CN112132264B - Regional exhaust emission prediction method and system based on space-time residual perception network - Google Patents
Regional exhaust emission prediction method and system based on space-time residual perception network Download PDFInfo
- Publication number
- CN112132264B CN112132264B CN202010953126.6A CN202010953126A CN112132264B CN 112132264 B CN112132264 B CN 112132264B CN 202010953126 A CN202010953126 A CN 202010953126A CN 112132264 B CN112132264 B CN 112132264B
- Authority
- CN
- China
- Prior art keywords
- time
- space
- data
- tail gas
- exhaust
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000008447 perception Effects 0.000 title claims abstract description 46
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000012544 monitoring process Methods 0.000 claims abstract description 32
- 230000007613 environmental effect Effects 0.000 claims abstract description 30
- 230000008859 change Effects 0.000 claims abstract description 17
- 238000007781 pre-processing Methods 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims description 44
- 230000000737 periodic effect Effects 0.000 claims description 32
- 238000009826 distribution Methods 0.000 claims description 27
- 230000006870 function Effects 0.000 claims description 24
- 238000012545 processing Methods 0.000 claims description 21
- 238000000605 extraction Methods 0.000 claims description 17
- 238000010586 diagram Methods 0.000 claims description 16
- 230000004927 fusion Effects 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 12
- 238000010276 construction Methods 0.000 claims description 11
- 230000036962 time dependent Effects 0.000 claims description 11
- 238000013507 mapping Methods 0.000 claims description 8
- 238000012935 Averaging Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 230000002123 temporal effect Effects 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 6
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000007526 fusion splicing Methods 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 2
- 238000012360 testing method Methods 0.000 abstract description 2
- 238000004590 computer program Methods 0.000 description 7
- 238000003860 storage Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 239000000428 dust Substances 0.000 description 2
- 239000003344 environmental pollutant Substances 0.000 description 2
- 231100000719 pollutant Toxicity 0.000 description 2
- 238000003915 air pollution Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 239000000779 smoke Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000012855 volatile organic compound Substances 0.000 description 1
Images
Classifications
-
- 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
- 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
- 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—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Educational Administration (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The regional exhaust emission prediction method and system based on the space-time residual error perception network can solve the technical problems that the existing methods are mostly based on test vehicle data, external influence factors are not enough to be considered, and relative errors are large. The method comprises the following steps: s100, acquiring historical tail gas space-time monitoring data and external environment data, and performing data preprocessing on the acquired data; s200, constructing a time sequence division set according to the change characteristics of the tail gas; s300, based on a pre-constructed and trained exhaust pollution space-time prediction model, predicting exhaust emission at the future t + k moment by using external environment characteristic data of the current moment t and historical exhaust space-time sequence data before the t-1 moment. The invention considers that the exhaust emission has space-time heterogeneity and is influenced by various external complex environmental factors through the space-time residual sensing network, and can realize regional exhaust prediction with higher precision on real telemetering data.
Description
Technical Field
The invention relates to the technical field of environmental detection, in particular to a regional exhaust emission prediction method and system based on a space-time residual perception network.
Background
According to the publication of '2012 Chinese environmental conditions' issued by the ministry of environmental protection, the national environmental quality is kept stable, but the situation is still severe. The northern city is influenced by large-scale weather, the air quality is obviously reduced, and the dust-haze area reaches 130 ten thousand square kilometers. The 'twelve five-element' plan for air pollution control in key areas, which is repeatedly implemented by state offices, indicates that PM2.5 pollution control is taken as a key point, and industrial smoke dust, construction flying dust, volatile organic compounds and motor vehicle tail gas pollution control work are well grasped. The motor vehicle exhaust is one of PM2.5 sources, and the real-time acquisition of the time-space distribution information of the urban area exhaust is of great benefit to the prevention and control of motor vehicle pollution and the environmental protection. A monitoring and early warning system is necessary to be established, and urban area tail gas space-time distribution at any moment is acquired, so that urban area tail gas pollution early warning can be provided, and decision support is provided for urban traffic planning of traffic municipal departments.
The method is characterized in that the actual traffic conditions of all regions in a city are different and the influence of environmental factors can affect the driving conditions of vehicles, the influence of tail gas on spatial diffusion is considered, the tail gas distribution in a given region can be influenced by the adjacent region, and the tail gas shows similar change characteristics in time due to the periodicity and trend change of vehicle flow in the region on a time scale. In addition, the tail gas emission has space-time heterogeneity and is influenced by various external complex environmental factors.
Disclosure of Invention
The invention provides a regional exhaust emission prediction method and system based on a space-time residual error perception network, which can solve the technical problems that the existing method is mostly based on test vehicle data, external influence factors are not enough to be considered, and relative errors are large.
In order to achieve the purpose, the invention adopts the following technical scheme:
the method comprises the following steps:
s100, acquiring historical tail gas space-time monitoring data and external environment data, and performing data preprocessing on the acquired data;
s200, constructing a time sequence division set according to the change characteristics of the tail gas;
s300, based on a pre-constructed and trained exhaust pollution space-time prediction model, predicting exhaust emission at the future t + k moment by using external environment characteristic data of the current moment t and historical exhaust space-time sequence data before the t-1 moment.
