CN113657023A - Near-surface ozone concentration inversion method based on combination of machine learning and deep learning - Google Patents

Near-surface ozone concentration inversion method based on combination of machine learning and deep learning Download PDF

Info

Publication number
CN113657023A
CN113657023A CN202110815597.5A CN202110815597A CN113657023A CN 113657023 A CN113657023 A CN 113657023A CN 202110815597 A CN202110815597 A CN 202110815597A CN 113657023 A CN113657023 A CN 113657023A
Authority
CN
China
Prior art keywords
neural network
deep
wide
ozone concentration
data
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
Application number
CN202110815597.5A
Other languages
Chinese (zh)
Inventor
马鹏飞
张玉环
周春艳
张建辉
王玉
张连华
张大为
李巍
赵少华
王中挺
朱海涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Satellite Application Center for Ecology and Environment of MEE
Original Assignee
Satellite Application Center for Ecology and Environment of MEE
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Satellite Application Center for Ecology and Environment of MEE filed Critical Satellite Application Center for Ecology and Environment of MEE
Priority to CN202110815597.5A priority Critical patent/CN113657023A/en
Publication of CN113657023A publication Critical patent/CN113657023A/en
Pending legal-status Critical Current

Links

Images

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
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Abstract

The invention discloses a near-ground ozone concentration inversion method based on combination of machine learning and Deep learning, which comprises the steps of training an inversion model by calling a training sample, and obtaining near-ground ozone concentration information of a corresponding area by inputting low-layer ozone profile in a troposphere obtained by satellite observation data on the basis of training the inversion model with good performance, wherein the inversion model is a W i de & Deep neural network, and specifically comprises a single-layer W i de neural network part and a Deep multi-layer neural network part, and an output layer adopts output of a log i st i cs regression i on comprehensive W i de and Deep part, so that the trained model can simultaneously obtain memory and generalization capabilities.

