CN108288109A - Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks - Google Patents

Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks Download PDF

Info

Publication number
CN108288109A
CN108288109A CN201810027031.4A CN201810027031A CN108288109A CN 108288109 A CN108288109 A CN 108288109A CN 201810027031 A CN201810027031 A CN 201810027031A CN 108288109 A CN108288109 A CN 108288109A
Authority
CN
China
Prior art keywords
space
time
residual error
data
network
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
CN201810027031.4A
Other languages
Chinese (zh)
Inventor
杨钰潇
李泽瑞
杜晓冬
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.)
Anhui Youth Tiancheng Technology Co Ltd
Original Assignee
Anhui Youth Tiancheng Technology Co Ltd
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 Anhui Youth Tiancheng Technology Co Ltd filed Critical Anhui Youth Tiancheng Technology Co Ltd
Priority to CN201810027031.4A priority Critical patent/CN108288109A/en
Publication of CN108288109A publication Critical patent/CN108288109A/en
Pending legal-status Critical Current

Links

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
    • G06N3/045—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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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

Abstract

The present invention provides a kind of motor-vehicle tail-gas concentration prediction methods based on LSTM depth space-time residual error networks, utilize the powerful data None-linear approximation ability of deep learning method and self-learning capability, it establishes and is based on LSTM depth space-time residual error network structures, realize the spatio-temporal prediction of urban mobile tail gas concentration, acquire urban mobile tail gas space-time data and external influence factors data, urban mobile tail gas space-time data is stored, training, prediction, directly carry out the forecast analysis of the double dimensions of time and space, consider weather simultaneously, the external conditions such as festivals or holidays, with one two layers fully-connected network simulate external influence factors feature, keep prediction result more comprehensively accurate.

