CN109978228A - A kind of PM2.5 concentration prediction method, apparatus and medium - Google Patents
A kind of PM2.5 concentration prediction method, apparatus and medium Download PDFInfo
- Publication number
- CN109978228A CN109978228A CN201910100028.5A CN201910100028A CN109978228A CN 109978228 A CN109978228 A CN 109978228A CN 201910100028 A CN201910100028 A CN 201910100028A CN 109978228 A CN109978228 A CN 109978228A
- Authority
- CN
- China
- Prior art keywords
- prediction
- neural network
- concentration
- data
- prediction method
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 50
- 238000013528 artificial neural network Methods 0.000 claims abstract description 47
- 238000012549 training Methods 0.000 claims abstract description 42
- 230000009467 reduction Effects 0.000 claims abstract description 7
- 238000005259 measurement Methods 0.000 claims abstract description 6
- 230000015654 memory Effects 0.000 claims description 21
- 230000006870 function Effects 0.000 claims description 13
- 238000005516 engineering process Methods 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 4
- 230000002123 temporal effect Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 description 9
- 238000012545 processing Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 230000000306 recurrent effect Effects 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 230000004913 activation Effects 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 3
- 239000000809 air pollutant Substances 0.000 description 2
- 231100001243 air pollutant Toxicity 0.000 description 2
- 238000003915 air pollution Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000011109 contamination Methods 0.000 description 1
- 230000001351 cycling effect Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000009545 invasion Effects 0.000 description 1
- 230000033001 locomotion Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 210000003733 optic disk Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000000714 time series forecasting Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/045—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
- G06N3/08—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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Tourism & Hospitality (AREA)
- Human Resources & Organizations (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Economics (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Strategic Management (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Educational Administration (AREA)
- Primary Health Care (AREA)
- Entrepreneurship & Innovation (AREA)
Abstract
The present invention discloses a kind of PM2.5 concentration prediction method, apparatus and medium, is related to pollution prediction technical field, and this method is based on CNN and two-way GRU neural network, builds PM2.5 prediction model based on one-dimensional convolutional neural networks CNN and two-way GRU neural network；Meteorological training data tensor feeding PM2.5 prediction model is trained；The one-dimensional convolutional neural networks CNN carries out local feature learning and dimensionality reduction to each input variable time series respectively, successively operates by convolution sum pondization, forms low-dimensional characteristic sequence；Characteristic sequence is inputted into two-way GRU neural network, two-way GRU neural network is from time positive sequence and time backward learning characteristic sequence；Climate measurement data tensor is sent into have trained in the PM2.5 prediction model completed and is predicted, PM2.5 prediction concentrations value is obtained；The model efficiently uses the speed of convolutional neural networks and the order sensibility of light weight characteristic and RNN, allows to check more data volumes in training, improves prediction accuracy.
Description
Technical field
The present invention relates to pollution prediction technical field more particularly to a kind of PM2.5 concentration prediction method, apparatus and Jie
Matter.
Background technique
Nowadays, many cities suffer from regular extensive haze invasion, affect the daily trip of people or even right
People's health causes serious harm.PM2.5 is the main component of haze, administers haze, improves the primary of air quality
Task is control PM2.5, and PM2.5 concentration prediction is the main contents of Air Quality Forecast.Effectively grasp the variation of PM2.5 concentration
Authentic communication is realized to efficient, the Accurate Prediction of prevention and control of air pollution, is of great significance.The variation of air pollution usually may be used
It is reflected by many meteorologic factors, such as temperature, humidity, wind direction, wind speed, snowfall, rainfall, therefore, makes full use of meteorological letter
Breath, establishing the PM2.5 prediction model comprising multiple relevant weather factors is a reliably solution.
From the point of view of existing literature, Air Quality Forecast method can substantially be divided into statistical method, shallow-layer machine learning side
Method and deep learning method.Statistical method includes that regression model, exponential smoothing model, ARIMA model and multi-ply linear return mould
Type etc., since these methods are limited to nonlinear data modeling ability, precision of prediction cannot be met the requirements；Shallow-layer machine learning
Method includes artificial neural network, decision tree, random forest, support vector machines, Bayesian network etc., these methods are appointed in prediction
Have excellent performance in business, but there is also this indicate scarce capacity, easily fall into locally optimal solution, convergence rate is slow the problems such as,
It is not suitable for handling big-sample data；In recent years, it is graduallyd mature as the promotion of GPU performance and deep learning are theoretical, depth
Learning algorithm shows powerful ability in terms of prediction task, and common model is Recognition with Recurrent Neural Network and its various changes
Body, such as shot and long term memory network (Long Short-Term Memory, LSTM), gating cycle unit (Gated Recurrent
Unit, GRU) etc..For Air Quality Forecast, a variety of meteorologic factors increase the understanding difficulty of PM2.5 concentration influence, study
It is still a great difficult problem to profound character representation, raising prediction accuracy.
Summary of the invention
The present invention provides a kind of PM2.5 concentration prediction method, apparatus and medium, comprehensive utilization aiming at the problem that background technique
The ability in feature extraction of convolutional neural networks and the time series forecasting ability of Recognition with Recurrent Neural Network, firstly, utilizing convolutional neural networks
Carrying out down-sampling to data reduces the scale and complexity of data, improves the extensive and learning ability of model entirety, then, will
Data after dimensionality reduction feed Recognition with Recurrent Neural Network, further excavate the information characteristics that different data sources provide in meteorological data, build
Non-linear relation between vertical Multivariate Time Series and air pollutants PM2.5 time series.The model efficiently uses convolution
The speed of neural network and the order sensibility of light weight characteristic and RNN allow to check more data volumes in training, improve pre-
Survey accuracy.
