CN109816163A - Dustiness prediction technique, forecasting system and computer readable storage medium - Google Patents
Dustiness prediction technique, forecasting system and computer readable storage medium Download PDFInfo
- Publication number
- CN109816163A CN109816163A CN201910035757.7A CN201910035757A CN109816163A CN 109816163 A CN109816163 A CN 109816163A CN 201910035757 A CN201910035757 A CN 201910035757A CN 109816163 A CN109816163 A CN 109816163A
- Authority
- CN
- China
- Prior art keywords
- dustiness
- presumptive area
- data
- distribution map
- space
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/045—Combinations of networks
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention proposes a kind of dustiness prediction technique, dustiness forecasting system and computer readable storage mediums.Wherein, dustiness prediction technique includes: to obtain the dustiness data of presumptive area；The dustiness space two-dimensional array distribution map of presumptive area is generated according to dustiness data；Dustiness space two-dimensional array distribution map is inputted into composite network model, generates the dustiness time series of presumptive area；It is predicted according to dustiness of the dustiness time series to presumptive area；Wherein, composite network model is made of convolutional neural networks and length memory network.The present invention comprehensively considers time point and the spatial geographical locations of different dustiness data, to reach optimal prediction effect, has not only guaranteed to carry out Accurate Prediction to the dustiness of presumptive area, but also simplify prediction process, accuracy rate is high.
Description
Technical field
The present invention relates to environmental pollutions to forecast field, in particular to a kind of dustiness prediction technique, a kind of dustiness
Forecasting system and a kind of computer readable storage medium.
Background technique
Environmental pollution refers to natural or artificial destruction, certain substance is added into environment, more than the self-purification capacity of environment
And the behavior of harm being generated, or due to artificial factor, pollution of the environment by harmful substance makes growth and breeding and the people of biology
The normal life of class is adversely affected.Since human factor makes the composition of environment or state change, Environmental Quality decline,
To upset and destroy the ecosystem and the normal working and living condition of the mankind.
In the related technology, it there are being acquired to air quality data, predicts.Utilize CNN (Convolutional
Neural Networks convolutional neural networks) model compression and extract data important feature, utilize LSTM (Long Stort-
Term Memory length memory network) it is that extraction time sequence signature is predicted with the air quality to certain area.On but
Stating technical solution only may be implemented to be acquired relevant air qualitative data, predict, have certain limitation, and predict
Process is relatively complicated, and prediction result accuracy is low.
Summary of the invention
The present invention is directed at least solve one of the technical problems existing in the prior art.
For this purpose, first aspect present invention proposes a kind of dustiness prediction technique.
Second aspect of the present invention proposes a kind of dustiness forecasting system.
Third aspect present invention proposes a kind of computer readable storage medium.
First aspect present invention proposes a kind of dustiness prediction technique, comprising: obtains the dustiness data of presumptive area；
The dustiness space two-dimensional array distribution map of presumptive area is generated according to dustiness data；Dustiness space two-dimensional array is distributed
Figure input composite network model, generates the dustiness time series of presumptive area；According to dustiness time series to presumptive area
Dustiness predicted；Wherein, composite network model is made of convolutional neural networks and length memory network.
The dustiness prediction technique that first aspect present invention provides selects convolutional neural networks and length memory network composition
Composite network model is predicted by dustiness of the composite network model realization to presumptive area.Presumptive area is obtained first
Dustiness data specifically obtain the dustiness data of diverse geographic location and different time；Then dustiness data are carried out
It arranges, generates dustiness space two-dimensional array distribution map, it specifically, can be by dustiness data topology at similar RGB (Red
The red green basket of Green Blue) image data information；After generating dustiness space two-dimensional array distribution map, by dustiness space two
Dimension group distribution map inputs composite network model, so that composite network model is generated according to dustiness space two-dimensional array distribution map
The dustiness time series of presumptive area；After getting dustiness time series, according to dustiness time series to fate
The dustiness in domain is predicted, to obtain the dustiness prediction result of presumptive area.
The present invention forms composite network model by CNN and LSTM, and dustiness space two-dimensional array distribution map is input to
It is trained, dustiness data space geographical location is inputted using CNN, using LSTM to dustiness in composite network model
Data time point inputs, and CNN and LSTM cooperate, and comprehensively considers time point and the space and geographical position of different dustiness data
It sets, to reach optimal prediction effect, had not only guaranteed to carry out Accurate Prediction to the dustiness of presumptive area, but also simplify prediction process.
Above-mentioned dustiness prediction technique according to the present invention, can also have following additional technical feature:
In the above-mentioned technical solutions, it is preferable that dustiness space two-dimensional array distribution map is inputted into composite network model, it is raw
At presumptive area dustiness time series the step of, specifically include: according to time sequencing by dustiness space two-dimensional array point
Butut inputs convolutional neural networks, generates dustiness Multidimensional numerical distribution map；Obtain the mesh in dustiness Multidimensional numerical distribution map
Mark dimensional characteristics；Target dimension feature is inputted into length memory network, generates the dustiness time series of presumptive area.
In the technical scheme, it after getting dustiness space two-dimensional array distribution map, will be polluted according to time sequencing
It spends space two-dimensional array distribution map and inputs convolutional neural networks, generate dustiness Multidimensional numerical distribution map；Generating, dustiness is more
After dimension group distribution map, the target dimension feature in dustiness Multidimensional numerical distribution map is extracted.Specifically it is more to extract dustiness
High-dimensional feature in dimension group distribution map；After extracting the high-dimensional feature in Multidimensional numerical distribution map, by high-dimensional spy
Sign input length memory network, generates the dustiness time series of presumptive area, so as to according to the dustiness time of presumptive area
Sequence predicts the dustiness of presumptive area.
In any of the above-described technical solution, it is preferable that target dimension feature is inputted length memory network, generates fate
It the step of dustiness time series in domain, specifically includes: target dimension feature is inputted into two-way time round-robin algorithm；According to the time
Arrangement carries out two-way sequence training arrangement to target dimension feature, generates the dustiness time series of presumptive area.
In the technical scheme, after getting the target dimension feature in dustiness Multidimensional numerical distribution map, according to when
Between arrangement the training arrangement of two-way sequence is carried out to target dimension feature, generate the dustiness time series of presumptive area.It i.e. will be high
Dimensional characteristics input in two-way time round-robin algorithm, carry out two-way sequence training arrangement according to time arrangement, then to later
Time data are predicted, time series analysis is done, and generate dustiness time series.
In any of the above-described technical solution, it is preferable that generate the dustiness space two of presumptive area according to dustiness data
It the step of dimension group distribution map, specifically includes: presumptive area is sequentially generated according to the geographical location of dustiness data and time
Dustiness data list；The dustiness space two-dimensional array distribution map of presumptive area is generated according to dustiness data list.
In the technical scheme, to guarantee that acquired dustiness data are more accurate, in diverse geographic location, when different
Between obtain different dustiness data, be then sequentially generated presumptive area according to the geographical location and time of multiple dustiness data
Dustiness data list, with according to dustiness data list generate presumptive area dustiness space two-dimensional array distribution map.
Specifically, classification, timing node and the geographical position of each dustiness data are shown in the dustiness space two-dimensional array distribution map
The information such as set.
In any of the above-described technical solution, it is preferable that dustiness data include: water pollution degree evidence, air pollution degree
According to, land pollution degree evidence.
In the technical scheme, to ensure the full forecast to presumptive area dustiness, therefore acquisition may influence to make a reservation for
A variety of data of regional pollution degree.Specifically, including but not limited to following dustiness data: water pollution degree evidence, air are dirty
Contaminate degree evidence, land pollution degree evidence.More specifically, including but not limited to following dustiness data: ph value of acid rain
(ratio that hydrogen ion concentration refers to hydrionic sum and the amount of total material in solution), nitrogen oxides, one
Carbonoxide, fluoride, lead and its compound, radiation level, the dustiness data such as soil pH value.
In any of the above-described technical solution, it is preferable that the dustiness for generating presumptive area according to dustiness data list is empty
Between two-dimensional array distribution map the step of, specifically further include: the dustiness data in dustiness data list are pre-processed；It is right
Pretreated dustiness data carry out batch quota and handle, so that being respectively positioned on target by pretreated dustiness data
In section；The dustiness space two-dimensional array distribution map of presumptive area is generated according to the dustiness data being located in target interval.
In the technical scheme, the pollution of presumptive area is sequentially generated in the geographical location and time for obtaining dustiness data
After spending data list, the dustiness data in dustiness data list are pre-processed.Specifically to abnormal dustiness
The Data Position time cleans, and washes too large or too small dustiness data, washes excessive with plain data gap
Dustiness data, enable clear data is used；Then, it carries out batch quota to the dustiness data cleaned to handle, so that dirty
Between dye degree data distribution 0-1, arithmetic speed faster, is easier to be fitted.
In any of the above-described technical solution, it is preferable that carried out according to dustiness of the dustiness time series to presumptive area
The step of prediction, specifically includes: by dustiness time series entrance loss function, obtaining the first calculated result；First is calculated
As a result stochastic gradient descent algorithm is inputted, the second calculated result is obtained；According to the second calculated result to the dustiness of presumptive area
It is predicted.
In the technical scheme, after the dustiness time series for generating presumptive area, dustiness time series is inputted
Loss function obtains the first calculated result, and the first calculated result is then inputted stochastic gradient descent algorithm, obtains the second calculating
As a result, to be predicted according to dustiness of second calculated result to presumptive area.Pass through loss function and stochastic gradient descent
The algorithm operation that iterates so that the value of loss function is smaller obtains optimum results, exports the prediction result of multi-tag, compares
The prediction result of legacy data is more accurate.
Second aspect of the present invention proposes a kind of dustiness forecasting system, comprising: memory, for storing computer journey
Sequence；Processor, for execute computer program with: obtain the dustiness data of presumptive area；It is generated according to dustiness data pre-
Determine the dustiness space two-dimensional array distribution map in region；Dustiness space two-dimensional array distribution map is inputted into composite network model,
Generate the dustiness time series of presumptive area；It is predicted according to dustiness of the dustiness time series to presumptive area；Its
In, composite network model is made of convolutional neural networks and length memory network.
Dustiness forecasting system provided by the invention includes mutually matched memory and processor, wherein processor choosing
Composite network model is formed with convolutional neural networks and length memory network, by the composite network model realization to presumptive area
Dustiness prediction.Specifically, the dustiness data of presumptive area are obtained first, specifically, obtain diverse geographic location and not
With the dustiness data of time；Then dustiness data are arranged, generates dustiness space two-dimensional array distribution map, specifically
Ground, can be by dustiness data topology at similar rgb image data information；Generating dustiness space two-dimensional array distribution map
Afterwards, dustiness space two-dimensional array distribution map is inputted into composite network model, so that composite network model is according to dustiness space
The dustiness time series of two-dimensional array distribution map generation presumptive area；After getting dustiness time series, according to pollution
Degree time series predicts the dustiness of presumptive area, to obtain the dustiness prediction result of presumptive area.
The present invention forms composite network model by CNN and LSTM, and dustiness space two-dimensional array distribution map is input to
It is trained, dustiness data space geographical location is inputted using CNN, using LSTM to dustiness in composite network model
Data time point inputs, and CNN and LSTM cooperate, and time and the space of different dustiness data is comprehensively considered, to reach
Optimal prediction effect had not only guaranteed to carry out Accurate Prediction to the dustiness of presumptive area, but also has simplified prediction process.
In the above-mentioned technical solutions, processor is specifically used for: being distributed dustiness space two-dimensional array according to time sequencing
Figure input convolutional neural networks, generate dustiness Multidimensional numerical distribution map；Obtain the target in dustiness Multidimensional numerical distribution map
Dimensional characteristics；Target dimension feature is inputted into two-way time round-robin algorithm；Target dimension feature is carried out according to time arrangement double
It arranges to sequence training, generates the dustiness time series of presumptive area；By dustiness time series entrance loss function, obtain
First calculated result；First calculated result is inputted into stochastic gradient descent algorithm, obtains the second calculated result；It is calculated according to second
As a result the pollution level of presumptive area is predicted.
In the technical scheme, it after getting dustiness space two-dimensional array distribution map, will be polluted according to time sequencing
It spends space two-dimensional array distribution map and inputs convolutional neural networks, generate dustiness Multidimensional numerical distribution map；Generating, dustiness is more
After dimension group distribution map, the target dimension feature in dustiness Multidimensional numerical distribution map is extracted.Specifically it is more to extract dustiness
High-dimensional feature in dimension group distribution map；After extracting the high-dimensional feature in Multidimensional numerical distribution map, arranged according to the time
Cloth carries out two-way sequence training arrangement to target dimension feature, generates the dustiness time series of presumptive area；Made a reservation for
After the dustiness time series in region, by dustiness time series entrance loss function, the first calculated result is obtained, then by
One calculated result input stochastic gradient descent algorithm, obtain the second calculated result, with according to the second calculated result to presumptive area
Dustiness predicted.It is iterated operation by loss function and stochastic gradient descent algorithm, so that the value of loss function
It is smaller, optimum results are obtained, the prediction result of multi-tag is exported, the result for comparing legacy data is more accurate.
Specifically, dustiness data include but is not limited to following dustiness data: water pollution degree evidence, air contamination
Data, land pollution degree evidence.More specifically, including but not limited to ph value of acid rain, nitrogen oxides, carbon monoxide, fluoride,
Lead and its compound, radiation level, the dustiness data such as soil pH value.
Third aspect present invention proposes a kind of computer readable storage medium, is stored thereon with computer program, calculates
The dustiness prediction technique such as any one of first aspect present invention is realized when machine program is executed by processor.
The computer readable storage medium that third invention of the present invention provides, is stored thereon with computer program, computer journey
The dustiness prediction technique such as any one of first aspect present invention is realized when sequence is executed by processor, therefore there is above-mentioned pollution
The whole beneficial effects for spending prediction technique, are no longer discussed one by one herein.
Additional aspect and advantage of the invention will become obviously in following description section, or practice through the invention
Recognize.
Detailed description of the invention
Above-mentioned and/or additional aspect of the invention and advantage will become from the description of the embodiment in conjunction with the following figures
Obviously and it is readily appreciated that, in which:
Fig. 1 shows the flow chart of the dustiness prediction technique of one embodiment of the invention；
Fig. 2 shows the flow charts of the dustiness prediction technique of another embodiment of the present invention；
Fig. 3 shows the flow chart of the dustiness prediction technique of a specific embodiment of the invention；
Fig. 4 shows the structural block diagram of the dustiness forecasting system of a specific embodiment of the invention.
Specific embodiment
To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real
Applying mode, the present invention is further described in detail.It should be noted that in the absence of conflict, the implementation of the application
Feature in example and embodiment can be combined with each other.
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention, still, the present invention may be used also
Implement in a manner of using other than the one described here, therefore, protection scope of the present invention is not by following public tool
The limitation of body embodiment.
The dustiness prediction technique provided according to some embodiments of the invention, dustiness are described referring to Fig. 1 to Fig. 4
Forecasting system and computer readable storage medium.
Fig. 1 shows the flow chart of the dustiness prediction technique of one embodiment of the invention.
As shown in Figure 1, the dustiness prediction technique includes:
S102 obtains the dustiness data of presumptive area；
S104 generates the dustiness space two-dimensional array distribution map of presumptive area according to dustiness data；
Dustiness space two-dimensional array distribution map is inputted composite network model, generates the dustiness of presumptive area by S106
Time series；
S108 is predicted according to dustiness of the dustiness time series to presumptive area.
The dustiness prediction technique that first aspect present invention provides selects convolutional neural networks and length memory network composition
Composite network model is predicted by dustiness of the composite network model realization to presumptive area.Presumptive area is obtained first
Dustiness data specifically obtain the dustiness data of diverse geographic location and different time；Then dustiness data are carried out
It arranges, generates dustiness space two-dimensional array distribution map, it specifically, can be by dustiness data topology at similar RGB image number
It is believed that breath；After generating dustiness space two-dimensional array distribution map, dustiness space two-dimensional array distribution map is inputted into composite web
Network model, so that composite network model generates the dustiness time sequence of presumptive area according to dustiness space two-dimensional array distribution map
Column；After getting dustiness time series, predicted according to dustiness of the dustiness time series to presumptive area, with
To the dustiness prediction result of presumptive area.
The present invention forms composite network model by CNN and LSTM, and dustiness space two-dimensional array distribution map is input to
It is trained, dustiness data space geographical location is inputted using CNN, using LSTM to dustiness in composite network model
Data time point inputs, and CNN and LSTM cooperate, and comprehensively considers time point and the space and geographical position of different dustiness data
It sets, to reach optimal prediction effect, had not only guaranteed to carry out Accurate Prediction to the dustiness of presumptive area, but also simplify prediction process.
In one embodiment of the invention, it is preferable that dustiness space two-dimensional array distribution map is inputted into composite network
Model, specifically includes the step of generating the dustiness time series of presumptive area: according to time sequencing by dustiness space two-dimensional
Array distribution map inputs convolutional neural networks, generates dustiness Multidimensional numerical distribution map；Obtain dustiness Multidimensional numerical distribution map
In target dimension feature；Target dimension feature is inputted into length memory network, generates the dustiness time series of presumptive area.
In this embodiment, after getting dustiness space two-dimensional array distribution map, according to time sequencing by dustiness
Space two-dimensional array distribution map inputs convolutional neural networks, generates dustiness Multidimensional numerical distribution map；Generating dustiness multidimensional
After array distribution map, the target dimension feature in dustiness Multidimensional numerical distribution map is extracted.Specifically extracting dustiness multidimensional
High-dimensional feature in array distribution map；After extracting the high-dimensional feature in Multidimensional numerical distribution map, by high-dimensional feature
Input length memory network, generate the dustiness time series of presumptive area, so as to according to dustiness time series to fate
The dustiness in domain is predicted.
In one embodiment of the invention, it is preferable that target dimension feature is inputted into length memory network, is generated predetermined
It the step of dustiness time series in region, specifically includes: target dimension feature is inputted into two-way time round-robin algorithm；According to when
Between arrangement the training arrangement of two-way sequence is carried out to target dimension feature, generate the dustiness time series of presumptive area.
In this embodiment, after getting the target dimension feature in dustiness Multidimensional numerical distribution map, according to the time
Arrangement carries out two-way sequence training arrangement to target dimension feature, generates the dustiness time series of presumptive area.I.e. by higher-dimension
Spend feature to input in two-way time round-robin algorithm, carry out two-way sequence training arrangement according to time arrangement, then to later when
Between data predicted, do time series analysis, generate dustiness time series.
In one embodiment of the invention, it is preferable that the dustiness space of presumptive area is generated according to dustiness data
It the step of two-dimensional array distribution map, specifically includes: presumptive area is sequentially generated according to the geographical location of dustiness data and time
Dustiness data list；The dustiness space two-dimensional array distribution map of presumptive area is generated according to dustiness data list.
In this embodiment, to guarantee that acquired dustiness data are more accurate, in diverse geographic location, different time
Different dustiness data are obtained, are then sequentially generated presumptive area according to the geographical location and time of multiple dustiness data
Dustiness data list, to generate the dustiness space two-dimensional array distribution map of presumptive area according to dustiness data list.Tool
Body, classification, timing node and the geographical location of each dustiness data are shown in the dustiness space two-dimensional array distribution map
Etc. information.
In one embodiment of the invention, it is preferable that dustiness data include: water pollution degree evidence, air contamination
Data, land pollution degree evidence.
In this embodiment, to ensure the full forecast to presumptive area dustiness, therefore acquisition may influence fate
A variety of data of domain dustiness.Specifically, including but not limited to following dustiness data: water pollution degree evidence, air pollution
Degree evidence, land pollution degree evidence.More specifically, including but not limited to following dustiness data: ph value of acid rain, nitrogen oxidation
Object, carbon monoxide, fluoride, lead and its compound, radiation level, the dustiness data such as soil pH value.
In one embodiment of the invention, it is preferable that the dustiness of presumptive area is generated according to dustiness data list
The step of space two-dimensional array distribution map, specifically further include: the dustiness data in dustiness data list are pre-processed；
It carries out batch quota to pretreated dustiness data to handle, so that being respectively positioned on mesh by pretreated dustiness data
It marks in section；The dustiness space two-dimensional array distribution of presumptive area is generated according to the dustiness data being located in target interval
Figure.
In this embodiment, the dustiness of presumptive area is sequentially generated in the geographical location and time for obtaining dustiness data
After data list, the dustiness data in dustiness data list are pre-processed.Specifically to abnormal pollution degree
It is cleaned according to the position time, washes too large or too small dustiness data, the excessive dirt with plain data gap
Degree evidence is contaminated, enable clear data is used；Then, it carries out batch quota to the dustiness data cleaned to handle, so that pollution
It spends between data distribution 0-1, arithmetic speed faster, is easier to be fitted.
In one embodiment of the invention, it is preferable that according to dustiness time series to the dustiness of presumptive area into
It the step of row prediction, specifically includes: by dustiness time series entrance loss function, obtaining the first calculated result；By the first meter
It calculates result and inputs stochastic gradient descent algorithm, obtain the second calculated result；Pollution according to the second calculated result to presumptive area
Degree is predicted.
In this embodiment, after the dustiness time series for generating presumptive area, dustiness time series is inputted and is damaged
Function is lost, the first calculated result is obtained, the first calculated result is then inputted into stochastic gradient descent algorithm, obtains the second calculating knot
Fruit, to be predicted according to dustiness of second calculated result to presumptive area.It is calculated by loss function and stochastic gradient descent
The method operation that iterates so that the value of loss function is smaller obtains optimum results, exports the prediction result of multi-tag, compares former
There is the prediction result of data more accurate.
Fig. 2 shows the flow charts of the dustiness prediction technique of another embodiment of the present invention.
As shown in Fig. 2, the dustiness prediction technique includes:
S202 obtains the dustiness data of presumptive area；
S204 pre-processes the dustiness data in dustiness data list；
S206 carries out batch quota to pretreated dustiness data and handles, so that by pretreated pollution
Degree evidence is respectively positioned in target interval；
S208 generates the dustiness space two-dimensional array point of presumptive area according to the dustiness data being located in target interval
Butut；
Dustiness space two-dimensional array distribution map is inputted convolutional neural networks according to time sequencing, generates pollution by S210
Spend Multidimensional numerical distribution map；
S212 obtains the target dimension feature in dustiness Multidimensional numerical distribution map；
Target dimension feature is inputted two-way time round-robin algorithm by S214；
S216 carries out two-way sequence training arrangement to target dimension feature according to time arrangement, generates the dirt of presumptive area
Dye degree time series；
Dustiness time series entrance loss function is obtained the first calculated result by S218；
First calculated result is inputted stochastic gradient descent algorithm, obtains the second calculated result by S220；
S222 is predicted according to dustiness of second calculated result to presumptive area.
In this embodiment, the dustiness data of presumptive area are obtained first, specifically different geographical in predeterminable area
The dustiness data of position and different time, including but not limited to following dustiness data: ph value of acid rain, nitrogen oxides, an oxygen
Change carbon, fluoride, lead and its compound, radiation level, the numerical value such as soil pH value.After getting dustiness data, according to
The geographical location of dustiness data and time are sequentially generated the dustiness data list of presumptive area, dustiness data column display
Dustiness data type, timing node and geographical location；After generating dustiness data list, to abnormal dustiness data bit
It sets the time to clean, washes too large or too small dustiness data, wash the dustiness excessive with plain data gap
Then data carry out batch quota to the dustiness data cleaned and handle so that the dustiness data distribution 0-1 cleaned it
Between, arithmetic speed faster, is easier to be fitted；After completing data cleansing, the dirt of presumptive area is generated according to dustiness data list
Dye degree space two-dimensional array distribution map specifically can be by dustiness data topology at similar rgb image data information, then
The dustiness space two-dimensional array distribution map is inputted into CNN sequentially in time, generates dustiness Multidimensional numerical distribution map；In life
After dustiness Multidimensional numerical distribution map, the high-dimensional feature in dustiness Multidimensional numerical distribution map is extracted, and by high-dimensional spy
Sign is input to LSTM, carries out two-way sequence training arrangement according to time arrangement, generates the dustiness time series of presumptive area；?
After the dustiness time series for generating presumptive area, by dustiness time series entrance loss function, the first calculated result is obtained,
Then the first calculated result is inputted into stochastic gradient descent algorithm, the second calculated result is obtained, according to the second calculated result pair
The dustiness of presumptive area is predicted.It is iterated operation by loss function and stochastic gradient descent algorithm, so that loss
The value of function is smaller, obtains optimum results, exports the prediction result of multi-tag, and the result for comparing legacy data is more accurate.
Fig. 3 shows the flow chart of the dustiness prediction technique of a specific embodiment of the invention.
As shown in figure 3, the dustiness prediction technique includes:
S302 collects data, obtains the concentration values of multi-pollutant from Hengyang；
S304 cleans abnormal data, and excessive or too small unsuitable data are removed from data set；
S306 chooses CNN-LSTM model and is trained；
S308 inputs CNN to multi-pollutant data with spatial information, and temporal information inputs LSTM；
S310 using softmax classifier and intersects entropy loss optimization, and accuracy (accuracy rate) reaches 88% or more.
The dustiness prediction technique that the specific embodiment provides mainly includes following procedure:
Data acquisition: acquisition data really pollute point detection device from Hengyang.Taking data is water pollution, and air is dirty
Dye, the numeric datas such as land pollution.The figure changed with space time is done in the pollutant emission different to diverse geographic location,
Then graphic sequence input model.
Data cleansing: to abnormal points of contamination, the position time is cleaned, wash too large or too small data point, with
The excessive point of plain data gap, enable clear data is used.Normalization Batch norm (batch quota) is done to remaining data
Processing, so that arithmetic speed faster, is easier to be fitted between data distribution 0-1.
Selection Model: it chooses CNN-LSTM model and is used for pollution prediction.The single accuracy rate with CNN LSTM model
Recall rate is all very poor, finally CNN is selected to extract, LSTM does the composite model of time series.
Model training: two-dimensional array distribution is made into according to location distribution to Hengyang Area pollutant data, then may be used
With two-dimensional array data topology at similar rgb image data information, then according to chronological order pollutant figure one by one
Input model first inputs CNN, the geographical location CNN multi-pollutant data is then extracted high dimensional feature, and high dimensional feature is successive
It inputs in Bi-LSTM (two-way time round-robin algorithm), carries out two-way sequence (sequence) sequence training row according to time arrangement
Then cloth is predicted time data later, time series analysis is done, and generates time series.
Model optimization: reselection softmax (loss function) classifier and cross entropy loss (intersect entropy loss
Function) after, iterate operation, so that the value of loss function is smaller, using SGD (stochastic gradient descent) algorithm, obtains excellent
Change as a result, output multi-tag prediction result, compare legacy data as a result, accuracy can promote 88% or more.
The present invention with CNN-LSTM complete in ground local pollution control point prediction, accuracy rate reaches 88% or more, can be
It is used in actual production, has reached expection.Later period can be changed to GRU (neural network) model to model structure, LSTM,
CNN can be changed to multilayer result and improve accuracy rate.
Second aspect of the present invention proposes a kind of dustiness forecasting system 400, as shown in Figure 4, comprising: memory 402 is used
In storage computer program；Processor 404, for execute computer program with: obtain the dustiness data of presumptive area；According to
The dustiness space two-dimensional array distribution map of dustiness data generation presumptive area；Dustiness space two-dimensional array distribution map is defeated
Enter composite network model, generates the dustiness time series of presumptive area；Dirt according to dustiness time series to presumptive area
Dye degree is predicted；Wherein, composite network model is made of convolutional neural networks and length memory network.
Dustiness forecasting system 400 provided by the invention includes mutually matched memory 402 and processor 404, wherein
Processor 404 selects convolutional neural networks and length memory network to form composite network model, real by the composite network model
Now the dustiness of presumptive area is predicted.Specifically, the dustiness data of presumptive area are obtained first, specifically, are obtained different
The dustiness data of geographical location and different time；Then dustiness data are arranged, and generates dustiness space two-dimensional
Array distribution map.It specifically, can be by dustiness data topology at similar rgb image data information；Generating dustiness space
After two-dimensional array distribution map, dustiness space two-dimensional array distribution map is inputted into composite network model, so that composite network model
The dustiness time series of presumptive area is generated according to dustiness space two-dimensional array distribution map；Getting dustiness time sequence
It after column, is predicted according to dustiness of the dustiness time series to presumptive area, to obtain the dustiness prediction of presumptive area
As a result.
The present invention forms composite network model by CNN and LSTM, and dustiness space two-dimensional array distribution map is input to
It is trained, dustiness data space geographical location is inputted using CNN, using LSTM to dustiness in composite network model
Data time point inputs, and CNN and LSTM cooperate, and time and the space of different dustiness data is comprehensively considered, to reach
Optimal prediction effect had not only guaranteed to carry out Accurate Prediction to the dustiness of presumptive area, but also has simplified prediction process.
In an embodiment of the invention, it is preferable that processor 404 is specifically used for: according to time sequencing that dustiness is empty
Between two-dimensional array distribution map input convolutional neural networks, generate dustiness Multidimensional numerical distribution map；Obtain dustiness Multidimensional numerical
Target dimension feature in distribution map；Target dimension feature is inputted into two-way time round-robin algorithm；It is arranged according to the time to target
Dimensional characteristics carry out two-way sequence training arrangement, generate the dustiness time series of presumptive area；Dustiness time series is defeated
Enter loss function, obtains the first calculated result；First calculated result is inputted into stochastic gradient descent algorithm, obtains the second calculating knot
Fruit；It is predicted according to pollution level of second calculated result to presumptive area.
In this embodiment, after getting dustiness space two-dimensional array distribution map, according to time sequencing by dustiness
Space two-dimensional array distribution map inputs convolutional neural networks, generates dustiness Multidimensional numerical distribution map；Generating dustiness multidimensional
After array distribution map, the target dimension feature in dustiness Multidimensional numerical distribution map is extracted.Specifically extracting dustiness multidimensional
High-dimensional feature in array distribution map；After extracting the high-dimensional feature in Multidimensional numerical distribution map, arranged according to the time
Two-way sequence training arrangement is carried out to target dimension feature, generates the dustiness time series of presumptive area；Obtaining fate
After the dustiness time series in domain, by dustiness time series entrance loss function, the first calculated result is obtained, then by first
Calculated result input stochastic gradient descent algorithm, obtain the second calculated result, with according to the second calculated result to presumptive area
Dustiness is predicted.It is iterated operation by loss function and stochastic gradient descent algorithm, so that the value of loss function is more
It is small, optimum results are obtained, the prediction result of multi-tag is exported, the result for comparing legacy data is more accurate.
Specifically, dustiness data include but is not limited to following dustiness data: water pollution degree evidence, air contamination
Data, land pollution degree evidence.More specifically, but be not limited to ph value of acid rain, nitrogen oxides, carbon monoxide, fluoride, lead and
Its compound, radiation level, the dustiness data such as soil pH value.
Third aspect present invention proposes a kind of computer readable storage medium, is stored thereon with computer program, calculates
The dustiness prediction technique such as any one of first aspect present invention is realized when machine program is executed by processor.
The computer readable storage medium that third invention of the present invention provides, is stored thereon with computer program, computer journey
The dustiness prediction technique such as any one of first aspect present invention is realized when sequence is executed by processor, therefore there is above-mentioned pollution
The whole beneficial effects for spending prediction technique, are no longer discussed one by one herein.
In the description of the present invention, term " multiple " then refers to two or more, unless otherwise restricted clearly, term
The orientation or positional relationship of the instructions such as "upper", "lower" is to be based on the orientation or positional relationship shown in the drawings, and is merely for convenience of retouching
It states the present invention and simplifies description, rather than the device or element of indication or suggestion meaning must have a particular orientation, with specific
Orientation construction and operation, therefore be not considered as limiting the invention；Term " connection ", " installation ", " fixation " etc. should all
It is interpreted broadly, for example, " connection " may be fixed connection or may be dismantle connection, or integral connection；It can be straight
Connect it is connected, can also be indirectly connected through an intermediary.It for the ordinary skill in the art, can be according to specific feelings
Condition understands the concrete meaning of above-mentioned term in the present invention.
In the description of this specification, the description of term " one embodiment ", " some embodiments ", " specific embodiment " etc.
Mean that particular features, structures, materials, or characteristics described in conjunction with this embodiment or example are contained at least one reality of the invention
It applies in example or example.In the present specification, schematic expression of the above terms are not necessarily referring to identical embodiment or reality
Example.Moreover, description particular features, structures, materials, or characteristics can in any one or more of the embodiments or examples with
Suitable mode combines.
These are only the preferred embodiment of the present invention, is not intended to restrict the invention, for those skilled in the art
For member, the invention may be variously modified and varied.All within the spirits and principles of the present invention, it is made it is any modification,
Equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of dustiness prediction technique characterized by comprising
Obtain the dustiness data of presumptive area；
The dustiness space two-dimensional array distribution map of the presumptive area is generated according to the dustiness data；
The dustiness space two-dimensional array distribution map is inputted into composite network model, when generating the dustiness of the presumptive area
Between sequence；
The dustiness of the presumptive area is predicted according to the dustiness time series；
Wherein, the composite network model is made of convolutional neural networks and length memory network.
2. dustiness prediction technique according to claim 1, which is characterized in that by the dustiness space two-dimensional array point
The step of Butut inputs composite network model, generates the dustiness time series of the presumptive area, specifically includes:
The dustiness space two-dimensional array distribution map is inputted into the convolutional neural networks according to time sequencing, generates dustiness
Multidimensional numerical distribution map；
Obtain the target dimension feature in the dustiness Multidimensional numerical distribution map；
The target dimension feature is inputted into the length memory network, generates the dustiness time series of the presumptive area.
3. dustiness prediction technique according to claim 2, which is characterized in that will be described in target dimension feature input
Length memory network, specifically includes the step of generating the dustiness time series of the presumptive area:
The target dimension feature is inputted into two-way time round-robin algorithm；
Two-way sequence training arrangement is carried out to the target dimension feature according to time arrangement, generates the pollution of the presumptive area
Spend time series.
4. dustiness prediction technique according to any one of claim 1 to 3, which is characterized in that according to the dustiness
Data generate the step of dustiness space two-dimensional array distribution map of the presumptive area, specifically include:
The dustiness data list of the presumptive area is sequentially generated according to the geographical location of the dustiness data and time；
The dustiness space two-dimensional array distribution map of the presumptive area is generated according to the dustiness data list.
5. dustiness prediction technique according to claim 4, which is characterized in that
The dustiness data include: water pollution degree evidence, air contamination data, land pollution degree evidence.
6. dustiness prediction technique according to claim 4, which is characterized in that generated according to the dustiness data list
The step of dustiness space two-dimensional array distribution map of the presumptive area, specifically further include:
Dustiness data in the dustiness data list are pre-processed；
It carries out batch quota to the pretreated dustiness data to handle, so that by the pretreated dustiness
Data are respectively positioned in target interval；
The dustiness space two-dimensional number of the presumptive area is generated according to the dustiness data being located in the target interval
Group distribution map.
7. dustiness prediction technique according to any one of claim 1 to 3, which is characterized in that according to the dustiness
The step of time series predicts the dustiness of the presumptive area, specifically includes:
By the dustiness time series entrance loss function, the first calculated result is obtained；
First calculated result is inputted into stochastic gradient descent algorithm, obtains the second calculated result；
The dustiness of the presumptive area is predicted according to second calculated result.
8. a kind of dustiness forecasting system characterized by comprising
Memory, for storing computer program；
Processor, for execute the computer program with:
Obtain the dustiness data of presumptive area；
The dustiness space two-dimensional array distribution map of the presumptive area is generated according to the dustiness data；
The dustiness space two-dimensional array distribution map is inputted into composite network model, when generating the dustiness of the presumptive area
Between sequence；
The dustiness of the presumptive area is predicted according to the dustiness time series；
Wherein, the composite network model is made of convolutional neural networks and length memory network.
9. dustiness forecasting system according to claim 8, which is characterized in that the processor is specifically used for:
The dustiness space two-dimensional array distribution map is inputted into the convolutional neural networks according to time sequencing, generates dustiness
Multidimensional numerical distribution map；
Obtain the target dimension feature in the dustiness Multidimensional numerical distribution map；
The target dimension feature is inputted into two-way time round-robin algorithm；
Two-way sequence training arrangement is carried out to the target dimension feature according to time arrangement, generates the pollution of the presumptive area
Spend time series；
By the dustiness time series entrance loss function, the first calculated result is obtained；
First calculated result is inputted into stochastic gradient descent algorithm, obtains the second calculated result；
The pollution level of the presumptive area is predicted according to second calculated result.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The dustiness prediction technique as described in any one of claims 1 to 7 is realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910035757.7A CN109816163A (en) | 2019-01-15 | 2019-01-15 | Dustiness prediction technique, forecasting system and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910035757.7A CN109816163A (en) | 2019-01-15 | 2019-01-15 | Dustiness prediction technique, forecasting system and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109816163A true CN109816163A (en) | 2019-05-28 |
Family
ID=66604166
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910035757.7A Pending CN109816163A (en) | 2019-01-15 | 2019-01-15 | Dustiness prediction technique, forecasting system and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109816163A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110796284A (en) * | 2019-09-20 | 2020-02-14 | 平安科技（深圳）有限公司 | Method and device for predicting pollution level of fine particulate matters and computer equipment |
| CN113496310A (en) * | 2021-06-16 | 2021-10-12 | 国家超级计算深圳中心(深圳云计算中心) | Atmospheric pollutant prediction method and system based on deep learning model |
| CN114414751A (en) * | 2021-12-15 | 2022-04-29 | 华能煤炭技术研究有限公司 | Underground water quality detection method and system |
| FR3120260A1 (en) * | 2021-03-01 | 2022-09-02 | Evaldepol | Method for determining the level of pollution in the subsoil of a site |
| CN115308370A (en) * | 2022-07-29 | 2022-11-08 | 合肥工业大学 | Air quality prediction method based on fusion of multiple neural networks |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108009674A (en) * | 2017-11-27 | 2018-05-08 | 上海师范大学 | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks |
| US20180300608A1 (en) * | 2017-04-12 | 2018-10-18 | Yodlee, Inc. | Neural Networks for Information Extraction From Transaction Data |
-
2019
- 2019-01-15 CN CN201910035757.7A patent/CN109816163A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180300608A1 (en) * | 2017-04-12 | 2018-10-18 | Yodlee, Inc. | Neural Networks for Information Extraction From Transaction Data |
| CN108009674A (en) * | 2017-11-27 | 2018-05-08 | 上海师范大学 | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110796284A (en) * | 2019-09-20 | 2020-02-14 | 平安科技（深圳）有限公司 | Method and device for predicting pollution level of fine particulate matters and computer equipment |
| FR3120260A1 (en) * | 2021-03-01 | 2022-09-02 | Evaldepol | Method for determining the level of pollution in the subsoil of a site |
| CN113496310A (en) * | 2021-06-16 | 2021-10-12 | 国家超级计算深圳中心(深圳云计算中心) | Atmospheric pollutant prediction method and system based on deep learning model |
| CN114414751A (en) * | 2021-12-15 | 2022-04-29 | 华能煤炭技术研究有限公司 | Underground water quality detection method and system |
| CN114414751B (en) * | 2021-12-15 | 2024-05-10 | 华能煤炭技术研究有限公司 | Underground water quality detection method and system |
| CN115308370A (en) * | 2022-07-29 | 2022-11-08 | 合肥工业大学 | Air quality prediction method based on fusion of multiple neural networks |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109816163A (en) | Dustiness prediction technique, forecasting system and computer readable storage medium | |
| Wang et al. | A multidimension cloud model-based approach for water quality assessment | |
| Anselin | Under the hood issues in the specification and interpretation of spatial regression models | |
| Tedesco et al. | A scenario for impacts of water availability loss due to climate change on riverine fish extinction rates | |
| CN108491970A (en) | A kind of Predict Model of Air Pollutant Density based on RBF neural | |
| CN107563122A (en) | The method of crime prediction of Recognition with Recurrent Neural Network is locally connected based on interleaving time sequence | |
| CN111950708B (en) | Neural network structure and method for finding daily life habits of college students | |
| CN110348435A (en) | A kind of object detection method and system based on clipping region candidate network | |
| CN111709318A (en) | High-resolution remote sensing image classification method based on generation countermeasure network | |
| CN115376318B (en) | Traffic data compensation method based on multi-attribute fusion neural network | |
| CN113570861B (en) | Traffic flow prediction method and device based on synthetic data | |
| Efremova et al. | AI-based evaluation of the SDGs: The case of crop detection with earth observation data | |
| CN114387270A (en) | Image processing method, image processing device, computer equipment and storage medium | |
| CN116383589A (en) | Spatial interpolation prediction method for soil nutrients | |
| Movahedi et al. | Interrelated patterns of electricity, gas, and water consumption in large-scale buildings | |
| CN107133693A (en) | Provinces and cities of China life expectancy sequential encryption estimation and its with noctilucence time and space usage method | |
| CN114462572A (en) | Air quality prediction method and device based on space-time fusion diagram | |
| CN110096651A (en) | Visual analysis method based on online social media individual center network | |
| López-Hernández | Second-order polynomial spatial error model. Global and local spatial dependence in unemployment in Andalusia | |
| Courtat | Walk on City Maps-Mathematical and Physical phenomenology of the City, a Geometrical approach | |
| Qi et al. | Social network-oriented learning agent for improving group intelligence coordination | |
| Jokić et al. | Time dynamics of the Dutch municipality network | |
| Martin | Spatial and temporal participation in recreational fishing | |
| Volchenkov | Analysis of urban complex networks | |
| Chen | The Application of tree-based ML algorithm in steel plates faults identification |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190528 |