Description

Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks
Technical field
The invention belongs to environmental monitoring technology fields, are related to a kind of motor-vehicle tail-gas concentration prediction method, specifically a kind of Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks.
Background technology
With social development and city progress, in recent years, vehicles number continues to increase in urban area, and many societies are asked Topic generates therewith, such as urban traffic jam is serious, traffic accident increases, tail gas pollution of motor-driven vehicle, drunk driving.Beijing, The big cities such as Shanghai, Guangzhou, motor vehicle have become the first of the pollutants such as discharge carbon monoxide, nitrogen oxides, hydrocarbon Big pollution sources.Since discharge of automobile exhaust fumes is mainly between 0.3 meter to 2 meters, the exactly respiration range of human body, to human body Health cost is very serious --- stimulation respiratory tract, make the immunity degradation of respiratory system, cause exposed population group's chronic bronchitis, The a series of symptoms such as bronchitis and the raising of dyspneic incidence, decline in pulmonary function.Strong carcinogen contained in tail gas Matter --- benezene material can cause lung cancer, thyroid cancer etc..
In order to improve the social concern of these motor-vehicle tail-gas generation, motor-driven vehicle in urban road is timely and accurately understood The emission behaviour of gas is established and is suitable for urban area motor-vehicle tail-gas concentration prediction model, may be implemented to each road of city road network Section exhaust emissions it is real-time prediction and trend estimation, for formulation city moving source Air Pollutant Emission supervision provide decision according to According to.When the pernicious gas constituent concentration in certain region reaches certain soiling value, platform can be sent out to government decision department restricting the number, The suggestions such as shunting, restricted driving, so as to reduce the air pollution caused by on-road vehicle.
There is nonlinearity, pollutant concentration to be influenced by the multiple elements of ambient enviroment for automotive emission, including Meteorological condition, air environment, geographical environment, condition of road surface, traffic flow factor etc., and the prior art can only be from time or space Single dimension carries out pollutant concentration prediction prediction, and prediction effect is undesirable.
Invention content
In view of the deficiencies of the prior art, the present invention provides a kind of motor-driven vehicles based on LSTM depth space-time residual error networks Motor-vehicle tail-gas concentration data is directly stored in the form of two-dimension time-space distribution grid trrellis diagram, is instructed by gas concentration prediction technique Practice, prediction, directly carry out the forecast analysis of the double dimensions of time-space, while considering the external conditions such as weather, festivals or holidays, with one A two layers of fully-connected network simulates external influence factors feature, keeps prediction result more comprehensively accurate.
The purpose of the present invention can be achieved through the following technical solutions：
Based on the motor-vehicle tail-gas concentration prediction method of LSTM depth space-time residual error networks, include the following steps：
Step S1 acquires in the specified time of target cities the urban mobile tail gas space-time data at specified time interval and outer Portion's influence factor data；
Step S2 pre-processes the step S1 total datas acquired；
Step S3 constructs the depth space-time residual error network structure based on LSTM；
Step S4, according to the depth space-time residual error based on LSTM constructed in the pretreated data of step S2 and step S3 Network structure constructs training dataset, validation data set and the test data of the depth space-time residual error network structure based on LSTM Collection；
Training dataset in step S4 is sent into the depth space-time residual error based on LSTM constructed in step S3 by step S5 In network structure, network training is carried out, is then used obtained by validation data set and the training of test data set pair in step S4 Parameter is adjusted, and steps up precision of prediction；
Step S6, by the depth space-time residual error network based on LSTM after network training, verification, test in step S5 Prediction model of the structure as urban mobile tail gas concentration, will be between specified time in pretreated target cities specified time Every urban mobile tail gas space-time data and external influence factors data input the network structure, finally can be obtained following a certain The moment city or its some region of automotive emission forecast result.
Further, the urban mobile tail gas space-time data acquired in the step S1 is specially to acquire one-year age It is interior, every 30 minutes urban mobile tail gas space-time datas, first by target cities region from the grid for being geographically divided into 64*64 Then lattice acquire every 30 minutes motor-vehicle tail-gas data of each grid to get to the matrix of a single pass similar picture
The external influence factors data acquired in the step S1 be and urban mobile tail gas space-time data same time Interior weather data, temperature record, air speed data and festivals or holidays data.
Further, the data prediction in the step S2 specifically includes one-hot coding, missing values processing and normalization Three aspects of processing.
Further, the depth space-time residual error network structure based on LSTM in the step S3 includes the simulated maneuver tailstock The external factor that the space-time residual error sub-network of gas concentration spatial character and time response, simulation external factor influence tail gas concentration Neural network and output Fusion Module.
Further, the space-time residual error sub-network, temporally above the distance at range prediction moment will count from the near to the distant According to current group, recent group and group at a specified future date is divided into, it is respectively fed to the identical space-time residual error sub-network of three structures；
Space-time residual error sub-network structure having the same under three times shares L+2 layers, and the 1st layer is convolutional layer Conv1 contains 64 3 × 3 convolution kernels, and the 2nd layer to L+1 layers are residual unit, and L+2 layers are convolutional layer Conv2, are contained 23 × 3 convolution kernels；
The list entries for currently organizing corresponding space-time residual error sub-network is
lcCurrently to organize the length of data sequence, c is to indicate to organize down in the recent period Mark, exports and is
The list entries for organizing corresponding space-time residual error sub-network in the recent period is
lpIt is the length of recent group data sequence, p is to indicate recent group Subscript, export and be
The list entries that long term organizes corresponding space-time residual error sub-network is
lqIt is the length of group data sequence at a specified future date, q is to indicate at a specified future date group Subscript, export and be
Further, the external factor neural network indicates that some time carves the outside of certain grid with the vector that a length is 12 Factor data, wherein vectorial preceding 5 values indicate that weather, the 6th value are temperature, and the 7th value is wind speed, and rear 5 values indicate section Holiday data；
