CN109214592A - A kind of Air Quality Forecast method of the deep learning of multi-model fusion - Google Patents
A kind of Air Quality Forecast method of the deep learning of multi-model fusion Download PDFInfo
- Publication number
- CN109214592A CN109214592A CN201811210072.3A CN201811210072A CN109214592A CN 109214592 A CN109214592 A CN 109214592A CN 201811210072 A CN201811210072 A CN 201811210072A CN 109214592 A CN109214592 A CN 109214592A
- Authority
- CN
- China
- Prior art keywords
- air quality
- model
- data
- quality data
- deep learning
- 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
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000013135 deep learning Methods 0.000 title claims abstract description 14
- 230000004927 fusion Effects 0.000 title claims abstract description 14
- 238000012360 testing method Methods 0.000 claims abstract description 27
- 230000007246 mechanism Effects 0.000 claims abstract description 15
- 238000013136 deep learning model Methods 0.000 claims abstract description 4
- 238000013528 artificial neural network Methods 0.000 claims description 18
- 241001269238 Data Species 0.000 claims description 6
- 238000013459 approach Methods 0.000 claims description 2
- 238000013499 data model Methods 0.000 claims 1
- 238000012549 training Methods 0.000 description 22
- 230000006870 function Effects 0.000 description 15
- 230000001537 neural effect Effects 0.000 description 6
- 230000000306 recurrent effect Effects 0.000 description 6
- 230000006403 short-term memory Effects 0.000 description 5
- 238000003915 air pollution Methods 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000007476 Maximum Likelihood Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 208000027418 Wounds and injury Diseases 0.000 description 2
- 239000000809 air pollutant Substances 0.000 description 2
- 231100001243 air pollutant Toxicity 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000006378 damage Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 208000014674 injury Diseases 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 210000004218 nerve net Anatomy 0.000 description 2
- 239000008188 pellet Substances 0.000 description 2
- 230000002123 temporal effect Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 101100149325 Escherichia coli (strain K12) setC gene Proteins 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000002844 melting Methods 0.000 description 1
- 230000008018 melting Effects 0.000 description 1
- QSHDDOUJBYECFT-UHFFFAOYSA-N mercury Chemical compound [Hg] QSHDDOUJBYECFT-UHFFFAOYSA-N 0.000 description 1
- 229910052753 mercury Inorganic materials 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000000465 moulding Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012887 quadratic function Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- General Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Game Theory and Decision Science (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Entrepreneurship & Innovation (AREA)
- Biomedical Technology (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Educational Administration (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention discloses a kind of Air Quality Forecast method of the deep learning of multi-model fusion, one, acquisition history air quality data and meteorological data;Two, missing values interpolation and normalized are carried out to history air quality data;Three, using deep learning model of the history air quality data building based on seq2seq as single factor test prediction model;Four, the seq2seq deep learning model based on double attention mechanism is constructed as multi-factor Estimation Model using history air quality data and meteorological data;Five, single factor test prediction model prediction result, multi-factor Estimation Model prediction result and the current weather data of air quality data are dissolved into xgboost boosted tree and carry out recurrence calculating, the predicted value of final air quality data is obtained, the method for the present invention can be improved the precision of prediction of air quality data.
Description
Technical field
The invention belongs to the crossing domains of Computer Subject and Environmental Science, and in particular to a kind of depth of multi-model fusion
The Air Quality Forecast method of study.
Background technique
Since social progress recent years, rapid industrial development result in a large amount of severe environmental pollution problem, especially
That air pollution is aggravating always, the pellet in air is more and more, the air quality that people increasingly live by
Gradually decline.In aerial pellet, category PM2.5 and PM10 is the most serious, not only generates to air quality very big
It influences, can also will cause very big injury to human body.In order to reduce injury of the air pollutants to human body, air quality is divided
Analysis has important practical significance with prediction.
For this problem of Air Quality Forecast, the researcher that there are many recent years is made that a large amount of contribution.He
Combine newest thought and new technology to realize quantitative research, make it possible to the reason of finding out air pollution and air quality
Basic trend.The wherein Effects of Factors there are many air qualities, and there is many uncertainty and complexity.Conventional
Artificial neural network, backpropagation (Back Propagation, BP) neural network and regression prediction method are not enough to accurately
Prediction, it is also difficult to learn its internal rule.
Wang Xin etc. is using LSTM (long short term memory network) Recognition with Recurrent Neural Network to fault time
Sequence is predicted;It wears Li Jie etc. and carries out PM2.5 short-term concentration dynamic forecasting using machine learning;Niu Yuxia etc. is calculated using heredity
Method and backpropagation (Back Propagation, BP) neural network carry out Air Quality Forecast;Model completes Xiang etc. and uses circulation mind
Air pollution space-time forecasting model is established through network (Recurrent Neural Network, RNN);Zheng Yi etc. establishes depth letter
Read the PM2.5 prediction model of network;It opens day etc. and carries out Air Quality Forecast using BP neural network.
Since monitoring station is when obtaining air quality data, generally because equipment fault or network delay Caton etc. are asked
Topic, causes air quality data more missing values occur.On traditional data prediction, especially missing values are handled, and are filled out
Mending the most of method of missing values is the methods of deletion, averaging method and neighbouring method, and the precision filled up is poor, however air quality
Data band has timing information, and it is poor to result in sequence data precision in training.Prediction model uses traditional list in the choice
One model, it may appear that the problems such as over-fitting or low precision, compared to the engineerings such as artificial neural network and traditional support vector machines
Model is practised, the prediction model based on deep learning can be in terms of the feature of capture data and precision of prediction preferably.For sequence
There are timing information in data, the extraction that traditional machine learning method is difficult timing information therein, and use deep learning
Technology can extract temporal aspect, to improve the precision of prediction.
Summary of the invention
In view of this, the present invention provides a kind of Air Quality Forecast method of the deep learning of multi-model fusion, it can
Improve the precision of prediction of air quality data.
Realize that technical scheme is as follows:
A kind of Air Quality Forecast method of the deep learning of multi-model fusion, comprising the following steps:
Step 1: obtaining history air quality data and meteorological data;
Step 2: carrying out missing values interpolation and normalized to history air quality data;
Step 3: being based on the depth of seq2seq (Sequence to Sequence) using the building of history air quality data
Learning model is spent as single factor test prediction model;
Step 4: using history air quality data and meteorological data building based on double attention (attention) mechanism
Seq2seq deep learning model as multi-factor Estimation Model;
Step 5: history air quality data input single factor test prediction model is obtained the prediction knot of single factor test prediction model
History air quality data and meteorological data input multi-factor Estimation Model are obtained the prediction knot of multi-factor Estimation Model by fruit
The prediction result of single factor test prediction model, the prediction result of multi-factor Estimation Model and current weather data are dissolved by fruit
Recurrence calculating is carried out in xgboost boosted tree, obtains the predicted value of final air quality data.
Further, air quality data includes PM2.5, PM10, NO2、CO、O3And SO2It is more than middle a kind of or two classes dense
Degree evidence.
Further, when air quality data is the concentration data of two classes or more, in step 5, air matter is successively predicted
One kind in data is measured, by the prediction result of the single factor test prediction model of such air quality data and multi-factor Estimation Model
Prediction result and current weather data, which are dissolved into xgboost boosted tree, carries out recurrence calculating, obtains such air quality number
According to predicted value.
Further, meteorological data includes temperature, air pressure, humidity, wind direction, wind speed and weather index.
Further, missing values interpolation is carried out using desired maximization approach.
Further, the input layer of single factor test prediction model is history air quality data, and the structure of hidden layer is yes
Seq2seq model, inside are coding and decoding structures, and coding and decoding unit therein is LSTM long short-term memory nerve net
Network, output layer are Air Quality Forecast value.
Further, the input layer of multi-factor Estimation Model is history air quality data and meteorological data, hidden layer knot
Structure be a kind of double attention mechanism seq2seq model, inside be a kind of coding-decoding (encoder-decoder) structure,
It is separately added into one layer of attention mechanism before coding (encoder) and decoding (decoder), output layer is air quality data
Predicted value.
The utility model has the advantages that
1, the present invention is in the building of single factor test prediction model and multi-factor Estimation Model, for the phase for influencing air quality
It closes the factor and makes analysis, then extract important feature, the input as model.
2, the present invention has used the seq2seq model of double attention mechanism in the selection of model, is to circulation nerve net
The extension of network (RNN), can be good at processing sequence data, can be right by the self-teaching of neural network rule therein
Following trend is made prediction.
3, the present invention is directed to the complexity of Air Quality Forecast, has merged the prediction result of multi-model, and common cooperation is to ask
Optimal solution is solved, to acquire optimal predicted value, completes the Accurate Prediction to air quality.
4, present invention incorporates the special self attributes of air quality, using EM EM algorithm to air quality number
According to Missing Data Filling is carried out, it can be good at the distribution of fitting data.
Detailed description of the invention
Fig. 1 is model structure of the present invention.
Fig. 2 is single factor test prediction model figure of the present invention.
Fig. 3 is multi-factor Estimation Model figure of the present invention.
Specific embodiment
The present invention will now be described in detail with reference to the accompanying drawings and examples.
The present invention needs to model air data, very important when before modeling to the selection of characteristic, especially
It is air quality data, the influence being related on space-time, such as time upper daily commuter time section, the week in each week
End, several days of the front and back of every month, each season (summer and winter air pollution level are different).Influence spatially simultaneously, than
As each monitoring station will receive the influence of the weather conditions of surrounding area, the influence of wind direction, the influence of humidity.Therefore, exist respectively
Model is established on time and Spatial Dimension, establishes unifactor model on time dimension first, is chosen single factor test (such as PM2.5) and is made
For the input of unifactor model, temporal trend is found out, obtains local rule;Then multifactor mould is established on Spatial Dimension
Type chooses the input of multifactor (surrounding area weather conditions etc.) as Multiple-Factor Model, obtains global rule;Finally using melting
Molding type combines the weather conditions of current time using the output of unifactor model and Multiple-Factor Model as input, calculates power
Weight, obtains optimal predicted value.As shown in Figure 1, the specific steps are as follows:
Step 1: obtaining historical weather data, including PM2.5 hourly, PM10, NO2, CO, O3, SO2, temperature, gas
Pressure, humidity, wind direction, wind speed, weather index, wherein temperature unit: DEG C, the barometric millimeter of mercury: hundred pas (hPa), humidity unit: percentage
Number (%), wind direction unit: the north starts clockwise angle definition, wind speed unit: m/s, weather index: fine day rains, cloudy
Deng.
Step 2: being pre-processed to historical data
2.1, since air quality data is all by outdoor sensor come in special circumstances or artificial feelings
Under condition, sensors for data inaccuracy or sensor stop collecting data, and air quality data is caused to have missing values.Cause
This needs to carry out missing values interpolation or completion to historical data when doing Air Quality Forecast.
Missing value complement is carried out using expectation maximization (expectation maximization algorithm, EM) method
Entirely.EM algorithm is the maximum likelihood and maximum a posteriori estimate to the probability parameter model containing hidden variable, wherein being handle
Air quality data is maximized as hidden variable, and according to likelihood formula, obtains the parameter expression of model, is then taken true
Data are iterated solution wherein parameter, finally calculate missing values.
2.2, historical data is normalized
Wherein, X is initial data, XminThat indicate is the minimum value in initial data, XmaxWhat is indicated is in initial data
Maximum value.
2.3, the historical data handled is proportionally divided, is divided into 80% training set and 20% test set.
The present invention carries out interpolation processing to missing values using greatest hope EM algorithm, and EM algorithm has time series data relatively good
Capability of fitting, very close to the distribution of initial data.
Step 3: building single factor test prediction model: model structure is divided into input layer, hidden layer, output layer, wherein input layer
Data be PM2.5, PM10, NO2、CO、O3, SO2Six kinds of concentration values are K × 6 as training data, the dimension of input layer.Output
Layer is PM2.5, PM10, NO2、CO、O3, SO2The predicted value of concentration, the dimension of output layer are 6 × T.
Wherein K indicate choose length of time series, that is, choose K time frame air quality data (PM2.5, PM10,
NO2、CO、O3, SO2), such as K=72h (hour).T indicates the air quality data in future time instance, that is, indicates prediction future T
The air quality data at moment, such as T=48h (hour).
Unifactor model hidden layer structure: this hidden layer uses seq2seq (Sequence to Sequence) model, makes
With coding-decoding (encoder-decoder) structure, wherein coding encoder structure is the long short-term memory an of multilayer
LSTM (Long Short-Term Memory) neural unit can calculate a context after coding (encoder)
(context) vector, for expressing the important feature of input data.Decoding (decoder) is partially mainly used to decode, and structure is
The LSTM unit of multilayer.
Step 4: building multi-factor Estimation Model: model structure is divided into input layer, hidden layer, output layer, wherein input layer
Data be entire historical weather data, including PM2.5, PM10, NO2、CO、O3、SO2, temperature, air pressure, humidity, wind direction, wind
12 kinds of data such as speed, weather index, the dimension of input layer are K × 12, and wherein K indicates the length of time series chosen, such as K=
72h (hour).Output layer is air quality data predicted value, and dimension is 6 × T, T indicates the air quality data of future time instance,
Including PM2.5, PM10, NO2、CO、O3、SO2Six kinds of concentration datas, such as T=48h (hour).
The hidden layer structure of Multiple-Factor Model: this model is a kind of seq2seq (the Sequence to of double attention mechanism
Sequence) model, inside be a kind of coding-decoding (encoder-decoder) structure, before the coding be added one layer note
Power (attention) mechanism of meaning, to obtain the contextual information before coding, the part coding encoder later is more than one
LSTM (Long Short-Term Memory) neural unit of layer can calculate one after coding (encoder)
Context vector, for expressing the important feature of input data.One layer of attention is also added before decoding at this time
(attention) mechanism.Decoding (decoder) is partially mainly used to export last result phase, and structure is the LSTM of multilayer
Unit.
Step 5: fusion forecasting model: use xgboost (eXtreme Gradient Boosting) boosted tree as
Training pattern successively trains PM2.5, PM10, NO2、CO、O3And SO2One of which in six kinds of air quality concentration datas, often
When secondary trained, the input of data is single factor test prediction model and the Multiple-Factor Model prediction of a certain air quality concentration data
Prediction result, while the weather conditions at nearest moment are also inputted, including temperature, air pressure, humidity, wind direction, wind speed, weather index
Corresponding labeled data is current each air quality data (PM2.5, PM10, NO when Deng, training2、CO、O3、SO2), such as
Training when input data be PM2.5 (unifactor model result), PM2.5 (Multiple-Factor Model result), temperature, air pressure, humidity,
Eight kinds of data characteristicses such as wind direction, wind speed, weather index (six kinds of current weather conditions), then labeled data is that current PM2.5 is dense
Degree evidence, and so on, remaining five class (PM10, NO can be obtained2、CO、O3、SO2) input data and labeled data, every
A kind of input data and labeled data, is input to xgboost boosted tree.Xgboost model training when inside establish n
Return subtree, that is, have n tree node, and each tree node has multiple leaf nodes, wherein leaf node be single factor test with it is more
Eight kinds of characteristic values of the inputs such as the predicted value of factor prediction model and nearest moment meteorological data, and each tree node is corresponding with
Which leaf node will the weight according to existing for eight characteristic values of input determined, the principle of node split be so that
Select the maximum feature of information gain as split point, the leaf node of subtree each in this way can have different degree (weight
Value), the training of each round adjusts leaf node weighted value according to objective function, forms new subtree, reaches after successive ignition
Optimal subtree, each children tree nodes are predicted values, and all children tree nodes predicted values are all added up, air can be obtained
Quality predictions.After training input data and labeled data input xgboost, it is pre- that six air qualities can be respectively obtained
Measured value, including PM2.5, PM10, NO2、CO、O3And SO2Six kinds of concentration datas.
Prediction result assessment
In order to which the result finally predicted is evaluated and analyzed, obtained prediction model is carried out using test set data
Detection and evaluation are used as evaluation index, root-mean-square error using root-mean-square error (root mean square error, RMSE)
Formula as shown by the equation:
Wherein,That indicate is predicted value, yiWhat is indicated is original value, and n indicates the sum of original value.
The unicity of evaluation index in order to prevent simultaneously, uses R-square as the evaluation index of regression analysis, wherein
The formula of R-square is as shown by the equation:
Wherein,That indicate is predicted value, yiWhat is indicated is original value,What is indicated is average value, and n indicates original value
Sum.
Embodiment 1
Technical solution in order to preferably explain the present invention chooses Beijing 36 air quality monitoring station in the present invention
The data of data and 18 weather stations, and sufficient sample data is obtained, specific implementation steps of the invention are as follows:
Step 1: obtaining data
The air quality data that the present invention uses is to have collected Beijing's in the January, 2017 of counting per hour in January, 2018
According to wherein air quality data mainly includes following several important air pollutants: PM2.5 (μ g/m3), PM10 (μ g/m3),
NO2(μ g/m3), CO (mg/m3), O3(μ g/m3) and SO2(μg/m3).And there are also weather meteorological datas, mainly include weather,
Temperature, air pressure, humidity, wind speed, wind direction.Wherein weather mainly includes fine day, snowy day, cloudy day, light rain, heavy rain, sand etc.;Temperature
It is the temperature value that weather station monitors, unit is degree Celsius;Air pressure refers to atmospheric pressure, and unit is hundred pas (hPa);Humidity refers in air
Vapor content, unit is percentage (%);Wind speed refers to that the speed for the wind that weather station monitors, unit are metre per second (m/s) (m/
s);Wind direction refers to the source of wind, such as north wind refers to the wind blown from north orientation south, wind direction by since the north clockwise angle determine
Justice.Such as the wind blown from south, wind direction is 180 degree, and the wind blown from east, wind direction is 90 degree.
Step 2: data prediction
For history air quality data, can inevitably there be monitoring instrument and go wrong, final data is caused to lack,
It enters data into before prediction model, first carries out the interpolation processing of missing values, the method for the interpolation that the present invention uses is maximum
It is expected that (EM) method, asks maximal possibility estimation and M to walk in sample and asks maximization to maximum likelihood result comprising E step.Given instruction
Practicing sample is { x1..., xm, independent between sample, target will find the implicit classification z of each sample, so that maximum likelihood function p
(x, z) is maximum.
It is to calculate log-likelihood function expectation for E step, maximal possibility estimation is carried out to hidden variable:
Qi(z(i)) :=p (z(i)|x(i);θ)
Wherein, QiTo give x(i)With parameter θ about z(i)Posterior probability, solve and how to select Q (z) that likelihood can be made
The function problem equal with its lower bound, is exactly to enable Q in facti(z(i)) it is z(i)Posterior probability.
It is to maximize the expected result of E step likelihood function for M step, it is newest to obtain new desired value
Desired value can be applied to missing values, carries out missing values and fills up.Q is selected in E stepiSeek the infimum of log-likelihood function:
E step and M walk continuous circulating repetition, until convergence.
Data after interpolation processing are normalized using min-max method for normalizing, and normalization can be fine
Ground reduces data in a certain range, and can be with the convergence rate of nondimensionalization, quickening model solution.The present invention is number
According to diminution between [0,1], normalized formula are as follows:
Wherein, X is original weather data, XminThat indicate is the minimum value in original weather data, XmaxWhat is indicated is former
Maximum value in beginning weather data.
The data that pretreatment is completed are divided, training dataset and test data set, ratio 8:2 are divided into.
Step 3: construction single factor test prediction model
Single factor test prediction model input data is PM2.5, PM10, NO2、CO、O3And SO2Six kinds of concentration as training data,
The dimension of input layer is K × 6, and K indicates the length of time series chosen.
Unifactor model uses seq2seq model, and internal structure is coding-decoding (encoder-decoder), gives
Sequence x={ x1,x2,…,xTCoded portion formula specifically:
ht=f (ht-1,xt)
Wherein, t is current time, htFor in the hidden state of t moment, f is LSTM encoder.
After encoder carries out coding completion, all htGroup is combined into vector c, as the input of decoder, lsb decoder
The formula divided specifically:
h(t)=f (h(t-1),y(t-1),c)
p(yt|y(t-1),y(t-2),...y(1), c) and=g (h(t),y(t-1),c)
Wherein, h(t)For the output of encoder, c is the last one state in encoder, and f and g are nonlinear activation function,
It is set as LSTM neural network unit.
Illustraton of model is partially made of 2 layers of Recognition with Recurrent Neural Network (RNN) as shown in Fig. 2, wherein encoding (encoder), each
Layer 64 LSTM neural unit of setting.And decode (decoder) and be partially made of 2 layers of Recognition with Recurrent Neural Network (RNN), each layer
64 LSTM neural units are set.
The training parameter setting of model is as shown in table 1:
1 model training parameter of table
Step 4: construction multi-factor Estimation Model
The input of multi-factor Estimation Model is entire historical weather data, including PM2.5, PM10, NO2、CO、O3、SO2, temperature
12 kinds of data such as degree, air pressure, humidity, wind direction, wind speed, weather index, the dimension of input layer are K × 12, and wherein K indicates selection
Length of time series.
Multiple-Factor Model is to improve it to the improvement of seq2seq model using the seq2seq model of double attention mechanism
The influence to other meteorologic factors to air quality can preferably be learnt afterwards, key message can be grabbed, to preferably learn
The sequential relationship between data is practised, there is stronger robustness.
Its encoder section gives K sequence dataA feedforward neural network is constructed,
Attention mechanism is added, specific formula is as follows:
Wherein ht-1And st-1For last hidden state and neuron
State, w and u are learning parameter.
Ensured using softmax function allThe sum of weight be equal to 1,
Wherein,It is attention weight of the k sequence in t moment, there are these to pay attention to weight, we can be adaptively
It extracts
Next it is encoded using the LSTM neural network of multilayer
Wherein f is LSTM neural network unit,It is the new sequence with attention weight, htFor the hiding shape of t moment
State.
The state h obtained after being encoded for encodert, attention mechanism is added, as follows:
Ensured using softmax function allThe sum of weight be equal to 1, it is as follows:
WhereinIt is the weight of i moment encoder output state.
Attention weight and encoder hidden state { h1,h2,…hTWeighted sum as context vector ct:
Its decoder section uses the LSTM of multilayer as decoder, once obtain weighted sum context vector c, so that it may
With by they and given target sequence (y1,y2,…,yT-1) combine:
WhereinWithIt is related to the size of decoder input, newlyIt is exactly new decoder states, state value is defeated
Enter in decoder, as follows:
Wherein g is LSTM neural network unit, dtFor last output state, as decoder section.
Illustraton of model is as shown in figure 3, be wherein added attention mechanism (attention) before coding (encoder), and encode
Part is made of 2 layers of Recognition with Recurrent Neural Network (RNN), 64 LSTM neural units of each layer of setting.Attention is added before decoding
Mechanism (attention), and decode (decoder) and be partially made of 2 layers of Recognition with Recurrent Neural Network (RNN), each layer is also provided with 64
It is a
LSTM neural unit.
Step 5: consolidated forecast model
Regression forecasting is carried out using xgboost (eXtreme Gradient Boosting) boosted tree, input is Dan Yin
The result of prime model and the result of Multiple-Factor Model and the meteorological condition at current time, specifically unifactor model is predicted
It as a result is p={ p1,p2…p6, the result of Multiple-Factor Model is z={ z1,z2…z6, p and z represent PM2.5, PM10, NO2,
CO, O3And SO2Six kinds of concentration datas, the meteorological condition g={ g at current time1, g2..., g6, g represents the temperature at nearest moment,
Air pressure, humidity, wind direction, wind speed, six kinds of data of weather index, in training or prediction, the data that are inputted in xgboost boosted tree
For xi={ pi,zi, g }, i is any one in 1~6, the y in trainingiIt is the current air mass data of mark, including
PM2.5, PM10, NO2, CO, O3, SO2, and the y in test or predictioniThe air quality data as predicted, such as x1=
{ pm2.5 (unifactor model result), pm2.5 (Multiple-Factor Model result), temperature, air pressure, humidity, wind direction, wind speed, weather refer to
Mark (weather conditions at six nearest moment), y1={ pm2.5 (current value in training set) }, xiAnd yiAnd so on.xgboost
Current meteorological condition is added in input data and is trained for boosted tree, and the weight of air quality data is adjusted flexibly, thus
Obtain accurate predicted value.
The method that Xgboost return calculating is as follows:
If there is k tree in xgboost,
What F was indicated in above formula is all function spaces returned in forest,For model predication value, fkWhat is indicated is i-th
Predicted value of the sample in kth tree.
The parameter required now is exactly the structure of each tree and the weight of every leaf, is in simple terms exactly to seek each subtree
fk, wherein setting:
Θ={ f1,f2,...fk}
However objective function obj () is to find a relatively good parameter Θ, formula is as follows:
Obj (Θ)=L (Θ)+Ω (Θ)
Wherein, L (Θ) is the fit solution that error function is used to express data, and Ω (Θ) is regularization term, for punishing
Complex model.
The step of training pattern is that each tree is cumulative, and until k tree stops, process is as follows:
Wherein,After representing i-th circulation, xiPredicted value, t indicate t wheel model training.
Error in training is as follows:
Wherein yiAs corresponding labeled data,When the data predicted, l indicates loss function, carries out costing bio disturbance, often
That sees has Squared Error Loss and Logistic loss.
Penalty term is as follows when training:
It finally obtains with Taylor's formula that the formula of objective function abbreviation is as follows:
Wherein setC is constant,
ft(x)=wq(x),w∈RT,q:Rd→ { 1,2 ..., T }, f (x) indicate the node predicted value of every stalk tree, and w is indicated
The weight of leaf, q indicate the structure of tree, and T indicates the leaf number of tree, RdIndicate that number of features is the data set of d.
Further switch to the minimum value for seeking above formula quadratic function, it may be assumed that
Wherein calculate information gain (Gain) formula are as follows:
Wherein L indicates that left subtree, R indicate right subtree, HLIndicate left subtree weight, HRIndicate right subtree weight, γ indicates to add
Enter the complexity cost that new leaf node introduces.
In this way by after training, xgboost boosted tree is divided by information gain (Gain), each subtree is established
ft(x)=wq(x), each data characteristics, i.e. xi={ pi, zi, g1, g2, g3, g4, g5, g6 } and it include single factor test and Multiple-Factor Model
The air quality data p obtainediAnd ziIt with current every meteorological data g, is mapped on the leaf node of each tree, according to more wheels
Training, the leaf node of each subtree can be adjusted according to weight w, and whole subtrees can also be updated at this time, every height
Tree node is a predicted value, and the foundation for adjusting weight is to obtain optimal value w according to objective function obj, finally all each
A subtree fkPredicted value add up to get the air quality data finally predicted is gone out.
In conclusion the above is merely preferred embodiments of the present invention, being not intended to limit the scope of the present invention.
All within the spirits and principles of the present invention, any modification, equivalent replacement, improvement and so on should be included in of the invention
Within protection scope.
Claims (7)
1. a kind of Air Quality Forecast method of the deep learning of multi-model fusion, which comprises the following steps:
Step 1: obtaining history air quality data and meteorological data;
Step 2: carrying out missing values interpolation and normalized to history air quality data;
Step 3: predicting mould using history air quality data deep learning model of the building based on seq2seq as single factor test
Type;
Step 4: constructing the seq2seq depth based on double attention mechanism using history air quality data and meteorological data
Model is practised as multi-factor Estimation Model;
Step 5: history air quality data input single factor test prediction model is obtained into the prediction result of single factor test prediction model,
History air quality data and meteorological data input multi-factor Estimation Model are obtained into the prediction result of multi-factor Estimation Model, it will
The prediction result of single factor test prediction model, the prediction result of multi-factor Estimation Model and current weather data are dissolved into
Recurrence calculating is carried out in xgboost boosted tree, obtains the predicted value of final air quality data.
2. a kind of Air Quality Forecast method of the deep learning of multi-model fusion as described in claim 1, which is characterized in that
Air quality data includes PM2.5, PM10, NO2、CO、O3And SO2Concentration datas more than middle a kind of or two classes.
3. a kind of Air Quality Forecast method of the deep learning of multi-model fusion as claimed in claim 2, which is characterized in that
When air quality data is the concentration data of two classes or more, in step 5, one kind in air quality data is successively predicted, it will
The prediction result of single factor test prediction model and the prediction result of multi-factor Estimation Model of such air quality data and current
Meteorological data, which is dissolved into xgboost boosted tree, carries out recurrence calculating, obtains the predicted value of such air quality data.
4. a kind of Air Quality Forecast method of the deep learning of multi-model fusion as described in claim 1, which is characterized in that
Meteorological data includes temperature, air pressure, humidity, wind direction, wind speed and weather index.
5. a kind of Air Quality Forecast method of the deep learning of multi-model fusion as described in claim 1, which is characterized in that
Missing values interpolation is carried out using desired maximization approach.
6. a kind of Air Quality Forecast method of the deep learning of multi-model fusion as described in claim 1, which is characterized in that
The input layer of single factor test prediction model is history air quality data, and for the structure of hidden layer to be seq2seq model, inside is to compile
Code and decoding structure, coding and decoding unit therein are LSTM long Memory Neural Networks in short-term, and output layer is that air quality is pre-
Measured value.
7. a kind of Air Quality Forecast method of the deep learning of multi-model fusion as described in claim 1, which is characterized in that
The input layer of multi-factor Estimation Model is history air quality data and meteorological data, and hidden layer structure is a kind of double attention machines
The seq2seq model of system, inside be coding-decoding structure, one layer of attention mechanism is separately added into before coding and decoding,
Output layer is air quality data predicted value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811210072.3A CN109214592B (en) | 2018-10-17 | 2018-10-17 | Multi-model-fused deep learning air quality prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811210072.3A CN109214592B (en) | 2018-10-17 | 2018-10-17 | Multi-model-fused deep learning air quality prediction method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109214592A true CN109214592A (en) | 2019-01-15 |
CN109214592B CN109214592B (en) | 2022-03-08 |
Family
ID=64980570
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811210072.3A Active CN109214592B (en) | 2018-10-17 | 2018-10-17 | Multi-model-fused deep learning air quality prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109214592B (en) |
Cited By (37)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109685290A (en) * | 2019-02-11 | 2019-04-26 | 南方电网科学研究院有限责任公司 | Deep learning-based power consumption prediction method, device and equipment |
CN109991685A (en) * | 2019-04-03 | 2019-07-09 | 北京市天元网络技术股份有限公司 | A kind of precipitation prediction technique and device based on more LSTM Model Fusions |
CN110009134A (en) * | 2019-03-08 | 2019-07-12 | 浙江大学 | The pulping energy consumption prediction technique of model is extracted based on seq2seq behavioral characteristics |
CN110209131A (en) * | 2019-05-07 | 2019-09-06 | 西安交通大学 | A kind of qualitative forecasting method based on error propagation network and promotion tree algorithm |
CN110333556A (en) * | 2019-06-03 | 2019-10-15 | 深圳中兴网信科技有限公司 | Air Quality Forecast method, apparatus, computer equipment and readable storage medium storing program for executing |
CN110428106A (en) * | 2019-08-05 | 2019-11-08 | 山东农业大学 | A kind of crop water requirement prediction technique based on machine learning |
CN110738349A (en) * | 2019-09-05 | 2020-01-31 | 国网浙江省电力有限公司杭州供电公司 | Power grid fault first-aid repair duration prediction method based on multi-model fusion |
CN110851796A (en) * | 2019-11-12 | 2020-02-28 | 北京工商大学 | Music copyright protection system based on block chain intelligent contract |
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 |
CN111141879A (en) * | 2020-02-21 | 2020-05-12 | 防灾科技学院 | Deep learning air quality monitoring method, device and equipment |
CN111160628A (en) * | 2019-12-13 | 2020-05-15 | 重庆邮电大学 | Air pollutant concentration prediction method based on CNN and double-attention seq2seq |
CN111209968A (en) * | 2020-01-08 | 2020-05-29 | 浙江师范大学 | Multi-meteorological factor mode forecast temperature correction method and system based on deep learning |
CN111243752A (en) * | 2020-01-16 | 2020-06-05 | 四川大学华西医院 | Prediction model for acute pancreatitis induced organ failure |
CN111553543A (en) * | 2020-05-18 | 2020-08-18 | 润联软件***(深圳)有限公司 | Power load prediction method based on TPA-Seq2Seq and related assembly |
CN111582551A (en) * | 2020-04-15 | 2020-08-25 | 中南大学 | Method and system for predicting short-term wind speed of wind power plant and electronic equipment |
CN111598156A (en) * | 2020-05-14 | 2020-08-28 | 北京工业大学 | PM based on multi-source heterogeneous data fusion2.5Prediction model |
CN111639787A (en) * | 2020-04-28 | 2020-09-08 | 北京工商大学 | Spatio-temporal data prediction method based on graph convolution network |
CN111784073A (en) * | 2020-07-16 | 2020-10-16 | 武汉空心科技有限公司 | Deep learning-based work platform task workload prediction method |
CN111798051A (en) * | 2020-07-02 | 2020-10-20 | 杭州电子科技大学 | Air quality space-time prediction method based on long-short term memory neural network |
CN111814964A (en) * | 2020-07-20 | 2020-10-23 | 江西省环境监测中心站 | Air pollution treatment method based on air quality condition prediction and storage medium |
CN112163527A (en) * | 2020-09-29 | 2021-01-01 | 华中科技大学 | Fusion model-based tobacco leaf baking state identification method, device and system |
CN112580859A (en) * | 2020-06-01 | 2021-03-30 | 北京理工大学 | Haze prediction method based on global attention mechanism |
CN112862168A (en) * | 2021-01-28 | 2021-05-28 | 中山大学 | Neural network multi-model combination-based population density prediction method and system |
CN113052353A (en) * | 2019-12-27 | 2021-06-29 | 中移雄安信息通信科技有限公司 | Air quality prediction and prediction model training method and device and storage medium |
CN113688822A (en) * | 2021-09-07 | 2021-11-23 | 河南工业大学 | Time sequence attention mechanism scene image identification method |
CN113837487A (en) * | 2021-10-13 | 2021-12-24 | 国网湖南省电力有限公司 | Power system load prediction method based on combined model |
CN113960925A (en) * | 2021-08-30 | 2022-01-21 | 中科苏州微电子产业技术研究院 | Building energy consumption control method and device based on artificial intelligence |
CN114429800A (en) * | 2020-10-15 | 2022-05-03 | 中国石油化工股份有限公司 | Methane hydrate generation rate prediction method and system based on model fusion |
CN114510850A (en) * | 2022-04-20 | 2022-05-17 | 四川国蓝中天环境科技集团有限公司 | Multi-model fusion calibration method and system for atmospheric six-parameter differentiation |
CN114518611A (en) * | 2021-12-24 | 2022-05-20 | 山东省青岛生态环境监测中心(中国环境监测总站黄海近岸海域环境监测分站) | Ozone forecasting method based on similar case discriminant analysis |
CN114547017A (en) * | 2022-04-27 | 2022-05-27 | 南京信息工程大学 | Meteorological big data fusion method based on deep learning |
CN114662791A (en) * | 2022-04-22 | 2022-06-24 | 重庆邮电大学 | Long time sequence pm2.5 prediction method and system based on space-time attention |
CN115079308A (en) * | 2022-07-04 | 2022-09-20 | 湖南省生态环境监测中心 | Air quality ensemble forecasting system and method thereof |
US20220316734A1 (en) * | 2021-04-14 | 2022-10-06 | Jiangnan University | Deep Spatial-Temporal Similarity Method for Air Quality Prediction |
CN115616978A (en) * | 2022-10-20 | 2023-01-17 | 重庆大学 | SQ-LSTMA-based thermal error prediction model, prediction method and control system |
CN115732041A (en) * | 2022-12-07 | 2023-03-03 | 中国石油大学(北京) | Carbon dioxide capture amount prediction model construction method, intelligent prediction method and device |
CN115794981A (en) * | 2022-12-14 | 2023-03-14 | 广西电网有限责任公司 | Method and system for counting meteorological data by using model |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180060665A1 (en) * | 2016-08-29 | 2018-03-01 | Nec Laboratories America, Inc. | Dual Stage Attention Based Recurrent Neural Network for Time Series Prediction |
CN107909206A (en) * | 2017-11-15 | 2018-04-13 | 电子科技大学 | A kind of PM2.5 Forecasting Methodologies based on deep structure Recognition with Recurrent Neural Network |
CN108197736A (en) * | 2017-12-29 | 2018-06-22 | 北京工业大学 | A kind of Air Quality Forecast method based on variation self-encoding encoder and extreme learning machine |
CN108510113A (en) * | 2018-03-21 | 2018-09-07 | 中南大学 | A kind of application of XGBoost in short-term load forecasting |
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 |
-
2018
- 2018-10-17 CN CN201811210072.3A patent/CN109214592B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180060665A1 (en) * | 2016-08-29 | 2018-03-01 | Nec Laboratories America, Inc. | Dual Stage Attention Based Recurrent Neural Network for Time Series Prediction |
CN107909206A (en) * | 2017-11-15 | 2018-04-13 | 电子科技大学 | A kind of PM2.5 Forecasting Methodologies based on deep structure Recognition with Recurrent Neural Network |
CN108197736A (en) * | 2017-12-29 | 2018-06-22 | 北京工业大学 | A kind of Air Quality Forecast method based on variation self-encoding encoder and extreme learning machine |
CN108510113A (en) * | 2018-03-21 | 2018-09-07 | 中南大学 | A kind of application of XGBoost in short-term load forecasting |
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 |
Cited By (52)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109685290A (en) * | 2019-02-11 | 2019-04-26 | 南方电网科学研究院有限责任公司 | Deep learning-based power consumption prediction method, device and equipment |
CN109685290B (en) * | 2019-02-11 | 2023-06-16 | 南方电网科学研究院有限责任公司 | Power consumption prediction method, device and equipment based on deep learning |
CN110009134A (en) * | 2019-03-08 | 2019-07-12 | 浙江大学 | The pulping energy consumption prediction technique of model is extracted based on seq2seq behavioral characteristics |
CN110009134B (en) * | 2019-03-08 | 2020-12-18 | 浙江大学 | Pulping energy consumption prediction method based on seq2seq dynamic feature extraction model |
CN109991685A (en) * | 2019-04-03 | 2019-07-09 | 北京市天元网络技术股份有限公司 | A kind of precipitation prediction technique and device based on more LSTM Model Fusions |
CN110209131A (en) * | 2019-05-07 | 2019-09-06 | 西安交通大学 | A kind of qualitative forecasting method based on error propagation network and promotion tree algorithm |
CN110333556A (en) * | 2019-06-03 | 2019-10-15 | 深圳中兴网信科技有限公司 | Air Quality Forecast method, apparatus, computer equipment and readable storage medium storing program for executing |
CN110428106A (en) * | 2019-08-05 | 2019-11-08 | 山东农业大学 | A kind of crop water requirement prediction technique based on machine learning |
CN110738349A (en) * | 2019-09-05 | 2020-01-31 | 国网浙江省电力有限公司杭州供电公司 | Power grid fault first-aid repair duration prediction method based on multi-model fusion |
CN110738349B (en) * | 2019-09-05 | 2023-07-11 | 国网浙江省电力有限公司杭州供电公司 | Power grid fault rush-repair duration prediction method based on multi-model fusion |
CN110851796A (en) * | 2019-11-12 | 2020-02-28 | 北京工商大学 | Music copyright protection system based on block chain intelligent contract |
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 |
CN111160628A (en) * | 2019-12-13 | 2020-05-15 | 重庆邮电大学 | Air pollutant concentration prediction method based on CNN and double-attention seq2seq |
CN113052353A (en) * | 2019-12-27 | 2021-06-29 | 中移雄安信息通信科技有限公司 | Air quality prediction and prediction model training method and device and storage medium |
CN113052353B (en) * | 2019-12-27 | 2022-10-18 | 中移雄安信息通信科技有限公司 | Air quality prediction and prediction model training method and device and storage medium |
CN111209968A (en) * | 2020-01-08 | 2020-05-29 | 浙江师范大学 | Multi-meteorological factor mode forecast temperature correction method and system based on deep learning |
CN111243752A (en) * | 2020-01-16 | 2020-06-05 | 四川大学华西医院 | Prediction model for acute pancreatitis induced organ failure |
CN111141879A (en) * | 2020-02-21 | 2020-05-12 | 防灾科技学院 | Deep learning air quality monitoring method, device and equipment |
CN111141879B (en) * | 2020-02-21 | 2023-02-03 | 防灾科技学院 | Deep learning air quality monitoring method, device and equipment |
CN111582551A (en) * | 2020-04-15 | 2020-08-25 | 中南大学 | Method and system for predicting short-term wind speed of wind power plant and electronic equipment |
CN111582551B (en) * | 2020-04-15 | 2023-12-08 | 中南大学 | Wind power plant short-term wind speed prediction method and system and electronic equipment |
CN111639787B (en) * | 2020-04-28 | 2024-03-15 | 北京工商大学 | Spatio-temporal data prediction method based on graph convolution network |
CN111639787A (en) * | 2020-04-28 | 2020-09-08 | 北京工商大学 | Spatio-temporal data prediction method based on graph convolution network |
CN111598156A (en) * | 2020-05-14 | 2020-08-28 | 北京工业大学 | PM based on multi-source heterogeneous data fusion2.5Prediction model |
CN111553543A (en) * | 2020-05-18 | 2020-08-18 | 润联软件***(深圳)有限公司 | Power load prediction method based on TPA-Seq2Seq and related assembly |
CN112580859A (en) * | 2020-06-01 | 2021-03-30 | 北京理工大学 | Haze prediction method based on global attention mechanism |
CN111798051B (en) * | 2020-07-02 | 2023-11-10 | 杭州电子科技大学 | Air quality space-time prediction method based on long-term and short-term memory neural network |
CN111798051A (en) * | 2020-07-02 | 2020-10-20 | 杭州电子科技大学 | Air quality space-time prediction method based on long-short term memory neural network |
CN111784073A (en) * | 2020-07-16 | 2020-10-16 | 武汉空心科技有限公司 | Deep learning-based work platform task workload prediction method |
CN111814964A (en) * | 2020-07-20 | 2020-10-23 | 江西省环境监测中心站 | Air pollution treatment method based on air quality condition prediction and storage medium |
CN112163527A (en) * | 2020-09-29 | 2021-01-01 | 华中科技大学 | Fusion model-based tobacco leaf baking state identification method, device and system |
CN114429800A (en) * | 2020-10-15 | 2022-05-03 | 中国石油化工股份有限公司 | Methane hydrate generation rate prediction method and system based on model fusion |
CN112862168A (en) * | 2021-01-28 | 2021-05-28 | 中山大学 | Neural network multi-model combination-based population density prediction method and system |
US11512864B2 (en) * | 2021-04-14 | 2022-11-29 | Jiangnan University | Deep spatial-temporal similarity method for air quality prediction |
US20220316734A1 (en) * | 2021-04-14 | 2022-10-06 | Jiangnan University | Deep Spatial-Temporal Similarity Method for Air Quality Prediction |
CN113960925A (en) * | 2021-08-30 | 2022-01-21 | 中科苏州微电子产业技术研究院 | Building energy consumption control method and device based on artificial intelligence |
CN113688822A (en) * | 2021-09-07 | 2021-11-23 | 河南工业大学 | Time sequence attention mechanism scene image identification method |
CN113837487A (en) * | 2021-10-13 | 2021-12-24 | 国网湖南省电力有限公司 | Power system load prediction method based on combined model |
CN114518611A (en) * | 2021-12-24 | 2022-05-20 | 山东省青岛生态环境监测中心(中国环境监测总站黄海近岸海域环境监测分站) | Ozone forecasting method based on similar case discriminant analysis |
CN114510850B (en) * | 2022-04-20 | 2022-06-21 | 四川国蓝中天环境科技集团有限公司 | Multi-model fusion calibration method and system for atmospheric six-parameter differentiation |
CN114510850A (en) * | 2022-04-20 | 2022-05-17 | 四川国蓝中天环境科技集团有限公司 | Multi-model fusion calibration method and system for atmospheric six-parameter differentiation |
CN114662791A (en) * | 2022-04-22 | 2022-06-24 | 重庆邮电大学 | Long time sequence pm2.5 prediction method and system based on space-time attention |
CN114547017A (en) * | 2022-04-27 | 2022-05-27 | 南京信息工程大学 | Meteorological big data fusion method based on deep learning |
CN114547017B (en) * | 2022-04-27 | 2022-08-05 | 南京信息工程大学 | Meteorological big data fusion method based on deep learning |
CN115079308A (en) * | 2022-07-04 | 2022-09-20 | 湖南省生态环境监测中心 | Air quality ensemble forecasting system and method thereof |
CN115079308B (en) * | 2022-07-04 | 2023-10-24 | 湖南省生态环境监测中心 | Air quality set forecasting system and method thereof |
CN115616978B (en) * | 2022-10-20 | 2024-06-28 | 重庆大学 | SQ-LSTMA-based thermal error prediction model, prediction method and control system |
CN115616978A (en) * | 2022-10-20 | 2023-01-17 | 重庆大学 | SQ-LSTMA-based thermal error prediction model, prediction method and control system |
CN115732041B (en) * | 2022-12-07 | 2023-10-13 | 中国石油大学(北京) | Carbon dioxide capture quantity prediction model construction method, intelligent prediction method and device |
CN115732041A (en) * | 2022-12-07 | 2023-03-03 | 中国石油大学(北京) | Carbon dioxide capture amount prediction model construction method, intelligent prediction method and device |
CN115794981B (en) * | 2022-12-14 | 2023-09-26 | 广西电网有限责任公司 | Method and system for counting meteorological data by using model |
CN115794981A (en) * | 2022-12-14 | 2023-03-14 | 广西电网有限责任公司 | Method and system for counting meteorological data by using model |
Also Published As
Publication number | Publication date |
---|---|
CN109214592B (en) | 2022-03-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109214592A (en) | A kind of Air Quality Forecast method of the deep learning of multi-model fusion | |
CN110135630B (en) | Short-term load demand prediction method based on random forest regression and multi-step optimization | |
CN110782093B (en) | PM fusing SSAE deep feature learning and LSTM2.5Hourly concentration prediction method and system | |
CN106650767B (en) | Flood forecasting method based on cluster analysis and real-time correction | |
CN108009674A (en) | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks | |
CN110580543A (en) | Power load prediction method and system based on deep belief network | |
CN108197736A (en) | A kind of Air Quality Forecast method based on variation self-encoding encoder and extreme learning machine | |
CN109583565A (en) | Forecasting Flood method based on the long memory network in short-term of attention model | |
CN106920007A (en) | PM based on second order Self-organized Fuzzy Neural Network2.5Intelligent Forecasting | |
CN109086926A (en) | A kind of track traffic for passenger flow prediction technique in short-term based on combination neural net structure | |
CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
CN102867217A (en) | Projection pursuit-based risk evaluation method for meteorological disasters of facility agriculture | |
CN109344999A (en) | A kind of runoff probability forecast method | |
CN112733996A (en) | GA-PSO (genetic Algorithm-particle swarm optimization) based hydrological time sequence prediction method for optimizing XGboost | |
CN111582541A (en) | Firefly algorithm-based inland inundation model prediction method | |
CN110942182A (en) | Method for establishing typhoon prediction model based on support vector regression | |
Li et al. | A method of rainfall runoff forecasting based on deep convolution neural networks | |
CN112561132A (en) | Water flow prediction model based on neural network | |
CN116310350A (en) | Urban scene semantic segmentation method based on graph convolution and semi-supervised learning network | |
Priatna et al. | Precipitation prediction using recurrent neural networks and long short-term memory | |
Khan et al. | Adaptive Error Curve Learning Ensemble Model for Improving Energy Consumption Forecasting. | |
CN114882373A (en) | Multi-feature fusion sandstorm prediction method based on deep neural network | |
CN114386654A (en) | Multi-scale numerical weather forecasting mode fusion weather forecasting method and device | |
CN116663915A (en) | Photovoltaic output ultra-short-term prediction method and device | |
Manokij et al. | Cascading Models of CNN and GRU with Autoencoder Loss for Precipitation Forecast in Thailand |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20240116 Address after: 230000 floor 1, building 2, phase I, e-commerce Park, Jinggang Road, Shushan Economic Development Zone, Hefei City, Anhui Province Patentee after: Dragon totem Technology (Hefei) Co.,Ltd. Address before: 100048, Fu Cheng Road, Beijing, Haidian District, No. 33 Patentee before: BEIJING TECHNOLOGY AND BUSINESS University |
|
TR01 | Transfer of patent right |