CN105488317A - Air quality prediction system and method - Google Patents
Air quality prediction system and method Download PDFInfo
- Publication number
- CN105488317A CN105488317A CN201410475151.2A CN201410475151A CN105488317A CN 105488317 A CN105488317 A CN 105488317A CN 201410475151 A CN201410475151 A CN 201410475151A CN 105488317 A CN105488317 A CN 105488317A
- Authority
- CN
- China
- Prior art keywords
- adjacent domain
- monitoring station
- forecast model
- measured
- place
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Air Conditioning Control Device (AREA)
Abstract
The invention provides a multi-prediction task based air quality prediction system. The system comprises a determination unit, a training unit and a prediction unit, wherein the determination unit is configured to determine an adjacent area of a to-be-tested place according to a distance threshold; the training unit is configured to train a prediction model to obtain optimal parameters of the prediction model, and the prediction model is constructed based on multiple prediction tasks and according to spatial features of the adjacent area; and the prediction unit is configured to predict the air quality of the to-be-tested place by utilizing the prediction model with the optimal parameters. The invention furthermore provides a multi-prediction task based air quality prediction method. According to the multi-prediction task based air quality prediction method and system, the relevance among similar tasks is fully utilized and the accuracy of air pollution prediction is improved.
Description
Technical field
The application relates to data analysis field, is specifically related to a kind of system and method for predicting air quality.
Background technology
Air pollution and air quality are more and more taken seriously.But because the usual cost of air pollutants checkout equipment is very high, troublesome poeration, so other environmental information (as temperature, humidity and visibility etc.) is utilized to predict to have high social benefit and economic worth to deleterious particle in air (as PM2.5 etc.).
Usually, by pollutant change informations such as historical information, neighbouring area information, pollutant levels prediction is carried out to future time section and the space without testing conditions on the spot.In order to prediction concentrations, depend on statistical method and go to find the similar website of concentration distribution of pollutants, the concentration treating survey station point carries out matching.
Summary of the invention
In the urban environment of reality, injurious factor is often more than a kind of, and the concentration change of these injurious factors has certain contact usually.The prior art of single modeling cannot make full use of this contact between each prediction task.In order to improve the predictive ability of model in global scope, the present invention carries out collaborative modeling to the multiple pollution factor in same space-time unique, improves the precision of prediction of each task.
According to a first aspect of the invention, providing a kind of system for predicting air quality based on multiple prediction task, comprising: determining unit, being configured to the adjacent domain determining place to be measured according to distance threshold; Training unit, be configured to training forecast model to obtain the optimized parameter of described forecast model, wherein, described forecast model is based on described multiple prediction task and build according to the space characteristics of described adjacent domain; And predicting unit, be configured to utilize the described forecast model with described optimized parameter to predict the air quality in place to be measured.
In one embodiment, the space characteristics of adjacent domain comprises numeric type characteristic sum Boolean type feature.
In one embodiment, numeric type feature comprises with the next item down or more item: wind speed, temperature, humidity, quantity of precipitation that monitoring station place in adjacent domain is measured, and the Distance geometry angle between monitoring station in adjacent domain.
In one embodiment, Boolean type feature comprises with the next item down or more item: whether the relative orientation of the monitoring station in adjacent domain meets specified conditions, and whether the relative distance of monitoring station in adjacent domain is greater than threshold value.
In one embodiment, training unit is configured to: for any two monitoring stations in adjacent domain, calculate space characteristics and the predicted value of any two monitoring stations; And for all monitoring stations in adjacent domain, according to space characteristics and the predicted value of any two monitoring stations calculated, the described optimized parameter of computational prediction model, make from multiple prediction task on the whole the absolute value sum of difference of the predicted value that obtained by the forecast model with optimized parameter and true measurement minimum.
In one embodiment, predicting unit is configured to: the space characteristics utilizing the adjacent domain in place to be measured, calculates the weighted sum of the predicted value of each monitoring station in adjacent domain, predicts the air quality in place to be measured thus.
In one embodiment, be directly proportional for the inverse of the weight of each monitoring station to the training error of this monitoring station, the true measurement of described training error and this monitoring station is relevant with the difference of predicted value obtained by the forecast model with optimized parameter.
According to a second aspect of the invention, providing a kind of method for predicting air quality based on multiple prediction task, comprising: the adjacent domain determining place to be measured according to distance threshold; Training forecast model is to obtain the optimized parameter of described forecast model, and wherein, described forecast model is based on described multiple prediction task and build according to the space characteristics of described adjacent domain; And utilize the described forecast model with described optimized parameter to predict the air quality in place to be measured.
In one embodiment, the space characteristics of adjacent domain comprises numeric type characteristic sum Boolean type feature.
In one embodiment, numeric type feature comprises with the next item down or more item: wind speed, temperature, humidity, quantity of precipitation that monitoring station place in adjacent domain is measured, and the Distance geometry angle between monitoring station in adjacent domain.
In one embodiment, Boolean type feature comprises with the next item down or more item: whether the relative orientation of the monitoring station in adjacent domain meets specified conditions, and whether the relative distance of monitoring station in adjacent domain is greater than threshold value.
In one embodiment, for any two monitoring stations in described adjacent domain, space characteristics and the predicted value of any two monitoring stations is calculated; And for all monitoring stations in adjacent domain, according to space characteristics and the predicted value of any two monitoring stations calculated, the optimized parameter of computational prediction model, make from multiple prediction task on the whole the absolute value sum of difference of the predicted value that obtained by the forecast model with optimized parameter and true measurement minimum.
In one embodiment, utilize the space characteristics of the adjacent domain in place to be measured, calculate the weighted sum of the predicted value of each monitoring station in adjacent domain, predict the air quality in place to be measured thus.
In one embodiment, be directly proportional for the inverse of the weight of each monitoring station to the training error of this monitoring station, the true measurement of described training error and this monitoring station is relevant with the difference of predicted value obtained by the forecast model with optimized parameter.
The present invention, by the modeling of multiple prediction task cooperation, takes full advantage of the relevance between similar tasks, thus improves the degree of accuracy of Air Pollution Forecast.
Accompanying drawing explanation
By hereafter detailed description with the accompanying drawing, above-mentioned and further feature of the present invention will become more apparent, wherein:
Fig. 1 shows the block diagram according to the system for predicting air quality of the present invention.
Fig. 2 shows according to the schematic diagram for determining adjacent domain of the present invention.
Fig. 3 shows the schematic diagram of the relation according to two example monitoring stations of the present invention and wind direction.
Fig. 4 shows according to the schematic diagram of place to be measured of the present invention with contiguous monitoring station.
Fig. 5 shows according to the schematic diagram of place to be measured of the present invention with contiguous monitoring station.
Fig. 6 shows the process flow diagram according to the method for predicting air quality of the present invention.
Embodiment
Below, in conjunction with the drawings to the description of specific embodiments of the invention, principle of the present invention and realization will become obvious.It should be noted that the present invention should not be limited to specific embodiment hereinafter described.In addition, in order to for simplicity, the detailed description of known technology unrelated to the invention is eliminated.
Fig. 1 shows the block diagram according to the system for predicting air quality of the present invention.As shown in Figure 1, system 10 comprises determining unit 110, training unit 120 and predicting unit 130.Below, the operation for predicting the unit in the system 10 of air quality is described in detail.
Determining unit 110 determines the adjacent domain in place to be measured according to distance threshold.By definition distance threshold, obtain the contiguous monitoring station in place to be measured.Distance threshold can rely on experience to determine.Distance threshold is larger, then the number of the contiguous monitoring station in place to be measured is more.Fig. 2 shows according to the schematic diagram for determining adjacent domain of the present invention.As shown in Figure 2, P represents place to be measured.After determining distance threshold, monitoring station A, B, C, D that those distance P are less than this distance threshold are seen as the contiguous monitoring station of P, and monitoring station E and F that distance P is greater than this distance threshold is not seen as the contiguous monitoring station of P, therefore the air pollution index of monitoring station E and F is not used in the pollution index of prediction P point.
Training unit 120 trains forecast model to obtain the optimized parameter of forecast model, and this forecast model is based on multiple prediction task and build according to the space characteristics of adjacent domain.In one embodiment of the invention, the space characteristics of adjacent domain can comprise numeric type characteristic sum Boolean type feature.Such as, numeric type feature can comprise with the next item down or more item: wind speed, temperature, humidity, quantity of precipitation that monitoring station place in adjacent domain is measured, and the Distance geometry angle between monitoring station in adjacent domain.Boolean type feature can comprise with the next item down or more item: whether the relative orientation of the monitoring station in adjacent domain meets specified conditions, and whether the relative distance of monitoring station in adjacent domain is greater than threshold value.Below, 3 examples that numeric type characteristic sum Boolean type feature is described by reference to the accompanying drawings.
