CN106529081A - PM2.5 real-time level prediction method and system based on neural net - Google Patents
PM2.5 real-time level prediction method and system based on neural net Download PDFInfo
- Publication number
- CN106529081A CN106529081A CN201611097948.9A CN201611097948A CN106529081A CN 106529081 A CN106529081 A CN 106529081A CN 201611097948 A CN201611097948 A CN 201611097948A CN 106529081 A CN106529081 A CN 106529081A
- Authority
- CN
- China
- Prior art keywords
- index
- prime
- pma
- real
- matrix
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The invention discloses a PM2.5 real-time level prediction method and system based on neural net. The method collects the historical concentration values of air contaminant indexes of PM2.5, O3, CO, PM10, SO2 and NO2 as well as the historical values of atmosphere temperature, moisture and wind force and the like, applies the historical data as the training set to train the neural net model, and build a neural net prediction model based on the neural net comprehensive atmospheric indexes. A mobile equipment terminal sends PM2.5 level real-time request to a server, substituting the real-time-acquired contaminant indexes and atmospheric indexes as the test data into the neural net prediction model for prediction and pushing. The method provides a PM2.5 level query option to the mobile terminal users in the cities with few PM2.5 monitoring points or without PM2.5 monitoring point, reduces the prediction cost of PM2.5, and at the same time conducts the real-time prediction accurate to day and time, thus having good universality.
Description
Technical field
The present invention relates to environmental pollution prediction field, the real-time grade of the more particularly to a kind of PM2.5 based on neutral net is pre-
Survey method and system.
Background technology
PM is the acronym of English particulate matter (particulate matter).During PM2.5 refers to air, kinetics are worked as
Particulate matter of the amount diameter less than or equal to 2.5 microns.Lives of the PM2.5 to the pollution of environment to people generates huge shadow
Ring.
The computational methods of PM2.5 mainly adopt physical method, and PM2.5 monitoring costs are higher, therefore at present in China, PM2.5
Observation station is less, and most cities do not have observation station.At present, the research to PM2.5 forecast analysis is more, and Forecasting Methodology is adopted mostly
Use linear method.Gene expression, Logistic regression models are such as based on, Linear Model for Prediction method, Ke Yiyou is mainly adopted
The concentration value of effect prediction PM2.5.But by analysis, affect the pollutant and atmospheric factor of PM2.5 that often there is nonlinear characteristic,
Linear model can not be simulated well.
Also research worker is predicted to PM2.5 using neutral net.As neutral net is a kind of statistical models,
With preferable generalization ability, can preferably simulating pollution thing and atmospheric factor change procedure, so many research worker choosings
Select to be simulated with neutral net and predict there is certain progress, but as prediction index is less, predictablity rate is relatively low.
Consideration is influential on PM2.5 to be not only pollutant, and atmospheric factor is possible to also to produce impact to PM2.5, has
Research worker may produce the atmospheric factor for affecting to PM2.5 to other and analyze, and analyze different index factors pre- to PM2.5
The impact of survey, but lack the algorithm model to specific PM2.5 grade forecasts, and the application of service is provided to mobile end subscriber.
Meanwhile, city that is less to monitoring point or not having monitoring point, it is impossible to provide real-time query service to mobile end subscriber.
The content of the invention
The technical problem to be solved is to provide a kind of real-time grade forecast sides of PM2.5 based on neutral net
Method, extraction cost are relatively low, be easier to the pollutant and the atmospheric factor that obtain as aggregative indicator daily or hour sets up prediction mould
Type.Meanwhile, real-time estimate service can be provided for mobile terminal user.The method is moved easily end subscriber and carries out PM2.5 to be looked in real time
Ask, while reducing forecast cost, establish daily general with hour PM2.5 forecast models.
For achieving the above object, the present invention provides following technical scheme:A kind of PM2.5 based on neutral net is in real time etc.
Level Forecasting Methodology, comprises the steps:
(1) offline history index PM2.5, O3 of pollutant in collection air, the concentration value of CO, PM10, SO2, NO2, structure
Build pollutant coefficient matrix PM:
Wherein, pollutant coefficient matrix PM first lists are shown as:To count all PM2.5 of natural law
Value, divided according to 1 grade of table;Remaining list is shown as:Si=(S1i…Smi)T, i=1 ... 5, be O3, CO, PM10,
SO2, NO2 concentration value;Wherein m is statistics natural law or hourage;
(2) gather air themperature, humidity, the offline history index of three air of wind-force, and with air in pollutant index
Combine, provide coefficient of colligation matrix PMA:
Coefficient of colligation matrix PMA and pollutant coefficient matrix PM differences are to add environmental coefficient matrix Q, QmiMatrix
It is the new temperature for adding, humidity, three air index matrixs of wind-force, wherein m is statistics natural law, i=1 ... 3, represents temperature, wet
Degree, three indexs of wind-force;
(3) using the method for Pearson's correlation coefficient, the O3 concentration values in coefficient of colligation matrix PMA, CO concentration values,
PM10 concentration values, SO2 concentration values, NO2 concentration values and air themperature index, humidity index, wind-force index substitute into pearson respectively
Between correlation coefficient parameter shown in the computational methods formula (3) of similarity:
Wherein sim (X, Y) represents the similarity between X, Y two indices, and i represents i-th sample of certain index, and n is sample
This sum;
(4) using the method for Pearson's correlation coefficient, data in PMA are substituted into formula (3), the index in PMA is carried out
Merge, the aggregative indicator matrix after merging is PMA '
Wherein,To count the value of all PM2.5 of natural law, divided according to 1 grade of table;Si=
(S1i…Smi)T, i≤5 are the concentration values after O3, CO, PM10, SO2, NO2 enter row index merging;Qi=(Q1i…Qmi)T, i≤3,
Be temperature, humidity, wind-force index merge after value;Wherein m is statistics natural law or hourage.
(5) make training set training neutral net with off-line data, set up the PM2.5 forecast models based on neutral net.Will
Select PM, PMA, PMA ' as the input layer data set of neutral net, it is predicted by BP neural network model:Wherein input layer is refreshing
The first data transfers by input of Jing are to hidden layer, then output layer are delivered to after amplified by data activation and by output layer
Output;Wherein, the excitation function between input layer and hidden layer adopts S function, such as shown in formula (5):
When actual error exceeds anticipation error, error amount changes the connection weight between each neuron along network back propagation
Value and threshold value, repetition training network, until meet anticipation error, it is determined that the mapping relations between input and output.In the present invention
Xuan Ze PM, PMA, PMA ' matrix is input data, PM2.5 grade forecasts value carries out experiment mapping for output;
(6) send the request of PM2.5 real-time queries from mobile terminal to server, user may be selected inquiry in real time " my god " it is or real
When " hour " inquired about;
(7) server obtains mobile terminal current time;
(8) according to the mobile terminal current time obtained in the index system and step (7) after merging in step (4), in real time
Collection pollutant index and air index;
(9) data in step (8) are substituted into into step (5), predicts real-time PM2.5 values;
(10) real time propelling movement will be predicted the outcome to mobile end subscriber.
The present invention gives a kind of real-time grade forecast systems of PM2.5 based on neutral net, including:
Data acquisition module, collects pollutant index and air index, and described pollutant index includes O3 concentration values, CO
Concentration value, PM10 concentration values, SO2 concentration values, NO2 concentration values, described air index include air themperature, humidity and wind-force;
Data processing module, builds pollutant coefficient matrix PM and coefficient of colligation square based on pollutant index and air index
Battle array PMA, then the calculating of similarity between pearson correlation coefficient indexs is utilized respectively to pollutant index and air index, carry out
Index merges, and after merging, obtains aggregative indicator coefficient matrix PMA ';
Models fitting module, will select pollutant coefficient matrix PM, coefficient of colligation matrix PMA, aggregative indicator coefficient matrix
PMA ', respectively as the input layer data set of neutral net, is predicted by BP neural network model;
Model evaluation module, using accuracy rate P as evaluation criterion, such as shown in formula (6):
Wherein, P is accuracy rate, pmiFor the number of the correctly predicted value of every kind of classification;M is PM2.5 value real data samples
Amount.
As the further optimization of such scheme, pollutant coefficient matrix PM is built:Pollutant index in collection air
The concentration value of PM2.5, O3, CO, PM10, SO2, NO2, builds pollutant coefficient matrix PM:
Wherein, pollutant coefficient matrix PM first lists are shown as:To count all PM2.5 of natural law
Value, divided according to 1 grade of table;Remaining list is shown as:Si=(S1i…Smi)T, i=1 ... 5 is O3 concentration values, CO dense
Angle value, PM10 concentration values, SO2 concentration values and NO2 concentration values.
As the further optimization of such scheme, with reference to air index parameter:Air themperature, humidity and wind-force, with air
In pollutant index combine, provide coefficient of colligation matrix PMA:
Coefficient of colligation matrix PMA and pollutant coefficient matrix PM differences are to add environmental coefficient matrix Q, environment system
Respectively row are represented by matrix number Q:Qi=(Q1i…Qmi)T, i=1 ... 3 is that the temperature that adds, humidity, three air of wind-force refer to
Mark.
As the further optimization of such scheme, using the method for Pearson's correlation coefficient, in coefficient of colligation matrix PMA
O3 concentration values, CO concentration values, PM10 concentration values, SO2 concentration values, NO2 concentration values and air themperature index, humidity index, wind
Power index substitutes into the computational methods formula (3) of similarity between pearson correlation coefficient parameters respectively, enters row index merging,
Aggregative indicator coefficient matrix PMA ' after merging, such as shown in formula (4);
Wherein,
Sim (X, Y) represents the similarity between X, Y two indices, and i represents i-th sample of certain index, and n is that sample is total
Number.
Used as the further optimization of such scheme, BP neural network model is predicted, input layer by be input into
Data, are delivered to hidden layer, then output layer are delivered to after amplified data activation and are exported by output layer;Wherein,
Excitation function between input layer and hidden layer adopts S function, such as shown in formula (5):
When actual error exceeds anticipation error, error amount changes the connection weight between each neuron along network back propagation
Value and threshold value, repetition training network, until meet anticipation error, it is determined that the mapping relations between input and output.
Compared with prior art, a kind of PM2.5 grade prediction techniques based on BP neural network model of the invention and it is
System has the advantages that:
(1) the real-time grade prediction techniques of a kind of PM2.5 based on neutral net of the invention, can carry for mobile terminal user
For daily or hour real-time PM2.5 grades inquiry and push;
(2) the real-time grade prediction techniques of a kind of PM2.5 based on neutral net of the invention, extraction cost are relatively low, relatively hold
The pollutant for easily obtaining are aggregative indicator with atmospheric factor, and by Similarity Measure between index, index are merged, is being carried
Predicted time complexity is reduced while height prediction is accurate;
(3) the real-time grade prediction techniques of a kind of PM2.5 based on neutral net of the invention, set up one kind daily or hour
The real-time grade forecast universal models of PM2.5.
(4) a kind of PM2.5 concentration prediction systems based on Pearson's correlation coefficient of the invention, extraction cost are relatively low, compared with
The pollutant for easily obtaining are aggregative indicator with atmospheric factor, and by Similarity Measure between index, index are merged,
Improve.
(5) a kind of PM2.5 concentration prediction systems based on Pearson's correlation coefficient of the invention, by O3, CO, PM10,
The air pollutants observation index that SO2, NO2 etc. are easily obtained is predicted to PM2.5, to reduce the input of PM2.5 equipment, in base
The atmospheric factors such as temperature, humidity, wind-force are added on the basis of this pollutant, the forecast model for setting up comprehensive air index is improved
Predictablity rate;The dependency between each index is analyzed while prediction, the influence factor higher with PM2.5 dependencys is analyzed,
Predicted time complexity is reduced while improving precision of prediction.
Description of the drawings
Fig. 1 is a kind of flow chart of the real-time grade prediction techniques of PM2.5 based on neutral net of the present invention.
Fig. 2 is index similar matrix.
Fig. 3 is PM-5-D with day as unit real-time estimate result schematic diagram.
Fig. 4 is PM-8-D with day as unit real-time estimate result schematic diagram.
Fig. 5 is PM-6-D with day as unit real-time estimate result schematic diagram.
Fig. 6 is PM-5-H with hour as unit real-time estimate result schematic diagram.
Fig. 7 is PM-8-H with hour as unit real-time estimate result schematic diagram.
Fig. 8 is PM-6-H with hour as unit real-time estimate result schematic diagram.
Specific embodiment
To make purpose, technical scheme and the advantage of invention of greater clarity, below by accompanying drawing and embodiment, to this
Inventive technique scheme is further elaborated.However, it should be understood that specific embodiment described herein is only to solve
Technical solution of the present invention is released, the scope of technical solution of the present invention is not limited to.
Referring to Fig. 1, Fig. 1 is a kind of flow chart of the real-time grade prediction techniques of PM2.5 based on neutral net of the present invention,
A kind of real-time grade prediction technique of the PM2.5 based on neutral net, comprises the steps:
(1) offline history index PM2.5, O3 of pollutant in collection air, the concentration value of CO, PM10, SO2, NO2, structure
Build pollutant coefficient matrix PM:
Wherein, pollutant coefficient matrix PM first lists are shown as:To count all PM2.5 of natural law
Value, divided according to 1 grade of table;Remaining list is shown as:Si=(S1i…Smi)T, i=1 ... 5, be O3, CO, PM10,
SO2, NO2 concentration value;Wherein m is statistics natural law or hourage;
(2) gather air themperature, humidity, the offline history index of three air of wind-force, and with air in pollutant index
Combine, provide coefficient of colligation matrix PMA:
Coefficient of colligation matrix PMA and pollutant coefficient matrix PM differences are to add environmental coefficient matrix Q, the present invention
Middle QmiMatrix is the new temperature for adding, humidity, three air index matrixs of wind-force, and to count natural law, i=1 ... 3 is represented wherein m
Temperature, humidity, three indexs of wind-force;
(3) using the method for Pearson's correlation coefficient, the O3 concentration values in coefficient of colligation matrix PMA, CO concentration values,
PM10 concentration values, SO2 concentration values, NO2 concentration values and air themperature index, humidity index, wind-force index substitute into pearson respectively
Between correlation coefficient parameter shown in the computational methods formula (3) of similarity:
Wherein sim (X, Y) represents the similarity between X, Y two indices, and i represents i-th sample of certain index, and n is sample
This sum;
(4) using the method for Pearson's correlation coefficient, data in PMA are substituted into formula (3), the index in PMA is carried out
Merge, the aggregative indicator matrix after merging is PMA '
Wherein,To count the value of all PM2.5 of natural law, divided according to 1 grade of table;Si=
(S1i…Smi)T, i≤5 are the concentration values after O3, CO, PM10, SO2, NO2 enter row index merging;Qi=(Q1i…Qmi)T, i≤3,
Be temperature, humidity, wind-force index merge after value;Wherein m is statistics natural law or hourage.
(5) make training set training neutral net with off-line data, set up the PM2.5 forecast models based on neutral net.Will
Select PM, PMA, PMA ' as the input layer data set of neutral net, it is predicted by BP neural network model:Wherein input layer is refreshing
The first data transfers by input of Jing are to hidden layer, then output layer are delivered to after amplified by data activation and by output layer
Output;Wherein, the excitation function between input layer and hidden layer adopts S function, such as shown in formula (5):
When actual error exceeds anticipation error, error amount changes the connection weight between each neuron along network back propagation
Value and threshold value, repetition training network, until meet anticipation error, it is determined that the mapping relations between input and output.In the present invention
Xuan Ze PM, PMA, PMA ' matrix is input data, PM2.5 grade forecasts value carries out experiment mapping for output;
(6) send the request of PM2.5 real-time queries from mobile terminal to server, user may be selected inquiry in real time " my god " it is or real
When " hour " inquired about;
(7) server obtains mobile terminal current time;
(8) according to the time obtained in the index system and step (7) after merging in step (4), Real-time Collection pollutant
Index and air index;
(9) data in step (8) are substituted into into (5), predicts real-time PM2.5 values;
(10) real time propelling movement will be predicted the outcome to mobile end subscriber.
Experiment collection from -2014 years on the 1st October 30 of September in 2014, the PM2.5 of 10 observation stations in Hefei, O3, CO,
PM10, SO2, NO2 per hour and 24 hour datas, the atmospheric temperature of same time, humidity, wind data.PM2.5 in experiment,
From pm25.in, atmosphere data comes from Chinese weather net to O3, CO, PM10, SO2, NO2 data.As data pick-up website limits
System, partial data are lost, and in this experiment, real data extracts data per hour and 24 hour datas that natural law is 57 days.Lead to herein
The mode of cross validation is crossed, the training and experimentation to parameter is verified.
According to ambient air quality index (AQI) technical stipulation (trying) standard, PM2.5 is divided standard according to concentration
For:Excellent, good, slight pollution, intermediate pollution, serious pollution, severe contamination, this experiment predict the outcome as grade, therefore empty with environment
Grade in makings volume index (AQI) technical stipulation (trying) standard corresponds to 1,2,3,4,5,6 grades, corresponding relation such as 1 institute of table
Show.
1 PM2.5 of table and air quality grade synopsis
PM2.5(ug/m3)
, using accuracy rate (P) as evaluation criterion, such as formula (6) is shown for the present invention:
Wherein P be accuracy rate, pmiFor the number of the correctly predicted value of every kind of classification, because being finally predicted as 5 classifications etc.
Level, so the value of n is 5 in this experiment;M is PM2.5 value real data sample sizes, and the value of P is the bigger the better.
In step (3), correlation analysiss are carried out to index, between observation index, whether there is linear relationship, if existing linear
Relation, then select pearson related algorithms to enter the similar calculating of row index.In stronger between PM2.5, PM10, CO, NO2, SO2
Linear relationship.Other index linear relationships are weaker, enter row index Similarity Measure to part index number using pearson algorithms and close
And.PMA of the number containing 8 aggregative indicatores is merged by formula (3), index related matrix is as shown in Figure 2.
PM10, CO, NO2 are higher with PM2.5 dependencys, while the autocorrelation of these three indexs is also higher, so by this
Three indexs merge, and retain and PM2.5 dependency highests PM10, delete O3, CO two indices.ECDC and after PMA ' matrixes
Also 6 aggregative indicatores.
PM, PMA, PMA are selected respectively ' tested, and analyze experimental result.
Experiment one:
The real-time estimate knot of O3, CO, PM10, SO2, NO2 with PM as matrix this 5 kinds of pollutant indexs in units of day
Really, it is designated as (PM-5-D) to predict the outcome in experiment, as shown in Figure 3.In Fig. 3, calculating predictablity rate by formula (6) is
77.19%, the accuracy rate is more than 60%, illustrates that 5 pollutant indexs are the principal elements for affecting PM2.5 values, as a result with one
Fixed reference value.
Experiment two:
With PMA as matrix except five kinds of O3, CO, PM10, SO2, NO2 pollution beyond the region of objective existences, add temperature, humidity, wind-force totally 8
Real-time estimate result of the comprehensive meteorological index in units of day, is designated as (PM-8-D) and predicts the outcome, as shown in Figure 4 in experiment.
In Fig. 4, predictablity rate is calculated for 84.21% by formula (6), be greatly improved compared with the predicting the outcome of Fig. 4.Can be with
Find out, after adding temperature, humidity, wind-force air index, as a result more accurately, illustrate the value of temperature, humidity, wind-force to PM2.5
There is large effect.
Experiment three:
PMA after index merging, with PMA ' as matrix, the real-time estimate result of totally 6 indexs in units of day, experiment
In be designated as (PM-6-D predicts the outcome), as shown in Figure 5:In Fig. 5, it is 83.46% to calculate predictablity rate by formula (6), with
Predicting the outcome for Fig. 5 is compared, and accuracy rate gap is little.As can be seen that after merging to PM10, O3, CO index, to the standard that predicts the outcome
Really rate affects little, but after leading indicator is extracted, data dimension reduces, and predicted time complexity is reduced.
Experiment four:
The real-time estimate knot of O3, CO, PM10, SO2, NO2 with PM as matrix this 5 kinds of pollutant indexs in units of hour
Really, it is designated as (PM-5-H) to predict the outcome in experiment, as shown in Figure 7.In Fig. 6, calculating predictablity rate by formula (6) is
85.71%, accuracy rate is higher, illustrates that 5 pollutant indexs are the principal elements for affecting PM2.5 values, as a result with certain ginseng
Examine value.
Experiment five:
With PMA as matrix except five kinds of O3, CO, PM10, SO2, NO2 pollution beyond the region of objective existences, add temperature, humidity, wind-force totally 8
Real-time estimate result of the comprehensive meteorological index in units of hour, is designated as (PM-8-H) and predicts the outcome, as shown in Figure 8 in experiment.
In Fig. 7, predictablity rate is calculated for 100% by formula (6), be greatly improved compared with the predicting the outcome of Fig. 4.Can see
Go out, after adding temperature, humidity, wind-force air index, as a result more accurately, illustrate that temperature, humidity, wind-force have to the value of PM2.5
Large effect.
Experiment six:
, after index merging, with PMA ' as matrix, the real-time estimate result of totally 6 indexs in units of hour, real for PMA
(PM-6 predicts the outcome) is designated as in testing.In Fig. 8, it is 85.71% to calculate predictablity rate by formula (6), and the prediction with Fig. 8 is tied
Fruit is compared, and only one of which time point has error, and accuracy rate gap is little.As can be seen that after merging to PM10, O3, CO index, it is right
The accuracy rate that predicts the outcome affects little, but after the high index of similarity is merged, data dimension reduces, predicted time complexity drop
It is low.
This paper presents a kind of real-time grade prediction technique of the PM2.5 based on neutral net, due to adopt average daily grade for
Predict the outcome, neutral net is selected as observation model.Experiment is proved:
(1) the method can for mobile terminal user provide daily or hour real-time PM2.5 grades inquiry and push;
(2) add the forecast model of the comprehensive meteorological index such as temperature, humidity, wind-force more accurate than the forecast model of 5 kinds of pollutant
Really rate is high, illustrates that temperature, humidity, wind-force have certain impact to PM2.5;
(3) similarity analysis are carried out to aggregative indicator matrix, find the factor higher with PM2.5 dependencys have PM10, CO,
NO2;
(4) can set up it is a kind of daily or hour the real-time grade forecast universal models of PM2.5.
In sum, this paper presents can provide daily for mobile terminal user or the real-time PM2.5 grades of hour are looked into
Ask and push, except the advantage for possessing neutral net PM2.5 forecast model low cost, while and improve predictablity rate, drop
Low predicted time.Although the prediction of PM2.5 can be carried out with higher accuracy rate by comprehensive meteorological index neutral net,
The space that accuracy rate is also further improved.Still there are data related to weather in a large number not disclose at present, and each city
PM2.5 observation stations are less, and data volume not enough, has certain impact to the accuracy predicted.In this regard, further work continues to extract dirt
Dye thing and air related data, while accuracy rate is improved, improve time complexity.While improved BP neural networks model,
Improve predictablity rate.
Moreover, it will be appreciated that although this specification is been described by according to embodiment, not each embodiment is only wrapped
Containing an independent technical scheme, this narrating mode of description is only that those skilled in the art should for clarity
Using description as an entirety, the technical scheme in each embodiment can also Jing it is appropriately combined, form those skilled in the art
Understandable other embodiment.
Claims (7)
1. real-time grade prediction techniques of a kind of PM2.5 based on neutral net, it is characterised in that:
(1) offline history index PM2.5, O3 of pollutant in collection air, the concentration value of CO, PM10, SO2, NO2, build dirty
Dye thing coefficient matrix PM:
Wherein, pollutant coefficient matrix PM first lists are shown as:To count all PM2.5's of natural law
Value, is divided according to 1 grade of table;Remaining list is shown as:Si=(S1i…Smi)T, i=1 ... 5, be O3, CO, PM10, SO2,
NO2 concentration values;Wherein m is statistics natural law or hourage;
(2) air themperature, humidity, the offline history index of three air of wind-force are gathered, and is mutually tied with the pollutant index in air
Close, provide coefficient of colligation matrix PMA:
Coefficient of colligation matrix PMA and pollutant coefficient matrix PM differences are to add environmental coefficient matrix Q, QmiMatrix is new
The temperature of addition, humidity, three air index matrixs of wind-force, wherein m are statistics natural law, and i=1 ... 3 represents temperature, humidity, wind
Three indexs of power;
(3) using the method for Pearson's correlation coefficient, the O3 concentration values in coefficient of colligation matrix PMA, CO concentration values, PM10 dense
Angle value, SO2 concentration values, NO2 concentration values and air themperature index, humidity index, wind-force index substitute into pearson phase relations respectively
Between number parameter shown in the computational methods formula (3) of similarity:
Wherein sim (X, Y) represents the similarity between X, Y two indices, and i represents i-th sample of certain index, and n is that sample is total
Number;
(4) using the method for Pearson's correlation coefficient, data in PMA are substituted into formula (3), the index in PMA are merged,
Aggregative indicator matrix after merging is PMA '
Wherein,To count the value of all PM2.5 of natural law, divided according to 1 grade of table;
Si=(S1i…Smi)T, i≤5 are the concentration values after O3, CO, PM10, SO2, NO2 enter row index merging;
Qi=(Q1i…Qmi)T, i≤3, be temperature, humidity, wind-force index merge after value;Wherein m is statistics natural law or hour
Number;
(5) make training set training neutral net with off-line data, set up the PM2.5 forecast models based on neutral net:
PM, PMA, PMA will be selected ' as the input layer data set of neutral net, it is predicted by BP neural network model:It is wherein defeated
Enter layer neuron by be input into data transfer to hidden layer, then be delivered to after amplified by data activation output layer and by
Output layer is exported;Wherein, the excitation function between input layer and hidden layer adopts S function, such as shown in formula (5):
When actual error exceeds anticipation error, error amount along network back propagation change connection weight between each neuron and
Threshold value, repetition training network, until meet anticipation error, it is determined that the mapping relations between input and output.Do not select in the present invention
Select PM, PMA, PMA ' matrix be input data, PM2.5 grade forecasts value for output carry out experiment mapping.
2. real-time grade prediction techniques of a kind of PM2.5 based on neutral net according to claim 1, it is characterised in that:
Also comprise the steps:
(6) send the request of PM2.5 real-time queries from mobile terminal to server, user may be selected inquiry in real time " my god " or in real time
" hour " is inquired about;
(7) server obtains mobile terminal current time;
(8) according to the mobile terminal current time obtained in the index system and step (7) after merging in step (4), Real-time Collection
Pollutant index and air index;
(9) data in step (8) are substituted into into step (5), predicts real-time PM2.5 values;
(10) real time propelling movement will be predicted the outcome to mobile end subscriber.
3. real-time grade forecast systems of a kind of PM2.5 based on neutral net, it is characterised in that include:
Data acquisition module, collects pollutant index and air index, and described pollutant index includes O3 concentration values, CO concentration
Value, PM10 concentration values, SO2 concentration values, NO2 concentration values, described air index include air themperature, humidity and wind-force;
Data processing module, builds pollutant coefficient matrix PM and coefficient of colligation matrix based on pollutant index and air index
PMA, then the calculating of similarity between pearson correlation coefficient indexs is utilized respectively to pollutant index and air index, referred to
Mark merges, and after merging, obtains aggregative indicator coefficient matrix PMA ';
Models fitting module, will select pollutant coefficient matrix PM, coefficient of colligation matrix PMA, and aggregative indicator coefficient matrix PMA ' divides
Not as the input layer data set of neutral net, it is predicted by BP neural network model;
Model evaluation module, using accuracy rate P as evaluation criterion, such as shown in formula (6):
Wherein, P is accuracy rate, pmiFor the number of the correctly predicted value of every kind of classification;M is PM2.5 value real data sample sizes.
4. real-time grade forecast systems of a kind of PM2.5 based on neutral net according to claim 3, it is characterised in that:
Build pollutant coefficient matrix PM:Pollutant index PM2.5, O3, the concentration value of CO, PM10, SO2, NO2 in collection air, structure
Build pollutant coefficient matrix PM:
Wherein, pollutant coefficient matrix PM first lists are shown as:To count all PM2.5's of natural law
Value, is divided according to 1 grade of table;Remaining list is shown as:Si=(S1i…Smi)T, i=1 ... 5 is O3 concentration values, CO concentration
Value, PM10 concentration values, SO2 concentration values and NO2 concentration values.
5. real-time grade forecast systems of a kind of PM2.5 based on neutral net according to claim 3 or 4, its feature exist
In:With reference to air index parameter:Air themperature, humidity and wind-force, are combined with the pollutant index in air, provide comprehensive system
Matrix number PMA:
Coefficient of colligation matrix PMA and pollutant coefficient matrix PM differences are to add environmental coefficient matrix Q, environmental coefficient square
Respectively row are represented by battle array Q:Qi=(Q1i…Qmi)T, i=1 ... 3, be add temperature, humidity, three air indexs of wind-force.
6. real-time grade forecast systems of a kind of PM2.5 based on neutral net according to claim 3 or 4 or 5, its feature
It is:Using the method for Pearson's correlation coefficient, the O3 concentration values in coefficient of colligation matrix PMA, CO concentration values, PM10 concentration
Value, SO2 concentration values, NO2 concentration values and air themperature index, humidity index, wind-force index substitute into pearson correlation coefficienies respectively
The computational methods formula (3) of similarity between parameter, enters row index and merges, the aggregative indicator coefficient matrix PMA ' after merging,
As shown in formula (4);
Wherein,
Sim (X, Y) represents the similarity between X, Y two indices, and i represents i-th sample of certain index, and n is total sample number.
7. real-time grade forecast systems of a kind of PM2.5 based on neutral net according to claim 3, it is characterised in that:
BP neural network model is predicted, the data that input layer will be input into, and is delivered to hidden layer, then by hidden layer by data
Activation is delivered to output layer and is exported by output layer after amplifying;Wherein, the excitation function between input layer and hidden layer adopts S letters
Shown in number, such as formula (5):
When actual error exceeds anticipation error, error amount along network back propagation change connection weight between each neuron and
Threshold value, repetition training network, until meet anticipation error, it is determined that the mapping relations between input and output.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611097948.9A CN106529081A (en) | 2016-12-03 | 2016-12-03 | PM2.5 real-time level prediction method and system based on neural net |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611097948.9A CN106529081A (en) | 2016-12-03 | 2016-12-03 | PM2.5 real-time level prediction method and system based on neural net |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106529081A true CN106529081A (en) | 2017-03-22 |
Family
ID=58354591
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611097948.9A Pending CN106529081A (en) | 2016-12-03 | 2016-12-03 | PM2.5 real-time level prediction method and system based on neural net |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106529081A (en) |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109142171A (en) * | 2018-06-15 | 2019-01-04 | 上海师范大学 | The city PM10 concentration prediction method of fused neural network based on feature expansion |
CN109615147A (en) * | 2018-12-27 | 2019-04-12 | 浙江大学 | A kind of following 72 hours air pollution forecasting method for early warning |
CN110187074A (en) * | 2019-06-12 | 2019-08-30 | 哈尔滨工业大学 | A kind of cross-country skiing racing track quality of snow prediction technique |
CN110610209A (en) * | 2019-09-16 | 2019-12-24 | 北京邮电大学 | Air quality prediction method and system based on data mining |
CN110796284A (en) * | 2019-09-20 | 2020-02-14 | 平安科技(深圳)有限公司 | Method and device for predicting pollution level of fine particulate matters and computer equipment |
CN110852493A (en) * | 2019-10-25 | 2020-02-28 | 天津大学 | Atmospheric PM2.5 concentration prediction method based on multiple model comparisons |
CN111077048A (en) * | 2019-11-27 | 2020-04-28 | 华南师范大学 | Opportunistic group intelligent air quality monitoring and evaluating method based on mobile equipment |
CN111157682A (en) * | 2020-01-06 | 2020-05-15 | 上海应用技术大学 | Air quality monitoring and predicting system and method |
CN111256745A (en) * | 2020-02-28 | 2020-06-09 | 芜湖职业技术学院 | Data calibration method for portable air quality monitor |
CN111289414A (en) * | 2020-03-12 | 2020-06-16 | 徐州工业职业技术学院 | PM2.5 pollution monitoring and predicting method |
CN111340292A (en) * | 2020-02-26 | 2020-06-26 | 福州大学 | Integrated neural network PM2.5 prediction method based on clustering |
CN111489015A (en) * | 2020-03-20 | 2020-08-04 | 天津大学 | Atmosphere O based on multiple model comparison and optimization3Concentration prediction method |
CN111581792A (en) * | 2020-04-23 | 2020-08-25 | 生态环境部环境规划院 | Atmospheric PM based on two-stage non-negative Lasso model2.5Concentration prediction method and system |
CN111882103A (en) * | 2020-06-15 | 2020-11-03 | 航天凯天环保科技股份有限公司 | Real-time online monitoring system for pollution source |
CN112766454A (en) * | 2020-12-04 | 2021-05-07 | 山东科技大学 | Air PM2.5 concentration prediction method, system, terminal and storage medium |
CN113970511A (en) * | 2021-10-21 | 2022-01-25 | 天津大学 | Air particulate matter data monitoring system and method based on BP neural network |
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 |
CN114861882A (en) * | 2022-05-07 | 2022-08-05 | 国网四川省电力公司电力科学研究院 | CO (carbon monoxide) 2 Space-time distribution reconstruction method and system |
WO2022217332A1 (en) * | 2021-04-14 | 2022-10-20 | Ecosystem Informatics Inc. | Monitoring an ambient air parameter using a trained model |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103207944A (en) * | 2013-02-04 | 2013-07-17 | 国家电网公司 | Electric power statistical index relevance analysis method |
CN105406476A (en) * | 2015-12-31 | 2016-03-16 | 国家电网公司 | Historical data-based power system stability fast judging method |
US20160117845A1 (en) * | 2014-10-27 | 2016-04-28 | King Fahd University Petroleum and Minerals | Contamination level estimation method for high voltage insulators |
CN106056210A (en) * | 2016-06-07 | 2016-10-26 | 浙江工业大学 | PM 2.5 concentration value prediction method based on hybrid neural network |
-
2016
- 2016-12-03 CN CN201611097948.9A patent/CN106529081A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103207944A (en) * | 2013-02-04 | 2013-07-17 | 国家电网公司 | Electric power statistical index relevance analysis method |
US20160117845A1 (en) * | 2014-10-27 | 2016-04-28 | King Fahd University Petroleum and Minerals | Contamination level estimation method for high voltage insulators |
CN105406476A (en) * | 2015-12-31 | 2016-03-16 | 国家电网公司 | Historical data-based power system stability fast judging method |
CN106056210A (en) * | 2016-06-07 | 2016-10-26 | 浙江工业大学 | PM 2.5 concentration value prediction method based on hybrid neural network |
Non-Patent Citations (3)
Title |
---|
张怡文 等: "基于神经网络的PM2.5预测模型", 《江苏师范大学学报(自然科学版)》 * |
王兰霞: "基于RBF神经网络的PM10污染预测研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅰ辑》 * |
秦珊珊: "悬浮颗粒物PM10与PM2.5的统计分析与预测", 《中国优秀硕士学位论文全文数据库 基础科学辑》 * |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109142171A (en) * | 2018-06-15 | 2019-01-04 | 上海师范大学 | The city PM10 concentration prediction method of fused neural network based on feature expansion |
CN109142171B (en) * | 2018-06-15 | 2021-08-03 | 上海师范大学 | Urban PM10 concentration prediction method based on feature expansion and fusing with neural network |
CN109615147A (en) * | 2018-12-27 | 2019-04-12 | 浙江大学 | A kind of following 72 hours air pollution forecasting method for early warning |
CN110187074A (en) * | 2019-06-12 | 2019-08-30 | 哈尔滨工业大学 | A kind of cross-country skiing racing track quality of snow prediction technique |
CN110610209A (en) * | 2019-09-16 | 2019-12-24 | 北京邮电大学 | Air quality prediction method and system based on data mining |
CN110796284A (en) * | 2019-09-20 | 2020-02-14 | 平安科技(深圳)有限公司 | Method and device for predicting pollution level of fine particulate matters and computer equipment |
WO2021051609A1 (en) * | 2019-09-20 | 2021-03-25 | 平安科技(深圳)有限公司 | Method and apparatus for predicting fine particulate matter pollution level, and computer device |
CN110852493A (en) * | 2019-10-25 | 2020-02-28 | 天津大学 | Atmospheric PM2.5 concentration prediction method based on multiple model comparisons |
CN111077048A (en) * | 2019-11-27 | 2020-04-28 | 华南师范大学 | Opportunistic group intelligent air quality monitoring and evaluating method based on mobile equipment |
CN111157682A (en) * | 2020-01-06 | 2020-05-15 | 上海应用技术大学 | Air quality monitoring and predicting system and method |
CN111340292A (en) * | 2020-02-26 | 2020-06-26 | 福州大学 | Integrated neural network PM2.5 prediction method based on clustering |
CN111340292B (en) * | 2020-02-26 | 2022-05-10 | 福州大学 | Integrated neural network PM2.5 prediction method based on clustering |
CN111256745A (en) * | 2020-02-28 | 2020-06-09 | 芜湖职业技术学院 | Data calibration method for portable air quality monitor |
CN111289414A (en) * | 2020-03-12 | 2020-06-16 | 徐州工业职业技术学院 | PM2.5 pollution monitoring and predicting method |
CN111489015A (en) * | 2020-03-20 | 2020-08-04 | 天津大学 | Atmosphere O based on multiple model comparison and optimization3Concentration prediction method |
CN111581792A (en) * | 2020-04-23 | 2020-08-25 | 生态环境部环境规划院 | Atmospheric PM based on two-stage non-negative Lasso model2.5Concentration prediction method and system |
CN111882103A (en) * | 2020-06-15 | 2020-11-03 | 航天凯天环保科技股份有限公司 | Real-time online monitoring system for pollution source |
CN112766454A (en) * | 2020-12-04 | 2021-05-07 | 山东科技大学 | Air PM2.5 concentration prediction method, system, terminal and storage medium |
WO2022217332A1 (en) * | 2021-04-14 | 2022-10-20 | Ecosystem Informatics Inc. | Monitoring an ambient air parameter using a trained model |
CN113970511A (en) * | 2021-10-21 | 2022-01-25 | 天津大学 | Air particulate matter data monitoring system and method based on BP neural network |
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 |
CN114861882A (en) * | 2022-05-07 | 2022-08-05 | 国网四川省电力公司电力科学研究院 | CO (carbon monoxide) 2 Space-time distribution reconstruction method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106529081A (en) | PM2.5 real-time level prediction method and system based on neural net | |
Li et al. | Prediction of PM2. 5 concentration based on the similarity in air quality monitoring network | |
CN108009674A (en) | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks | |
CN107688906A (en) | The transmission line of electricity meteorological element NO emissions reduction analysis system and method for multi-method fusion | |
CN105069537B (en) | A kind of construction method of combination type air Quality Prediction | |
CN106651036A (en) | Air quality forecasting system | |
CN110346518B (en) | Traffic emission pollution visualization early warning method and system thereof | |
CN107885928A (en) | Consider the stepstress acceleration Degradation Reliability analysis method of measurement error | |
CN112598438A (en) | Outdoor advertisement recommendation system and method based on large-scale user portrait | |
CN112035448A (en) | Neural network short-term rainfall forecasting method integrating water vapor and meteorological elements of foundation GNSS | |
CN110211325A (en) | A kind of area road icing high precision monitor early warning system based on meteorological big data | |
CN104834944B (en) | A kind of urban area air quality method of estimation based on coorinated training | |
CN112149887A (en) | PM2.5 concentration prediction method based on data space-time characteristics | |
CN107798418A (en) | A kind of traffic accident frequency Forecasting Methodology based on traffic analysis cell | |
CN115100819B (en) | Landslide hazard early warning method and device based on big data analysis and electronic equipment | |
CN109921938A (en) | Fault detection method under a kind of cloud computing environment | |
CN106157616A (en) | A kind of magnitude of traffic flow short-term prediction device | |
CN116523129A (en) | Road surface anti-skid performance prediction method based on XGBOOST and fractal dimension | |
CN114330120A (en) | 24-hour PM prediction based on deep neural network2.5Method of concentration | |
CN110716998A (en) | Method for spatializing fine-scale population data | |
CN113836808A (en) | PM2.5 deep learning prediction method based on heavy pollution feature constraint | |
CN104569340B (en) | Underground environment quality determination method and device | |
CN115983329A (en) | Method, device, equipment and storage medium for predicting air quality and meteorological conditions | |
CN107992454A (en) | A kind of air quality grade Forecasting Methodology based on online ordinal regression | |
CN114814092A (en) | IP index measuring method based on BP neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170322 |
|
RJ01 | Rejection of invention patent application after publication |