Further, the construction steps of the tail gas pollution space-time prediction model in the step S300 are as follows:
s301, acquiring historical tail gas space-time monitoring data and external environment data, and performing data preprocessing on the acquired data;
s302, constructing a time sequence division set according to the change characteristics of the tail gas;
s303, constructing a regional exhaust pollution emission prediction model of the deep space-time residual perception network according to the exhaust division sequence data and the external environment factor data;
s304, training the deep space-time residual perception network by utilizing the preprocessed tail gas monitoring data and external environment data to obtain a tail gas pollution space-time prediction model.
Further, the step S100 of acquiring historical exhaust gas space-time monitoring data and external environment data, and the data preprocessing performed on the acquired data specifically includes:
s101, acquiring historical tail gas space-time monitoring data of a vehicle and road network traffic, meteorological environment and urban interest point distribution external environment data by using non-contact measured tail gas remote sensing monitoring equipment;
and S102, performing missing value completion, abnormal value abandonment and data normalization processing on the obtained monitoring data.
Further, the step S200 of constructing a time sequence division set according to the tail gas variation characteristic specifically includes:
s201, according to the sequence length l of the proximity time segment c Constructing proximity time segments
S202, according to the length l of the periodic time segment sequence p Constructing periodic time slices
p is the time interval of the periodic time segment;
s203, according to the trend time segment sequence length l s Constructing trending time segments
s is the time interval of the trending time segment sequence.
Further, the step S303 of constructing a regional exhaust pollution emission prediction model of the deep space-time residual perception network according to the exhaust division sequence data and the external environment factor data includes:
s3031, extracting time-dependent characteristics by dividing a proximity time segment H c Periodic time segment H p And trending time segment H s Respectively sending the shallow layer feature extraction products into the convolution layer units with the same structure to carry out shallow layer feature extraction;
the spatial and temporal distribution characteristics of the tail gas of the three time-division segments obtained by convolution operation are as follows,
wherein ×, f denotes a convolution operation, f denotes an activation function, in particular a linear rectifier unit ReLU, f (z) = max (0,z); w (1) ，b (1) Respectively obtaining a weight matrix to be learned and a bias vector parameter of the first layer convolution layer; h c (1) ，H p (1) ，H s (1) Respectively is a characteristic diagram of a first layer convolution layer proximity time segment, a periodic time segment and a trend time segment;
the layer output is then fed to a trending module, a periodicity module, and a proximity module, respectively, to extract the time dependence of the exhaust gas distribution;
wherein the time-dependent extraction steps are as follows:
regarding the proximity time characteristic, considering that the tail gas changes in a short time are similar, the proximity characteristic diagram is kept as the original input;
for the periodic time characteristics, extracting periodic invariance characteristics on the time change of the tail gas by introducing a self-attention mechanism (self-attention);
for the trend time characteristics, averaging the trend time segment characteristic layers by introducing average pooling operation to obtain a trend characteristic subgraph;
wherein the content of the first and second substances,
the time-dependent extraction operation is as follows:
wherein
Representing residual join operations, g being a linear embedding function, W θ ，/>
Respectively, an embedded weight matrix to be learned, f AP Is an averaging pooling operation, based on the measured values of the measured values>
Respectively carrying out time dependency processing on the proximity time segment, the periodic time segment and the trend time segment, and sending the feature graphs into a residual convolution unit for processing after front-end fusion;
s3032, fusing external environment characteristics;
mapping an external environment characteristic input x to an internal characteristic space representation z through an encoder, and then reconstructing z to an output y through a decoder;
the specific fusion steps include:
firstly, performing front-end fusion splicing on a proximity time segment, a periodic time segment and a trend time segment through a characteristic graph extracted by time dependence, and then sending the front-end fusion spliced characteristic graph to a stacked convolution residual error unit for processing;
performing front-end fusion operation on the time-dependent extracted feature graph is recorded as follows:
wherein
And b (2) Respectively, learning parameters to be optimized;
for the tail gas space-time residual error network part, the time correlation characteristics are fused and output H through the front end of the time processing component st And is combined withDesigning a residual convolution unit to extract spatial dependency, and recording the output of a space-time residual network as follows:
as for the external environmental factors, it is preferable that, external environmental factor E at time t t Including road network structure information E road Weather environmental factor E weather Traffic flow factor E traffic And point of interest information E POI Having different data dimension structures, learning deep layer characteristics of tail gas space-time distribution influenced by external environmental factors by stacking a plurality of self-encoders, and mapping hidden layer characteristics to network input layer X by utilizing full connection layer t High-dimensional feature vectors of the same dimension;
the two parts are fused at the back end, and the final prediction result of the t moment is output by utilizing the tanh activation function and recorded as
Wherein X Res Is a space-time residual network part output, X Ext Is an external environmental factor feature extraction network output, W st And W Ext Respectively corresponding weight parameter matrixes to be learned; the tanh activation function will eventually fuse the results
Mapping to [ -1,1]To (c) to (d);
by minimizing the predicted value
With the true value X t The square error (MSE) between them is taken as the loss function of the space-time residual perception network model training and is recorded as: />
Where θ is all the parameters to be learned in the spatio-temporal residual perception network model.
Further, the encoder portion in S3032 is written as:
z＝σ(Wx+b)
w and b are weight and bias parameters of the encoder respectively, and sigma is a sigmoid activation function;
the corresponding decoder is written as:
y＝σ(W′z+b′)
wherein W 'and b' are the weight and bias parameters of the decoder, respectively; the self-encoder replicates the similar input of training data by minimizing the reconstruction error y-x.
Further, in S304, the training of the deep space-time residual perception network is performed by using the preprocessed exhaust monitoring data and the external environment data, and the obtained exhaust pollution space-time prediction model specifically includes:
initializing a training data set
Training samples ({ H) are created in sequence according to the available time stamp sequence t (1 ≦ t ≦ n-1) c ，H p ，H s ，E t }，X t ) And is stored in a training set D, in which,
l c ，l p ，l s respectively, a proximity time segment length, a periodic time segment length, and a trending time segment length, p, s are the corresponding time spans of the periodic time segment and the trending time segment, respectively, { X 0 ，X 1 ，…X n-1 And { E } and 0 ，E 1 ，…E n-1 are pretreatment, respectivelyThe subsequent historical observation sequence and external environment factor sequence;
initializing all parameters theta to be learned in a deep space-time residual perception network model to be trained;
randomly selecting a batch of training samples D from the training set D batch Using each training sample batch D batch And training the model by a minimum loss function, and updating the learning parameter theta to reach a training termination condition to obtain a depth space-time residual perception network model after training.
Further, the S300, based on a pre-constructed and trained exhaust pollution space-time prediction model, specifically predicting exhaust emission at a future time t + k by using external environment feature data at the current time t and historical exhaust space-time sequence data before the time t-1, includes:
according to external characteristic data E of current time t t And a historical observation data set { H } of the exhaust gas before the time t-1 C ，H P ，H S And predicting the regional exhaust emission at the time t.
On the other hand, the invention also discloses a regional tail gas emission prediction system based on the space-time residual error perception network, which comprises the following units:
the data acquisition unit is used for acquiring historical tail gas space-time monitoring data and external environment data and carrying out data preprocessing on the acquired data;
the time sequence division set construction unit is used for constructing a time sequence division set according to the tail gas change characteristics;
and the prediction unit is used for predicting the exhaust emission at the future t + k moment by utilizing the external environment characteristic data of the current moment t and the historical exhaust space-time sequence data before the t-1 moment based on a pre-constructed and trained exhaust pollution space-time prediction model.
Further, the following subunits are included:
the model construction unit is used for constructing a regional exhaust pollution emission prediction model of the deep space-time residual perception network according to the exhaust division sequence data and the external environment factor data;
and the model training unit is used for training the deep space-time residual perception network by utilizing the preprocessed tail gas monitoring data and the external environment data to obtain a tail gas pollution space-time prediction model.
According to the technical scheme, the regional exhaust emission prediction method and system based on the space-time residual sensing network are different from the traditional prediction method based on the standard rasterization space-time network model, the space-time residual sensing network considers that the exhaust emission has space-time heterogeneity and is influenced by various external complex environmental factors, and the regional exhaust emission prediction with higher precision can be realized on real telemetering data.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a schematic diagram of a front-end and back-end feature fusion module of the present invention;
FIG. 3 is a front-end and back-end fusion block diagram of spatio-temporal sequence prediction according to the present invention;
FIG. 4 is a 7 AM distribution plot of the spatiotemporal variation distribution of NOx pollution during early peak hours; FIG. 5 is a graph of the 8 am distribution of the spatiotemporal variation distribution of NOx pollution during early peak hours;
fig. 6 is a NOx 24 hour change prediction curve and a truth curve.
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, the method for predicting regional exhaust emissions based on the space-time residual perception network according to this embodiment includes:
the method comprises the following steps:
s100, acquiring historical tail gas space-time monitoring data and external environment data, and performing data preprocessing on the acquired data;
s200, constructing a time sequence division set according to the change characteristics of the tail gas;
s300, based on a pre-constructed and trained exhaust pollution space-time prediction model, predicting exhaust emission at the future t + k moment by using external environment characteristic data of the current moment t and historical exhaust space-time sequence data before the t-1 moment.
The construction steps of the tail gas pollution space-time prediction model in the S300 are as follows:
s301, acquiring historical tail gas space-time monitoring data and external environment data, and performing data preprocessing on the acquired data;
s302, constructing a time sequence division set according to the tail gas change characteristics;
s303, constructing a regional exhaust pollution emission prediction model of the deep space-time residual perception network according to the exhaust division sequence data and the external environment factor data;
s304, training the deep space-time residual perception network by utilizing the preprocessed tail gas monitoring data and external environment data to obtain a tail gas pollution space-time prediction model.
The following is a detailed description:
the step S101: acquiring historical tail gas space-time monitoring data and external environment data, and performing data preprocessing on the acquired data, wherein the method specifically comprises the following subdivision steps of S101 to S102:
s101, acquiring historical tail gas space-time monitoring data of a vehicle and external environment data such as road network traffic, meteorological environment, urban interest point distribution and the like by using non-contact measured tail gas remote sensing monitoring equipment;
and S102, performing missing value completion, abnormal value abandonment and data normalization processing on the obtained monitoring data.
The step S200: constructing a time sequence division set according to the variation characteristics of the tail gas, and specifically comprising the following subdivision steps S201 to S203:
s201, according to the sequence length l of the proximity time segment c Constructing proximity time segments
S202, according to the length l of the periodic time segment sequence p Constructing periodic time slices
p is the time interval (one week) of the periodic time slices.
S203, according to the trend time segment sequence length l s Constructing trending time segments
s is the time interval (one month) of the trending temporal segment sequence.
The step S303: the regional exhaust pollution emission prediction model of the deep space-time residual perception network is constructed according to the exhaust division sequence data and the external environmental factor data, and the method specifically comprises the following subdivision steps S3031-S3032:
s3031, extracting time-dependent characteristics by dividing a proximity time segment H c Periodic time segment H p And trending time segment H s And respectively sending the shallow layer feature extraction products into the convolution layer units with the same structure to carry out shallow layer feature extraction. The spatial and temporal distribution characteristics of the tail gas of the three time-division segments obtained by convolution operation are as follows,
where ×, represents convolution operation, f represents an activation function, such as the linear rectification unit ReLU, f (z) = max (0,z); w (1) ，b (1) Respectively obtaining a weight matrix to be learned and a bias vector parameter of the first layer convolution layer; h c (1) ，H p (1) ，H s (1) Respectively, the first layer convolution layer proximity time segment, the periodicity time segment and the trend time segment. The layer output is then fed to a trending module, a periodicity module, and a proximity module, respectively, to extract the time dependence of the exhaust gas distribution, the time dependence extraction being shown in FIG. 2. The time dependency extraction is detailed as follows, and regarding the proximity time characteristic, considering that the tail gas changes in a short time are similar, the proximity characteristic diagram is kept as the original input; for the periodic time characteristics, extracting periodic invariance characteristics on the time change of the tail gas by introducing a self-attention mechanism (self-attention); and for the trend time characteristics, averaging the trend time segment characteristic layers by introducing an average pooling operation to obtain a trend characteristic subgraph. The time-dependent extraction operation is as follows:
wherein
Representing residual join operations, g being a linear embedding function, W θ ，/>
Respectively, an embedded weight matrix to be learned, f AP Is an averaging pooling operation, based on the measured values of the measured values>
Respectively, the characteristic graphs of the proximity time segment, the periodic time segment and the trend time segment after time dependency processing are sent to a residual convolution unit for processing after front end fusion.
As shown in fig. 2. FIG. 2 a temporal dependency extraction module.
And S3032, fusing external environment characteristics. The extrinsic ambient feature input x is mapped to the intrinsic feature space representation z by the encoder, and then reconstructed to the output y by the decoder. The general encoder section can be written as:
z＝σ(Wx+b)
where W and b are the weight and bias parameters of the encoder, respectively, and σ is the sigmoid activation function. The corresponding decoder can be written as:
y＝σ(W′z+b′)
where W 'and b' are the weight and bias parameters of the decoder, respectively. The self-encoder replicates the similar input of training data by minimizing the reconstruction error y-x.
FIG. 3 is a front-end fusion block diagram of spatio-temporal sequence prediction, in which the proximity time segment, the periodic time segment and the trend time segment are subjected to front-end fusion splicing by a time-dependent extracted feature diagram and then sent to a stacked convolution residual unit for processing. Front-end fusion is generally used for fusing data with similar structural features, and performing front-end fusion operation on the time-dependent extracted feature graph can be written as:
wherein
And b (2) Respectively, the learning parameters to be optimized.
For the tail gas space-time residual error network part, the time correlation characteristics are fused and output H through the front end of the time processing component st And designing a residual convolution unit to extract spatial dependency, wherein the output of a space-time residual network can be recorded as:
for external environmental factorsConsidering that the spatial and temporal distribution of the tail gas in the region can be influenced by external complex environmental factors, such as road network traffic information, traffic flow information, urban functional interest points, meteorological environment and the like, and the external environmental factor E at the moment t t Including road network structure information E road Weather environmental factor E weather Traffic flow factor E traffic And point of interest information E POI Having different data dimension structures, learning deep layer characteristics of tail gas space-time distribution influenced by external environmental factors by stacking a plurality of self-encoders, and mapping hidden layer characteristics to a network input layer X by utilizing a full connection layer t High-dimensional feature vectors of the same dimension. The two parts, namely the external environmental factor characteristic and the tail gas space-time distribution characteristic, are fused at the rear end, and then the final prediction result of the t moment is output by utilizing the tanh activation function and recorded as
Wherein X Res Is a space-time residual network part output, X Ext Is an external environmental factor feature extraction network output, W st And W Ext Respectively, corresponding weight parameter matrices to be learned. the tanh activation function will eventually fuse the results
Mapping to [ -1,1]In the meantime. By minimizing the predictor->
With the true value X t The square error (MSE) between them is taken as the loss function of the space-time residual perception network model training and is recorded as:
where θ is all the parameters to be learned in the spatio-temporal residual aware network model.
The above step S304: and training the deep space-time residual perception network by utilizing the preprocessed tail gas monitoring data and external environment data to obtain a tail gas pollution space-time prediction model.
The method specifically comprises the following steps:
initializing a training data set
Training samples ({ H) are created in sequence according to the available time stamp sequence t (1 ≦ t ≦ n-1) c ，H p ，H s ，E t }，X t ) And is stored in a training set D, in which,
l c ，l p ，l s respectively, a proximity time segment length, a periodic time segment length, and a trending time segment length, p, s are the corresponding time spans of the periodic time segment and the trending time segment, respectively, { X 0 ，X 1 ，…X n-1 And { E } and 0 ，E 1 ，…E n-1 respectively representing a preprocessed historical observation sequence and an external environment factor sequence;
initializing all parameters theta to be learned in a deep space-time residual perception network model to be trained;
randomly selecting a batch of training samples D from the training set D batch Using each training sample batch D batch Training the model through a minimum loss function, and updating a learning parameter theta to reach a training termination condition to obtain a depth space-time residual perception network model after training is completed;
the step S300: based on a trained exhaust pollution space-time prediction model, the exhaust emission at the future t + k moment is predicted by using external environment characteristic data of the current moment t and historical exhaust space-time sequence data before the t-1 moment, and the method specifically comprises the following steps:
according to external characteristic data E of current time t t And a historical observation data set { H } of the exhaust gas before the time t-1 C ，H P ，H S And predicting the zone exhaust emission at the time t, as shown in the figure.
As shown in fig. 4 and 5 in particular, fig. 4 is a 7 am distribution plot of the spatio-temporal variation distribution of NOx pollution during early peak hours, and fig. 5 is a 8 am distribution plot, it can be seen that the tail gas spatial distribution in urban areas during early peak hours is significantly increased due to the increased commute volume in the areas as people start to go home to work.
FIG. 6 is a NOx 24-hour change prediction curve and a true value curve, a solid line represents a predicted value of each pollutant, a dotted line represents a true measured value of each pollutant, the predicted values can be well fitted to the true values, and effectiveness of space-time residual perception network processing area moving source pollution space-time distribution prediction problems designed by the embodiment is proved.
Therefore, in the embodiment, the pollution time sequence data and the environmental data are rasterized into the similar multi-channel image sequence data, so that the exhaust emission is associated with the geographic environment information, and a deep space-time residual perception network model is adopted for processing, so that the high-precision exhaust prediction is realized on the real telemetering data.
On the other hand, the invention also discloses a regional tail gas emission prediction system based on the space-time residual error perception network, which comprises the following units:
the data acquisition unit is used for acquiring historical tail gas space-time monitoring data and external environment data and carrying out data preprocessing on the acquired data;
the time sequence division set construction unit is used for constructing a time sequence division set according to the tail gas change characteristics;
and the prediction unit is used for predicting the exhaust emission at the future t + k moment by utilizing the external environment characteristic data of the current moment t and the historical exhaust space-time sequence data before the t-1 moment based on a pre-constructed and trained exhaust pollution space-time prediction model.
Further, the following subunits are included:
the model construction unit is used for constructing a regional exhaust pollution emission prediction model of the deep space-time residual perception network according to the exhaust division sequence data and the external environment factor data;
and the model training unit is used for training the deep space-time residual perception network by utilizing the preprocessed tail gas monitoring data and the external environment data to obtain a tail gas pollution space-time prediction model.
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 so forth) 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 regional exhaust emission prediction method based on a space-time residual error perception network is characterized by comprising the following steps:
the method comprises the following steps:
s100, acquiring historical tail gas space-time monitoring data and external environment data, and performing data preprocessing on the acquired data;
s200, constructing a time sequence division set according to the change characteristics of the tail gas;
s300, based on a pre-constructed and trained exhaust pollution space-time prediction model, predicting exhaust emission at a future t + k moment by using external environment characteristic data of the current moment t and historical exhaust space-time sequence data before the t-1 moment;
the construction steps of the tail gas pollution space-time prediction model in the S300 are as follows:
s301, acquiring historical tail gas space-time monitoring data and external environment data, and performing data preprocessing on the acquired data;
s302, constructing a time sequence division set according to the tail gas change characteristics;
s303, constructing a regional exhaust pollution emission prediction model of the deep space-time residual perception network according to the exhaust division sequence data and the external environment factor data;
s304, training a deep space-time residual perception network by utilizing the preprocessed tail gas monitoring data and external environment data to obtain a tail gas pollution space-time prediction model;
the S303, according to the tail gas division sequence data and the external environmental factor data, constructing a regional tail gas pollution emission prediction model of the deep space-time residual perception network, comprises the following steps:
s3031, extracting time-dependent characteristics by dividing a proximity time segment H c Periodic time segment H p And trending time segment H s Respectively sending the shallow layer feature extraction products into the convolution layer units with the same structure to carry out shallow layer feature extraction;
the spatial and temporal distribution characteristics of the tail gas of the three time-division segments obtained by convolution operation are as follows,
wherein ×, f denotes a convolution operation, f denotes an activation function, in particular a linear rectifier unit ReLU, f (z) = max (0,z); w (1) ,b (1) Respectively obtaining a weight matrix to be learned and a bias vector parameter of the first layer convolution layer; h c (1) ,H p (1) ,H s (1) Respectively, first layer convolution layer proximity timeFeature maps of segments, periodic time segments, and trending time segments;
the layer output is then fed to a trending module, a periodicity module, and a proximity module, respectively, to extract the time dependence of the exhaust gas distribution;
wherein the time-dependent extraction steps are as follows:
regarding the proximity time characteristic, considering that the tail gas changes are similar in a short time, the proximity characteristic diagram is kept as the original input;
for the periodic time characteristics, extracting periodic invariance characteristics on the time change of the tail gas by introducing a self-attention mechanism (self-attention);
for the trend time characteristics, averaging the trend time segment characteristic layers by introducing average pooling operation to obtain a trend characteristic subgraph;
wherein the content of the first and second substances,
the time-dependent extraction operation is as follows:
wherein
Denotes the residual join operation, g is the linear embedding function, W θ ,/>
Respectively, an embedded weight matrix to be learned, f AP Is an averaging pooling operation, based on a number of pooled frames>
Respectively carrying out time dependency processing on the proximity time segment, the periodic time segment and the trend time segment, and sending the feature graphs into a residual convolution unit for processing after front-end fusion;
s3032, fusing external environment characteristics;
mapping the external environment characteristic input x to an internal characteristic space representation z through an encoder, and then reconstructing z to an output y through a decoder;
the specific fusion steps include:
firstly, performing front-end fusion splicing on a proximity time segment, a periodic time segment and a trend time segment through a characteristic graph extracted by time dependence, and then sending the front-end fusion spliced characteristic graph to a stacked convolution residual error unit for processing;
the front-end fusion operation performed on the feature map extracted by the time dependency is recorded as:
wherein
And b (2) Respectively, learning parameters to be optimized;
for the tail gas space-time residual error network part, the time correlation characteristics are fused and output H through the front end of the time processing component st And designing a residual convolution unit to extract spatial dependency, wherein the output of a space-time residual network can be recorded as:
as for the external environmental factor, the external environmental factor E set at time t t Including road network structure information E road Weather environmental factor E weather Traffic flow factor E traffic And point of interest information E POI Having different data dimension structures, learning deep layer characteristics of tail gas space-time distribution influenced by external environmental factors by stacking a plurality of self-encoders, and mapping hidden layer characteristics to network input layer X by utilizing full connection layer t High-dimensional feature vectors of the same dimension;
the two parts, namely the external environmental factor characteristic and the tail gas space-time distribution characteristic, are fused at the back end and then input by utilizing the tanh activation functionThe final prediction at time t is recorded as
Wherein X Res Is a space-time residual network part output, X Ext Is an external environmental factor feature extraction network output, W st And W Ext Respectively corresponding weight parameter matrixes to be learned; the tanh activation function will eventually fuse the results
Mapping to [ -1,1]To (c) to (d);
by minimizing predicted values
And true value X t The square error (MSE) between them is taken as the loss function of the space-time residual perception network model training and is recorded as:
where θ is all the parameters to be learned in the spatio-temporal residual perceptual network model.
2. The regional exhaust emission prediction method based on the space-time residual error perception network according to claim 1, characterized in that: the S100, acquiring historical tail gas space-time monitoring data and external environment data, and preprocessing the acquired data specifically comprises the following steps:
s101, acquiring historical tail gas space-time monitoring data of a vehicle and road network traffic, meteorological environment and urban interest point distribution external environment data by using non-contact measured tail gas remote sensing monitoring equipment;
and S102, performing missing value completion, abnormal value abandonment and data normalization processing on the obtained monitoring data.
3. The regional exhaust emission prediction method based on the space-time residual error perception network according to claim 2, characterized in that: the S200 construction of the time sequence division set according to the tail gas variation characteristics specifically comprises:
s201, according to the sequence length l of the proximity time segment c Constructing proximity time segments
S202, according to the length l of the periodic time segment sequence p Constructing periodic time slices
p is the time interval of the periodic time segment;
s203, according to the trend time segment sequence length l s Constructing trending time segments
s is the time interval of the trending time segment sequence.
4. The regional exhaust emission prediction method based on the space-time residual error perception network according to claim 3, characterized in that: the encoder part in S3032 is recorded as:
z＝σ(Wx+b)
w and b are weight and bias parameters of the encoder respectively, and sigma is a sigmoid activation function;
the corresponding decoder is written as:
y＝σ(W′z+b′)
where W 'and b' are the weight and bias parameters of the decoder, respectively; the self-encoder replicates similar inputs of training data by minimizing the reconstruction error y-x.
5. The regional exhaust emission prediction method based on the space-time residual error perception network according to claim 4, characterized in that: the S304 trains the deep space-time residual perception network by using the pretreated exhaust gas monitoring data and the external environment data, and the obtained exhaust pollution space-time prediction model specifically includes:
initializing a training data set
Training samples ({ H) are created in sequence according to the available time stamp sequence t (1 ≦ t ≦ n-1) c ，H p ，H s ，E t }，X t ) And is stored in a training set D, in which,
l c ，l p ，l s respectively, a proximity time segment length, a periodic time segment length, and a trending time segment length, p, s are the corresponding time spans of the periodic time segment and the trending time segment, respectively, { X 0 ，X 1 ，…X n-1 And { E } and 0 ，E 1 ，…E n-1 respectively a preprocessed historical observation sequence and an external environment factor sequence;
initializing all parameters theta to be learned in a deep space-time residual perception network model to be trained;
randomly selecting a batch of training samples D from the training set D batch Using each training sample batch D batch And training the model by a minimum loss function, and updating the learning parameter theta to reach a training termination condition to obtain a depth space-time residual perception network model after training.
6. The regional exhaust emission prediction method based on the space-time residual error perception network according to claim 5, characterized in that: the S300 is based on a pre-constructed and trained exhaust pollution space-time prediction model, and specifically comprises the following steps of predicting the exhaust emission at the future t + k moment by using external environment characteristic data of the current moment t and historical exhaust space-time sequence data before the t-1 moment:
according to external characteristic data E of current time t t And a historical observation data set { H } of the exhaust gas before the time t-1 C ,H P ,H S And predicting the exhaust emission of the area at the time t.
7. A regional exhaust emission prediction system based on a space-time residual error perception network is used for realizing the regional exhaust emission prediction method based on the space-time residual error perception network, which is characterized in that: the method comprises the following units:
the data acquisition unit is used for acquiring historical tail gas space-time monitoring data and external environment data and carrying out data preprocessing on the acquired data;
the time sequence division set construction unit is used for constructing a time sequence division set according to the tail gas change characteristics;
and the prediction unit is used for predicting the exhaust emission at the future t + k moment by utilizing the external environment characteristic data of the current moment t and the historical exhaust space-time sequence data before the t-1 moment based on a pre-constructed and trained exhaust pollution space-time prediction model.
8. The system according to claim 7, wherein the prediction system for regional exhaust emissions based on space-time residual error perception network is characterized in that,
the device also comprises the following subunits:
the model construction unit is used for constructing a regional exhaust pollution emission prediction model of the deep space-time residual perception network according to the exhaust division sequence data and the external environment factor data;
and the model training unit is used for training the deep space-time residual sensing network by utilizing the pretreated tail gas monitoring data and the external environment data to obtain a tail gas pollution space-time prediction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010953126.6A CN112132264B (en) | 2020-09-11 | 2020-09-11 | Regional exhaust emission prediction method and system based on space-time residual perception network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010953126.6A CN112132264B (en) | 2020-09-11 | 2020-09-11 | Regional exhaust emission prediction method and system based on space-time residual perception network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112132264A CN112132264A (en) | 2020-12-25 |
| CN112132264B true CN112132264B (en) | 2023-04-07 |
Family
ID=73846209
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010953126.6A Active CN112132264B (en) | 2020-09-11 | 2020-09-11 | Regional exhaust emission prediction method and system based on space-time residual perception network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112132264B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112819218B (en) * | 2021-01-27 | 2022-05-06 | 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) | High-resolution urban mobile source pollution space-time prediction method, system and storage medium |
| CN112991765B (en) * | 2021-02-03 | 2022-05-10 | 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) | Method, terminal and storage medium for updating road high-emission source recognition model |
| CN113128776B (en) * | 2021-04-26 | 2023-07-07 | 中国科学技术大学先进技术研究院 | Multi-vehicle type diesel vehicle emission prediction method and system with data self-migration |
| 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 |
| CN113469405B (en) * | 2021-05-18 | 2024-04-02 | 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) | Subimage segmentation-based cross-city tail gas migration prediction method, system and medium |
| CN113505160A (en) * | 2021-07-21 | 2021-10-15 | 宁波工程学院 | Motor vehicle exhaust emission estimation method based on convolution deep neural network |
| CN114741948B (en) * | 2022-03-09 | 2024-03-12 | 西北工业大学 | Aero-engine degradation trend prediction method based on residual stacking convolution network of sequence reconstruction |
| CN114743072B (en) * | 2022-05-24 | 2023-01-31 | 中国科学院计算机网络信息中心 | Training method of short-term time sequence prediction model |
| 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 |
| CN116577673B (en) * | 2023-07-12 | 2023-09-12 | 深圳先进储能材料国家工程研究中心有限公司 | Distributed neural network-based energy storage power station fault diagnosis method and system |
| CN116777120B (en) * | 2023-08-16 | 2023-10-27 | 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) | Urban road mobile source carbon emission calculation method based on road network OD pairs |
| CN117349650B (en) * | 2023-09-20 | 2024-04-12 | 同济大学 | Method for extracting foundation pit structural data multi-element space-time fusion and evolution features |
| CN117390938B (en) * | 2023-12-13 | 2024-03-08 | 中交第四航务工程勘察设计院有限公司 | High-pile wharf operation and maintenance period structure health monitoring method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106844857A (en) * | 2016-12-21 | 2017-06-13 | 环境保护部信息中心 | A kind of method and apparatus for simulating water pollution accident contaminated zone spatial and temporal distributions |
| CN108288109A (en) * | 2018-01-11 | 2018-07-17 | 安徽优思天成智能科技有限公司 | Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks |
| CN109165690A (en) * | 2018-09-03 | 2019-01-08 | 安徽优思天成智能科技有限公司 | A kind of method and system of city road network tail gas pollution forecast of distribution |
| CN109214570A (en) * | 2018-09-03 | 2019-01-15 | 安徽优思天成智能科技有限公司 | A kind of urban area tail gas pollution prediction technique based on depth space-time correction model |
| CN111243269A (en) * | 2019-12-10 | 2020-06-05 | 福州市联创智云信息科技有限公司 | Traffic flow prediction method based on depth network integrating space-time characteristics |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10921245B2 (en) * | 2018-06-08 | 2021-02-16 | Ball Aerospace & Technologies Corp. | Method and systems for remote emission detection and rate determination |
-
2020
- 2020-09-11 CN CN202010953126.6A patent/CN112132264B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106844857A (en) * | 2016-12-21 | 2017-06-13 | 环境保护部信息中心 | A kind of method and apparatus for simulating water pollution accident contaminated zone spatial and temporal distributions |
| CN108288109A (en) * | 2018-01-11 | 2018-07-17 | 安徽优思天成智能科技有限公司 | Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks |
| CN109165690A (en) * | 2018-09-03 | 2019-01-08 | 安徽优思天成智能科技有限公司 | A kind of method and system of city road network tail gas pollution forecast of distribution |
| CN109214570A (en) * | 2018-09-03 | 2019-01-15 | 安徽优思天成智能科技有限公司 | A kind of urban area tail gas pollution prediction technique based on depth space-time correction model |
| CN111243269A (en) * | 2019-12-10 | 2020-06-05 | 福州市联创智云信息科技有限公司 | Traffic flow prediction method based on depth network integrating space-time characteristics |
Non-Patent Citations (2)
| Title |
|---|
| "Deep spatiotemporal residual early-late fusion network for city region vehicle emission pollution prediction";Zhenyi Xu 等;《Neurocomputing》;20190510;第183-199页 * |
| "城市公交尾气排放特性与预测模型研究";王梦迪;《中国优秀博硕士学位论文全文数据库(硕士)-工程科技Ⅰ辑》;20200615;第2020年卷(第6期);B027-898 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112132264A (en) | 2020-12-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112132264B (en) | Regional exhaust emission prediction method and system based on space-time residual perception network | |
| CN108921200B (en) | Method, apparatus, device and medium for classifying driving scene data | |
| Xu et al. | Real-time prediction of taxi demand using recurrent neural networks | |
| CN115843347A (en) | Generating autonomous vehicle simulation data from recorded data | |
| CN111400620A (en) | User trajectory position prediction method based on space-time embedded Self-orientation | |
| Klischat et al. | Scenario factory: Creating safety-critical traffic scenarios for automated vehicles | |
| CN113222328A (en) | Air quality monitoring equipment point arrangement and site selection method based on road section pollution similarity | |
| CN111310828B (en) | Target detection model fine-tuning method and device for ADAS scene | |
| CN111428558A (en) | Vehicle detection method based on improved YO L Ov3 method | |
| US11620838B2 (en) | Systems and methods for answering region specific questions | |
| CN112633602B (en) | Traffic congestion index prediction method and device based on GIS map information | |
| CN114331542A (en) | Method and device for predicting charging demand of electric vehicle | |
| CN115291515A (en) | Automatic driving simulation test system and evaluation method based on digital twinning | |
| CN114360239A (en) | Traffic prediction method and system for multilayer space-time traffic knowledge map reconstruction | |
| Yang et al. | Feature selection in conditional random fields for map matching of GPS trajectories | |
| CN114202120A (en) | Urban traffic travel time prediction method aiming at multi-source heterogeneous data | |
| CN113744541B (en) | Road network discharge loss space-time distribution reconstruction method and system for confrontation graph convolution network | |
| de la Peña et al. | AD PerDevKit: An autonomous driving perception development kit using CARLA simulator and ROS | |
| CN117058024A (en) | Transformer-based efficient defogging semantic segmentation method and application thereof | |
| CN115239027B (en) | Method and device for forecasting air quality check set | |
| CN117237475A (en) | Vehicle traffic track generation method and device based on diffusion generation model | |
| CN115985132A (en) | Roadside parking availability prediction method based on graph neural network | |
| Han et al. | Evaluate good bus driving behavior with lstm | |
| CN113344239B (en) | Traffic congestion condition prediction method and system based on two-stage spectral clustering | |
| US20220289217A1 (en) | Vehicle-in-virtual-environment (vve) methods and systems for autonomous driving system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |