CN106056210B - A kind of PM2.5 concentration value Forecasting Methodologies based on hybrid neural networks - Google Patents
A kind of PM2.5 concentration value Forecasting Methodologies based on hybrid neural networks Download PDFInfo
- Publication number
- CN106056210B CN106056210B CN201610401881.7A CN201610401881A CN106056210B CN 106056210 B CN106056210 B CN 106056210B CN 201610401881 A CN201610401881 A CN 201610401881A CN 106056210 B CN106056210 B CN 106056210B
- Authority
- CN
- China
- Prior art keywords
- network
- data
- training
- concentration values
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
A kind of PM2.5 concentration value Forecasting Methodologies based on hybrid neural networks, include the following steps:Step 1, the acquisition of four class sample datas.The four classes sample data includes PM2.5 concentration values historical data, PM2.5 concentration value indexs of correlation historical data, meteorological historical data and PM2.5 component resolving data;Step 2, using first nerves network initial forecast PM2.5 concentration values;Step 3, using nervus opticus network re prediction PM2.5 concentration values;Step 4 finally predicts PM2.5 concentration values using third nerve network, exports the final predicted value of PM2.5 concentration values.The present invention is outside the data using PM2.5 concentration values historical data, PM2.5 concentration value index of correlation historical datas and meteorological these three classifications of historical data, PM2.5 component resolving data are also introduced, the Change and Development rule of energy accurate description PM2.5 concentration values improves precision of prediction.
Description
Technical field
Electric powder prediction more particularly to one kind the present invention relates to air particle PM2.5 concentration values is based on mixing god
The Forecasting Methodology of PM2.5 concentration values through network.
Background technology
Nowadays air pollution has become focus of concern.In densely populated city, air pollution is tight
People's health and life have been influenced again.In air pollution regulations, (diameter is less than or equal to 2.5 microns of particle to PM2.5
Object) concentration value have become weigh air quality significant Testing index.It is dense to future time section PM2.5 according to historical data
The prediction of angle value has become studying a question with stronger academic significance and application value.
To solve the above-mentioned problems, Shi Xuhua et al. is in patent《A kind of regional air PM2.5 concentration prediction methods》In, lead to
It crosses and establishes the concentration value prediction that Support vector regression model carries out PM2.5.Wang Shuqiang et al. is in patent《A kind of air quality
PM2.5 Forecasting Methodologies and system》In, it is predicted by establishing the concentration value that tensor regression model is supported to carry out PM2.5.Ma Tiancheng etc.
People is in paper《Fuzzy neural network PM2.5 concentration predictions based on modified PSO》In, using a kind of modified PSO optimizations
Fuzzy neural network predicts the concentration value of PM2.5.Zhang Yiwen et al. is in paper《PM2.5 prediction models based on neutral net》
In, by the prediction for establishing neural fusion PM2.5 concentration values.Jing Tao et al. is in paper《T controlled distribution genetic algorithms are excellent
Change the PM2.5 mass concentrations prediction of BP neural network》In, by establishing based on t controlled distribution genetic algorithm optimization BP nerve nets
Network model predicts PM2.5 mass concentrations.Sun Rong bases et al. are in paper《It is a kind of based on BP neural network innovatory algorithm
PM2.5 Forecasting Methodologies》In, by combining principal component analysis, terminating the concentration of coaching method and BP neural network mould to PM2.5 in advance
Value is predicted.Chen Qiang et al. is in paper《Zhengzhou City PM2.5 concentration spatial-temporal distribution characteristic and prediction model research》In, by building
Vertical BP neural network predicts the concentration value of Zhengzhou City locality PM2.5.
Through document investigation and analysis, the PM2.5 concentration value Forecasting Methodologies having proposed at present are required to substantial amounts of historical data and make
For system input sample.These historical datas can be divided into three classifications.Primary sources are PM2.5 concentration value historical datas.
Secondary sources are PM2.5 concentration value indexs of correlation historical data (such as AQI, PM10, NO2、CO、SO2、O3).Three classes data
It is meteorological historical data (such as temperature, relative humidity, air pressure, wind speed, precipitation etc.).It is existing but through literature survey
PM2.5 concentration value Forecasting Methodologies are not accounted for using the 4th class data, i.e., PM2.5 component resolvings data (for example possess by motor vehicle
Amount, industrial gas emission amount, power consumption).For different cities or area, PM2.5 Crack causes have larger difference.Such as
Three classes data before fruit only considers, then designed PM2.5 concentration value forecasting systems are difficult accurate simulation locality PM2.5 concentration values
Change and Development rule.
The content of the invention
In order to overcome existing PM2.5 concentration values prediction mode that can not describe the Change and Development rule of PM2.5 concentration values, prediction
The relatively low deficiency of precision, the present invention using PM2.5 concentration values historical data, PM2.5 concentration value index of correlation historical datas and
Outside the data of meteorological these three classifications of historical data, PM2.5 component resolving data are also introduced, a kind of accurate description is provided
The Change and Development rule of PM2.5 concentration values, the PM2.5 concentration values prediction side based on hybrid neural networks for improving precision of prediction
Method.
The technical solution adopted by the present invention to solve the technical problems is:
A kind of PM2.5 concentration value Forecasting Methodologies based on hybrid neural networks, described method includes following steps:
Step 1, the acquisition of four class sample datas.The four classes sample data includes PM2.5 concentration values historical data, PM2.5
Concentration value index of correlation historical data, meteorological historical data and PM2.5 component resolving data.
Step 2, using first nerves network initial forecast PM2.5 concentration values, process is as follows:
Step 2.1 creates four layers of neutral net for including input layer, hidden layer, articulamentum and output layer, and setting is hidden
Number of nodes containing layer and output layer.The node number of the hidden layer provides estimate, the empirical equation using empirical equation
It is as follows:
In above formula, m and n are respectively the neuron number of input layer and output layer, and a is the constant between [0,10].
Step 2.2, training function, contiguous function and the output function for setting hidden layer, articulamentum and output layer respectively, if
Determine anticipation error minimum value, maximum iteration and the learning rate of network.
Step 2.3, using the PM2.5 concentration values historical data as first nerves network inputs data, and by first god
It is divided into first nerves network training data and first nerves network testing data through network inputs data, by the first nerves net
Network training data normalized, normalization formula are as follows:
In above formula, x treats normalized data, x to be describedminAnd xmaxMinimum value in respectively described new sample data
And maximum, y are the data after normalization, are distributed in [0.1,0.9] section.
The first nerves network training data are input in the first nerves network created training the by step 2.4
One neutral net, calculation error, and according to error transfer factor first nerves network weight.
Step 2.5 judges whether the first nerves network restrains, when error is less than anticipation error minimum value, algorithm
Convergence.Terminate algorithm when reaching maximum iteration, the first nerves network training is completed.
The first nerves network testing data is inputted the trained first nerves network by step 2.6, is obtained
The initial predicted value of PM2.5 concentration values.
Step 2.7, by the initial predicted value of the PM2.5 concentration values and the PM2.5 concentration value index of correlation history
Data and meteorological historical data are incorporated as nervus opticus network inputs data.
Step 3, using nervus opticus network re prediction PM2.5 concentration values, process is as follows:
Step 3.1, create a three-layer neural network for include input layer, hidden layer and output layer, setting hidden layer with
The node number of output layer.
The training function of step 3.2, the setting nervus opticus network, anticipation error minimum value, the maximum of setting network
Iterations and learning rate.
The nervus opticus network inputs data are divided into nervus opticus network training data and nervus opticus by step 3.3
The nervus opticus network training data are normalized network testing data.
Nervus opticus network training data after the normalization are input to the nervus opticus net created by step 3.4
Network, and training nervus opticus network, calculation error, and according to error transfer factor nervus opticus network weight.
Step 3.5 judges whether the nervus opticus network restrains, when error is less than anticipation error minimum value, algorithm
Convergence.Terminate algorithm when reaching maximum iteration, the nervus opticus network training is completed.
The nervus opticus network testing data is input in the nervus opticus network that the training is completed by step 3.6,
The re prediction value of PM2.5 concentration values is exported, by the re prediction value of the PM2.5 concentration values and PM2.5 component resolving data
It is incorporated as third nerve network inputs data.
Step 4 finally predicts PM2.5 concentration values using third nerve network, and process is as follows:
Step 4.1, create a three-layer neural network for include input layer, hidden layer and output layer, setting hidden layer with
The node number of output layer.
The training function of step 4.2, the setting third nerve network, anticipation error minimum value, the maximum of setting network
Iterations and learning rate.
The third nerve network data is divided into third nerve network training data and third nerve network by step 4.3
The third nerve network training data are normalized test data.
Third nerve network training data after the normalization are input to the third nerve net created by step 4.4
Network, and training third nerve network, calculation error, and according to error transfer factor third nerve network weight.
Step 4.5 judges whether the third nerve network restrains, when error is less than anticipation error minimum value, algorithm
Convergence.Terminate algorithm when reaching maximum iteration, the third nerve network training is completed.
The third nerve network testing data is input in the third nerve network that the training is completed by step 4.6,
Export the final predicted value of PM2.5 concentration values.
Further, in the step 1, the PM2.5 concentration values index of correlation historical data includes AQI, and (air quality refers to
Number), PM10 (Particulate Matter 10), SO2(sulfur dioxide), CO (carbon monoxide), CO2(carbon dioxide), O3It is (smelly
Oxygen), the meteorology historical data includes temperature on average, dew point, relative humidity, pressure, wind speed, precipitation, the PM2.5 ingredients
Parsing data includes vehicle guaranteeding organic quantity, industrial gas emission amount and power consumption etc..
The present invention technical concept be:PM2.5 concentration value historical datas are being used, PM2.5 concentration value indexs of correlation are gone through
History data (AQI, PM10, NO2、CO、SO2、O3) and meteorological historical data (temperature, relative humidity, air pressure, wind speed, precipitation etc.)
Outside this three classes data, emphasis introduces PM2.5 component resolvings data (vehicle guaranteeding organic quantity, industrial gas emission amount, electricity consumption
Amount) this 4th class data carries out the prediction of PM2.5 concentration values.Further, this four classes data is passed through into three neutral nets point
Stage forecast PM2.5 concentration values.
Beneficial effects of the present invention are mainly manifested in:Technical scheme can accurately simulation locality PM2.5 it is dense
The changing rule of angle value, effective precision of prediction that must improve current PM2.5 concentration values realize that " localization " and " compartmentalization " is accurate
Prediction.
Description of the drawings
Fig. 1 is a kind of PM2.5 concentration value Forecasting Methodology schematic diagrames based on hybrid neural networks.
Fig. 2 is the training flow chart of ELman neutral nets.
Fig. 3 is the training flow chart of BP neural network.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings.
With reference to Fig. 1~Fig. 3, a kind of PM2.5 concentration value Forecasting Methodologies based on hybrid neural networks, the described method includes such as
Lower step:
Step 1, the acquisition of four class sample datas.The four classes sample data includes PM2.5 (Particulate Matter
2.5) concentration value historical data, PM2.5 concentration value indexs of correlation historical data, meteorological historical data and PM2.5 component resolving numbers
According to.Further, the PM2.5 concentration values index of correlation historical data includes AQI (air quality index), PM10
(Particulate Matter 10)、SO2(sulfur dioxide), CO (carbon monoxide), CO2(carbon dioxide), O3(ozone), it is described
Meteorological historical data includes temperature on average, dew point, relative humidity, pressure, wind speed, precipitation, the PM2.5 component resolvings data
Including vehicle guaranteeding organic quantity, industrial gas emission amount and power consumption.
Four class data of present invention acquisition Hangzhou.The Hangzhou AQI of 2015 (air quality index), PM2.5
(Particulate Matter 2.5)、PM10(Particulate Matter 10)、SO2(sulfur dioxide), CO (an oxidation
Carbon), CO2(carbon dioxide), O3(ozone) is collected in Chinese air quality on-line monitoring analysis platform website, Hangzhou 2015
Temperature on average, dew point, relative humidity, pressure, wind speed, precipitation WEATHER UNDERGROUND websites collect, Hangzhou
Vehicle guaranteeding organic quantity, industrial gas emission amount and the power consumption of 2014 is collected in Hangzhou Information Statistics net website.
In the present invention, the four class sample datas such as table 1 that is collected into:
Table 1
Step 2, the first Elman neutral net initial forecast PM2.5 concentration values, process are as follows:
Step 2.1 creates four layers of neutral net for including input layer, hidden layer, articulamentum and output layer, and setting is hidden
Number of nodes containing layer and output layer is respectively 6 and 1.The node number of the hidden layer provides estimate using empirical equation, institute
It is as follows to state empirical equation:
In formula:M and n is respectively the neuron number of input layer and output layer, and a is the constant between [0,10];
Step 2.2, training function, contiguous function and the output function for setting hidden layer, articulamentum and output layer respectively, institute
Trained function is stated as Sigmoid functions, articulamentum and output layer functions are purelin functions.The anticipation error of setting network is most
Small value is 10-5, maximum iteration 104It is 0.1 with learning rate;
The PM2.5 concentration values historical data is divided into first by step 2.3 as the first Elman Neural Network Datas
Elman neural metwork trainings data and the first Elman neutral net test data two parts, by the first Elman nerve nets
Network training data normalized, normalization formula are as follows:
In formula:X treats normalized data, x to be describedminAnd xmaxMinimum value in respectively described new sample data and
Maximum, y are the data after normalization, are distributed in [0.1,0.9] section;
The first Elman neural metwork training data are input to the first Elman nerve nets created by step 2.4
The first Elman neutral nets of training, calculation error, according to error transfer factor weights in network;
Step 2.5 judges whether the first Elman neutral nets restrain, when error is less than anticipation error minimum value,
Algorithmic statement;Or terminating algorithm when reaching maximum iteration, the first Elman neural metwork trainings are completed;
The first Elman neutral nets test data is inputted the trained first Elman nerves by step 2.6
Network obtains the predicted value of preliminary PM2.5 concentration values;
Step 2.7, by the predicted value of the preliminary PM2.5 concentration values and the PM2.5 concentration value index of correlation history
Data and meteorological historical data are incorporated as the second BP neural network data;
Step 3, the second BP neural network re prediction PM2.5 concentration values, process are as follows:
Step 3.1 creates a three-layer neural network for including input layer, hidden layer and output layer, and setting is implicit respectively
The node number of layer and output layer is 12 and 1.
Step 3.2 sets the training function of second BP neural network as Sigmoid functions, the expectation of setting network
Error minimum value is 10-5, maximum iteration 104It is 0.1 with learning rate;
The second BP neural network data are divided into the second BP neural network training data and the 2nd BP god by step 3.3
Through network testing data, the second BP neural network training data is normalized;
Step 3.4 will tell that the second BP neural network training data after normalizing is input to the 2nd BP god created
Through the second BP neural network of training, calculation error, according to error transfer factor weights in network;
Step 3.5 judges whether second BP neural network restrains, and when error is less than anticipation error minimum value, calculates
Method restrains;Or terminating algorithm when reaching maximum iteration, the second BP neural network training is completed;
The second BP neural network test data is input to the 2nd BP nerve nets that the training is completed by step 3.6
In network, the re prediction value of PM2.5 concentration values is exported, by the re prediction value of the PM2.5 concentration values and PM2.5 component resolvings
Data are incorporated as the 3rd BP neural network data;
Step 4, the 3rd BP neural network finally predict PM2.5 concentration values, and process is as follows:
Step 4.1 creates a three-layer neural network for including input layer, hidden layer and output layer, and setting is implicit respectively
The node number of layer and output layer is 4 and 1.
Step 4.2 sets the training function of the 3rd BP neural network as Sigmoid functions, the expectation of setting network
Error minimum value is 10-5, maximum iteration 104It is 0.1 with learning rate;
The 3rd BP neural network data are divided into the 3rd BP neural network training data and the 3rd BP god by step 4.3
Through network testing data, the 3rd BP neural network training data is normalized;
Step 4.4 will tell that the 3rd BP neural network training data after normalizing is input to the 3rd BP god created
Through the 3rd BP neural network of training, calculation error, according to error transfer factor weights in network;
Step 4.5 judges whether the 3rd BP neural network restrains, and when error is less than anticipation error minimum value, calculates
Method restrains;Or terminating algorithm when reaching maximum iteration, the 3rd BP neural network training is completed;
The third nerve network testing data is input to the 3rd BP neural network that the training is completed by step 4.6
In, the final predicted value of output PM2.5 concentration values.
The predicted value of PM2.5 concentration values is obtained according to above operation, such as table 2:
Table 2.
Claims (2)
1. a kind of PM2.5 concentration value Forecasting Methodologies based on hybrid neural networks, it is characterised in that:The described method includes following steps
Suddenly:
Step 1, the acquisition of four class sample datas, the four classes sample data include PM2.5 concentration values historical data, PM2.5 concentration
It is worth index of correlation historical data, meteorological historical data and PM2.5 component resolving data;
Step 2, using first nerves network initial forecast PM2.5 concentration values, process is as follows:
Step 2.1 creates four layers of neutral net for including input layer, hidden layer, articulamentum and output layer, sets hidden layer
With the number of nodes of output layer, the node number of the hidden layer provides estimate M using empirical equation, and the empirical equation is such as
Under:
<mrow>
<mi>M</mi>
<mo>=</mo>
<msqrt>
<mrow>
<mi>n</mi>
<mo>+</mo>
<mi>m</mi>
</mrow>
</msqrt>
<mo>+</mo>
<mi>a</mi>
</mrow>
In above formula, m and n are respectively the neuron number of input layer and output layer, and a is the constant between [0,10];
Step 2.2, training function, contiguous function and the output function for setting hidden layer, articulamentum and output layer respectively, set net
Anticipation error minimum value, maximum iteration and the learning rate of network;
Step 2.3, using the PM2.5 concentration values historical data as first nerves network inputs data, and by first nerves net
Network input data is divided into first nerves network training data and first nerves network testing data, and the first nerves network is instructed
Practice data normalization processing, normalization formula is as follows:
<mrow>
<mi>y</mi>
<mo>=</mo>
<mn>0.1</mn>
<mo>+</mo>
<mfrac>
<mrow>
<mn>0.8</mn>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>-</mo>
<msub>
<mi>x</mi>
<mi>min</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
</msub>
<mo>-</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
</mfrac>
</mrow>
In above formula, x is to treat normalized data, xminAnd xmaxMinimum value and maximum in respectively new sample data, y are
Data after normalization are distributed in [0.1,0.9] section;
The first nerves network training data are input to the first god of training in the first nerves network created by step 2.4
Through network, calculation error, and according to error transfer factor first nerves network weight;
Step 2.5 judges whether the first nerves network restrains, when error is less than anticipation error minimum value, algorithmic statement;
Terminate algorithm when reaching maximum iteration, the first nerves network training is completed;
The first nerves network testing data is inputted the trained first nerves network by step 2.6, obtains PM2.5
The initial predicted value of concentration value;
Step 2.7, by the initial predicted value of the PM2.5 concentration values and the PM2.5 concentration value index of correlation historical datas
Nervus opticus network inputs data are incorporated as with meteorological historical data;
Step 3, using nervus opticus network re prediction PM2.5 concentration values, process is as follows:
Step 3.1 creates a three-layer neural network for including input layer, hidden layer and output layer, sets hidden layer and output
The node number of layer;
The training function of step 3.2, the setting nervus opticus network, anticipation error minimum value, the greatest iteration of setting network
Number and learning rate;
The nervus opticus network inputs data are divided into nervus opticus network training data and nervus opticus network by step 3.3
The nervus opticus network training data are normalized test data;
Nervus opticus network training data after the normalization are input to the nervus opticus network created by step 3.4, and
And training nervus opticus network, calculation error, and according to error transfer factor nervus opticus network weight;
Step 3.5 judges whether the nervus opticus network restrains, when error is less than anticipation error minimum value, algorithmic statement;
Terminate algorithm when reaching maximum iteration, the nervus opticus network training is completed;
The nervus opticus network testing data is input in the nervus opticus network that the training is completed, output by step 3.6
The re prediction value of PM2.5 concentration values merges the re prediction value of the PM2.5 concentration values and PM2.5 component resolvings data
As third nerve network inputs data;
Step 4 finally predicts PM2.5 concentration values using third nerve network, and process is as follows:
Step 4.1 creates a three-layer neural network for including input layer, hidden layer and output layer, sets hidden layer and output
The node number of layer;
The training function of step 4.2, the setting third nerve network, anticipation error minimum value, the greatest iteration of setting network
Number and learning rate;
The third nerve network data is divided into third nerve network training data and third nerve network test by step 4.3
The third nerve network training data are normalized data;
Third nerve network training data after the normalization are input to the third nerve network created by step 4.4, and
Training third nerve network, calculation error, and according to error transfer factor third nerve network weight;
Step 4.5 judges whether the third nerve network restrains, when error is less than anticipation error minimum value, algorithmic statement;
Terminate algorithm when reaching maximum iteration, the third nerve network training is completed;
The third nerve network testing data is input in the third nerve network that the training is completed, output by step 4.6
The final predicted value of PM2.5 concentration values.
2. a kind of PM2.5 concentration value Forecasting Methodologies based on hybrid neural networks as described in claim 1, it is characterised in that:
In the step 1, the PM2.5 concentration values index of correlation historical data includes air quality index AQI, PM10, SO2、CO、CO2
And O3, it is described meteorology historical data include temperature on average, dew point, relative humidity, pressure, wind speed and precipitation, the PM2.5 into
Decomposing analysis data includes vehicle guaranteeding organic quantity, industrial gas emission amount and power consumption.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610401881.7A CN106056210B (en) | 2016-06-07 | 2016-06-07 | A kind of PM2.5 concentration value Forecasting Methodologies based on hybrid neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610401881.7A CN106056210B (en) | 2016-06-07 | 2016-06-07 | A kind of PM2.5 concentration value Forecasting Methodologies based on hybrid neural networks |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106056210A CN106056210A (en) | 2016-10-26 |
CN106056210B true CN106056210B (en) | 2018-06-01 |
Family
ID=57171107
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610401881.7A Active CN106056210B (en) | 2016-06-07 | 2016-06-07 | A kind of PM2.5 concentration value Forecasting Methodologies based on hybrid neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106056210B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
FR3103605A1 (en) * | 2019-11-22 | 2021-05-28 | Elichens | Method for estimating the concentration of analyte in the air in the vicinity of a path traveled by means of transport |
Families Citing this family (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10599977B2 (en) | 2016-08-23 | 2020-03-24 | International Business Machines Corporation | Cascaded neural networks using test ouput from the first neural network to train the second neural network |
CN106529081A (en) * | 2016-12-03 | 2017-03-22 | 安徽新华学院 | PM2.5 real-time level prediction method and system based on neural net |
CN106529746A (en) * | 2016-12-29 | 2017-03-22 | 南京恩瑞特实业有限公司 | Method for dynamically fusing, counting and forecasting air quality based on dynamic and thermal factors |
CN106599520B (en) * | 2016-12-31 | 2018-08-03 | 中国科学技术大学 | A kind of air pollutant concentration forecasting procedure based on LSTM-RNN models |
CN106611090B (en) * | 2016-12-31 | 2018-04-10 | 中国科学技术大学 | A kind of road side air pollutant concentration Forecasting Methodology based on reconstruct deep learning |
CN106650825B (en) * | 2016-12-31 | 2020-05-12 | 中国科学技术大学 | Motor vehicle exhaust emission data fusion system |
CN107192690B (en) * | 2017-05-19 | 2019-04-23 | 重庆大学 | Near infrared spectrum Noninvasive Blood Glucose Detection Methods and its detection network model training method |
US10713783B2 (en) | 2017-06-01 | 2020-07-14 | International Business Machines Corporation | Neural network classification |
US11138724B2 (en) | 2017-06-01 | 2021-10-05 | International Business Machines Corporation | Neural network classification |
CN107368894B (en) * | 2017-07-28 | 2020-11-10 | 国网河南省电力公司电力科学研究院 | Atmospheric pollution prevention and control electricity consumption data analysis platform based on big data sharing |
CN107545122A (en) * | 2017-09-27 | 2018-01-05 | 重庆长安汽车股份有限公司 | A kind of simulation system of the vehicle gaseous effluent based on neutral net |
CN108399470B (en) * | 2018-02-09 | 2021-10-08 | 苏州科技大学 | Indoor PM2.5 prediction method based on multi-example genetic neural network |
CN108426812B (en) * | 2018-04-08 | 2020-07-31 | 浙江工业大学 | PM2.5 concentration value prediction method based on memory neural network |
TWI662422B (en) * | 2018-04-23 | 2019-06-11 | 國家中山科學研究院 | Air quality prediction method based on machine learning model |
CN109376903B (en) * | 2018-09-10 | 2021-12-17 | 浙江工业大学 | PM2.5 concentration value prediction method based on game neural network |
CN109670646B (en) * | 2018-12-21 | 2022-04-08 | 浙江工业大学 | PM2.5 concentration value prediction method based on mixed threshold neural network |
CN109916788B (en) * | 2019-01-14 | 2020-05-19 | 南京大学 | PM pair for distinguishing emission change and meteorological condition change of different areas2.5Method for influencing concentration |
CN110046771B (en) * | 2019-04-25 | 2021-04-16 | 河南工业大学 | PM2.5 concentration prediction method and device |
CN112529344B (en) * | 2019-09-18 | 2023-09-05 | 中国科学院沈阳计算技术研究所有限公司 | Algorithm for optimizing air quality value based on Elman neural network |
CN110796284B (en) * | 2019-09-20 | 2022-05-17 | 平安科技(深圳)有限公司 | Method and device for predicting pollution level of fine particulate matters and computer equipment |
CN111174824B (en) * | 2019-12-27 | 2022-04-12 | 北京首钢自动化信息技术有限公司 | Control platform that acid mist discharged |
CN113281229B (en) * | 2021-02-09 | 2022-11-29 | 北京工业大学 | Multi-model self-adaptive atmosphere PM based on small samples 2.5 Concentration prediction method |
CN113011660A (en) * | 2021-03-23 | 2021-06-22 | 上海应用技术大学 | Air quality prediction method, system and storage medium |
CN114814092A (en) * | 2022-04-12 | 2022-07-29 | 上海应用技术大学 | IP index measuring method based on BP neural network |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104535465A (en) * | 2015-01-07 | 2015-04-22 | 东北大学 | PM2.5 concentration detection method and device based on neural network |
CN104729965A (en) * | 2015-01-28 | 2015-06-24 | 东北大学 | PM2.5 concentration detection method based on interzone radial basis function nerve network |
-
2016
- 2016-06-07 CN CN201610401881.7A patent/CN106056210B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104535465A (en) * | 2015-01-07 | 2015-04-22 | 东北大学 | PM2.5 concentration detection method and device based on neural network |
CN104729965A (en) * | 2015-01-28 | 2015-06-24 | 东北大学 | PM2.5 concentration detection method based on interzone radial basis function nerve network |
Non-Patent Citations (1)
Title |
---|
《基于混合神经网络的臭氧浓度软测量》;刘敏等;《计算机测量与控制》;20031130;第660-662页 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
FR3103605A1 (en) * | 2019-11-22 | 2021-05-28 | Elichens | Method for estimating the concentration of analyte in the air in the vicinity of a path traveled by means of transport |
Also Published As
Publication number | Publication date |
---|---|
CN106056210A (en) | 2016-10-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106056210B (en) | A kind of PM2.5 concentration value Forecasting Methodologies based on hybrid neural networks | |
CN108426812A (en) | A kind of PM2.5 concentration value prediction techniques based on Memory Neural Networks | |
Drewil et al. | Air pollution prediction using LSTM deep learning and metaheuristics algorithms | |
Xu et al. | Research and application of a hybrid model based on dynamic fuzzy synthetic evaluation for establishing air quality forecasting and early warning system: A case study in China | |
CN106611090B (en) | A kind of road side air pollutant concentration Forecasting Methodology based on reconstruct deep learning | |
Kang et al. | Application of BP neural network optimized by genetic simulated annealing algorithm to prediction of air quality index in Lanzhou | |
CN106651100B (en) | Air quality evaluation system and method based on vehicle-mounted monitoring points optimized through Internet of vehicles | |
Becerra-Rico et al. | Airborne particle pollution predictive model using Gated Recurrent Unit (GRU) deep neural networks | |
CN108268935B (en) | PM2.5 concentration value prediction method and system based on time sequence recurrent neural network | |
CN106765959A (en) | Heat-air conditioner energy-saving control method based on genetic algorithm and depth B P neural network algorithms | |
Gadhavi et al. | Student final grade prediction based on linear regression | |
CN110533248A (en) | The Predict Model of Air Pollutant Density of fusion machine learning and LSTM | |
CN109118020A (en) | A kind of subway station energy consumption short term prediction method and its forecasting system | |
CN109242136A (en) | A kind of micro-capacitance sensor wind power Chaos-Genetic-BP neural network prediction technique | |
Al_Janabi et al. | Pragmatic method based on intelligent big data analytics to prediction air pollution | |
Li et al. | Research on application of a hybrid heuristic algorithm in transportation carbon emission | |
CN111077048A (en) | Opportunistic group intelligent air quality monitoring and evaluating method based on mobile equipment | |
CN108364098B (en) | Method for measuring influence of weather characteristics on user sign-in | |
CN104750953A (en) | Method for simulating middle-and-small-scale airborne substance atmosphere transportation ensemble diffusion | |
CN109887290A (en) | Traffic flow forecasting method based on equilibrium index exponential smoothing and stack self-encoding encoder | |
Lu et al. | A deep belief network based model for urban haze prediction | |
Pang et al. | Study on forecasting the parameters of gas environment of metro station | |
CN112949938B (en) | Wind power climbing event direct forecasting method for improving training sample class imbalance | |
CN106845756A (en) | A kind of large-sized photovoltaic inverter impedance Identification Errors Index Assessment method | |
Zhang et al. | Diagnosis of contamination discharge state of porcelain insulators based on GA-CNN |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |