CN107909206A - A kind of PM2.5 Forecasting Methodologies based on deep structure Recognition with Recurrent Neural Network - Google Patents
A kind of PM2.5 Forecasting Methodologies based on deep structure Recognition with Recurrent Neural Network Download PDFInfo
- Publication number
- CN107909206A CN107909206A CN201711130537.XA CN201711130537A CN107909206A CN 107909206 A CN107909206 A CN 107909206A CN 201711130537 A CN201711130537 A CN 201711130537A CN 107909206 A CN107909206 A CN 107909206A
- Authority
- CN
- China
- Prior art keywords
- mrow
- data
- msub
- training
- group
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a kind of PM2.5 Forecasting Methodologies based on deep structure Recognition with Recurrent Neural Network, utilize the mass data of collection, according to deep learning and the prediction model of the PM2.5 of Recognition with Recurrent Neural Network the Theory Construction deep structure, pass through the extraction and training of data characteristics, realize the prediction of haze weather, it is intended to improve the efficiency and precision of haze prediction, prevents and administer to propose convictive decision-making foundation for haze.Prediction model does not almost have requirement for data structure, as long as energy self study when data are sufficiently large so that deep learning is very suitable for the needs of internet big data application instantly.
Description
Technical field
The invention belongs to environmental project in detection technique field, more specifically, it is related to one kind and is followed based on deep structure
The PM2.5 Forecasting Methodologies of ring neutral net.
Background technology
Air quality is always the significant problem of relation mankind's future destiny, with social progress, the urgency of car ownership
Increase severely plus cause people's particle content of inhaling in air to be substantially increased, problem of environmental pollution is on the rise.With air quality
Continuous deterioration, haze weather phenomenon is more and more, and harm is increasing.Haze is a kind of hazard weather phenomenon.Inhalable
Grain thing PM2.5, is only haze weather main reason, not only air quality is had a serious impact, it is important that human body is good for
Health threatens huge.
Forecasting research for air quality has many thought and method, in numerous methods, based on system work
The thought of journey, and effectively combine new theory and new method to environmental quality, especially haze, realizes quantitative research and effectively pre-
Survey is Main Trends of The Development.
Due to the influence of a large amount of uncertain and complicated sexual factors such as climate, temperature, mankind's activity, all kinds of weather datas
Time series there is the characteristic such as nonlinearity, uncertainty, conventional prediction method is difficult to grasp change therein
Rule and variation characteristic.
The neutral net of shallow-layer is solving the problems, such as the simple or more positive effect of limitation, but due to modeling and table
Show that ability is limited, run into some in real life it is more complicated be involved in the problems, such as that natural sign fulfillment capability is limited.
The neutral net of depth, has multiple stealthy layers, is taken out than traditional neural network with more the advantage in structure, feature
As ability can be strong.Deep neural network uses a kind of brand-new coding mode, it is not necessary to is directly to solve the problems, such as algorithm for design
And programming, it is only necessary to program for training process, oneself can just learn to the correct side solved the problems, such as during network retraining
Method, in the case where data volume is guaranteed, simple algorithm can obtain special effect plus complicated data.
Simultaneously because the tremendous increase of chip process performance, increases and recent machine learning for trained data explosion
There is remarkable progress with signal, information processing research, these all allow deep learning method efficiently use the non-linear of complexity
Function and nonlinear compound function come the character representation that learns to be distributed and be layered and can substantially effectively using mark and
The data of non-mark.
Recognition with Recurrent Neural Network (RNN) is a kind of depth network that can be used for unsupervised (and having supervision) study, and depth is even
It can reach consistent with the length of list entries, under unsupervised learning pattern, RNN is used to pre- according to previous data sample
Following data sequence is surveyed, and classification information is not used in learning process, therefore RNN is very suitable for sequence data modeling.
The content of the invention
It is an object of the invention to overcome the deficiencies of the prior art and provide a kind of based on deep structure Recognition with Recurrent Neural Network
PM2.5 Forecasting Methodologies, by combining the basic theory of Recognition with Recurrent Neural Network, net structure and flow principle are pre- to build PM2.5
Model is surveyed, so as to fulfill the prediction of PM2.5.
For achieving the above object, a kind of PM2.5 Forecasting Methodologies based on deep structure Recognition with Recurrent Neural Network of the present invention,
It is characterised in that it includes following steps：
(1), acquisition historical weather data, including temperature hourly, illumination, wind speed, rainfall, SO2, O3, NO,
PM10, PM2.5 data target, wherein, temperature unit：DEG C, Light Units：Lm/ ㎡, wind speed unit：M/s, rainfall unit：
Mm, SO2, O3, NO, PM10, PM2.5 are concentration datas；
(2), data prediction
(2.1), completion processing is carried out to missing historical weather data
The historical weather data lacked using averaging method completion：
Wherein, XtRepresent the missing historical weather data at current time, Xt-1Represent the missing weather history number of previous moment
According to Xt+1The missing historical weather data of later moment in time before expression；
(2.2), all historical weather datas are normalized
Will be so historical weather data be normalized between -1~1 according to equation below；
Wherein, X' represents the historical weather data after normalization, and X represents the historical weather data before normalization,Represent
Historical weather data average, XmaxRepresent historical weather data maximum, XminRepresent historical weather data minimum value；
(3), the historical weather data after the completion of pretreatment is proportionally divided into training data and test data；
(4), the PM2.5 prediction models based on deep learning theory and Recognition with Recurrent Neural Network construction deep structure
(4.1), deep layer Recognition with Recurrent Neural Network prediction model is built：One layer of input layer, multilayer hidden layer, one layer of input layer,
Model depth is more than N layers, and input is training data, and output is the predicted value of PM2.5 concentration；
(4.2), set input layer dimension as K × (H-1), output layer dimension is 1 × T, input layer and hidden layer, hidden layer with
The activation primitive of hidden layer, hidden layer and output layer uses Tanh functions；
Wherein, the depth that K expressions Recognition with Recurrent Neural Network is unfolded in temporal sequence, i.e. K time frame, each time frame are defeated
Enter one group of historical weather data；H represents data target number, and T represents the prediction model output data number of Recognition with Recurrent Neural Network,
Represent the PM2.5 concentration with the K bars historical data prediction following T moment, i.e., in order to input the weather data at preceding K moment, in advance
The PM2.5 concentration datas at T moment after measuring；
(4.3), the loss function used in PM2.5 prediction models is selected
Using mean square error as loss function in PM2.5 prediction models：
Wherein, q represents iterations, and t is output vector dimension, yi,jRepresent the actual value of training data, yi,j' represent instruction
Practice the predicted value of data；
(4.4), the parameter in PM2.5 prediction models is updated using small lot stochastic gradient descent algorithm
(4.4.1), initiation parameter θ0；
(4.4.2), by training data, according to time series, often m training data is divided into one group, recycles small lot random
Gradient descent algorithm calculates the Grad of each training data in first group of training data, and then Grad, which is weighted, averagely asks
With obtain the downward gradient of this group of training dataI represents i-th group of training data,Represent the corresponding input of the τ training data, output data in i-th group；
(4.4.3), the downward gradient of this group of training data update the parameter in PM2.5 prediction models, parameter more new formula
For：
Wherein, θi-1Represent the target component after the completion of one group of data training, θiAfter the completion of representing this group of training data
Target component, η represent learning rate；
(4.4.4), after the completion of the target component renewal after the completion of this group of training data, return to step (4.4.2) carries out
The training and renewal of next group of training data, until error amount is less than setting expected error value or last group of training data instruction
Practice and terminate when completing, then update and preserve final argument, obtain the PM2.5 prediction models of training completion；
(5), whether judge PM2.5 prediction models reaches trained stop condition
Test set data are inputted into a K groups data into trained PM2.5 prediction models according to time series,
Export T predicted value, then error judgment will be carried out between each predicted value and actual value, if error in allowed band,
Think that prediction model completes training, otherwise return to step (4) re -training, until reaching stop condition；
(6), the prediction of PM2.5 is carried out using PM2.5 prediction models
Current K groups weather data is inputted to PM2.5 prediction models, exports T PM2.5 predicted value.
What the goal of the invention of the present invention was realized in：
A kind of PM2.5 Forecasting Methodologies based on deep structure Recognition with Recurrent Neural Network of the present invention, using the mass data of collection,
According to deep learning and the prediction model of the PM2.5 of Recognition with Recurrent Neural Network the Theory Construction deep structure, pass through carrying for data characteristics
Take and train, realize the prediction of haze weather, it is intended to improve the efficiency and precision of haze prediction, prevent and administer to propose for haze
Convictive decision-making foundation.Prediction model does not almost have requirement for data structure, as long as data can be learnt by oneself when sufficiently large
Practise so that deep learning is very suitable for the needs of internet big data application instantly.
Meanwhile a kind of PM2.5 Forecasting Methodologies based on deep structure Recognition with Recurrent Neural Network of the present invention are also with beneficial below
Effect：
(1) initial data is transformed to higher level abstract table by some simple nonlinear models by prediction model
Reach, recombinant multilayer conversion, study extracts extremely complex Function feature method.The prediction of traditional shallow structure is evaded
Model is the expression ability in the case of finite sample and unit of account to complicated function is limited the problem of.
(2) core of prediction model is difference lies in multiple hidden layers, and each layer feature extraction is not artificial participation design, and
It is to learn by oneself to obtain from data using general learning process, data structure is not required, simplifies data handling procedure,
Raising efficiency.
(3) it can be achieved with the prediction of the PM2.5 concentration of this area using the data of different regions, according to actual conditions and want
Seek the parameter for redefining PM2.2 prediction models, it is not necessary to rebuild network again, therefore there can be flexibility and removable
Plant property.
Brief description of the drawings
Fig. 1 is a kind of PM2.5 Forecasting Methodology flow charts based on deep structure Recognition with Recurrent Neural Network of the present invention；
Embodiment
The embodiment of the present invention is described below in conjunction with the accompanying drawings, so as to those skilled in the art preferably
Understand the present invention.Requiring particular attention is that in the following description, when known function and the detailed description of design perhaps
When can desalinate the main contents of the present invention, these descriptions will be ignored herein.
Embodiment
Fig. 1 is a kind of PM2.5 Forecasting Methodology flow charts based on deep structure Recognition with Recurrent Neural Network of the present invention.
In the present embodiment, as shown in Figure 1, a kind of PM2.5 predictions based on deep structure Recognition with Recurrent Neural Network of the present invention
Method, comprises the following steps：
S1, acquisition historical weather data, including temperature hourly, illumination, wind speed, rainfall, SO2, O3, NO, PM10,
PM2.5 data targets, wherein, temperature unit：DEG C, Light Units：Lm/ ㎡, wind speed unit：M/s, rainfall unit：Mm, SO2,
O3, NO, PM10, PM2.5 are concentration datas；
In the present embodiment, apply for obtaining the weather history number in May, 2014 in May, 2017 in China Meteorological Administration
According to data message includes the temperature of each hour, illumination, wind speed, rainfall, SO2, O3, NO, PM10, PM2.5 data targets
(wherein temperature unit：DEG C, Light Units：Lm/ ㎡, wind speed unit：M/s, rainfall unit：Mm, SO2, O3, NO, PM10,
PM2.5 is concentration data), amount to 9 indexs, amount to 26280 × 9 data, ensure that the circulation nerve net of deep structure
Network predicts the data volume of PM2.5 models.
S2, data prediction
S2.1, carry out completion processing to missing historical weather data
The weather data of collection is the historical data based on time series, collects data there are few missing data, uses
The method completion missing data of average, ensures the integrality of data.
The formula of historical weather data using averaging method completion missing is：
Wherein, XtRepresent the missing historical weather data at current time, Xt-1Represent the missing weather history number of previous moment
According to Xt+1The missing historical weather data of later moment in time before expression；
S2.2, be normalized all historical weather datas
Will be so historical weather data be normalized between -1~1 according to equation below；
Wherein, X' represents the historical weather data after normalization, and X represents the historical weather data before normalization,Represent
Historical weather data average, XmaxRepresent historical weather data maximum, XminRepresent historical weather data minimum value；
S3, be divided into training data according to 70%, 30% ratio by the historical weather data after the completion of pretreatment and test number
According to；
S4, the PM2.5 prediction models based on deep learning theory and Recognition with Recurrent Neural Network construction deep structure
In S4.1, the present embodiment, deep layer Recognition with Recurrent Neural Network prediction model is built：One layer of input layer, eight layers of hidden layer, one
Layer input layer, model depth are 10 layers, and the number of nodes of input layer is 9, and the number of nodes of hidden layer is 50, and the number of nodes of output layer is
5, input is training data, and output is the predicted value of PM2.5 concentration；
S4.2, set input layer dimension as K × 9, and output layer dimension is 1 × T, input layer and hidden layer, and hidden layer is with hiding
The activation primitive of layer, hidden layer and output layer uses Tanh functions；
Wherein, the depth that K expressions Recognition with Recurrent Neural Network is unfolded in temporal sequence, i.e. K time frame, each time frame are defeated
Enter one group of historical weather data；T represents the prediction model output data number of Recognition with Recurrent Neural Network, and K bar historical datas are used in expression
The PM2.5 concentration at prediction following T moment, i.e., in order to input the weather data at preceding K moment, T moment after predicting
PM2.5 concentration datas；
The loss function used in S4.3, selection PM2.5 prediction models
In the present embodiment, in PM2.5 prediction models by data according to time series, concentrate from training data and chosen by section
Go out subset MiAs a small lot data set, subset MiMiddle sample point number is m, comprising inputting and marking output data, is remembered
For Mi(Xi,Yi), input data Xi, authentic signature output data is Yi, neural metwork training output data is to be expressed as Yi',
Correspondence represents as follows：
Represent to input τ in temporal sequenceMatrix, that is, obtain
Row vector, τ=1,2 ..., m-k+1.
Each subset MiTraining loss function be chosen to be mean square error, represent to predict mould in given training data subset
The error metrics of the parameter vector of type, are denoted asRepresented with equation below：
Wherein, q represents iterations, and t is output vector dimension, yi,jRepresent the actual value of training data, yi,j' represent instruction
Practice the predicted value of data；
S4.4, using small lot stochastic gradient descent algorithm update PM2.5 prediction models in parameter
S4.4.1, initiation parameter θ0；
S4.4.2, by training data, according to time series, often m training data is divided into one group, recycles small lot boarding steps
The Grad that descent algorithm calculates each training data in first group of training data is spent, then q Grad is weighted flat
Sum, obtain the downward gradient of this group of training dataI represents i-th group of training data,Represent the corresponding input of the τ training data, output data in i-th group；
S4.4.3, the downward gradient of this group of training data update the parameter in PM2.5 prediction models, parameter more new formula
For：
Wherein, θi-1Represent the target component after the completion of one group of data training, θiAfter the completion of representing this group of training data
Target component, η represent learning rate；
In the present embodiment, calculated with reference to embodiment in step S3, PM2.5 prediction models using small lot stochastic gradient descent
During method training parameter, each subset MiIn need iteration (m-k+1) secondary, use K datas every time, it is rightDerivation obtains
Average summation is weighted to the gradient of each parameter, and to m-k+1 gradient, the decline ladder as a small lot training
Degree, then carries out parameter renewal, is specially：
S4.4.4, when after the completion of this group of training data target component renewal after the completion of, return to step S4.4.2 is carried out down
The training and renewal of one group of training data, until less than setting expected error value or reaching last group training when downward gradient
Data training terminates when completing, and then updates and preserves final argument, obtains the PM2.5 prediction models of training completion；
S5, judge PM2.5 prediction models whether reach trained stop condition
Test set data are inputted into a K groups data into trained PM2.5 prediction models according to time series,
Export T predicted value, then error judgment will be carried out between each predicted value and actual value, if error in allowed band,
Think that prediction model completes training, otherwise return to step S4 re -trainings, until reaching stop condition；
S6, the prediction using PM2.5 prediction models progress PM2.5
Current K groups weather data is inputted to PM2.5 prediction models, exports T PM2.5 predicted value.
Although the illustrative embodiment of the present invention is described above, in order to the technology of the art
Personnel understand the present invention, it should be apparent that the invention is not restricted to the scope of embodiment, to the common skill of the art
For art personnel, if various change appended claim limit and definite the spirit and scope of the present invention in, these
Change is it will be apparent that all utilize the innovation and creation of present inventive concept in the row of protection.
Claims (1)
1. a kind of PM2.5 Forecasting Methodologies based on deep structure Recognition with Recurrent Neural Network, it is characterised in that comprise the following steps：
(1), acquisition historical weather data, including temperature hourly, illumination, wind speed, rainfall, SO2, O3, NO, PM10,
PM2.5 data targets, wherein, temperature unit：DEG C, Light Units：Lm/ ㎡, wind speed unit：M/s, rainfall unit：Mm, SO2,
O3, NO, PM10, PM2.5 are concentration datas；
(2), data prediction
(2.1), completion processing is carried out to missing historical weather data
The historical weather data lacked using averaging method completion：
<mrow>
<msub>
<mi>X</mi>
<mi>t</mi>
</msub>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mn>2</mn>
</mfrac>
<mrow>
<mo>(</mo>
<msub>
<mi>X</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>+</mo>
<msub>
<mi>X</mi>
<mrow>
<mi>t</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
Wherein, XtRepresent the missing historical weather data at current time, Xt-1Represent the missing historical weather data of previous moment,
Xt+1The missing historical weather data of later moment in time before expression；
(2.2), all historical weather datas are normalized
Will be so historical weather data be normalized between -1~1 according to equation below；
<mrow>
<msup>
<mi>X</mi>
<mo>&prime;</mo>
</msup>
<mo>=</mo>
<mfrac>
<mrow>
<mi>X</mi>
<mo>-</mo>
<mover>
<mi>X</mi>
<mo>&OverBar;</mo>
</mover>
</mrow>
<mrow>
<msub>
<mi>X</mi>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>X</mi>
<mi>min</mi>
</msub>
</mrow>
</mfrac>
</mrow>
Wherein, X' represents the historical weather data after normalization, and X represents the historical weather data before normalization,Represent history day
Gas data mean value, XmaxRepresent historical weather data maximum, XminRepresent historical weather data minimum value；
(3), the historical weather data after the completion of pretreatment is proportionally divided into training data and test data；
(4), the PM2.5 prediction models based on deep learning theory and Recognition with Recurrent Neural Network construction deep structure
(4.1), deep layer Recognition with Recurrent Neural Network prediction model is built：One layer of input layer, multilayer hidden layer, one layer of input layer, model
Depth is more than N layers, and input is training data, and output is the predicted value of PM2.5 concentration；
(4.2), input layer dimension is set as K × (H-1), and output layer dimension is 1 × T, input layer and hidden layer, and hidden layer is with hiding
The activation primitive of layer, hidden layer and output layer uses Tanh functions；
Wherein, K represents the depth that Recognition with Recurrent Neural Network is unfolded in temporal sequence, i.e. K time frame, each time frame input one
Group historical weather data；H represents data target number, and T represents the prediction model output data number of Recognition with Recurrent Neural Network, represents
The PM2.5 concentration at following T moment is predicted with K bars historical data, i.e., in order to input the weather data at preceding K moment, is predicted
The PM2.5 concentration datas at T moment afterwards；
(4.3), the loss function used in PM2.5 prediction models is selected
Using mean square error as loss function in PM2.5 prediction models：
<mrow>
<mi>L</mi>
<mi>o</mi>
<mi>s</mi>
<mi>s</mi>
<msub>
<mrow>
<mo>(</mo>
<mi>y</mi>
<mo>,</mo>
<msup>
<mi>y</mi>
<mo>&prime;</mo>
</msup>
<mo>)</mo>
</mrow>
<mrow>
<mi>b</mi>
<mi>a</mi>
<mi>t</mi>
<mi>c</mi>
<mi>h</mi>
</mrow>
</msub>
<mo>=</mo>
<mi>J</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&theta;</mi>
<mn>0</mn>
</msub>
<mo>;</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>&tau;</mi>
</msubsup>
<mo>;</mo>
<msubsup>
<mi>y</mi>
<mi>i</mi>
<mi>&tau;</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mn>2</mn>
<mi>q</mi>
<mi>t</mi>
</mrow>
</mfrac>
<mo>&lsqb;</mo>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>q</mi>
</msubsup>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>t</mi>
</msubsup>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>y</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>-</mo>
<msup>
<msub>
<mi>y</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>&prime;</mo>
</msup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>&rsqb;</mo>
</mrow>
Wherein, t is output vector dimension, yi,jRepresent the actual value of training data, yi,j' represent training data predicted value；
(4.4), the parameter in PM2.5 prediction models is updated using small lot stochastic gradient descent algorithm
(4.4.1), initiation parameter θ0；
(4.4.2), by training data, according to time series, often m training data is divided into one group, recycles small lot stochastic gradient
Descent algorithm calculates the Grad of each training training data in first group of training data, and then Grad, which is weighted, averagely asks
With obtain the downward gradient of this group of training dataI represents i-th group of training data,
Represent the corresponding input of the τ training data, output data in i-th group；
(4.4.3), the downward gradient of this group of training data update the parameter in PM2.5 prediction models, and parameter more new formula is：
<mrow>
<msub>
<mi>&theta;</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<msub>
<mi>&theta;</mi>
<mrow>
<mi>i</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>-</mo>
<mfrac>
<mi>&eta;</mi>
<mi>q</mi>
</mfrac>
<msub>
<mo>&dtri;</mo>
<msub>
<mi>&theta;</mi>
<mi>i</mi>
</msub>
</msub>
<mi>J</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&theta;</mi>
<mn>0</mn>
</msub>
<mo>;</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>&tau;</mi>
</msubsup>
<mo>;</mo>
<msubsup>
<mi>y</mi>
<mi>i</mi>
<mi>&tau;</mi>
</msubsup>
<mo>)</mo>
</mrow>
</mrow>
Wherein, θi-1Represent the target component after the completion of one group of data training, θiRepresent the target after the completion of this group of training data
Parameter, η represent learning rate；
(4.4.4), after the completion of the target component renewal after the completion of this group of training data, return to step (4.4.2) carries out next
The training and renewal of group training data, until error amount has been trained less than setting expected error value or last group of training data
Into when terminate, then update and preserve final argument, obtain training completion PM2.5 prediction models；
(5), whether judge PM2.5 prediction models reaches trained stop condition
Test set data are inputted into a K groups data into trained PM2.5 prediction models according to time series, output
T predicted value, then error judgment will be carried out between each predicted value and actual value, if error is in allowed band, then it is assumed that
Prediction model completes training, otherwise return to step (4) re -training, until reaching stop condition；
(6), the prediction of PM2.5 is carried out using PM2.5 prediction models
Current K groups weather data is inputted to PM2.5 prediction models, exports T PM2.5 predicted value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711130537.XA CN107909206B (en) | 2017-11-15 | 2017-11-15 | PM2.5 prediction method based on deep structure recurrent neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711130537.XA CN107909206B (en) | 2017-11-15 | 2017-11-15 | PM2.5 prediction method based on deep structure recurrent neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107909206A true CN107909206A (en) | 2018-04-13 |
| CN107909206B CN107909206B (en) | 2021-06-04 |
Family
ID=61845600
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711130537.XA Active CN107909206B (en) | 2017-11-15 | 2017-11-15 | PM2.5 prediction method based on deep structure recurrent neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107909206B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108564326A (en) * | 2018-04-19 | 2018-09-21 | 安吉汽车物流股份有限公司 | Prediction technique and device, computer-readable medium, the logistics system of order |
| CN108957418A (en) * | 2018-05-30 | 2018-12-07 | 西安电子科技大学 | A kind of radar target identification method based on Recognition with Recurrent Neural Network model |
| CN109214592A (en) * | 2018-10-17 | 2019-01-15 | 北京工商大学 | A kind of Air Quality Forecast method of the deep learning of multi-model fusion |
| CN109242166A (en) * | 2018-08-25 | 2019-01-18 | 中科绿建（天津）科技发展有限公司 | A kind of environmental forecasting prevention and control system based on multiple dimensioned deep neural network |
| CN109447373A (en) * | 2018-11-16 | 2019-03-08 | 上海海事大学 | Haze method is predicted based on the LSTM neural network of python platform |
| CN109613178A (en) * | 2018-11-05 | 2019-04-12 | 广东奥博信息产业股份有限公司 | A kind of method and system based on recurrent neural networks prediction air pollution |
| CN109934130A (en) * | 2019-02-28 | 2019-06-25 | 中国空间技术研究院 | The in-orbit real-time fault diagnosis method of satellite failure and system based on deep learning |
| CN110059082A (en) * | 2019-04-17 | 2019-07-26 | 东南大学 | A kind of weather prediction method based on 1D-CNN and Bi-LSTM |
| CN110334382A (en) * | 2019-05-09 | 2019-10-15 | 电子科技大学 | A kind of automotive window based on Recognition with Recurrent Neural Network hazes condition predicting method |
| CN110543931A (en) * | 2019-07-26 | 2019-12-06 | 浙江工业大学 | PM2.5 concentration value prediction method based on time correlation network |
| CN110766219A (en) * | 2019-10-21 | 2020-02-07 | 成都理工大学工程技术学院 | Haze prediction method based on BP neural network |
| CN111103220A (en) * | 2019-12-31 | 2020-05-05 | 西安交通大学 | Method and system for predicting and regulating concentration of atmospheric pollutants |
| CN111292523A (en) * | 2018-12-06 | 2020-06-16 | 中国信息通信科技集团有限公司 | Network intelligent system |
| CN112990531A (en) * | 2020-12-23 | 2021-06-18 | 山西大学 | Haze prediction method based on feature-enhanced ConvLSTM |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106384166A (en) * | 2016-09-12 | 2017-02-08 | 中山大学 | Deep learning stock market prediction method combined with financial news |
| CN106599520A (en) * | 2016-12-31 | 2017-04-26 | 中国科学技术大学 | LSTM-RNN model-based air pollutant concentration forecast method |
| CN106845371A (en) * | 2016-12-31 | 2017-06-13 | 中国科学技术大学 | A kind of city road network automotive emission remote sensing monitoring system |
| CN106952161A (en) * | 2017-03-31 | 2017-07-14 | 洪志令 | A kind of recent forward prediction method of stock based on shot and long term memory depth learning network |
| CN107239859A (en) * | 2017-06-05 | 2017-10-10 | 国网山东省电力公司电力科学研究院 | The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term |
-
2017
- 2017-11-15 CN CN201711130537.XA patent/CN107909206B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106384166A (en) * | 2016-09-12 | 2017-02-08 | 中山大学 | Deep learning stock market prediction method combined with financial news |
| CN106599520A (en) * | 2016-12-31 | 2017-04-26 | 中国科学技术大学 | LSTM-RNN model-based air pollutant concentration forecast method |
| CN106845371A (en) * | 2016-12-31 | 2017-06-13 | 中国科学技术大学 | A kind of city road network automotive emission remote sensing monitoring system |
| CN106952161A (en) * | 2017-03-31 | 2017-07-14 | 洪志令 | A kind of recent forward prediction method of stock based on shot and long term memory depth learning network |
| CN107239859A (en) * | 2017-06-05 | 2017-10-10 | 国网山东省电力公司电力科学研究院 | The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term |
Non-Patent Citations (3)
| Title |
|---|
| XIAOLU LI 等: "Influence of Social-economic Activities on Air Pollutants in Beijing, China", 《OPEN GEOSCIENCES》 * |
| 艾洪福 等: "基于BP人工神经网络的雾霾天气预测研究", 《计算机仿真》 * |
| 范竣翔 等: "基于RNN的空气污染时空预报模型研究", 《测绘科学》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108564326A (en) * | 2018-04-19 | 2018-09-21 | 安吉汽车物流股份有限公司 | Prediction technique and device, computer-readable medium, the logistics system of order |
| CN108564326B (en) * | 2018-04-19 | 2021-12-21 | 安吉汽车物流股份有限公司 | Order prediction method and device, computer readable medium and logistics system |
| CN108957418A (en) * | 2018-05-30 | 2018-12-07 | 西安电子科技大学 | A kind of radar target identification method based on Recognition with Recurrent Neural Network model |
| CN109242166A (en) * | 2018-08-25 | 2019-01-18 | 中科绿建（天津）科技发展有限公司 | A kind of environmental forecasting prevention and control system based on multiple dimensioned deep neural network |
| CN109214592A (en) * | 2018-10-17 | 2019-01-15 | 北京工商大学 | A kind of Air Quality Forecast method of the deep learning of multi-model fusion |
| CN109613178A (en) * | 2018-11-05 | 2019-04-12 | 广东奥博信息产业股份有限公司 | A kind of method and system based on recurrent neural networks prediction air pollution |
| CN109447373A (en) * | 2018-11-16 | 2019-03-08 | 上海海事大学 | Haze method is predicted based on the LSTM neural network of python platform |
| CN111292523A (en) * | 2018-12-06 | 2020-06-16 | 中国信息通信科技集团有限公司 | Network intelligent system |
| CN109934130A (en) * | 2019-02-28 | 2019-06-25 | 中国空间技术研究院 | The in-orbit real-time fault diagnosis method of satellite failure and system based on deep learning |
| CN110059082A (en) * | 2019-04-17 | 2019-07-26 | 东南大学 | A kind of weather prediction method based on 1D-CNN and Bi-LSTM |
| CN110334382A (en) * | 2019-05-09 | 2019-10-15 | 电子科技大学 | A kind of automotive window based on Recognition with Recurrent Neural Network hazes condition predicting method |
| CN110543931A (en) * | 2019-07-26 | 2019-12-06 | 浙江工业大学 | PM2.5 concentration value prediction method based on time correlation network |
| CN110766219A (en) * | 2019-10-21 | 2020-02-07 | 成都理工大学工程技术学院 | Haze prediction method based on BP neural network |
| CN111103220A (en) * | 2019-12-31 | 2020-05-05 | 西安交通大学 | Method and system for predicting and regulating concentration of atmospheric pollutants |
| CN112990531A (en) * | 2020-12-23 | 2021-06-18 | 山西大学 | Haze prediction method based on feature-enhanced ConvLSTM |
| CN112990531B (en) * | 2020-12-23 | 2021-11-19 | 山西大学 | Haze prediction method based on feature-enhanced ConvLSTM |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107909206B (en) | 2021-06-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107909206A (en) | A kind of PM2.5 Forecasting Methodologies based on deep structure Recognition with Recurrent Neural Network | |
| CN107682216B (en) | A kind of network traffics protocol recognition method based on deep learning | |
| CN106548645B (en) | Vehicle route optimization method and system based on deep learning | |
| CN106779151B (en) | A kind of line of high-speed railway wind speed multi-point multi-layer coupling prediction method | |
| CN103581188B (en) | A kind of network security situation prediction method and system | |
| CN106910351A (en) | A kind of traffic signals self-adaptation control method based on deeply study | |
| CN103679611B (en) | Operation method of city comprehensive emergency intelligent simulation system based on case-based reasoning | |
| CN108009674A (en) | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks | |
| CN110794842A (en) | Reinforced learning path planning algorithm based on potential field | |
| CN107564025A (en) | A kind of power equipment infrared image semantic segmentation method based on deep neural network | |
| CN109214599B (en) | Method for predicting link of complex network | |
| CN108172301A (en) | A kind of mosquito matchmaker's epidemic Forecasting Methodology and system based on gradient boosted tree | |
| CN110473592B (en) | Multi-view human synthetic lethal gene prediction method | |
| CN102622515B (en) | A kind of weather prediction method | |
| CN104217258B (en) | A kind of electric load sigma-t Forecasting Methodology | |
| CN103105246A (en) | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm | |
| CN104636801A (en) | Transmission line audible noise prediction method based on BP neural network optimization | |
| CN110164129B (en) | Single-intersection multi-lane traffic flow prediction method based on GERNN | |
| CN107563122A (en) | The method of crime prediction of Recognition with Recurrent Neural Network is locally connected based on interleaving time sequence | |
| CN106991666A (en) | A kind of disease geo-radar image recognition methods suitable for many size pictorial informations | |
| CN106408120B (en) | Local area landslide prediction device and method | |
| CN115951014A (en) | CNN-LSTM-BP multi-mode air pollutant prediction method combining meteorological features | |
| CN106650933A (en) | Deep neural network optimizing method based on coevolution and back propagation | |
| CN104424507A (en) | Prediction method and prediction device of echo state network | |
| CN102564496A (en) | Micro-analysis method for transformer device based on BP nerve network and manual shoal |
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 |