To achieve the goals above, the present invention proposes a kind of PM2.5 concentration prediction method, and this method is based on CNN and two-way
GRU neural network, includes the following steps:
S10, PM2.5 prediction model is built based on one-dimensional convolutional neural networks CNN and two-way GRU neural network；
The PM2.5 prediction model, specifically:
The input terminal of the one-dimensional convolutional neural networks CNN connects meteorological data tensor, one-dimensional convolutional neural networks CNN's
Output end connects the input terminal of two-way GRU neural network, and the output end of two-way GRU neural network accesses full articulamentum, last
Layer is comprising one for generating the neuron of PM2.5 concentration prediction value；
S20, meteorological training data tensor feeding PM2.5 prediction model is trained；
Specifically:
The one-dimensional convolutional neural networks CNN carries out local feature learning to each input variable time series respectively and drops
Dimension is successively operated by convolution sum pondization, forms low-dimensional characteristic sequence；Characteristic sequence is inputted into two-way GRU neural network, it is two-way
GRU neural network is from time positive sequence and time backward learning characteristic sequence；
S30, Climate measurement data tensor is sent into have trained in the PM2.5 prediction model completed and is predicted, obtain
PM2.5 prediction concentrations value.
Preferably, before step S10 further include:
S01, meteorological data is obtained, and meteorological data is pre-processed and standardized；
S02, construction batch data generator, generate neural network and input tensor.
Preferably, the meteorological data, including but not limited to: PM2.5 concentration, dew point, temperature, air pressure, wind direction, wind
Speed, snowfall and rainfall.
Preferably, the batch data generator specifies prediction input comprising time step number, prediction target not
The time step number come and each batch include sample number, and return is a tuple, i.e. the one of multivariable input data crowd
Amount, corresponding target PM2.5 concentration array.
Preferably, the local feature learning and dimensionality reduction, specifically: local feature learning is carried out using convolutional layer, is made
Sub-sampling is carried out with layer pond layer, reduces the length of each one-dimensional input.
Preferably, the PM2.5 prediction model uses mean absolute error as loss function, each in training
Backpropagation operation is carried out according to mean absolute error value in small lot, and mean square error is selected to refer to as the error assessment of model
Mark, measures the quality of prediction.
Preferably, the two-way GRU neural network, including two GRU neural networks, respectively from time positive sequence and time
Backward handles input time sequence, and the resetting door and update door of GRU neural network constantly adjust inherent parameters in a large amount of training,
Make it from the temporal dependence relationship between learning data in the data that one-dimensional convolutional neural networks CNN is extracted.
Preferably, it is described by meteorological training data tensor feeding PM2.5 prediction model be trained, model layer with
Introduce Dropout technology extensively between layer to prevent over-fitting.
The present invention also proposes a kind of PM2.5 concentration prediction device, comprising:
Processor；
Memory is coupled to the processor and is stored with instruction, and the instruction is executing reality by the processor
Now the step of PM2.5 concentration prediction method.
The present invention also proposes that a kind of computer-readable storage medium, the computer-readable storage medium are stored with
The step of application program of PM2.5 concentration prediction method, the application program realizes PM2.5 concentration prediction method as mentioned.
The present invention proposes a kind of PM2.5 concentration prediction method, apparatus and medium, has the following beneficial effects:
(1) PM2.5 concentration prediction technology of the invention can preferably learn localized variation mode, efficiently use convolutional Neural
The speed of network and the order sensibility of light weight characteristic and Recognition with Recurrent Neural Network allow to check data letter earlier under equal conditions
Breath, prediction accuracy are higher than traditional prediction method；
(2) time series is handled from time positive sequence and time backward using two-way GRU, can capture may be by unidirectional GRU
The mode neglected improves the feature learning ability of time series, to promote prediction accuracy.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
The structure shown according to these attached drawings obtains other attached drawings.
Fig. 1 is PM2.5 concentration prediction method flow diagram in first preferred embodiment of the invention；
Fig. 2 is PM2.5 prediction model structural schematic diagram in first preferred embodiment of the invention；
Fig. 3 is two-way GRU neural network working principle diagram in first preferred embodiment of the invention；
Fig. 4 is PM2.5 concentration prediction apparatus structure schematic diagram in first preferred embodiment of the invention；
Fig. 5 is computer storage medium structural schematic diagram in first preferred embodiment of the invention；
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
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 description, it is clear that described embodiment is only a part of the embodiments of the present invention, instead of all the embodiments.Base
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts it is all its
His embodiment, shall fall within the protection scope of the present invention.
It is to be appreciated that if relating to directionality instruction (such as up, down, left, right, before and after ...) in the embodiment of the present invention,
Then directionality instruction be only used for explain under a certain particular pose (as shown in the picture) between each component relative positional relationship,
Motion conditions etc., if the particular pose changes, directionality instruction is also correspondingly changed correspondingly.
In addition, being somebody's turn to do " first ", " second " etc. if relating to the description of " first ", " second " etc. in the embodiment of the present invention
Description be used for description purposes only, be not understood to indicate or imply its relative importance or implicitly indicate indicated skill
The quantity of art feature." first " is defined as a result, the feature of " second " can explicitly or implicitly include at least one spy
Sign.It in addition, the technical solution between each embodiment can be combined with each other, but must be with those of ordinary skill in the art's energy
It is enough realize based on, will be understood that the knot of this technical solution when conflicting or cannot achieve when occurs in the combination of technical solution
Conjunction is not present, also not the present invention claims protection scope within.