Description

Near-surface ozone concentration inversion method based on combination of machine learning and deep learning
Technical Field
The invention relates to the technical field of near-surface ozone concentration inversion, in particular to a near-surface ozone concentration inversion method based on combination of machine learning and deep learning.
Background
Ozone monitoring in regional and global areas is a vital social problem, and the overall pollution condition of the region cannot be accurately evaluated due to the fact that ground monitoring sites are rare, and particularly in rural and remote areas, the space coverage range is not enough. The satellite remote sensing can provide ozone concentration information of large regional scales relative to ground monitoring sites, at present, most remote sensing monitoring applications developed in the aspect of environmental assessment directly utilize ozone column concentration to assess urban ozone pollution conditions, and ozone heavily polluted areas caused by artificial emission cannot be accurately positioned, so near-ground ozone concentration monitoring of large regional scales is urgently required to be developed and used as important supplement of ground observation and atmosphere simulation, and the ozone pollution process is more comprehensively known.
For the above reasons, the present inventors have conducted intensive studies on the existing near-surface ozone concentration inversion method, and have awaited designing a near-surface ozone concentration inversion method based on a combination of machine learning and deep learning, which can solve the above problems.
Disclosure of Invention
In order to overcome the problems, the inventor of the present invention has conducted intensive research to design a near-ground ozone concentration inversion method based on a combination of machine learning and Deep learning, the method includes training an inversion model by calling a training sample, and obtaining near-ground ozone concentration information of a corresponding region by inputting a low-layer ozone profile in a troposphere obtained by satellite observation data on the basis of training the inversion model with good performance, wherein the inversion model is a Wide & Deep neural network, and specifically, the inversion model is composed of a shallow (or single-layer) Wide neural network portion and a Deep multi-layer neural network portion, and an output layer adopts logistics regression to synthesize outputs of the Wide and Deep portions, so that the trained model can obtain memory and generalization capabilities at the same time, thereby completing the present invention.
Specifically, the invention aims to provide a near-surface ozone concentration inversion method based on the combination of machine learning and depth learning, which is characterized in that,
the method comprises the following steps:
s1, taking a training sample to train an inversion model;
and S2, taking satellite remote sensing information of the monitored area and inputting the satellite remote sensing information into the inversion model to obtain near-ground ozone concentration information of the monitored area.
Wherein, step S1 includes the following substeps:
s11, establishing a Wide & Deep neural network;
and S12, substituting the training samples into the Wide & Deep neural network to train the Wide & Deep neural network to obtain an inversion model.
Wherein the Wide & Deep neural network comprises a shallow or single-layer Wide neural network part and a Deep multi-layer neural network part;
in step S12, the input data of the Wide neural network portion includes basic feature data and combined features;
in step S12, the input data of the Deep multi-layer neural network portion includes basic feature data;
the basic feature data includes aod055 data, ERA5 feature data, TROPOMI Level2 feature data, and temporal feature data.
The basic characteristic data is obtained by calling satellite data.
Performing characteristic transformation on the basic characteristic data through the following formula (I) to obtain a combined basic characteristic:
Figure BDA0003169920370000031
wherein phi isk(X) denotes a combination base feature, d denotes a feature space, and XiRepresenting the ith basic feature, ckiRepresenting a base feature selection item;
preferably, the combined features as input to the wide neural network portion are obtained from the combined base features by a gbdt model.
In S12, the training sample target value is the ozone concentration value detected by the ground base station;
the output data of the Wide & Deep neural network is as follows:
Figure BDA0003169920370000032
wherein, the
Figure BDA0003169920370000033
Representing the output of the Wide neural network portion,
Figure BDA0003169920370000034
representing the output of the activation function of the last layer of the Deep multi-layer neural network part;
sigma represents a final sigmoid activation function of the Wide & Deep neural network;
b represents a constant.
Wherein, in S2, the satellite remote sensing information includes a low-zone ozone profile in the troposphere.
The invention has the advantages that:
(1) according to the near-ground ozone concentration inversion method based on the combination of machine learning and Deep learning, the method adopts a Wide & Deep combined machine learning model for the first time, has the advantages that the traditional machine learning is convenient to fit artificial experience, and has the self-learning and ultrahigh-dimensional characteristic data processing capabilities of a neural network, so that a near-ground ozone concentration product with a large area scale is obtained, and data support is provided for near-ground ozone pollution cause analysis;
(2) according to the near-surface ozone concentration inversion method based on the combination of machine learning and Deep learning, multiple parameters influencing the near-surface ozone concentration are analyzed and processed through the Wide & Deep combined model, and the high-precision near-surface ozone concentration is obtained through inversion.
Drawings
FIG. 1 shows an ozone concentration information scattergram obtained in an embodiment of the present invention;
fig. 2 shows an ozone concentration information scattergram obtained by the XGBoost method.
Detailed Description
The invention is explained in more detail below with reference to the figures and examples. The features and advantages of the present invention will become more apparent from the description.