The external factor neural network is made of two full articulamentums, and first layer is the embeding layer of every sub- factor, After have an activation primitive, the second layer, which carries out dimension conversion, makes it export XExtAnd XtDimension is identical.
Further, the output Fusion Module carry out current group, recent group, three space-time residual error sub-networks of group at a specified future date it is defeated The fusion gone out, using a kind of matrix fusion method based on parameter, formula is as follows：
Wherein XResIt is the output after three space-time residual error sub-network fusions, Wc、Wp、WqIt is respectively by three sections of time effects Can learning parameter, ° indicate Hadamard product calculations, i.e. corresponding element multiplication in two matrixes；
Then by the result X after fusionResWith the output X of neural networkExtIt carries out directly being added fusion, result after being added The output of overall network, the i.e. predicted value of t time intervals are obtained by tanh activation primitives
For wherein tanh as activation primitive, expression formula is as follows：
Wherein x is the data of input.Result is exported on [- 1,1] section.
Beneficial effects of the present invention：The present invention provides a kind of motor-vehicle tail-gas based on LSTM depth space-time residual error networks Concentration prediction method is based on using the powerful data None-linear approximation ability of deep learning method and self-learning capability, foundation LSTM depth space-time residual error network structures realize the spatio-temporal prediction of urban mobile tail gas concentration, acquire urban mobile tail gas Urban mobile tail gas space-time data stored, trained, predicted by space-time data and external influence factors data, directly into The forecast analysis of the double dimensions of row time-space, while considering the external conditions such as weather, festivals or holidays, with two layers full connection Network analog external influence factors feature keeps prediction result more comprehensively accurate.
Description of the drawings
Fig. 1 is flow chart of the method for the present invention.
Fig. 2 is that the present invention is based on the depth space-time residual error schematic network structures of LSTM.
Fig. 3 is that the present invention is based on the schematic diagrames of the residual unit of LSTM.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation describes, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on Embodiment in the present invention, those of ordinary skill in the art are obtained all other without creative efforts Embodiment shall fall within the protection scope of the present invention.
As shown in Figure 1, the present invention provides a kind of urban mobile tail gas based on LSTM depth space-time residual error networks is dense Prediction technique is spent, following steps are specifically included：
Step S1 acquires in the specified time of target cities the urban mobile tail gas space-time data at specified time interval and outer Portion's influence factor data.
Wherein, urban mobile tail gas space-time data is specially to acquire in one-year age, every 30 minutes urban automobiles Then tail gas space-time data acquires every 30 points of each grid first by target cities region from the grid for being geographically divided into 64*64 The motor-vehicle tail-gas data of clock are to get to the matrix of a single pass similar picture
Wherein, external influence factors data are acquisition and the day destiny in urban mobile tail gas space-time data same time According to, temperature record, air speed data and festivals or holidays data.Specifically, weather data is divided into 16 classes：Fine, cloudy, clear to cloudy, mist, rain Press from both sides snow, thunder shower, light rain, heavy rain, moderate rain, heavy rain, heavy snow, slight snow, moderate snow, rain, hail, cloudy.Festivals or holidays data totally 18 class： New Year, New Year's Eve, the Lantern Festival, imperial new line, Valentine's Day, the Ching Ming Festival, International Working Woman's Day, the Mother's Day, the Dragon Boat Festival, International Labour Day, Children's Day, father Section, the seventh evening of the seventh moon in lunarcalendar, the Mid-autumn Festival, National Day, the Double Ninth Festival, All Saints' Day, Christmas Day.
Step S2 pre-processes the step S1 total datas acquired, specifically includes one-hot coding, missing values processing With three aspects of normalized.
One-hot coding (One-hot Encoder)：One-hot coding is carried out to 16 kinds of weather datas and 18 kinds of festivals or holidays data, It is translated into binary vector.
Missing values are filled up：For the data of excalation, it is averaged with rear m data using preceding m of missing data Method filled up, ensure the completeness and adequate of data, ensure the accuracy and confidence level of prediction result.The present embodiment In, temperature, the air speed data in the emission data and external factor of missing are filled up, preferably m=30.
Normalized：To temperature, the wind speed number in urban mobile tail gas space-time data and external influence factors data According to being normalized, ensure that the input data of different data range plays identical effect.Specifically, with each grid For unit, to the region, 1 year total data is normalized as the following formula.
WhereinIt is the data after i-th of grid normalization, x(i)Original number before being i-th of grid normalization According to,WithIt is the maximum value and minimum value in all data of i-th of grid respectively.
Step S3 constructs the depth space-time residual error network structure based on LSTM.
Wherein, as shown in Fig. 2, the depth space-time residual error network structure based on LSTM includes that simulated maneuver tailstock gas concentration is empty Between the space-time residual error sub-network of characteristic and time response, simulation external factor external factor neural network that tail gas concentration is influenced And output Fusion Module.
Wherein, space-time residual error sub-network, temporally above the distance at range prediction moment from the near to the distant divides data successively For current group, recent group and group at a specified future date, the identical space-time residual error sub-network of three structures is respectively fed to model motor-vehicle tail-gas The spatial character of concentration, single space-time residual error sub-network is L+2 layers shared, and first layer and last layer are convolutional layer, from second Layer is residual unit to L+1 layers, and the structure chart of residual unit is as shown in Figure 3.
For currently organizing corresponding space-time residual error sub-network, first convolutional layer Conv1 has 64 3 × 3 convolution kernels, Its input/output relation meets following formula：
WhereinIt is the emission data of the 0th moment input,It is the emission data of the 1st moment input,For Weight,For amount of bias, * representing matrix multiplyings.
Layer is residual unit from the second layer to L+1, and the input/output relation of residual unit meets following formula：
WhereinResidual error function, i.e. the combination of "+LSTM layers of ReLu functions ", as shown in Figure 3.Refer to first of residual error The unit all parameters to be learnt.
The last one convolutional layer Conv2 has 23 × 3 convolution kernels.
Then it is L+2 layers shared currently to organize corresponding space-time residual error sub-network, list entries is lcCurrently to organize the length of data sequence, c is the subscript for indicating currently to organize, and exports and isSimilarly, it organizes in the recent period corresponding The list entries of space-time residual error sub-networklpIt is the length of recent group data sequence, P is the subscript for indicating to organize in the recent period, exports and isThe list entries that long term organizes corresponding space-time residual error sub-network islqIt is the length of group data sequence at a specified future date, q is to indicate group subscript at a specified future date, exports and is
Wherein, to the one-hot coding of weather and festivals or holidays data and to temperature, wind before external factor neural network is passed through The normalization pretreatment of speed, the vector for being 12 with a length indicate that some time carves the external factor data of certain grid.Wherein, vectorial Preceding 5 values indicate that weather, the 6th value are temperature, and the 7th value is wind speed, and rear 5 values indicate festivals or holidays data.
External factor neural network is made of two full articulamentums, as shown in Fig. 2, first layer is the insertion of every sub- factor Layer, followed by activation primitive, the second layer, which carries out dimension conversion, makes it export XExtAnd XtDimension is identical.
Wherein, output Fusion Module carries out the fusion of current group, recent group and group at a specified future date three space-time residual error networks output, Using a kind of matrix fusion method based on parameter, formula is as follows：
Wherein XResIt is the output after three space-time residual error network integrations, Wc、Wp、WqIt is respectively by three sections of time effects Can learning parameter,.Indicate Hadamard product calculations, i.e. corresponding element multiplication in two matrixes.
Then by the result X after fusionResWith the output X of neural networkExtIt carries out directly being added fusion, result after being added The output of overall network, the i.e. predicted value of t time intervals are obtained by tanh activation primitives
For wherein tanh as activation primitive, expression formula is as follows：
Wherein x is the data of input.Result is exported on [- 1,1] section.
Step S4, according to the depth space-time residual error based on LSTM constructed in the pretreated data of step S2 and step S3 Network structure constructs training dataset, validation data set and the test data of the depth space-time residual error network structure based on LSTM Collection.
Specifically, pretreated data in step S2 are divided into training dataset, validation data set and test data set Three data sets, three parts data proportion are followed successively by 70%, 10%, 20%, the depth based on LSTM after being respectively used to Training, verification and the test of space-time residual error network model.The above value is preferred, can be adjusted according to actual conditions.
Training dataset in step S4 is sent into the depth space-time residual error based on LSTM constructed in step S3 by step S5 In network structure, network training is carried out, is then used obtained by validation data set and the training of test data set pair in step S4 Parameter is adjusted, and steps up precision of prediction.
In step S5, when training the depth space-time residual error network based on LSTM, the predicted value of t time intervals isIt is real The tail gas concentration value on border is Xt, the mean square error that Loss functions are the two is defined, is shown below, the target of network training is to make The equal error is minimum.
In step S5, when depth space-time residual error network of the training based on LSTM, using back propagation and Adam algorithms, Middle Adam algorithms store the exponential damping average value of previous squared gradient, and the exponential damping for maintaining previous gradient is average Value, be arranged initial learning rate be 0.01, weight decaying take 0.0005, mini-batches sizes be 32, backpropagation when A length of 20 time step of spacer step, i.e., per 20K iteration after learning rate divided by 10.
In step S5, validation data set is during network training per 1000 progress one-time authentications of iteration, final ratio Compared with testloss and trainloss.When testloss is no longer reduced, network training is terminated, is indicated for motor-vehicle tail-gas The depth space-time residual error network structure training based on LSTM of concentration prediction is completed.Parameter value in above-mentioned training process is Preferred value can be adjusted according to actual conditions.
In step S5, test data set carries out defeated after the completion of the depth space-time residual error network structure training based on LSTM Enter, according to the city region motor-vehicle tail-gas concentration data at a certain moment in future that historical data is predicted, checks the value With the gap of desired value, and then the parameter in the depth space-time residual error network structure based on LSTM is adjusted, and is stepped up pre- Survey precision.
Step S6, by the depth space-time residual error network based on LSTM after network training, verification, test in step S5 Prediction model of the structure as urban mobile tail gas concentration, will be between specified time in pretreated target cities specified time Every urban mobile tail gas space-time data and external influence factors data input the network structure, finally can be obtained following a certain The moment city or its some region of automotive emission forecast result.
Existing pollutant concentration prediction technique can only realize the prediction of single dimension on time shaft, and the present invention is with space-time number The residual error list based on LSTM is added on the basis of extracting the internal characteristics between data with convolutional neural networks according to as input Member avoids the phenomenon that efficiency is greatly reduced or even is not easy to restrain when training caused by due to the convolution number of plies is excessive；At the same time, lead to External influence factors sub-network is crossed, making prediction result not only functionally realizes the prediction of the double dimensions of time-space, more fully Influence of the external factor such as weather, temperature, wind speed to motor-driven vehicle gas concentration is embodied, there is certain social value and reality Meaning.
In the description of this specification, the description of reference term " one embodiment ", " example ", " specific example " etc. means Particular features, structures, materials, or characteristics described in conjunction with this embodiment or example are contained at least one implementation of the present invention In example or example.In the present specification, schematic expression of the above terms may not refer to the same embodiment or example. Moreover, particular features, structures, materials, or characteristics described can be in any one or more of the embodiments or examples to close Suitable mode combines.
Above content is only to structure of the invention example and explanation, affiliated those skilled in the art couple Described specific embodiment does various modifications or additions or substitutes by a similar method, without departing from invention Structure or beyond the scope defined by this claim, is within the scope of protection of the invention.

Claims (7)

1. the motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks, which is characterized in that including following step Suddenly：
Step S1 acquires the urban mobile tail gas space-time data at specified time interval and external shadow in the specified time of target cities Ring factor data；
Step S2 pre-processes the step S1 total datas acquired；
Step S3 constructs the depth space-time residual error network structure based on LSTM；
Step S4, according to the depth space-time residual error network based on LSTM constructed in the pretreated data of step S2 and step S3 Structure constructs training dataset, validation data set and the test data set of the depth space-time residual error network structure based on LSTM；
Training dataset in step S4 is sent into the depth space-time residual error network based on LSTM constructed in step S3 by step S5 In structure, network training is carried out, then uses the parameter of the validation data set and test data set pair training gained in step S4 It is adjusted, steps up precision of prediction；
Step S6, by the depth space-time residual error network structure based on LSTM after network training, verification, test in step S5 As the prediction model of urban mobile tail gas concentration, by specified time interval in pretreated target cities specified time Urban mobile tail gas space-time data and external influence factors data input the network structure, and the following a certain moment finally can be obtained The city or its some region of automotive emission forecast result.
2. special according to the motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks in claim 1 Sign is：The urban mobile tail gas space-time data acquired in the step S1 is specially to acquire in one-year age, every 30 minutes Urban mobile tail gas space-time data, first then target cities region is acquired from the grid for being geographically divided into 64*64 Each every 30 minutes motor-vehicle tail-gas data of grid are to get to the matrix of a single pass similar picture
The external influence factors data acquired in the step S1 be in urban mobile tail gas space-time data same time Weather data, temperature record, air speed data and festivals or holidays data.
3. special according to the motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks in claim 1 Sign is：Data prediction in the step S2 specifically includes one-hot coding, missing values processing and three sides of normalized Face.
4. special according to the motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks in claim 1 Sign is：The depth space-time residual error network structure based on LSTM in the step S3 includes simulated maneuver tail gas concentration space The external factor neural network that the space-time residual error sub-network of characteristic and time response, simulation external factor influence tail gas concentration with And output Fusion Module.
5. special according to the motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks in claim 4 Sign is：The space-time residual error sub-network, temporally above data are divided by the distance at range prediction moment from the near to the distant Current group, recent group and group at a specified future date, are respectively fed to the identical space-time residual error sub-network of three structures；
Space-time residual error sub-network structure having the same under three times shares L+2 layers, and the 1st layer is convolutional layer Conv1, is contained There are 64 3 × 3 convolution kernels, the 2nd layer to L+1 layers are residual unit, and L+2 layers are convolutional layer Conv2, contain 23 × 3 Convolution kernel；
The list entries for currently organizing corresponding space-time residual error sub-network islcFor long term The length of group data sequence, c is the subscript for indicating currently to organize, and exports and is
The list entries for organizing corresponding space-time residual error sub-network in the recent period islpIt is The length of group data sequence in the recent period, p is the subscript for indicating to organize in the recent period, exports and is
The list entries that long term organizes corresponding space-time residual error sub-network islqIt is The length of long term group data sequence, q are the subscripts for indicating group at a specified future date, export and are
6. special according to the motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks in claim 4 Sign is：The external factor neural network indicates that some time carves the external factor data of certain grid with the vector that a length is 12, Wherein, vectorial preceding 5 values indicate that weather, the 6th value are temperature, and the 7th value is wind speed, and rear 5 values indicate festivals or holidays data；
The external factor neural network is made of two full articulamentums, and first layer is the embeding layer of every sub- factor, followed by Activation primitive, the second layer, which carries out dimension conversion, makes it export XExtAnd XtDimension is identical.
7. special according to the motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks in claim 4 Sign is：The output Fusion Module carries out current group, recent group, the fusion at a specified future date for organizing the output of three space-time residual error sub-networks, Using a kind of matrix fusion method based on parameter, formula is as follows：
Wherein XResIt is the output after three space-time residual error sub-network fusions, Wc、Wp、WqBeing respectively can by three sections of time effects Learning parameter,Indicate Hadamard product calculations, i.e. corresponding element multiplication in two matrixes；
Then by the result X after fusionResWith the output X of neural networkExtIt carries out directly being added fusion, result after being added passes through Tanh activation primitives obtain the output of overall network, the i.e. predicted value of t time intervals
For wherein tanh as activation primitive, expression formula is as follows：
Wherein x is the data of input.Result is exported on [- 1,1] section.
CN201810027031.4A 2018-01-11 2018-01-11 Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks Pending CN108288109A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810027031.4A CN108288109A (en) 2018-01-11 2018-01-11 Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810027031.4A CN108288109A (en) 2018-01-11 2018-01-11 Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks

Publications (1)

Publication Number Publication Date
CN108288109A true CN108288109A (en) 2018-07-17

Family

ID=62835034

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810027031.4A Pending CN108288109A (en) 2018-01-11 2018-01-11 Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks

Country Status (1)

Country Link
CN (1) CN108288109A (en)

Cited By (26)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109086946A (en) * 2018-09-11 2018-12-25 东南大学 A kind of polluted gas emitted smoke method of conventional energy resource and new energy public transit vehicle
CN109492822A (en) * 2018-11-24 2019-03-19 上海师范大学 Air pollutant concentration time-space domain interaction prediction method
CN109948716A (en) * 2019-03-25 2019-06-28 中国民航大学 A kind of airport delay prediction technique based on region residual sum LSTM network
CN109991685A (en) * 2019-04-03 2019-07-09 北京市天元网络技术股份有限公司 A kind of precipitation prediction technique and device based on more LSTM Model Fusions
CN110046787A (en) * 2019-01-15 2019-07-23 重庆邮电大学 A kind of urban area charging demand for electric vehicles spatio-temporal prediction method
CN110738355A (en) * 2019-09-19 2020-01-31 河源职业技术学院 urban waterlogging prediction method based on neural network
CN110738367A (en) * 2019-10-12 2020-01-31 东北大学秦皇岛分校 traffic exhaust emission prediction method based on deep residual error network
WO2020043473A1 (en) * 2018-08-31 2020-03-05 International Business Machines Corporation Data prediction
CN110929793A (en) * 2019-11-27 2020-03-27 谢国宇 Time-space domain model modeling method and system for ecological environment monitoring
CN111144666A (en) * 2020-01-02 2020-05-12 吉林大学 Ocean thermocline prediction method based on deep space-time residual error network
CN111222666A (en) * 2018-11-26 2020-06-02 中兴通讯股份有限公司 Data calculation method and device
CN111260121A (en) * 2020-01-12 2020-06-09 桂林电子科技大学 Urban-range pedestrian flow prediction method based on deep bottleneck residual error network
CN111291860A (en) * 2020-01-13 2020-06-16 哈尔滨工程大学 Anomaly detection method based on convolutional neural network feature compression
CN111415050A (en) * 2020-04-27 2020-07-14 新智数字科技有限公司 Short-term load prediction method and short-term load prediction model training method and device
CN111625999A (en) * 2020-05-29 2020-09-04 中南林业科技大学 Forest fire early warning model and system based on deep learning technology
CN111882122A (en) * 2020-07-16 2020-11-03 成都市高博汇科信息科技有限公司 Traffic accident black point prediction method based on deep learning and space-time big data
CN112132264A (en) * 2020-09-11 2020-12-25 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) Regional exhaust emission prediction method and system based on space-time residual perception network
CN112215408A (en) * 2020-09-24 2021-01-12 交控科技股份有限公司 Rail transit passenger flow volume prediction method and device
CN112598170A (en) * 2020-12-18 2021-04-02 中国科学技术大学 Vehicle exhaust emission prediction method and system based on multi-component fusion time network
CN112734028A (en) * 2020-12-28 2021-04-30 三峡大学 Modeling method for prediction model of concentration of dissolved gas in transformer oil
CN113077081A (en) * 2021-03-26 2021-07-06 航天科工智能运筹与信息安全研究院(武汉)有限公司 Traffic pollution emission prediction method
CN113095535A (en) * 2020-01-08 2021-07-09 普天信息技术有限公司 Flow prediction method and device based on deep space-time residual error network
CN113222236A (en) * 2021-04-30 2021-08-06 中国科学技术大学先进技术研究院 Data distribution self-adaptive cross-regional exhaust emission prediction method and system
CN113222206A (en) * 2021-01-29 2021-08-06 太原理工大学 Traffic state prediction method based on ResLS-C deep learning combination
CN113222247A (en) * 2021-05-11 2021-08-06 润联软件系统(深圳)有限公司 Chemical raw material price and purchase quantity prediction method, device and related equipment
CN113283588A (en) * 2021-06-03 2021-08-20 青岛励图高科信息技术有限公司 Near-shore single-point wave height forecasting method based on deep learning

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080243342A1 (en) * 1995-12-12 2008-10-02 Automotive Technologies International, Inc. Side Curtain Airbag With Inflator At End
WO2016177722A1 (en) * 2015-05-05 2016-11-10 Medizinische Universität Wien Computerized device and method for processing image data
CN106599520A (en) * 2016-12-31 2017-04-26 中国科学技术大学 LSTM-RNN model-based air pollutant concentration forecast method
CN106845371A (en) * 2016-12-31 2017-06-13 中国科学技术大学 A kind of city road network automotive emission remote sensing monitoring system
CN107562784A (en) * 2017-07-25 2018-01-09 同济大学 Short text classification method based on ResLCNN models

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080243342A1 (en) * 1995-12-12 2008-10-02 Automotive Technologies International, Inc. Side Curtain Airbag With Inflator At End
WO2016177722A1 (en) * 2015-05-05 2016-11-10 Medizinische Universität Wien Computerized device and method for processing image data
CN106599520A (en) * 2016-12-31 2017-04-26 中国科学技术大学 LSTM-RNN model-based air pollutant concentration forecast method
CN106845371A (en) * 2016-12-31 2017-06-13 中国科学技术大学 A kind of city road network automotive emission remote sensing monitoring system
CN107562784A (en) * 2017-07-25 2018-01-09 同济大学 Short text classification method based on ResLCNN models

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
JUNBO ZHANG ET AL: "Deep Spatio-Temporal Residual Networks for Citywide Crowd Flows Prediction", 《PROCEEDINGS OF THE THIRTY-FIRST AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE》 *
ZERUILI ET AL: "Remote sensing and artificialneural network estimation of on-road vehicle emissions", 《IN PROCEEDINGS OF INTERNATIONAL CONFERENCE ON ADVANCED ROBOTICS AND MECHATRONICS》 *

Cited By (36)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020043473A1 (en) * 2018-08-31 2020-03-05 International Business Machines Corporation Data prediction
CN109086946A (en) * 2018-09-11 2018-12-25 东南大学 A kind of polluted gas emitted smoke method of conventional energy resource and new energy public transit vehicle
CN109086946B (en) * 2018-09-11 2021-09-17 东南大学 Method for predicting emission of polluted gas of conventional energy and new energy public transport vehicle
CN109492822A (en) * 2018-11-24 2019-03-19 上海师范大学 Air pollutant concentration time-space domain interaction prediction method
CN109492822B (en) * 2018-11-24 2021-08-03 上海师范大学 Air pollutant concentration time-space domain correlation prediction method
CN111222666A (en) * 2018-11-26 2020-06-02 中兴通讯股份有限公司 Data calculation method and device
CN110046787A (en) * 2019-01-15 2019-07-23 重庆邮电大学 A kind of urban area charging demand for electric vehicles spatio-temporal prediction method
CN109948716A (en) * 2019-03-25 2019-06-28 中国民航大学 A kind of airport delay prediction technique based on region residual sum LSTM network
CN109948716B (en) * 2019-03-25 2023-04-07 中国民航大学 Airport delay prediction method based on regional residual error and LSTM network
CN109991685A (en) * 2019-04-03 2019-07-09 北京市天元网络技术股份有限公司 A kind of precipitation prediction technique and device based on more LSTM Model Fusions
CN110738355B (en) * 2019-09-19 2023-07-04 河源职业技术学院 Urban waterlogging prediction method based on neural network
CN110738355A (en) * 2019-09-19 2020-01-31 河源职业技术学院 urban waterlogging prediction method based on neural network
CN110738367A (en) * 2019-10-12 2020-01-31 东北大学秦皇岛分校 traffic exhaust emission prediction method based on deep residual error network
CN110738367B (en) * 2019-10-12 2022-06-28 东北大学秦皇岛分校 Traffic tail gas emission prediction method based on deep residual error network
CN110929793A (en) * 2019-11-27 2020-03-27 谢国宇 Time-space domain model modeling method and system for ecological environment monitoring
CN111144666A (en) * 2020-01-02 2020-05-12 吉林大学 Ocean thermocline prediction method based on deep space-time residual error network
CN113095535A (en) * 2020-01-08 2021-07-09 普天信息技术有限公司 Flow prediction method and device based on deep space-time residual error network
CN111260121A (en) * 2020-01-12 2020-06-09 桂林电子科技大学 Urban-range pedestrian flow prediction method based on deep bottleneck residual error network
CN111291860A (en) * 2020-01-13 2020-06-16 哈尔滨工程大学 Anomaly detection method based on convolutional neural network feature compression
CN111415050B (en) * 2020-04-27 2023-12-05 新奥新智科技有限公司 Short-term load prediction method, short-term load prediction model training method and device
CN111415050A (en) * 2020-04-27 2020-07-14 新智数字科技有限公司 Short-term load prediction method and short-term load prediction model training method and device
CN111625999A (en) * 2020-05-29 2020-09-04 中南林业科技大学 Forest fire early warning model and system based on deep learning technology
CN111882122A (en) * 2020-07-16 2020-11-03 成都市高博汇科信息科技有限公司 Traffic accident black point prediction method based on deep learning and space-time big data
CN112132264B (en) * 2020-09-11 2023-04-07 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) Regional exhaust emission prediction method and system based on space-time residual perception network
CN112132264A (en) * 2020-09-11 2020-12-25 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) Regional exhaust emission prediction method and system based on space-time residual perception network
CN112215408A (en) * 2020-09-24 2021-01-12 交控科技股份有限公司 Rail transit passenger flow volume prediction method and device
CN112598170B (en) * 2020-12-18 2022-10-28 中国科学技术大学 Vehicle exhaust emission prediction method and system based on multi-component fusion time network
CN112598170A (en) * 2020-12-18 2021-04-02 中国科学技术大学 Vehicle exhaust emission prediction method and system based on multi-component fusion time network
CN112734028A (en) * 2020-12-28 2021-04-30 三峡大学 Modeling method for prediction model of concentration of dissolved gas in transformer oil
CN113222206B (en) * 2021-01-29 2022-05-13 太原理工大学 Traffic state prediction method based on ResLS-C deep learning combination
CN113222206A (en) * 2021-01-29 2021-08-06 太原理工大学 Traffic state prediction method based on ResLS-C deep learning combination
CN113077081A (en) * 2021-03-26 2021-07-06 航天科工智能运筹与信息安全研究院(武汉)有限公司 Traffic pollution emission prediction method
CN113222236A (en) * 2021-04-30 2021-08-06 中国科学技术大学先进技术研究院 Data distribution self-adaptive cross-regional exhaust emission prediction method and system
CN113222247A (en) * 2021-05-11 2021-08-06 润联软件系统(深圳)有限公司 Chemical raw material price and purchase quantity prediction method, device and related equipment
CN113283588A (en) * 2021-06-03 2021-08-20 青岛励图高科信息技术有限公司 Near-shore single-point wave height forecasting method based on deep learning
CN113283588B (en) * 2021-06-03 2022-04-19 青岛励图高科信息技术有限公司 Near-shore single-point wave height forecasting method based on deep learning

Similar Documents

Publication Publication Date Title
CN108288109A (en) Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks
CN108364087A (en) A kind of spatio-temporal prediction method of urban mobile tail gas concentration
CN109214592A (en) A kind of Air Quality Forecast method of the deep learning of multi-model fusion
CN103148862B (en) Low carbon discharge constraint influence considered traffic mode and path selection method
CN110321932B (en) Full-city air quality index estimation method based on deep multi-source data fusion
CN101976429A (en) Cruise image based imaging method of water-surface aerial view
CN108133295A (en) A kind of motor-driven vehicle gas concentration continuous time Forecasting Methodology for target road section
CN105117595B (en) A kind of private car trip data integrated approach based on floating car data
CN105740643A (en) Self-adaptive PM<2.5>concentration speculating method based on city region grid
CN110346518B (en) Traffic emission pollution visualization early warning method and system thereof
CN110738367B (en) Traffic tail gas emission prediction method based on deep residual error network
CN112989715B (en) Multi-signal-lamp vehicle speed planning method for fuel cell vehicle
CN105678034A (en) Pedestrian crossing time model and signalized intersection pedestrian crosswalk width optimizing method
CN113299078B (en) Multi-mode traffic trunk line signal coordination control method and device based on multi-agent cooperation
CN110110576A (en) A kind of traffic scene thermal infrared semanteme generation method based on twin semantic network
CN115063978A (en) Bus arrival time prediction method based on digital twins
CN114692355B (en) Motor vehicle pollution diffusion simulation method and system by coupling WRF (write driver software) and OpenFOAM (open-world automatic learning machine) models
CN113515798B (en) Urban three-dimensional space expansion simulation method and device
Cao et al. Analysis of spatiotemporal changes in cultural heritage protected cities and their influencing factors: Evidence from China
CN113744541A (en) Road network discharge loss space-time distribution reconstruction method and system for confrontation graph convolution network
CN113255956A (en) Urban atmospheric pollution prediction method
Lingqiu et al. A LSTM based bus arrival time prediction method
CN115329425A (en) Air quality driven building group form optimization design method and system
CN115144835A (en) Method for inverting weather radar reflectivity by satellite based on neural network
CN114330871A (en) Method for predicting urban road conditions by combining public transport operation data with GPS data

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20180717

RJ01 Rejection of invention patent application after publication