The characteristic that the present invention is directed to the dynamic instability of air pollutants PM2.5 concentration-time sequence and relies on for a long time, together
When by multiple meteorological factor influences such as temperature, pressure, wind speed, devise a kind of based on CNN and two-way GRU neural network
PM2.5 concentration prediction method.The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings；
In first preferred embodiment of the invention, as shown in Figure 1, including the following steps:
S01, meteorological data is obtained, and meteorological data is pre-processed and standardized；
In embodiments of the present invention, data set derives from embassy in Beijing, the U.S., which includes January 1 in 2010
Day to meteorological data hourly during on December 31st, 2014 and PM2.5 contamination data, include: PM2.5 concentration, dew point, temperature
8 features such as degree, air pressure, wind direction, wind speed, snowfall, rainfall reject the large-scale data missing point in data, last total
Data volume is 43800 row data, and using preceding 30000 row data as training set, 30001-38000 row is as verifying collection, 38001-
43800 rows are as test set.Wind direction feature contains 4 attributes in data: NW, CV, SE, NE need to encode that as floating number
According to, it is assigned a value of -10,0,10,20 respectively, finally, entire data set is made standardization, that is, subtracts the mean value of each feature,
Again divided by the variance of each feature:
Wherein, xi meanFor the mean value of ith feature variable, σi xIndicate the variance of ith feature variable；Mean value in the formula
Calculating with variance is just for training set, because verifying collection and the distribution of test set are unknown in reality；
S02, construction batch data generator, generate neural network and input tensor；
Tensor is the Data Structures of machine learning system, is popularization of the matrix to any dimension, and tensor generator is
Continuous grey iterative generation batch data is learnt and is predicted for model as input in model training and test process, the generation
The specified prediction input of device includes how many sample in following how many a time steps, each batch comprising how many a time steps, prediction target
This number, return is a tuple (batch of multivariable input data, corresponding PM2.5 concentration array)；
In embodiments of the present invention, mode input data are a tuple (sample, targets), wherein " sample " is comprising 3
The array (sample number of each batch, the past time step for being included, variable number) of a dimension, " target " are comprising 2 dimensions
The array (sample number of each batch, corresponding target PM2.5 concentration value) of degree, here, setting each training batch includes
72 samples (past 3 days), the past time step for being included are 36, prediction target is in following 3 time steps, variable number 8,
That is: according to the historical information of 8 variables such as the PM2.5 concentration of 36 hours of past, temperature, the PM2.5 of following 3 hours of prediction is dense
Angle value；
S10, PM2.5 prediction model is built based on one-dimensional convolutional neural networks CNN and two-way GRU neural network；
The PM2.5 prediction model, as shown in Fig. 2, specifically:
The input terminal of the one-dimensional convolutional neural networks CNN connects meteorological data tensor, one-dimensional convolutional neural networks CNN's
Output end connects the input terminal of two-way GRU neural network, and the output end of two-way GRU neural network accesses full articulamentum, last
Layer is comprising one for generating the neuron of PM2.5 concentration prediction value；
S20, meteorological training data tensor feeding PM2.5 prediction model is trained；Specifically:
S201, the one-dimensional convolutional neural networks CNN carry out local feature to each input variable time series respectively
Simultaneously dimensionality reduction is practised, is successively operated by convolution sum pondization, low-dimensional characteristic sequence is formed；
In the embodiment of the present invention, input of the input tensor comprising 8 variables as one-dimensional convolutional neural networks CNN, one
It ties up convolutional neural networks CNN and local feature learning and dimensionality reduction is carried out to each input variable time series respectively, use convolutional layer
Local feature learning is carried out, sub-sampling is carried out using layer pond layer, reduces the length of one-dimensional input.Wherein the first convolutional layer includes
80 convolution kernels (1*3 dimensional vector), activation primitive are " ReLu ", are followed by maximum pond layer, and input tensor is reduced half,
Into the second convolutional layer (parameter is with the first convolutional layer), final characteristic sequence is obtained after convolution operation；
S202, characteristic sequence is inputted to two-way GRU neural network, two-way GRU neural network is inverse from time positive sequence and time
Sequence learning characteristic sequence；
In the embodiment of the present invention, characteristic sequence feeds two-way GRU neural network, and two-way GRU neural network is from time positive sequence
With time backward learning characteristic sequence.Fig. 3 is the working principle of two-way GRU, and input feature vector sequence passes through two GRU neural networks
Learnt simultaneously from time positive sequence and time backward, the feature learnt merges again is output to next layer.This part packet
GRU layers two-way containing two layers, every layer of neuron number is 64.
The resetting door (r) of GRU and update door (z) constantly adjust inherent parameters in a large amount of training, make it from convolutional Neural
Temporal dependence relationship in the information that network extracts between learning data, wherein update door and specify which information that can remain into down
One state, and reset door and specify how the information of original state is combined with new input information；
The workflow of GRU unit is as follows:
Setting σ is activation primitive, and x (t) is input, and h (t-1) is the output of previous moment；
Firstly, calculating the activation for updating door and resetting door according to input and previous output:
zt=σ (Wz* [x (t), h (t-1)]) (2)
rt=σ (Wr* [x (t), h (t-1)]) (3)
Candidate output is calculated according to resetting door activation value:
The output of GRU unit is the linear interpolation of previous moment output and candidate output:
Wherein, WZ、WrAnd WhIt is the weight for updating door, resetting door and candidate output respectively；
S203, full articulamentum is accessed after two-way GRU, the present embodiment is 3 layers of full articulamentum, every layer of neuron number difference
Are as follows: 64,16,1, the last layer only includes a neuron, for generating PM2.5 concentration prediction value；
To prevent model over-fitting, introduce Dropout technology extensively between the layers, including convolutional layer and pond layer it
Between, it is between GRU layers and internal, which is in the training process at random given up the output feature of place layer with certain probability, should
Technology is in the training process at random given up the output feature of place layer with certain probability；In model training, according to verifying damage
Lose the current best model of dynamical save, that is, when completing a wheel training, if verifying penalty values do not improve, not overlay model
File, what is saved always in this way is best model in the training process；In model training, in the event of loss platform,
Local minimum is jumped out by increasing or reducing learning rate；In model training, if evaluation index MSE is in certain specific round
Do not improved, it is automatic to interrupt training, and it is stored in best model obtained in training process；
Probability used by the embodiment of the present invention of giving up is between 0.2 to 0.5.Model uses RMSprop optimizer, this
It is a kind of autoadapted learning rate method, is suitable for processing cycle neural network；
In the embodiment of the present invention, training is carried out with small lot, and tensor generator in the training process criticize by continuous grey iterative generation
Data are measured, each batch includes 72 samples, and maximum training round is set as 100 wheel of training on entire training set.Model makes
With mean absolute error (Mean Absolute Error, MAE) be used as loss function, training in each small lot basis
MAE value carries out backpropagation operation, meanwhile, select the error of mean square error (Mean Square Error, MSE) as model
Evaluation index measures the quality of prediction.Calculation is as follows:
Wherein, n is sample total, yiFor true value, yi' it is predicted value.
In model training, according to the verifying loss current best model of dynamical save, that is, when completing a wheel training, if tested
Card penalty values do not improve, then not overlay model file, what is saved always in this way is best model in the training process；?
In model training, in the event of loss platform, local minimum is jumped out by increasing or reducing learning rate；In model training
In, it is automatic to interrupt training if evaluation index MSE is not improved in certain specific round, and save in the training process
Obtained best model.
S30, Climate measurement data tensor is sent into have trained in the PM2.5 prediction model completed and is predicted, obtain
PM2.5 prediction concentrations value.
In the embodiment of the present invention, the model saved is loaded, by test set data input model, it is pre- to generate prediction PM2.5
Measured value.The embodiment of the present invention compared multi-layer perception (MLP) MLP, long-term memory network LSTM, gating cycle unit GRU, bidirectional gate
Cycling element Bi-GRU is controlled, to ensure the reasonability compared, each reference model is all made of the consistent hidden layer knot of same the method for the present invention
Structure and number of nodes record each comfortable training set, verifying collection, loss function value and evaluation function value on test set, such as following table institute
Show, numerical value is smaller, and to represent performance more excellent.
Table 1
It is found in BiGRU is compared with GRU, two-way GRU ratio GRU has lower error amount, illustrates to utilize two-way GRU energy
Prediction effect is promoted, this is because two-way GRU handles time series from time positive sequence and time backward, can capture may coverlet
The mode neglected to GRU improves the feature learning ability of time series.It is found in compared with other 4 models, this hair
A kind of bright PM2.5 concentration prediction method accuracy based on CNN and two-way GRU neural network is obviously improved, this is illustrated
Convolutional neural networks can help GRU to obtain better estimated performance, because CNN utilizes its local feature learning ability and dimensionality reduction
Ability obtains the sequence pattern being more conducive to for GRU processing.
The present invention also proposes a kind of PM2.5 concentration prediction device；
In second preferred embodiment of the invention, as shown in Figure 4, comprising:
Processor；
Memory is coupled to the processor and is stored with instruction, and the instruction is executing reality by the processor
Now the step of PM2.5 concentration prediction method, such as:
S01, meteorological data is obtained, and meteorological data is pre-processed and standardized；
S02, construction batch data generator, generate neural network and input tensor；
S10, PM2.5 prediction model is built based on one-dimensional convolutional neural networks CNN and two-way GRU neural network；
S20, meteorological training data tensor feeding PM2.5 prediction model is trained；
S30, Climate measurement data tensor is sent into have trained in the PM2.5 prediction model completed and is predicted, obtain
PM2.5 prediction concentrations value.
In the embodiment of the present invention, the detail of each step has hereinbefore been elaborated, and is no longer repeated herein；
In the embodiment of the present invention, the PM2.5 concentration prediction device internal processor can be made of integrated circuit,
Such as can be made of the integrated circuit of single package, it is also possible to be encapsulated by multiple identical functions or different function integrated
Circuit is formed, including one or more central processing unit (Central Processing unit, CPU), microprocessor,
Digital processing chip, graphics processor and combination of various control chips etc..Processor is taken using various interfaces and connection
All parts by running or execute the program being stored in memory or unit, and are called and are stored in memory
Data, to execute the various functions and processing data of PM2.5 concentration prediction；
Memory is mounted on and executes in PM2.5 concentration prediction device, and transporting for storing program code and various data
The access realized high speed during row, be automatically completed program or data.The memory includes read-only memory (Read-
Only Memory, ROM), random access memory (Random Access Memory, RAM), programmable read only memory
(Programmable Read-Only Memory, PROM), Erasable Programmable Read Only Memory EPROM (Erasable
Programmable Read-Only Memory, EPROM), disposable programmable read-only memory (One-time
Programmable Read-Only Memory, OTPROM), electronics erasing type can make carbon copies read-only memory
(Electrically-Erasable Programmable Read-Only Memory, EEPROM), CD-ROM (Compact
Disc Read-Only Memory, CD-ROM) or other disc memories, magnetic disk storage, magnetic tape storage or can
For carrying or any other computer-readable medium of storing data.
The present invention also proposes a kind of computer-readable storage medium；
In third preferred embodiment of the invention, as shown in Figure 5, comprising: the computer-readable storage medium is stored with
The step of application program of PM2.5 concentration prediction method, the application program realizes PM2.5 concentration prediction method as mentioned,
Such as:
S01, meteorological data is obtained, and meteorological data is pre-processed and standardized；
S02, construction batch data generator, generate neural network and input tensor；
S10, PM2.5 prediction model is built based on one-dimensional convolutional neural networks CNN and two-way GRU neural network；
S20, meteorological training data tensor feeding PM2.5 prediction model is trained；
S30, Climate measurement data tensor is sent into have trained in the PM2.5 prediction model completed and is predicted, obtain
PM2.5 prediction concentrations value.
In the embodiment of the present invention, the detail of each step has hereinbefore been elaborated, and is no longer repeated herein；
In the description of embodiments of the present invention, it should be noted that in flow chart or described otherwise above herein
Any process or method description be construed as, indicate to include one or more for realizing specific logical function or mistake
Module, segment or the part of the code of the executable instruction of the step of journey, and the range packet of the preferred embodiment of the present invention
Include other realization, wherein sequence shown or discussed can not be pressed, including according to related function by it is basic simultaneously
Mode or in the opposite order, Lai Zhihang function, this should be managed by the embodiment of the present invention person of ordinary skill in the field
Solution.
Expression or logic and/or step described otherwise above herein in flow charts, for example, being considered use
In the order list for the executable instruction for realizing logic function, may be embodied in any computer-readable medium, for
Instruction execution system, device or equipment (such as computer based system, including the system of processing module or other can be from instruction
Execute system, device or equipment instruction fetch and the system that executes instruction) use, or combine these instruction execution systems, device or
Equipment and use.For the purpose of this specification, " computer-readable medium ", which can be, any may include, store, communicating, propagating
Or transfer program uses for instruction execution system, device or equipment or in conjunction with these instruction execution systems, device or equipment
Device.The more specific example (non-exhaustive list) of computer-readable medium include the following: there are one or more wirings
Electrical connection section (electronic device), portable computer diskette box (magnetic device), random access memory (RAM), read-only memory
(ROM), erasable edit read-only storage (EPROM or flash memory), fiber device and portable optic disk is read-only deposits
Reservoir (CDROM).In addition, computer-readable medium can even is that the paper that can print described program on it or other are suitable
Medium because can then be edited for example by carrying out optical scanner to paper or other media, interpret or when necessary with
Other suitable methods are handled electronically to obtain described program, are then stored in computer storage.
The above description is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all at this
Under the inventive concept of invention, using equivalent structure transformation made by description of the invention and accompanying drawing content, or directly/use indirectly
It is included in other related technical areas in scope of patent protection of the invention.
Claims (10)
1. a kind of PM2.5 concentration prediction method, which is characterized in that this method is based on CNN and two-way GRU neural network, including such as
Lower step:
S10, PM2.5 prediction model is built based on one-dimensional convolutional neural networks CNN and two-way GRU neural network；
The PM2.5 prediction model, specifically:
The input terminal of the one-dimensional convolutional neural networks CNN connects meteorological data tensor, the output of one-dimensional convolutional neural networks CNN
End connects the input terminal of two-way GRU neural network, and the output end of two-way GRU neural network accesses full articulamentum, the last layer packet
Containing one for generating the neuron of PM2.5 concentration prediction value；
S20, meteorological training data tensor feeding PM2.5 prediction model is trained；
Specifically:
The one-dimensional convolutional neural networks CNN carries out local feature learning and dimensionality reduction to each input variable time series respectively,
It is successively operated by convolution sum pondization, forms low-dimensional characteristic sequence；Characteristic sequence is inputted into two-way GRU neural network, two-way GRU
Neural network is from time positive sequence and time backward learning characteristic sequence；
S30, Climate measurement data tensor is sent into have trained in the PM2.5 prediction model completed and is predicted, it is pre- to obtain PM2.5
Survey concentration value.
2. PM2.5 concentration prediction method according to claim 1, which is characterized in that before step S10 further include:
S01, meteorological data is obtained, and meteorological data is pre-processed and standardized；
S02, construction batch data generator, generate neural network and input tensor.
3. PM2.5 concentration prediction method according to claim 2, which is characterized in that the meteorological data, including but not
It is limited to: PM2.5 concentration, dew point, temperature, air pressure, wind direction, wind speed, snowfall and rainfall.
4. PM2.5 concentration prediction method according to claim 2, which is characterized in that the batch data generator,
Specified prediction input includes sample number in following time step number and each batch comprising time step number, prediction target,
Return is a tuple, i.e. multivariable input data a batch, corresponding target PM2.5 concentration array.
5. PM2.5 concentration prediction method according to claim 1, which is characterized in that the local feature learning and drop
Dimension, specifically: local feature learning is carried out using convolutional layer, sub-sampling is carried out using layer pond layer, reduces each one-dimensional input
Length.
6. PM2.5 concentration prediction method according to claim 1, which is characterized in that the PM2.5 prediction model makes
It uses mean absolute error as loss function, backpropagation is carried out according to mean absolute error value in each small lot in training
Operation, and error assessment index of the mean square error as model is selected, measure the quality of prediction.
7. PM2.5 concentration prediction method according to claim 1, which is characterized in that the two-way GRU neural network,
Including two GRU neural networks, input time sequence, the weight of GRU neural network are handled from time positive sequence and time backward respectively
It sets door and updates door and constantly adjust inherent parameters in a large amount of training, the data for extracting it from one-dimensional convolutional neural networks CNN
Temporal dependence relationship between middle learning data.
8. PM2.5 concentration prediction method according to claim 1, which is characterized in that described by meteorological training data
Amount is sent into PM2.5 prediction model and is trained, and introduces Dropout technology extensively between layers in model to prevent to intend
It closes.
9. a kind of PM2.5 concentration prediction device characterized by comprising
Processor；
Memory is coupled to the processor and is stored with instruction, and the instruction is executing the power of realization by the processor
Benefit require any one of 1 to 8 described in PM2.5 concentration prediction method the step of.
10. a kind of computer-readable storage medium, which is characterized in that the computer-readable storage medium is stored with
The application program of PM2.5 concentration prediction method, the application program realize such as PM2.5 described in any item of the claim 1 to 8
The step of concentration prediction method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910100028.5A CN109978228B (en) | 2019-01-31 | 2019-01-31 | PM2.5 concentration prediction method, device and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910100028.5A CN109978228B (en) | 2019-01-31 | 2019-01-31 | PM2.5 concentration prediction method, device and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109978228A true CN109978228A (en) | 2019-07-05 |
| CN109978228B CN109978228B (en) | 2023-12-12 |
Family
ID=67076864
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910100028.5A Active CN109978228B (en) | 2019-01-31 | 2019-01-31 | PM2.5 concentration prediction method, device and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109978228B (en) |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110533248A (en) * | 2019-09-02 | 2019-12-03 | 中科格物智信(天津)科技有限公司 | The Predict Model of Air Pollutant Density of fusion machine learning and LSTM |
| CN110599236A (en) * | 2019-08-14 | 2019-12-20 | 同济大学 | Short-time parking demand prediction method based on GRU model |
| CN110738354A (en) * | 2019-09-18 | 2020-01-31 | 北京建筑大学 | Method and device for predicting particulate matter concentration, storage medium and electronic equipment |
| CN110793896A (en) * | 2019-12-03 | 2020-02-14 | 承德石油高等专科学校 | Short-term prediction method for dust concentration in tail gas |
| CN110807577A (en) * | 2019-10-15 | 2020-02-18 | 中国石油天然气集团有限公司 | Pollution emission prediction method and device |
| CN111178610A (en) * | 2019-12-24 | 2020-05-19 | 中信百信银行股份有限公司 | Multi-layer convolution and GRU (generalized regression Unit) -based capacity prediction method and device |
| CN111210081A (en) * | 2020-01-09 | 2020-05-29 | 中国人民解放军国防科技大学 | Bi-GRU-based PM2.5 data processing and prediction method |
| CN111222704A (en) * | 2020-01-10 | 2020-06-02 | 中国人民解放军国防科技大学 | Bi-GRU-based vehicle position prediction method |
| CN111242369A (en) * | 2020-01-09 | 2020-06-05 | 中国人民解放军国防科技大学 | PM2.5 data prediction method based on multiple fusion convolution GRU |
| CN111259539A (en) * | 2020-01-14 | 2020-06-09 | 中山大学 | Simulation method and system for road PM2.5 refined pollution distribution and computer storage medium |
| CN111275168A (en) * | 2020-01-17 | 2020-06-12 | 南京信息工程大学 | Air quality prediction method of bidirectional gating circulation unit based on convolution full connection |
| CN111340292A (en) * | 2020-02-26 | 2020-06-26 | 福州大学 | Integrated neural network PM2.5 prediction method based on clustering |
| CN111345817A (en) * | 2020-02-25 | 2020-06-30 | 广州视源电子科技股份有限公司 | QRS complex position determination method, device, equipment and storage medium |
| CN111507533A (en) * | 2020-04-22 | 2020-08-07 | 中科三清科技有限公司 | Air pollutant concentration prediction method and device, electronic equipment and storage medium |
| CN111507415A (en) * | 2020-04-21 | 2020-08-07 | 南京信息工程大学 | Multi-source atmospheric data clustering method based on distribution density |
| CN111582588A (en) * | 2020-05-12 | 2020-08-25 | 中国人民解放军国防科技大学 | Building energy consumption prediction method based on triple convolution fusion GRU |
| CN112333194A (en) * | 2020-11-09 | 2021-02-05 | 国网上海市电力公司 | GRU-CNN-based comprehensive energy network security attack detection method |
| CN112734017A (en) * | 2021-04-01 | 2021-04-30 | 成都信息工程大学 | Method for establishing precipitation prediction model |
| CN112882683A (en) * | 2019-11-29 | 2021-06-01 | 阿里巴巴集团控股有限公司 | Random number generator judging method, random number processing method, device and equipment |
| CN112986492A (en) * | 2019-12-17 | 2021-06-18 | 华为技术有限公司 | Method and device for establishing gas concentration prediction model |
| CN113052618A (en) * | 2019-12-26 | 2021-06-29 | 华为技术有限公司 | Data prediction method and related equipment |
| CN113496310A (en) * | 2021-06-16 | 2021-10-12 | 国家超级计算深圳中心(深圳云计算中心) | Atmospheric pollutant prediction method and system based on deep learning model |
| WO2021236362A1 (en) * | 2020-05-19 | 2021-11-25 | Merln, Llc | Computing technologies for preserving signals in data inputs with moderate to high levels of variances in data sequence lengths for artificial neural network model training |
| CN114609008A (en) * | 2022-03-30 | 2022-06-10 | 中国科学院地理科学与资源研究所 | Training method of gas concentration prediction model and gas concentration prediction method |
| CN115034514A (en) * | 2022-07-06 | 2022-09-09 | 吉林大学 | Small sample time sequence prediction method and system based on meta-learning |
Citations (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105894125A (en) * | 2016-04-22 | 2016-08-24 | 牛东晓 | Transmission and transformation project cost estimation method |
| 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 |
| CN106845621A (en) * | 2017-01-18 | 2017-06-13 | 山东大学 | Dense population number method of counting and system based on depth convolutional neural networks |
| CN107274383A (en) * | 2017-05-17 | 2017-10-20 | 南京邮电大学 | A kind of haze visibility detecting method based on deep learning |
| US20170367651A1 (en) * | 2016-06-27 | 2017-12-28 | Facense Ltd. | Wearable respiration measurements system |
| CN107944450A (en) * | 2017-11-16 | 2018-04-20 | 深圳市华尊科技股份有限公司 | A kind of licence plate recognition method and device |
| CN108009674A (en) * | 2017-11-27 | 2018-05-08 | 上海师范大学 | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks |
| CN108108836A (en) * | 2017-12-15 | 2018-06-01 | 清华大学 | A kind of ozone concentration distribution forecasting method and system based on space-time deep learning |
| CN108197736A (en) * | 2017-12-29 | 2018-06-22 | 北京工业大学 | A kind of Air Quality Forecast method based on variation self-encoding encoder and extreme learning machine |
| JP2018147495A (en) * | 2018-03-20 | 2018-09-20 | ヤフー株式会社 | Prediction device, prediction method, prediction program, information processing device, information processing method, and information processing program |
| CN108717409A (en) * | 2018-05-16 | 2018-10-30 | 联动优势科技有限公司 | A kind of sequence labelling method and device |
| CN108860211A (en) * | 2018-05-25 | 2018-11-23 | 中车青岛四方机车车辆股份有限公司 | A kind of wrong report recognition methods and device based on shaft temperature sensor |
| CN108875587A (en) * | 2018-05-24 | 2018-11-23 | 北京飞搜科技有限公司 | Target distribution detection method and equipment |
| CN108985237A (en) * | 2018-07-20 | 2018-12-11 | 安徽农业大学 | A kind of detection method and its system of the wheat scab based on depth mixing |
| CN109063416A (en) * | 2018-07-23 | 2018-12-21 | 太原理工大学 | Gene expression prediction technique based on LSTM Recognition with Recurrent Neural Network |
| CN109063908A (en) * | 2018-07-30 | 2018-12-21 | 浙江鸿程计算机系统有限公司 | A kind of city AQI prediction and space fine granularity AQI hierarchical estimation method based on depth multi-task learning |
| CN109086803A (en) * | 2018-07-11 | 2018-12-25 | 南京邮电大学 | A kind of haze visibility detection system and method based on deep learning and the personalized factor |
| CN109147826A (en) * | 2018-08-22 | 2019-01-04 | 平安科技（深圳）有限公司 | Music emotion recognition method, device, computer equipment and computer storage medium |
-
2019
- 2019-01-31 CN CN201910100028.5A patent/CN109978228B/en active Active
Patent Citations (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105894125A (en) * | 2016-04-22 | 2016-08-24 | 牛东晓 | Transmission and transformation project cost estimation method |
| US20170367651A1 (en) * | 2016-06-27 | 2017-12-28 | Facense Ltd. | Wearable respiration measurements system |
| 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 |
| CN106845621A (en) * | 2017-01-18 | 2017-06-13 | 山东大学 | Dense population number method of counting and system based on depth convolutional neural networks |
| CN107274383A (en) * | 2017-05-17 | 2017-10-20 | 南京邮电大学 | A kind of haze visibility detecting method based on deep learning |
| CN107944450A (en) * | 2017-11-16 | 2018-04-20 | 深圳市华尊科技股份有限公司 | A kind of licence plate recognition method and device |
| CN108009674A (en) * | 2017-11-27 | 2018-05-08 | 上海师范大学 | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks |
| CN108108836A (en) * | 2017-12-15 | 2018-06-01 | 清华大学 | A kind of ozone concentration distribution forecasting method and system based on space-time deep learning |
| CN108197736A (en) * | 2017-12-29 | 2018-06-22 | 北京工业大学 | A kind of Air Quality Forecast method based on variation self-encoding encoder and extreme learning machine |
| JP2018147495A (en) * | 2018-03-20 | 2018-09-20 | ヤフー株式会社 | Prediction device, prediction method, prediction program, information processing device, information processing method, and information processing program |
| CN108717409A (en) * | 2018-05-16 | 2018-10-30 | 联动优势科技有限公司 | A kind of sequence labelling method and device |
| CN108875587A (en) * | 2018-05-24 | 2018-11-23 | 北京飞搜科技有限公司 | Target distribution detection method and equipment |
| CN108860211A (en) * | 2018-05-25 | 2018-11-23 | 中车青岛四方机车车辆股份有限公司 | A kind of wrong report recognition methods and device based on shaft temperature sensor |
| CN109086803A (en) * | 2018-07-11 | 2018-12-25 | 南京邮电大学 | A kind of haze visibility detection system and method based on deep learning and the personalized factor |
| CN108985237A (en) * | 2018-07-20 | 2018-12-11 | 安徽农业大学 | A kind of detection method and its system of the wheat scab based on depth mixing |
| CN109063416A (en) * | 2018-07-23 | 2018-12-21 | 太原理工大学 | Gene expression prediction technique based on LSTM Recognition with Recurrent Neural Network |
| CN109063908A (en) * | 2018-07-30 | 2018-12-21 | 浙江鸿程计算机系统有限公司 | A kind of city AQI prediction and space fine granularity AQI hierarchical estimation method based on depth multi-task learning |
| CN109147826A (en) * | 2018-08-22 | 2019-01-04 | 平安科技（深圳）有限公司 | Music emotion recognition method, device, computer equipment and computer storage medium |
Non-Patent Citations (8)
| Title |
|---|
| CLARK_XU: "Keras系列之LSTM和GRU实践", 《HTTPS://BLOG.CSDN.NET/FINDINGXU/ARTICLE/DETAILS/86696904》 * |
| V. ATHIRA 等: "DeepAirNet: Applying recurrent networks for air quality prediction", 《PROCEDIA COMPUT. SCI》, vol. 132 * |
| 张兰霞 等: "\"基于双向GRU神经网络和双层注意力机制的中文文本中人物关系抽取研究\"", 《计算机应用与软件》, vol. 35, no. 11 * |
| 金欢欢 等: "基于生成少数类技术的深度自动睡眠分期模型", 《计算机应用》, vol. 38, no. 9 * |
| 陈立国 等: "\"基于GRU型循环神经网络的随机域名检测"", 《计算机系统应用》, vol. 27, no. 8 * |
| 黄婕: ""基于RNN-CNN集成深度学习模型的PM2.5小时浓度预测研究"", 《中国优秀硕士学位论文全文数据库 工程科技I辑》 * |
| 黄婕: ""基于RNN-CNN集成深度学习模型的PM2.5小时浓度预测研究"", 《中国优秀硕士学位论文全文数据库 工程科技I辑》, 15 December 2018 (2018-12-15) * |
| 黎镭 等: ""GRU递归神经网络对股票收盘价的预测研究"", 《计算机与现代化》, no. 11, pages 296 - 299 * |
Cited By (33)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110599236A (en) * | 2019-08-14 | 2019-12-20 | 同济大学 | Short-time parking demand prediction method based on GRU model |
| CN110533248A (en) * | 2019-09-02 | 2019-12-03 | 中科格物智信(天津)科技有限公司 | The Predict Model of Air Pollutant Density of fusion machine learning and LSTM |
| CN110738354B (en) * | 2019-09-18 | 2021-02-05 | 北京建筑大学 | Method and device for predicting particulate matter concentration, storage medium and electronic equipment |
| CN110738354A (en) * | 2019-09-18 | 2020-01-31 | 北京建筑大学 | Method and device for predicting particulate matter concentration, storage medium and electronic equipment |
| CN110807577A (en) * | 2019-10-15 | 2020-02-18 | 中国石油天然气集团有限公司 | Pollution emission prediction method and device |
| CN112882683A (en) * | 2019-11-29 | 2021-06-01 | 阿里巴巴集团控股有限公司 | Random number generator judging method, random number processing method, device and equipment |
| CN110793896B (en) * | 2019-12-03 | 2022-04-08 | 承德石油高等专科学校 | Short-term prediction method for dust concentration in tail gas |
| CN110793896A (en) * | 2019-12-03 | 2020-02-14 | 承德石油高等专科学校 | Short-term prediction method for dust concentration in tail gas |
| CN112986492A (en) * | 2019-12-17 | 2021-06-18 | 华为技术有限公司 | Method and device for establishing gas concentration prediction model |
| CN111178610A (en) * | 2019-12-24 | 2020-05-19 | 中信百信银行股份有限公司 | Multi-layer convolution and GRU (generalized regression Unit) -based capacity prediction method and device |
| CN113052618A (en) * | 2019-12-26 | 2021-06-29 | 华为技术有限公司 | Data prediction method and related equipment |
| CN111242369A (en) * | 2020-01-09 | 2020-06-05 | 中国人民解放军国防科技大学 | PM2.5 data prediction method based on multiple fusion convolution GRU |
| CN111242369B (en) * | 2020-01-09 | 2023-06-02 | 中国人民解放军国防科技大学 | PM2.5 data prediction method based on multiple fusion convolution GRU |
| CN111210081A (en) * | 2020-01-09 | 2020-05-29 | 中国人民解放军国防科技大学 | Bi-GRU-based PM2.5 data processing and prediction method |
| CN111222704A (en) * | 2020-01-10 | 2020-06-02 | 中国人民解放军国防科技大学 | Bi-GRU-based vehicle position prediction method |
| CN111259539A (en) * | 2020-01-14 | 2020-06-09 | 中山大学 | Simulation method and system for road PM2.5 refined pollution distribution and computer storage medium |
| CN111275168A (en) * | 2020-01-17 | 2020-06-12 | 南京信息工程大学 | Air quality prediction method of bidirectional gating circulation unit based on convolution full connection |
| CN111345817A (en) * | 2020-02-25 | 2020-06-30 | 广州视源电子科技股份有限公司 | QRS complex position determination method, device, equipment and storage medium |
| CN111340292B (en) * | 2020-02-26 | 2022-05-10 | 福州大学 | Integrated neural network PM2.5 prediction method based on clustering |
| CN111340292A (en) * | 2020-02-26 | 2020-06-26 | 福州大学 | Integrated neural network PM2.5 prediction method based on clustering |
| CN111507415B (en) * | 2020-04-21 | 2023-07-25 | 南京信息工程大学 | Multi-source atmosphere data clustering method based on distribution density |
| CN111507415A (en) * | 2020-04-21 | 2020-08-07 | 南京信息工程大学 | Multi-source atmospheric data clustering method based on distribution density |
| CN111507533A (en) * | 2020-04-22 | 2020-08-07 | 中科三清科技有限公司 | Air pollutant concentration prediction method and device, electronic equipment and storage medium |
| CN111582588B (en) * | 2020-05-12 | 2022-07-05 | 中国人民解放军国防科技大学 | Building energy consumption prediction method based on triple convolution fusion GRU |
| CN111582588A (en) * | 2020-05-12 | 2020-08-25 | 中国人民解放军国防科技大学 | Building energy consumption prediction method based on triple convolution fusion GRU |
| WO2021236362A1 (en) * | 2020-05-19 | 2021-11-25 | Merln, Llc | Computing technologies for preserving signals in data inputs with moderate to high levels of variances in data sequence lengths for artificial neural network model training |
| CN112333194B (en) * | 2020-11-09 | 2022-08-09 | 国网上海市电力公司 | GRU-CNN-based comprehensive energy network security attack detection method |
| CN112333194A (en) * | 2020-11-09 | 2021-02-05 | 国网上海市电力公司 | GRU-CNN-based comprehensive energy network security attack detection method |
| CN112734017A (en) * | 2021-04-01 | 2021-04-30 | 成都信息工程大学 | Method for establishing precipitation prediction model |
| CN113496310A (en) * | 2021-06-16 | 2021-10-12 | 国家超级计算深圳中心(深圳云计算中心) | Atmospheric pollutant prediction method and system based on deep learning model |
| CN114609008A (en) * | 2022-03-30 | 2022-06-10 | 中国科学院地理科学与资源研究所 | Training method of gas concentration prediction model and gas concentration prediction method |
| CN114609008B (en) * | 2022-03-30 | 2023-09-01 | 中国科学院地理科学与资源研究所 | Training method of gas concentration prediction model and gas concentration prediction method |
| CN115034514A (en) * | 2022-07-06 | 2022-09-09 | 吉林大学 | Small sample time sequence prediction method and system based on meta-learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109978228B (en) | 2023-12-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109978228A (en) | A kind of PM2.5 concentration prediction method, apparatus and medium | |
| CN110276439A (en) | Time Series Forecasting Methods, device and storage medium based on attention mechanism | |
| CN111814956B (en) | Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction | |
| CN108805222A (en) | A kind of deep learning digital handwriting body recognition methods based on ARM platforms | |
| CN110083125A (en) | A kind of machine tool thermal error modeling method based on deep learning | |
| CN111275168A (en) | Air quality prediction method of bidirectional gating circulation unit based on convolution full connection | |
| CN114462718A (en) | CNN-GRU wind power prediction method based on time sliding window | |
| CN111784777A (en) | SMT material quantity statistical method and statistical system based on convolutional neural network | |
| CN115907436B (en) | Quality coupling prediction-based water resource water environment regulation and control method and system | |
| CN109615147A (en) | A kind of following 72 hours air pollution forecasting method for early warning | |
| CN111310918B (en) | Data processing method, device, computer equipment and storage medium | |
| CN114912673A (en) | Water level prediction method based on whale optimization algorithm and long-term and short-term memory network | |
| CN114758180B (en) | Knowledge distillation-based lightweight flower identification method | |
| CN115310782A (en) | Power consumer demand response potential evaluation method and device based on neural turing machine | |
| CN113449919B (en) | Power consumption prediction method and system based on feature and trend perception | |
| CN111292121A (en) | Garden load prediction method and system based on garden image | |
| CN107038450A (en) | Unmanned plane policing system based on deep learning | |
| CN111783688B (en) | Remote sensing image scene classification method based on convolutional neural network | |
| CN116186594B (en) | Method for realizing intelligent detection of environment change trend based on decision network combined with big data | |
| CN116911178A (en) | Method and system for predicting capacity of small and medium-sized reservoirs based on weather forecast | |
| CN116402194A (en) | Multi-time scale load prediction method based on hybrid neural network | |
| CN115330082A (en) | PM2.5 concentration prediction method of LSTM network based on attention mechanism | |
| CN115640901A (en) | Small sample load prediction method based on hybrid neural network and generation countermeasure | |
| CN112052386B (en) | Information recommendation method, device and storage medium | |
| CN114970357A (en) | Energy-saving effect evaluation method, system, device and storage medium |
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 |