The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. While the various aspects of the embodiments are presented in drawings, the drawings are not necessarily drawn to scale unless specifically indicated.
According to the near-surface ozone concentration inversion method based on the combination of machine learning and deep learning, which is provided by the invention, the method comprises the following steps:
s1, taking a training sample to train an inversion model;
and S2, taking satellite remote sensing information of the monitored area and inputting the satellite remote sensing information into the inversion model to obtain near-ground ozone concentration information of the monitored area.
Wherein, step S1 includes the following substeps:
s11, establishing a Wide & Deep neural network;
and S12, substituting the training samples into the Wide & Deep neural network to train the Wide & Deep neural network to obtain an inversion model.
The training sample comprises a tropospheric mid-low ozone profile corresponding to an area, TROPOMI Level2 characteristic data, ERA5 characteristic data and aod055 data for estimating PM 2.5; in addition, near-surface ozone concentration data obtained by a surface base station as a training sample target value is also included.
In a preferred embodiment, the Wide & Deep neural network comprises a shallow or single layer Wide neural network portion and a Deep multi-layer neural network portion;
the shallow layer or single layer Wide neural network part is preferably a single layer neural network, and the transfer function of the shallow layer or single layer Wide neural network part is a single layer activation function, which is specifically as follows:
Figure BDA0003169920370000051
in step S12, the input data of the Wide neural network portion includes basic feature data and combined features;
in step S12, the input data of the Deep multi-layer neural network portion includes basic feature data; namely, the basic feature data needs to be respectively transmitted to the Wide neural network part and the Deep multi-layer neural network part.
The basic feature data comprises aod055 data, ERA5 feature data, TROPOMI Level2 feature data and time feature data, wherein the aod055 data is PM2.5 feature data, namely data for estimating PM 2.5;
wherein the ERA5 characteristic data includes'd 2m','t2m', 'pev', 'str', 'skt', 'sp', 'bld', 'blh', 'tcc', 'r','t', 'u', 'v', 'tp'; the d2m represents the 2m dew point temperature, t2m represents the 2m temperature, pev represents the potential amount of evapotranspiration, str represents the surface thermal radiation, skt represents the air temperature, sp represents the ground pressure, bld represents the boundary layer energy dissipation, blh represents the boundary layer height, tc represents the total cloud cover, r represents the relative humidity, t represents the temperature, u represents the wind in the longitudinal direction, v represents the wind in the latitudinal direction, and tp represents the total precipitation.
The TROPOMI Level2 characteristic data include 'Ozone (O3) total column', 'Ozone (O3) topotherapeutic column', 'Ozone (O3) profile', 'Ozone (O3) topotherapeutic profile'; the Ozone (O3) total column represents total Ozone column amount, the Ozone (O3) topospheric column represents tropospheric Ozone column concentration, the Ozone (O3) profile represents Ozone profile, and the Ozone (O3) topospheric profile represents tropospheric Ozone profile.
The temporal feature data includes 'day _ sin', 'day _ cos', 'month _ sin', 'month _ cos', which represent sine-wise quantized features of time "day", day _ cos represents cosine-vectorized features of time "day", month _ sin represents sine-vectorized features of time "month", and month _ cos represents cosine-vectorized features of time "month".
The basic characteristic data is obtained by calling satellite data.
Preferably, before inputting data into the Wide & Deep neural network, performing characteristic transformation on basic characteristic data to obtain combined basic characteristics, screening out the combined characteristics through a gbdt model, and inputting the obtained combined characteristics and the basic characteristics into the Wide neural network part;
wherein the characteristic transformation is performed by the following formula (one):
Figure BDA0003169920370000061
wherein phi isk(X) denotes the k-th combined basis feature, d denotes the feature space, XiRepresenting the ith basic feature, ckiRepresenting a basic feature selection item, which takes a value of 0 or 1, taking a value of 1 only when k is equal to i, and taking the other itemsA value of 0;
in the application, the combined basic features enumerate all the second-order and third-order combined features by a polynomial combination method, and input into a gbdt model to select the first 16 dimensions with larger weights as the combined features input into the wide neural network part.
Preferably, the combination features in the present application include: 'aod055_ d2m _ str', 'aod055_ d2m', 'aod055_ str _ sp', 'aod055_ sp','d2m _ t2m','d 2m _ pev _ r _ t','t2m _ v', 'pev _ blh', 'pev _ r _ t', 'str _ skt', 'skt _ blh', 'sp _ t', 'bld _ t', 'bld _ blh _ t', 'blh _ r', and 'blh _ t'. The combined signature includes two or three sets of basic signature data, such as the combined signature 'aod055_ d2m _ str', which includes aod055 data, 2m dew point temperature data, and surface thermal radiation data.
The Wide neural network part can obtain the capability of learning shallow layer intersection among the features by inputting the combined features obtained by the feature intersection function, and the shallow layer feature intersection capability further ensures that a Wide & Deep combined model formed by the Wide neural network can learn artificial priori knowledge.
In a preferred embodiment, the Deep multi-layer neural network part is constructed by adding multiple layers of fully-connected networks on the basis of a Deep DNN feedforward neural network;
the multilayer fully-connected network is defined as:
a(l+1)＝f(W(l)a(l)+h(l))
wherein, a(l+1)Represents the L +1 layer output;
a(l)the output of the L-th layer is represented,
h(l)representing bias items, belonging to the target of model training and learning,
W(l)representing parameter items, belonging to the target of model training and learning,
f () represents an activation function, belonging to the goal of model training learning.
In the application, the output of each layer of fully-connected network can be used as the input of the next layer of fully-connected network, so that the ability of Deep crossing among learning features is obtained, and the Deep crossing ensures that a Wide & Deep combined model composed of Deep neural networks has the ability of independently exploring and learning ultrahigh-dimensional feature spaces.
In a preferred embodiment, in S12, the training sample target value is the ozone concentration value obtained by the ground base station;
the output data of the Wide & Deep neural network is as follows:
Figure BDA0003169920370000071
fitting the ozone by the above formula, wherein the output data is the fitted ozone concentration, and the error between the output data and the actual ozone is the target to be reduced in the model iterative learning;
wherein, the
Figure BDA0003169920370000072
Representing the output of the Wide neural network portion, wherein X represents a base feature and φ (X) represents a composite feature; that is to say that the first and second electrodes,
Figure BDA0003169920370000081
represents the output of the Wide neural network portion obtained by the basis feature X and the combined feature phi (X) together,
Figure BDA0003169920370000082
representing the output of the activation function of the last layer of the Deep multi-layer neural network part; wherein a is(lf)Representing the last layer output of a multi-layer fully-connected network.
Sigma represents a final sigmoid activation function of the Wide & Deep neural network;
b represents a constant, is a parameter needing to be learned in model training, and is selected and set according to training data.
Further, preferably, the Wide neural network part and the Deep multi-layer neural network part are independent from each other during training, that is, data are respectively input, output values are respectively obtained after processing, the result of fitting the two output values is used as the output result of the model, and then the model parameters are compared with the actual ozone, and the model parameters are adjusted by using the feedback of the comparison result.
Preferably, the judgment condition of the completion of the training is that the number of training iterations exceeds 1 ten thousand or that fitting ozone information obtained by the training iterations satisfies the following formula:
fitting ozone-actual ozone/actual ozone is less than or equal to 0.05;
the actual ozone is near-surface ozone concentration data obtained through a ground base station.
In a preferred embodiment, in S2, satellite remote sensing information of the monitored area is retrieved, a low-and-low-layer ozone profile in the troposphere of the monitored area is obtained according to the retrieved information, and the low-and-low-layer ozone profile in the troposphere is input into the inverse model obtained in S1, so that the inverse model can output corresponding near-surface ozone concentration information of the monitored area.
Examples
Training an inversion model:
step 1, a training sample is called from satellite data, and the training sample specifically comprises aod055 data, ERA5 characteristic data, TROPOMI Level2 characteristic data, time characteristic data and near-ground ozone concentration data obtained by a ground base station. Part of the training samples are shown in the following table:
Figure BDA0003169920370000091
step 2, establishing a Wide & Deep neural network, wherein input data of the Wide neural network part in the Wide & Deep neural network comprise basic characteristic data and combined characteristics
Input data of a Deep multilayer neural network part in the Wide & Deep neural network comprise basic characteristic data;
performing characteristic transformation on basic characteristic data through a formula (I) below, and screening out combined characteristics through a gbdt model;
Figure BDA0003169920370000092
the transfer function in the Wide neural network portion is:
Figure BDA0003169920370000093
the multi-layer fully-connected network in the Deep multi-layer neural network part is defined as:
a(l+1)＝f(W(l)a(l)+h(l))
wherein, a(l+1)Represents the L +1 layer output;
a(l)the output of the L-th layer is represented,
h(l)a bias term is represented as a function of,
W(l)the items of the parameters are represented by,
f () represents an activation function.
The output data of the Wide & Deep neural network is as follows:
Figure BDA0003169920370000101
wherein, the
Figure BDA0003169920370000102
Represents the output of the Wide neural network portion obtained by the basis feature X and the combined feature phi (X) together,
Figure BDA0003169920370000103
representing the output of the activation function of the last layer of the Deep multi-layer neural network part;
sigma represents Wide&Final sigmoid activation function of Deep neural network, and the solution of the activation function is
Figure BDA0003169920370000104
b represents a constant.
After 1 ten thousand inversion training, completing the training of an inversion model, calling a low-layer ozone profile in a troposphere in the Jingjin Ji area, and inputting the low-layer ozone profile into the inversion model to obtain corresponding near-surface ozone concentration information of the area.
The ozone concentration data obtained by the ground observation station in the area is called, and two groups of ozone concentration data are compared to obtain the ozone concentration situation of the near-ground ozone concentration inversion method based on the combination of machine learning and depth learning, which is provided by the application, is shown in fig. 1, wherein fig. 1 is a scatter diagram, each point is a group of ozone concentration information, the abscissa is the ozone concentration obtained by inversion, the ordinate is the ozone concentration observed by the observation station, and R2 in the diagram reaches 0.9.
Using the same data as in the above example, the concentration of ozone obtained by the conventional method, i.e., XGBoost, is shown in fig. 2, and it is known that R2 is 0.74.
The method has the advantages that the accuracy rate is obviously improved.
The present invention has been described above in connection with preferred embodiments, which are merely exemplary and illustrative. On the basis of the above, the invention can be subjected to various substitutions and modifications, and the substitutions and the modifications are all within the protection scope of the invention.

