim 4, characterized in that: in the step S35, a gradient penalty term is added to the loss function of the feature discriminator in the training process, so that the loss function L of the feature discriminatorDComprises the following steps:
Figure FDA0003301602680000087
where λ is the tuning parameter.
6. The method for estimating urban mobile source pollutant emission according to claim 5, characterized in that: in step S35, λ is an adjustment parameter, and its value is 10.
7. The method for estimating urban mobile source pollutant emission according to claim 5, characterized in that: in step S35, an estimated mean square error term is added to the loss function of the generator, so that the loss function L of the feature generatorGComprises the following steps:
Figure FDA0003301602680000088
the feature generator and feature arbiter are alternately trained using an Adam optimizer until the two loss functions converge to a stationary state.
8. A system for estimating urban mobile source pollution emission by using incomplete information is characterized in that: comprises the following units of a first unit, a second unit,
the data acquisition and processing unit is used for acquiring motor vehicle driving related data in urban traffic and urban traffic network data preprocessing and constructing a data space;
and the traffic emission value calculating unit is used for dividing the road sections into 2 types according to the data density degree obtained on the road sections: the information missing road section which loses traffic mode information due to sparse acquired data and the information complete road section which has dense enough data; constructing a continuous speed field and an inferred flow value for the information complete road section through the motor vehicle driving related data, and calculating a corresponding traffic emission value through a microscopic emission factor model according to the continuous speed field and the inferred flow value;
and the discharge distribution information filling unit is used for respectively extracting the topological characteristics of the road network and the time sequence dependency of the motor vehicle discharge sequence on the information complete road section by utilizing the graph convolution unit and the gate control linear unit, thereby reconstructing the discharge sequence on the information missing road section and filling the missing discharge distribution information.
CN202111192115.1A 2021-10-13 2021-10-13 Method and system for estimating urban mobile source pollution emission by using incomplete information Pending CN114036135A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111192115.1A CN114036135A (en) 2021-10-13 2021-10-13 Method and system for estimating urban mobile source pollution emission by using incomplete information

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111192115.1A CN114036135A (en) 2021-10-13 2021-10-13 Method and system for estimating urban mobile source pollution emission by using incomplete information

Publications (1)

Publication Number Publication Date
CN114036135A true CN114036135A (en) 2022-02-11

Family

ID=80141172

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111192115.1A Pending CN114036135A (en) 2021-10-13 2021-10-13 Method and system for estimating urban mobile source pollution emission by using incomplete information

Country Status (1)

Country Link
CN (1) CN114036135A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114626169A (en) * 2022-03-03 2022-06-14 北京百度网讯科技有限公司 Traffic network optimization method, device, equipment, readable storage medium and product
CN115422694A (en) * 2022-11-03 2022-12-02 深圳市城市交通规划设计研究中心股份有限公司 Road section missing flow calculation method, electronic equipment and storage medium
CN115964621A (en) * 2023-03-17 2023-04-14 中国科学技术大学 Regional road network exhaust emission data completion method
CN116910569A (en) * 2023-09-13 2023-10-20 中电车联信安科技有限公司 Vehicle carbon emission big data monitoring method, system, device and storage medium
CN117576918A (en) * 2024-01-17 2024-02-20 四川国蓝中天环境科技集团有限公司 Urban road flow universe prediction method based on multi-source data

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114626169A (en) * 2022-03-03 2022-06-14 北京百度网讯科技有限公司 Traffic network optimization method, device, equipment, readable storage medium and product
CN115422694A (en) * 2022-11-03 2022-12-02 深圳市城市交通规划设计研究中心股份有限公司 Road section missing flow calculation method, electronic equipment and storage medium
CN115964621A (en) * 2023-03-17 2023-04-14 中国科学技术大学 Regional road network exhaust emission data completion method
CN115964621B (en) * 2023-03-17 2023-05-09 中国科学技术大学 Regional road network tail gas emission data complement method
CN116910569A (en) * 2023-09-13 2023-10-20 中电车联信安科技有限公司 Vehicle carbon emission big data monitoring method, system, device and storage medium
CN116910569B (en) * 2023-09-13 2023-12-19 中电车联信安科技有限公司 Vehicle carbon emission big data monitoring method, system, device and storage medium
CN117576918A (en) * 2024-01-17 2024-02-20 四川国蓝中天环境科技集团有限公司 Urban road flow universe prediction method based on multi-source data
CN117576918B (en) * 2024-01-17 2024-04-02 四川国蓝中天环境科技集团有限公司 Urban road flow universe prediction method based on multi-source data

Similar Documents

Publication Publication Date Title
CN114036135A (en) Method and system for estimating urban mobile source pollution emission by using incomplete information
Liao et al. Deep learning for air quality forecasts: a review
Yu et al. Real-time traffic speed estimation with graph convolutional generative autoencoder
Kim et al. Urban traffic flow prediction system using a multifactor pattern recognition model
CN111080070B (en) Urban land utilization cellular automaton simulation method based on space errors
US20160125307A1 (en) Air quality inference using multiple data sources
CN113591380B (en) Traffic flow prediction method, medium and equipment based on graph Gaussian process
Torrens Calibrating and validating cellular automata models of urbanization
Lin et al. A spatial-temporal hybrid model for short-term traffic prediction
Ou et al. Learn, assign, and search: real-time estimation of dynamic origin-destination flows using machine learning algorithms
CN113723715B (en) Method, system, equipment and storage medium for automatically matching public transport network with road network
CN108090624B (en) Urban ecological safety simulation and prediction method for improving cellular automaton
Ye et al. Short-term prediction of available parking space based on machine learning approaches
Huang et al. Physics-informed deep learning for traffic state estimation: Illustrations with LWR and CTM models
Xu et al. Understanding the usage patterns of bicycle-sharing systems to predict users’ demand: a case study in Wenzhou, China
CN113779169B (en) Space-time data stream model self-enhancement method
Iskandaryan et al. Graph neural network for air quality prediction: A case study in madrid
Chen et al. A novel predictor for exploring PM2. 5 spatiotemporal propagation by using convolutional recursive neural networks
Zhan et al. Prediction of air quality in major cities of China by deep learning
Wang et al. Abnormal trajectory detection based on geospatial consistent modeling
Zhao et al. Traffic emission estimation under incomplete information with spatiotemporal convolutional GAN
Li et al. MGC-GAN: Multi-graph convolutional generative adversarial networks for accurate citywide traffic flow prediction
Li et al. The parallel and precision adaptive method of marine lane extraction based on QuadTree
Zhong et al. Estimating link flows in road networks with synthetic trajectory data generation: Inverse reinforcement learning approach
Wu et al. Hierarchical Bayesian network based incremental model for flood prediction

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