As shown in Figure 3, from the angle of monitoring station A, A comprises one group of numeric type feature with contiguous monitoring station B.Such as, this numeric type feature can be distance between A and B or angle, and other features (such as distance be multiplied by angle, distance be multiplied by angle and be multiplied by wind speed etc.).Meanwhile, A also comprises the numeric type feature irrelevant with other monitoring stations, such as wind speed, temperature, humidity, quantity of precipitation etc.The angle that it should be noted that between A and B take A as the angle that the wind direction of starting point calculates for benchmark, and span is more than or equal to 0 degree and is less than or equal to 180 degree.
In addition, monitoring station A also has Boolean type feature (i.e. 0-1 feature).This stack features can be used for describing the relation of monitoring station A to other monitoring stations.Such as, Boolean type feature can describe monitoring station B and whether be in the direction such as due east, Zheng Xi, due south, positive north of monitoring station A (such as, for due east feature, when value is the direction, due east that 1 interval scale B is in A, when value is that 0 interval scale B is not or not the direction, due east of A), or whether the distance of monitoring station B and monitoring station A exceedes certain threshold value (such as, the distance being 1 interval scale monitoring station B and monitoring station A when value exceedes certain threshold value, and the distance being 0 interval scale monitoring station B and monitoring station A when value is less than certain threshold value).
Training unit 120, for any two monitoring stations in adjacent domain, calculates space characteristics and the predicted value of any two monitoring stations.Then, training unit 120 is for all monitoring stations in described adjacent domain, according to space characteristics and the predicted value of any two monitoring stations calculated, calculate the described optimized parameter of described forecast model, make from described multiple prediction task on the whole the absolute value sum of difference of the predicted value that obtained by the described forecast model with described optimized parameter and true measurement minimum.Below, shown in composition graphs 2, scene describes the concrete operations of training unit 120.
Table 1 shows the eigenwert relevant with monitoring station B with monitoring station A that training unit 120 calculates.Particularly, the first row of table 1 describes the feature of A-> B, and the second line description is from the feature of B-> A.Last of table 1 is classified as the desired value that each row of data needs prediction.Here, target of prediction is the difference of the pollution index between two monitoring station A and B.
Feature | Distance | Angle | North | Northeast | East | The southeast | South | Southwest | West | Northwest | Difference |
A->B | 23 | 0.52 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 11 |
B->A | 23 | 2.62 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | -11 |
Table 1
Training unit 120 in a similar way, can calculate other monitoring stations eigenwert each other in adjacent domain.Such as, for the scene of Fig. 2, training unit 120 also calculates A-> C, A-> D, B-> C, B-> D, C-> A, C-> B, C-> D, D-> A, D-> B, the eigenwert of D-> C.
In one embodiment, training unit 120 adopts following forecast model:
f
i(X)=λ
11x
1+λ
12x
2+...+λ
inx
n
Wherein, i is the id of task, the air pollutants prediction task that different i is just corresponding different.λ represents the parameter of forecast model, represents often kind of feature to the significance level predicted the outcome.X representation feature value.Here, eigenwert has n item.
F
i(X) treat measured value corresponding to representation feature value vector X, corresponding to y (X) representation eigenvalue vector X, treat the actual value of measured value.In the present embodiment, optimization aim can be: try to achieve one group of λ, make | f (X)-y (X) | and minimum, wherein X={X
1, X
2..., X
1, l be training sample number (for Fig. 2 scene l=12), X
irepresent i-th feature value vector.
Therefore, one group of parameter of finally trying to achieve has to be made
minimum character.Wherein, m is task number, and l is training sample sum.Y
i(X
j) represent vectorial X
jthe actual value of corresponding task i.Wherein Section 1
the model obtained for making training and actual value close.Section 2
for retaining the correlativity between multitask.Its specific formula for calculation can be such as:
Its Middle molecule λ
ijbe the parameter value of task i in a jth feature, denominator is all parameter sums of task i.I.e. λ '
ijbe the normalization mean value of a parameter.Matrix
formula can be such as:
describe the similarity between different task submodel, the similarity namely in matrix often between row.More similar between this similarity larger explanation task.In order to control the impact of this similarity on front portion matching actual value, a factor mu can be introduced to control this weight.
Predicting unit 130 utilizes the forecast model with optimized parameter to predict the air quality in place to be measured.Preferably, predicting unit 130 utilizes the space characteristics of the adjacent domain in place to be measured, calculates the weighted sum of the predicted value of each monitoring station in described adjacent domain, predicts the air quality in place to be measured thus.Wherein, be directly proportional for the inverse of the weight of each monitoring station to the training error of this monitoring station, the true measurement of described training error and this monitoring station is relevant with the difference of predicted value obtained by the forecast model with optimized parameter.
Below, the exemplary scene shown in composition graphs 4, describes the operation of training unit 120 and predicting unit 130 in detail.
As shown in Figure 4, surrounding's existence 4 contiguous monitoring station A, B, C, D of place P to be measured.A place P to be measured and stack features value can be calculated between its all contiguous monitoring station A, B, C, D.Because the actual value of each contiguous monitoring station is all known, by calculated difference and actual value and the estimated value that just can calculate place P to be measured.Merged by the value estimated each point, just obtain final predicting the outcome.
Particularly, if diff
ai () represents the true difference of the air quality of monitoring station A and place i.F
ai () represents the prediction difference of the air quality of monitoring station A and place i, index represents air pollution index.
diff
A(i)=|index(A)-index(i)|
The penalty values of definition monitoring station A is as follows:
Wherein, the penalty values loss (A) of monitoring station A represents that monitoring station A is close to the training error sum of monitoring station to other each.This value is larger, illustrates that the deflection forecast made for benchmark with monitoring station A gets over out of true, thus its when final prediction place P to be measured, proportion just should be less.φ be default on the occasion of, its objective is the generation preventing from being removed by zero.Such as, φ can get certain value between 0 to 1.
The weight of monitoring station A point can be the inverse of the penalty values of monitoring station A and the ratio of summation reciprocal:
Wherein, denominator represents the sum reciprocal of the penalty values of all monitoring stations in adjacent domain.
The final index that predicts the outcome (P) is the weighted sum of each predicted value, as follows:
Wherein, f
i(P) prediction difference of the air quality of place i and place P to be measured is represented.
The whole training process of training unit 120 of arthmetic statement below:
Being described in detail as follows of above-mentioned algorithm:
Input: data matrix X, number of tasks m, eigenwert number n, data instance number l, learning rate η;
Export: model parameter λ
i..., λ
n
Detailed process:
1: all model parameter λ of initialization
i..., λ
n, compose a random value namely to all parameters.
2: m task is circulated
3: the penalty values calculating each task, namely calculate the error between actual value and predicted value.
4: the parameter of n feature is circulated
5: the renewal Grad of trying to achieve each parameter according to error.
6: according to the Grad of trying to achieve, each parameter value is upgraded, total error is diminished.Wherein, learning rate is the step-length of Gradient Descent,
it is the direction of Gradient Descent.
7: end loop
8: end loop
9-11: detect and whether restrain, if model is restrained, finishing iteration process.Criterion is: if the error of the front model of error ratio that the model after upgrading draws is large, model will be restrained.Convergence represents the position that model has reached optimum, then what obtain to other direction findings is not optimum.
12: return model parameter λ
i..., λ
n.
Below, the exemplary scene shown in composition graphs 5, provides the concrete operations example of system 10 for the value of two prediction tasks (task 1:PM2.5 and task 2:PM10).
After setpoint distance threshold value, determining unit 110 determines surrounding's existence 3 contiguous monitoring station A, B, C of place P to be measured, as shown in Figure 5.
Then, training unit 120 extracts the eigenwert (the Boolean type characteristic sum numeric type feature namely above mentioned) of A, B, C 3.Then, training unit 120 extracts the difference of the air pollution index between A, B, C.Like this, the training data in following table 2 is obtained.
Table 2
Training unit 120 makes overall predicated error minimum one group of parameter value (that is, obtaining the optimization model based on A, B, C 3) according to process computation as described above.Afterwards, again training data is input in optimization model, obtains the predicted value last two row of 3 (see the following form) of air quality difference.
Table 3
Afterwards, training unit 120 obtains the training error of each monitoring station, as shown in table 4 below.
Feature | Task 1 (PM2.5) training error | Task 2 (PM10) training error |
A->B,C | 2 | 3 |
B->A,C | 2 | 1 |
C->A,B | 2 | 2 |
Table 4
Further, training unit 120 obtains the weighted value of each monitoring station, as shown in table 5 below.
Monitoring station | Task 1 (PM2.5) weighted value | Task 2 (PM10) weighted value |
A | 1/3 | 2/11 |
B | 1/3 | 6/11 |
C | 1/3 | 3/11 |
Table 5
Then, predicting unit 130 calculates the predicted value (that is, predicting the difference of the air quality between P and other monitoring stations, last two row seen the following form in 6) that geodetic point P is treated in each monitoring station.
Feature | Distance | Angle | North | Northeast | East | The southeast | South | Southwest | West | Northwest | Task 1:PM2.5 | Task 2:PM10 |
A->P | 26 | 0.26 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | -2 | -14 |
B->P | 20 | 1.4 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | -13 | 13 |
C->P | 15 | 0.84 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 8 | -1 |
Table 6
Finally, predicting unit 130 merges these predicted values according to the weighted value generated before, obtains net result as follows:
PM2.5(P)=(2/11)*(-2+37)+(6/11)*(-13+48)+(3/11)*(8+29)=35.54
PM10(P)=(2/11)*(-14+45)+(6/11)*(13+16)+(3/11)*(-1+31)=29.64
The Air Quality Forecast system of the present embodiment, by the modeling of multiple prediction task cooperation, takes full advantage of the relevance between similar tasks, thus improves the degree of accuracy of Air Pollution Forecast.
Fig. 6 shows the process flow diagram of the method for predicting air quality according to the present invention's example embodiment.As shown in Figure 6, method 60 starts in step S610 place.
In step S620, determine the adjacent domain in place to be measured according to distance threshold.
In step S630, training forecast model is to obtain the optimized parameter of forecast model, and wherein forecast model builds according to the space characteristics of adjacent domain based on multiple prediction task.Such as, the space characteristics of adjacent domain can comprise numeric type characteristic sum Boolean type feature.Preferably, numeric type feature can comprise with the next item down or more item: wind speed, temperature, humidity, quantity of precipitation that monitoring station place in adjacent domain is measured, and the Distance geometry angle between monitoring station in adjacent domain.Boolean type feature can comprise with the next item down or more item: whether the relative orientation of the monitoring station in adjacent domain meets specified conditions, and whether the relative distance of monitoring station in adjacent domain is greater than threshold value.
Preferably, for any two monitoring stations in adjacent domain, calculate space characteristics and the predicted value of described any two monitoring stations.Then, for all monitoring stations in adjacent domain, according to space characteristics and the predicted value of any two monitoring stations calculated, the optimized parameter of computational prediction model, make from described multiple prediction task on the whole the absolute value sum of difference of the predicted value that obtained by the forecast model with optimized parameter and true measurement minimum.
In step S640, utilize the forecast model with optimized parameter to predict the air quality in place to be measured.Preferably, the space characteristics of the adjacent domain in place to be measured can be utilized, calculate the weighted sum of the predicted value of each monitoring station in adjacent domain, predict the air quality in place to be measured thus.Preferably, be directly proportional for the inverse of the weight of each monitoring station to the training error of this monitoring station, the true measurement of training error and this monitoring station is relevant with the difference of predicted value obtained by the forecast model with optimized parameter.
Finally, method 60 terminates in step S650 place.
Should be appreciated that, the above embodiment of the present invention can be realized by the combination of both software, hardware or software and hardware.Such as, various assemblies in equipment in above-described embodiment can be realized by multiple device, these devices include but not limited to: mimic channel, digital circuit, general processor, digital signal processing (DSP) circuit, programmable processor, special IC (ASIC), field programmable gate array (FPGA), programmable logic device (PLD) (CPLD), etc.
In addition, those skilled in the art will appreciate that the initial parameter described in the embodiment of the present invention can store in the local database, also can be stored in distributed data base or can be stored in remote data base.
In addition, embodiments of the invention disclosed herein can realize on computer program.More specifically, this computer program is following a kind of product: have computer-readable medium, on computer-readable medium, coding has computer program logic, and when performing on the computing device, this computer program logic provides relevant operation to realize technique scheme of the present invention.When performing at least one processor of computing system, computer program logic makes the operation (method) of processor execution described in the embodiment of the present invention.This set of the present invention is typically provided as Downloadable software image, shared data bank etc. in other media or one or more module arranging or be coded in software, code and/or other data structures on the computer-readable medium of such as light medium (such as CD-ROM), floppy disk or hard disk etc. or the firmware on such as one or more ROM or RAM or PROM chip or microcode.Software or firmware or this configuration can be installed on the computing device, perform technical scheme described by the embodiment of the present invention to make the one or more processors in computing equipment.
Although below show the present invention in conjunction with the preferred embodiments of the present invention, one skilled in the art will appreciate that without departing from the spirit and scope of the present invention, various amendment, replacement and change can be carried out to the present invention.Therefore, the present invention should not limited by above-described embodiment, and should be limited by claims and equivalent thereof.
Claims (14)
1., based on the system for predicting air quality of multiple prediction task, comprising:
Determining unit, is configured to: the adjacent domain determining place to be measured according to distance threshold;
Training unit, be configured to: training forecast model is to obtain the optimized parameter of described forecast model, wherein, described forecast model is based on described multiple prediction task and build according to the space characteristics of described adjacent domain, and the space characteristics of described adjacent domain comprises numeric type characteristic sum Boolean type feature; And
Predicting unit, is configured to: utilize the described forecast model with described optimized parameter to predict the air quality in place to be measured.
2. system according to claim 1, wherein, described numeric type feature comprises with the next item down or more item: wind speed, temperature, humidity, quantity of precipitation that monitoring station place in described adjacent domain is measured, and the Distance geometry angle between monitoring station in described adjacent domain.
3. system according to claim 1, wherein, described Boolean type feature comprises with the next item down or more item: whether the relative orientation of the monitoring station in described adjacent domain meets specified conditions, and whether the relative distance of monitoring station in described adjacent domain is greater than threshold value.
4. system according to claim 1, wherein, described training unit is configured to:
For any two monitoring stations in described adjacent domain, calculate space characteristics and the predicted value of described any two monitoring stations; And
For all monitoring stations in described adjacent domain, according to space characteristics and the predicted value of any two monitoring stations calculated, calculate the described optimized parameter of described forecast model, make from described multiple prediction task on the whole the absolute value sum of difference of the predicted value that obtained by the described forecast model with described optimized parameter and true measurement minimum.
5. system according to claim 4, wherein, described training unit is configured to:
The described optimized parameter of described forecast model is also calculated based on the similarity between each task in described multiple prediction task.
6. system according to claim 1, wherein, described predicting unit is configured to: the space characteristics utilizing the adjacent domain in place to be measured, calculates the weighted sum of the predicted value of each monitoring station in described adjacent domain, predicts the air quality in place to be measured thus.
7. system according to claim 6, wherein, be directly proportional for the inverse of the weight of each monitoring station to the training error of this monitoring station, the true measurement of described training error and this monitoring station is relevant with the difference of predicted value obtained by the described forecast model with described optimized parameter.
8., based on the method for predicting air quality of multiple prediction task, comprising:
The adjacent domain in place to be measured is determined according to distance threshold;
Training forecast model is to obtain the optimized parameter of described forecast model, wherein, described forecast model is based on described multiple prediction task and build according to the space characteristics of described adjacent domain, and the space characteristics of described adjacent domain comprises numeric type characteristic sum Boolean type feature; And
Utilize the described forecast model with described optimized parameter to predict the air quality in place to be measured.
9. method according to claim 8, wherein, described numeric type feature comprises with the next item down or more item: wind speed, temperature, humidity, quantity of precipitation that monitoring station place in described adjacent domain is measured, and the Distance geometry angle between monitoring station in described adjacent domain.
10. method according to claim 8, wherein, described Boolean type feature comprises with the next item down or more item: whether the relative orientation of the monitoring station in described adjacent domain meets specified conditions, and whether the relative distance of monitoring station in described adjacent domain is greater than threshold value.
11. methods according to claim 8, wherein,
For any two monitoring stations in described adjacent domain, calculate space characteristics and the predicted value of described any two monitoring stations; And
For all monitoring stations in described adjacent domain, according to space characteristics and the predicted value of any two monitoring stations calculated, calculate the described optimized parameter of described forecast model, make from described multiple prediction task on the whole the absolute value sum of difference of the predicted value that obtained by the described forecast model with described optimized parameter and true measurement minimum.
12. methods according to claim 11, wherein, also calculate the described optimized parameter of described forecast model based on the similarity between each task in described multiple prediction task.
13. methods according to claim 8, wherein, utilize the space characteristics of the adjacent domain in place to be measured, calculate the weighted sum of the predicted value of each monitoring station in described adjacent domain, predict the air quality in place to be measured thus.
14. methods according to claim 13, wherein, be directly proportional for the inverse of the weight of each monitoring station to the training error of this monitoring station, the true measurement of described training error and this monitoring station is relevant with the difference of predicted value obtained by the described forecast model with described optimized parameter.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410475151.2A CN105488317B (en) | 2014-09-17 | 2014-09-17 | System and method for predicting air quality |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410475151.2A CN105488317B (en) | 2014-09-17 | 2014-09-17 | System and method for predicting air quality |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105488317A true CN105488317A (en) | 2016-04-13 |
CN105488317B CN105488317B (en) | 2018-11-16 |
Family
ID=55675292
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410475151.2A Active CN105488317B (en) | 2014-09-17 | 2014-09-17 | System and method for predicting air quality |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105488317B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106384205A (en) * | 2016-09-30 | 2017-02-08 | 百度在线网络技术(北京)有限公司 | Modeling method used for collecting operation input duration and modeling device thereof |
CN107884511A (en) * | 2016-09-30 | 2018-04-06 | 日本电气株式会社 | For predicting the apparatus and method of IAQ |
CN108537383A (en) * | 2018-04-09 | 2018-09-14 | 山东建筑大学 | A kind of room air prediction technique based on Model Fusion |
CN110210681A (en) * | 2019-06-11 | 2019-09-06 | 西安电子科技大学 | A kind of prediction technique of the monitoring station PM2.5 value based on distance |
CN111144625A (en) * | 2019-12-10 | 2020-05-12 | 北京蛙鸣信息科技发展有限公司 | Air quality prediction method and system based on adjacent space data principal component elements |
CN111222672A (en) * | 2018-11-27 | 2020-06-02 | 百度在线网络技术(北京)有限公司 | Air Quality Index (AQI) prediction method and device |
WO2020135886A1 (en) * | 2018-12-29 | 2020-07-02 | 中科三清科技有限公司 | Air pollutant forecasting method and apparatus, and electronic device |
CN112561199A (en) * | 2020-12-23 | 2021-03-26 | 北京百度网讯科技有限公司 | Weather parameter prediction model training method, weather parameter prediction method and device |
CN112633295A (en) * | 2020-12-22 | 2021-04-09 | 深圳集智数字科技有限公司 | Prediction method and device for loop task, electronic equipment and storage medium |
WO2024077876A1 (en) * | 2022-10-12 | 2024-04-18 | 华院计算技术(上海)股份有限公司 | Adaptation-based local dynamic coke quality prediction method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102192927A (en) * | 2010-11-05 | 2011-09-21 | 重庆大学 | Air-quality monitoring system based on electronic nose technique, and monitoring method thereof |
CN103514366A (en) * | 2013-09-13 | 2014-01-15 | 中南大学 | Urban air quality concentration monitoring missing data recovering method |
CN104008278A (en) * | 2014-05-14 | 2014-08-27 | 昆明理工大学 | PM2.5 concentration prediction method based on feature vectors and least square support vector machine |
-
2014
- 2014-09-17 CN CN201410475151.2A patent/CN105488317B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102192927A (en) * | 2010-11-05 | 2011-09-21 | 重庆大学 | Air-quality monitoring system based on electronic nose technique, and monitoring method thereof |
CN103514366A (en) * | 2013-09-13 | 2014-01-15 | 中南大学 | Urban air quality concentration monitoring missing data recovering method |
CN104008278A (en) * | 2014-05-14 | 2014-08-27 | 昆明理工大学 | PM2.5 concentration prediction method based on feature vectors and least square support vector machine |
Non-Patent Citations (6)
Title |
---|
MARKEY JOHNSON ET AL: "Evaluation of land-use regression models used to predict air quality concentrations in an urban area", 《ATMOSPHERIC ENVIRONMENT》 * |
W.Z.LU ET AL: "Application of evolutionary neural network method in predicting pollutant levels in downtown area of Hong Kong", 《NEUROCOMPUTING》 * |
司瑶冰: "呼和浩特市大气污染物变化特征及空气质量预报方法研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑》 * |
王敏等: "基于BP人工神经网络的城市PM2.5浓度空间预测", 《环境污染与防治》 * |
王玉璟: "空间插值算法的研究及其在空气质量监测中的应用", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
许亚辉等: "基于反距离加权空间插值方法的南京老城区PM2.5的空间分布研究", 《中国科技信息》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106384205A (en) * | 2016-09-30 | 2017-02-08 | 百度在线网络技术(北京)有限公司 | Modeling method used for collecting operation input duration and modeling device thereof |
CN107884511A (en) * | 2016-09-30 | 2018-04-06 | 日本电气株式会社 | For predicting the apparatus and method of IAQ |
CN107884511B (en) * | 2016-09-30 | 2022-01-21 | 日本电气株式会社 | Apparatus and method for predicting indoor air quality |
CN108537383A (en) * | 2018-04-09 | 2018-09-14 | 山东建筑大学 | A kind of room air prediction technique based on Model Fusion |
CN111222672B (en) * | 2018-11-27 | 2023-01-31 | 百度在线网络技术(北京)有限公司 | Air Quality Index (AQI) prediction method and device |
CN111222672A (en) * | 2018-11-27 | 2020-06-02 | 百度在线网络技术(北京)有限公司 | Air Quality Index (AQI) prediction method and device |
WO2020135886A1 (en) * | 2018-12-29 | 2020-07-02 | 中科三清科技有限公司 | Air pollutant forecasting method and apparatus, and electronic device |
CN110210681A (en) * | 2019-06-11 | 2019-09-06 | 西安电子科技大学 | A kind of prediction technique of the monitoring station PM2.5 value based on distance |
CN111144625A (en) * | 2019-12-10 | 2020-05-12 | 北京蛙鸣信息科技发展有限公司 | Air quality prediction method and system based on adjacent space data principal component elements |
CN112633295A (en) * | 2020-12-22 | 2021-04-09 | 深圳集智数字科技有限公司 | Prediction method and device for loop task, electronic equipment and storage medium |
CN112561199A (en) * | 2020-12-23 | 2021-03-26 | 北京百度网讯科技有限公司 | Weather parameter prediction model training method, weather parameter prediction method and device |
CN112561199B (en) * | 2020-12-23 | 2024-06-21 | 北京百度网讯科技有限公司 | Weather parameter prediction model training method, weather parameter prediction method and device |
WO2024077876A1 (en) * | 2022-10-12 | 2024-04-18 | 华院计算技术(上海)股份有限公司 | Adaptation-based local dynamic coke quality prediction method |
Also Published As
Publication number | Publication date |
---|---|
CN105488317B (en) | 2018-11-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105488317A (en) | Air quality prediction system and method | |
CN105488316A (en) | Air quality prediction system and method | |
Xu et al. | Air quality early-warning system for cities in China | |
Dervilis et al. | On robust regression analysis as a means of exploring environmental and operational conditions for SHM data | |
Nguyen et al. | Adaptive response surface method based on a double weighted regression technique | |
Ma et al. | Calibration of microsimulation with heuristic optimization methods | |
Ding et al. | Complex network theory applied to the growth of Kuala Lumpur’s public urban rail transit network | |
Jiang et al. | A novel hybrid strategy for PM2. 5 concentration analysis and prediction | |
Lu et al. | Integrating dynamic Bayesian network and physics-based modeling for risk analysis of a time-dependent power distribution system during hurricanes | |
Abadi et al. | An analysis of the sustainability of basin water resources using Vensim model | |
Xu et al. | Pipe break prediction based on evolutionary data-driven methods with brief recorded data | |
Shahbazi et al. | A novel regression imputation framework for Tehran air pollution monitoring network using outputs from WRF and CAMx models | |
CN107908889B (en) | Multi-objective optimization deployment method for water quality monitoring network with part of monitoring points selected in advance | |
Jeong et al. | Prediction of flexible pavement deterioration in relation to climate change using fuzzy logic | |
Juszczyk et al. | Forecasting of sports fields construction costs aided by ensembles of neural networks | |
Guse et al. | Identifying the connective strength between model parameters and performance criteria | |
Vrachimis et al. | Leakage detection and localization in water distribution systems: A model invalidation approach | |
Chebbi et al. | Optimal extension of rain gauge monitoring network for rainfall intensity and erosivity index interpolation | |
Chen et al. | Probabilistic forecasting of drought: a hidden Markov model aggregated with the RCP 8.5 precipitation projection | |
Lee et al. | In search for modeling predictive control of indoor air quality and ventilation energy demand in subway station | |
Wang et al. | Bayesian approaches for evaluating wind‐resistant performance of long‐span bridges using structural health monitoring data | |
Bertola et al. | A methodology for measurement-system design combining information from static and dynamic excitations for bridge load testing | |
Mostafaei et al. | Fully automated operational modal identification of regular and irregular buildings with ensemble learning | |
Yoon et al. | Optimal decision making in post-hazard bridge recovery strategies for transportation networks after seismic events | |
Mawada et al. | Climate change impacts on the extreme rainfall for selected sites in North Western England |
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 |