Claims (7)

1. A near-surface ozone concentration inversion method based on the combination of machine learning and deep learning is characterized in that,
the method comprises the following steps:
s1, taking a training sample to train an inversion model;
and S2, taking satellite remote sensing information of the monitored area and inputting the satellite remote sensing information into the inversion model to obtain near-ground ozone concentration information of the monitored area.
2. The near-surface ozone concentration inversion method based on the combination of machine learning and deep learning according to claim 1,
step S1 includes the following substeps:
s11, establishing a Wide & Deep neural network;
and S12, substituting the training samples into the Wide & Deep neural network to train the Wide & Deep neural network to obtain an inversion model.
3. The near-surface ozone concentration inversion method based on the combination of machine learning and deep learning according to claim 2,
the Wide & Deep neural network comprises a shallow or single-layer Wide neural network part and a Deep multi-layer neural network part;
in step S12, the input data of the Wide neural network portion includes basic feature data and combined features;
in step S12, the input data of the Deep multi-layer neural network portion includes basic feature data;
the basic feature data includes aod055 data, ERA5 feature data, TROPOMI Level2 feature data, and temporal feature data.
4. The near-surface ozone concentration inversion method based on the combination of machine learning and deep learning according to claim 3,
the basic characteristic data is obtained by calling satellite data.
5. The near-surface ozone concentration inversion method based on the combination of machine learning and deep learning according to claim 2,
performing characteristic transformation on the basic characteristic data through the following formula (I) to obtain a combined basic characteristic:
Figure FDA0003169920360000021
wherein phi isk(X) denotes a combination base feature, d denotes a feature space, and XiRepresenting the ith basic feature, ckiRepresenting basic featuresSelecting an item;
preferably, the combined features as inputs to the wide neural network portion are obtained from the combined base features by a gbdt model.
6. The near-surface ozone concentration inversion method based on the combination of machine learning and deep learning according to claim 2,
in S12, the training sample target value is the ozone concentration value detected by the ground base station;
the output data of the Wide & Deep neural network is as follows:
Figure FDA0003169920360000022
wherein, the
Figure FDA0003169920360000023
Representing the output of the Wide neural network portion,
Figure FDA0003169920360000024
representing the output of the activation function of the last layer of the Deep multi-layer neural network part;
sigma represents a final sigmoid activation function of the Wide & Deep neural network;
b represents a constant.
7. The near-surface ozone concentration inversion method based on the combination of machine learning and deep learning according to claim 1,
at S2, the satellite telemetry information includes a low-zone ozone profile in the troposphere.
CN202110815597.5A 2021-07-19 2021-07-19 Near-surface ozone concentration inversion method based on combination of machine learning and deep learning Pending CN113657023A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110815597.5A CN113657023A (en) 2021-07-19 2021-07-19 Near-surface ozone concentration inversion method based on combination of machine learning and deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110815597.5A CN113657023A (en) 2021-07-19 2021-07-19 Near-surface ozone concentration inversion method based on combination of machine learning and deep learning

Publications (1)

Publication Number Publication Date
CN113657023A true CN113657023A (en) 2021-11-16

Family

ID=78489625

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110815597.5A Pending CN113657023A (en) 2021-07-19 2021-07-19 Near-surface ozone concentration inversion method based on combination of machine learning and deep learning

Country Status (1)

Country Link
CN (1) CN113657023A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114357894A (en) * 2022-03-11 2022-04-15 阿里巴巴达摩院(杭州)科技有限公司 Atmospheric pollutant processing method, storage medium and computer terminal
CN117236528A (en) * 2023-11-15 2023-12-15 成都信息工程大学 Ozone concentration forecasting method and system based on combined model and factor screening

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112052627A (en) * 2020-08-21 2020-12-08 海南星瞰信息咨询中心(有限合伙) Method, device, medium and equipment for estimating near-surface ozone space distribution
CN112163375A (en) * 2020-09-28 2021-01-01 中国科学院空天信息创新研究院 Long-time sequence near-surface ozone inversion method based on neural network
CN112579922A (en) * 2020-12-24 2021-03-30 东北大学 Interest point recommendation method based on user track

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112052627A (en) * 2020-08-21 2020-12-08 海南星瞰信息咨询中心(有限合伙) Method, device, medium and equipment for estimating near-surface ozone space distribution
CN112163375A (en) * 2020-09-28 2021-01-01 中国科学院空天信息创新研究院 Long-time sequence near-surface ozone inversion method based on neural network
CN112579922A (en) * 2020-12-24 2021-03-30 东北大学 Interest point recommendation method based on user track

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
HENG-TZE CHENG等: "Wide & Deep Learning for Recommender Systems", 《PROCEEDINGS OF THE 1ST WORKSHOP ON DEEP LEARNING FOR RECOMMENDER SYSTEMS》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114357894A (en) * 2022-03-11 2022-04-15 阿里巴巴达摩院(杭州)科技有限公司 Atmospheric pollutant processing method, storage medium and computer terminal
CN117236528A (en) * 2023-11-15 2023-12-15 成都信息工程大学 Ozone concentration forecasting method and system based on combined model and factor screening
CN117236528B (en) * 2023-11-15 2024-01-23 成都信息工程大学 Ozone concentration forecasting method and system based on combined model and factor screening

Similar Documents

Publication Publication Date Title
CN109492830B (en) Mobile pollution source emission concentration prediction method based on time-space deep learning
CN106920007B (en) PM based on second-order self-organizing fuzzy neural network2.5Intelligent prediction method
CN104616060B (en) Insulator dirty degree Forecasting Methodology based on BP neural network and fuzzy logic
Dedovic et al. Forecasting PM10 concentrations using neural networks and system for improving air quality
CN107766879A (en) The MLP electric network fault cause diagnosis methods of feature based information extraction
CN113657023A (en) Near-surface ozone concentration inversion method based on combination of machine learning and deep learning
CN112330050A (en) Power system load prediction method considering multiple features based on double-layer XGboost
CN105844334B (en) A kind of temperature interpolation method based on radial base neural net
Assas et al. Use of the artificial neural network and meteorological data for predicting daily global solar radiation in Djelfa, Algeria
CN114694767B (en) PM2.5 concentration prediction method based on space-time diagram ordinary differential equation network
Oh et al. Predicting Concentration of PM10 Using Optimal Parameters of Deep Neural Network.
Boutahir et al. Machine learning and deep learning applications for solar radiation predictions review: morocco as a case of study
CN116013426A (en) Site ozone concentration prediction method with high space-time resolution
CN115438582A (en) PM2.5 concentration prediction method combining multiple elements and attention
CN113011455B (en) Air quality prediction SVM model construction method
Nazeri Tahroudi et al. Application of copula‐based approach as a new data‐driven model for downscaling the mean daily temperature
CN107016440A (en) The multiresolution deep neural network intelligent diagnosing method of machine driving failure
Benmouiza Hourly solar irradiation forecast using hybrid local gravitational clustering and group method of data handling methods
CN113108949B (en) Model fusion-based sonde temperature sensor error prediction method
Bani-Hani et al. Prediction of energy gains from Jordanian wind stations using artificial neural network
Meena et al. Constructing the downscale precipitation using ANN model over the Kshipra river basin, Madhya Pradesh
Chen et al. A comparative study on multi-regression analysis and BP neural network of PM2. 5 index
CN111062118A (en) Multilayer soft measurement modeling system and method based on neural network prediction layering
Willard Machine Learning Techniques for Time Series Regression in Unmonitored Environmental Systems
Shambuyi et al. ANN-Based Model for Daily Solar Radiation Prediction with A Low Number of Hidden Neurons And Optimal Inputs

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination