CN109472396A - Mountain fire prediction technique based on depth e-learning - Google Patents
Mountain fire prediction technique based on depth e-learning Download PDFInfo
- Publication number
- CN109472396A CN109472396A CN201811209832.9A CN201811209832A CN109472396A CN 109472396 A CN109472396 A CN 109472396A CN 201811209832 A CN201811209832 A CN 201811209832A CN 109472396 A CN109472396 A CN 109472396A
- Authority
- CN
- China
- Prior art keywords
- data
- matrix
- depth
- mountain fire
- meteorological
- 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 52
- 238000012545 processing Methods 0.000 claims abstract description 18
- 239000011159 matrix material Substances 0.000 claims description 105
- 238000007500 overflow downdraw method Methods 0.000 claims description 24
- 230000002123 temporal effect Effects 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 12
- 230000004927 fusion Effects 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000006116 polymerization reaction Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 102000008297 Nuclear Matrix-Associated Proteins Human genes 0.000 claims description 3
- 108010035916 Nuclear Matrix-Associated Proteins Proteins 0.000 claims description 3
- 210000000299 nuclear matrix Anatomy 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 abstract description 5
- 238000000605 extraction Methods 0.000 abstract description 4
- 230000010354 integration Effects 0.000 abstract description 4
- 239000000446 fuel Substances 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000007246 mechanism Effects 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
- 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Development Economics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Educational Administration (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of mountain fire prediction techniques based on depth e-learning that can be improved forecasting efficiency and forecasting accuracy.The mountain fire prediction technique based on depth e-learning learns mass historical data by the Automatic Feature Extraction network on building time and Spatial Dimension, reaches the analysis of mountain fire prediction and warning.Bottleneck of traditional mode of learning in mass data is overcome, GPU and distribution can be efficiently used quickly to be trained;The mode of traditional artificial feature has been abandoned, has realized feature automation;Depth network has been used flexibly to carry out effective integration to different dimensions feature, to improve the forecasting efficiency and forecasting accuracy of mountain fire risk profile.It is suitble to promote and apply in technical field of data processing.
Description
Technical field
The present invention relates to technical field of data processing, especially a kind of mountain fire prediction technique based on depth e-learning.
Background technique
In natural calamity, mountain fire is a kind of very special and very damaging disaster;Ecology can not only be caused
Significant impact, also can be to the people of spot, animal and some " assets " cause heavy losses;Such as the tower in power equipment, line
Cable, substation;Signal tower in communication etc..And the genesis mechanism of mountain fire is extremely complex, just can under many factors interaction
Mountain fire occurs.
With the development of the technologies such as Internet of Things, digitlization, Internet of Things, make it possible to get largely by reasonable manner
Data can carry out prediction, pre- to natural calamity by way of non-experiment in conjunction with the promotion of present computing capability
It is alert etc..Secondly, present data are not only big data, and data diversification, data are many and diverse.Similar natural calamity is asked
Topic, not only contains geographical space temperature, further comprises time dimension, the two dimensions it is effective combine modeling be also it is present very
The difficult point of more traditional mathematics models.Therefore, the present invention mainly utilizes current hot spot technology " deep learning " to solve time and sky
Between the problem of modeling, and the characteristics of combine mountain fire, applied in mountain fire prediction and warning.
Traditional characteristic extracting mode is to extract respectively in room and time dimension, then carry out subsequent analysis and build
Mould, there are certain subjectivities for the mode of this feature extraction, and are not easily found effective feature to support subsequent mould
It is poor to will lead to modelling effect for type.
Traditional machine learning model is the mode based on manual features, and feature spatially cannot be carried out to effectively benefit
With being modeled on time dimension, traditional time series cannot establish long-term time dependence, and time series is mainly still
Based on statistics.The machine learning algorithm of this quasi-tradition designs upper no space or the concept of time, is all based on the study of feature
Process.
On Fusion Features, each feature can be typically added on each sample, even if there is feature of overall importance, this
Kind mode is inflexible, when different types of feature or inconsistent feature are also possible that problem when merging.
Traditional mode of learning is on the basis of certain data volume, and data volume is too small or big city excessively goes wrong.Work as number
It is too small according to measuring, it is very low to the confidence level of result.But traditional mode of learning is again extremely difficult to the processing of mass data, has certain
Bottleneck.
Summary of the invention
Technical problem to be solved by the invention is to provide it is a kind of can be improved forecasting efficiency and forecasting accuracy based on
The mountain fire prediction technique of depth e-learning.
The technical solution adopted by the present invention to solve the technical problems are as follows: should the mountain fire prediction side based on depth e-learning
Method, comprising the following steps:
1), mountain fire risk forecast model is established using depth online learning methods;
The mountain fire risk forecast model includes convolutional layer, fused layer, LSTM layers, full articulamentum;
2), for the needs of mountain fire risk profile task, acquisition is since current time forward in the phase of history time
Remotely-sensed data and meteorological data;
3), handle to obtain the gas as unit of day by temporal resolution fusion method to the meteorological data of 2) step acquisition
Image data;
4), the meteorological data as unit of day for obtaining remotely-sensed data and 3) step passes through spatial data fusion method
Realize the fusion matching of spatial data;
5), by space is all converted to by remotely-sensed data that 4) step process obtains, meteorological data, rate is 500m* respectively
500m;Temporal resolution is 1 day, and format is the data of tiff format;
6) data that step 5) obtains, are obtained into batch data by following processing, concrete processing procedure is as described below:
A, tiff data are resolved to matrix form by the parsing of tiff data;Missing Data Filling, filling are carried out to data again
Mode be to be filled by adjacent point: it is specific as follows shown in:
x′I, j=xI-1, j|xI, j-1;
B, the remotely-sensed data of matrix form is divided into multiple remotely-sensed data matrixes, and records the longitude and latitude in each point
Position;
C, meteorological data is corresponded to according to longitude and latitude position, is generated identical as remotely-sensed data matrix size in b step
Meteorological data matrix;
D, meteorological data matrix and remotely-sensed data matrix are merged and ultimately generates large-scale matrix, the format of matrix are as follows:
[time step number, matrix line number, matrix columns, characteristic];
E, target sample matrix is generated by history fire point data, is generated in the size and step d of target sample matrix
Large-scale matrix is consistent;
F, the data of the large-scale matrix generated in the data of target sample matrix and step d are mapped, and will be entire
Data are divided into different size of data set, can form batch data;
7), by the convolutional layer of batch data input mountain fire risk forecast model, and the same day is extracted from batch data
Meteorological data inputs the full articulamentum of mountain fire risk forecast model, and final mountain fire Occurrence forecast probability can be obtained.
Further, the temporal resolution fusion method is described in detail below in 3) step:
By the d days meteorological index m in observation data m hourlyd,i(i ∈ 0 ..., 23) daily carry out polymerization generation accordingly
Statistical indicator, comprising: same day maximum value max_md, minimum value min_md, very poor value range_md, as daily amount total_md;
Wherein, max_md=max (md,i)(i∈0,…,23)
min_md=min (md,i)(i∈0,…,23)
range_md=max (md,i)-min(md,i)(i∈0,…,23)
total_md=sum (md,i)(i∈0,…,23)
Maximum value max_md, minimum value min_mdFor temperature, humidity, rainfall, wind speed, very poor value range_mdSystem is used for
Temperature, humidity, as daily amount total_mdFor rainfall level.
Further, the spatial data fusion method is used close to matching method or the interpolation side krige in 4) step
Method.
Further, in 1) step, using depth online learning methods establish mountain fire risk forecast model include with
Lower step:
A, for the needs of mountain fire risk profile task, acquisition is distant in the phase of history time forward since current time
Feel data and meteorological data;
B, handle to obtain the meteorology as unit of day by temporal resolution fusion method to the meteorological data of step A acquisition
Data;
C, the meteorological data as unit of day for obtaining remotely-sensed data and step B is real by spatial data fusion method
The fusion matching of existing spatial data;
D, the remotely-sensed data, the meteorological data that handle by step C are all converted to space rate is 500m* respectively
500m;Temporal resolution is 1 day, and format is the data of tiff format;
E, the data that step D is obtained are obtained into batch data by following processing, concrete processing procedure is as described below:
A, tiff data are resolved to matrix form by the parsing of tiff data;Missing Data Filling, filling are carried out to data again
Mode be to be filled by adjacent point: it is specific as follows shown in:
x′I, j=xI-1, j|xI, j-1;
B, the remotely-sensed data of matrix form is divided into multiple remotely-sensed data matrixes, and records the longitude and latitude in each point
Position;
C, meteorological data is corresponded to according to longitude and latitude position, is generated identical as remotely-sensed data matrix size in b step
Meteorological data matrix;
D, meteorological data matrix and remotely-sensed data matrix are merged and ultimately generates large-scale matrix, the format of matrix are as follows:
[time step number, matrix line number, matrix columns, characteristic];
E, target sample matrix is generated by history fire point data, is generated in the size and step d of target sample matrix
Large-scale matrix is consistent;
F, the data of the large-scale matrix generated in the data of target sample matrix and step d are mapped, and will be entire
Data are divided into different size of data set, can form batch data;
F, based on batch data construct depth network model, the depth network model include convolutional layer, fused layer,
LSTM layers, full articulamentum;
The convolutional layer is stacked by multiple convolution units, and each convolution unit obtains with the following method: first
Batch data are subjected to convolutional calculation as input matrix and obtain convolution results, the convolutional calculation formula are as follows: C=A*B,
Middle * indicates that convolution algorithm, A indicate input matrix and a B convolution nuclear matrix, and the convolution algorithm uses following discrete calculation
Formula:Wherein m, n are the sizes values of convolution kernel.wm,nFor m in convolution kernel, the position n
Value, b be bigoted item;Then obtained convolution results are subjected to maximum pond and obtain convolution unit
The fused layer merges obtained convolution unit using ADD method, specifically, the volume that convolutional layer is obtained
Product unit carries out flatten expansion, and new feature is added, and the shape of the new feature is consistent with the matrix size of input,
Stack is carried out according still further to time dimension, obtains fusion matrix [batch_size, time_steps, depth], wherein depth=
w*h;
Described LSTM layers includes LSTMCell layers, dropper layers and full-connect layers, will merge matrix [batch_
Size, time_steps, depth] it inputs after LSTMCell layers and successively passes through at dropper layers and full-connect layers again
It manages and is exported to obtain LSTM data using result function, the result function is using dimension tanh function output, the dimension
Tanh function is as described below:
The full articulamentum merges the same day meteorological data for obtaining LSTM data and extracting from batch data
Processing;
G, depth network model is trained using more epoch, and carries out the instruction of depth network model using more GPU
Practice, CPU is responsible for the update of parameter, and GPU carries out the training of network, obtains final mountain fire prediction model.
Further, the temporal resolution fusion method is described in detail below in step B:
By the d days meteorological index m in observation data m hourlyd,i(i ∈ 0 ..., 23) daily carry out polymerization generation accordingly
Statistical indicator, comprising: same day maximum value max_md, minimum value min_md, very poor value range_md, as daily amount total_md;
Wherein, max_md=max (md,i)(i∈0,…,23)
min_md=min (md,i)(i∈0,…,23)
range_md=max (md,i)-min(md,i)(i∈0,…,23)
total_md=sum (md,i)(i∈0,…,23)
Maximum value max_md, minimum value min_mdFor temperature, humidity, rainfall, wind speed, very poor value range_mdSystem is used for
Temperature, humidity, as daily amount total_mdFor rainfall level.
Further, the spatial data fusion method is used close to matching method or the interpolation side krige in step C
Method.
Beneficial effects of the present invention: building time and space dimension should be passed through based on the mountain fire prediction technique of depth e-learning
Automatic Feature Extraction network on degree, and mass historical data is learnt, reach the analysis of mountain fire prediction and warning.Overcome biography
Bottleneck of the system mode of learning in mass data, can efficiently use GPU and distribution quickly to be trained;Traditional people is abandoned
The mode of work feature realizes feature automation;Depth network has been used flexibly to carry out effective integration to different dimensions feature,
To improve the forecasting efficiency and forecasting accuracy of mountain fire risk profile.
Specific embodiment
The mountain fire prediction technique based on depth e-learning, comprising the following steps:
1), mountain fire risk forecast model is established using depth online learning methods;
The mountain fire risk forecast model includes convolutional layer, fused layer, LSTM layers, full articulamentum;
2), for the needs of mountain fire risk profile task, acquisition is since current time forward in the phase of history time
Remotely-sensed data and meteorological data;The remotely-sensed data includes: Fuel loads FMC, combustible load FL, fuel type
FT;Elevation, the gradient, slope aspect;The meteorological data includes temperature, humidity, rainfall, wind speed, wind direction;
3), handle to obtain the gas as unit of day by temporal resolution fusion method to the meteorological data of 2) step acquisition
Image data;The temporal resolution fusion method is described in detail below:
By the d days meteorological index m in observation data m hourlyd,i(i ∈ 0 ..., 23) daily carry out polymerization generation accordingly
Statistical indicator, comprising: same day maximum value max_md, minimum value min_md, very poor value range_md, as daily amount total_md;
Wherein, max_md=max (md,i)(i∈0,…,23)
min_md=min (md,i)(i∈0,…,23)
range_md=max (md,i)-min(md,i)(i∈0,…,23)
total_md=sum (md,i)(i∈0,…,23)
Maximum value max_md, minimum value min_mdFor temperature, humidity, rainfall, wind speed, very poor value range_mdSystem is used for
Temperature, humidity, as daily amount total_mdFor rainfall level;
4), the meteorological data as unit of day for obtaining remotely-sensed data and 3) step passes through spatial data fusion method
Realize the fusion matching of spatial data;The spatial data fusion method is used close to matching method or krige interpolation method;
5), by space is all converted to by remotely-sensed data that 4) step process obtains, meteorological data, rate is 500m* respectively
500m;Temporal resolution is 1 day, and format is the data of tiff format;
6) data that step 5) obtains, are obtained into batch data by following processing, concrete processing procedure is as described below:
A, tiff data are resolved to matrix form by the parsing of tiff data;Missing Data Filling, filling are carried out to data again
Mode be to be filled by adjacent point: it is specific as follows shown in:
x′I, j=xI-1, j|xI, j-1;
B, the remotely-sensed data of matrix form is divided into multiple remotely-sensed data matrixes, and records the longitude and latitude in each point
Position;
C, meteorological data is corresponded to according to longitude and latitude position, is generated identical as remotely-sensed data matrix size in b step
Meteorological data matrix;
D, meteorological data matrix and remotely-sensed data matrix are merged and ultimately generates large-scale matrix, the format of matrix are as follows:
[time step number, matrix line number, matrix columns, characteristic];
E, target sample matrix is generated by history fire point data, is generated in the size and step d of target sample matrix
Large-scale matrix is consistent;
F, the data of the large-scale matrix generated in the data of target sample matrix and step d are mapped, and will be entire
Data are divided into different size of data set, can form batch data;During batch data assembling, positive example data are less,
In order to solve proportional imbalance, all positive example data be assembled into each batch in each use;
7), by the convolutional layer of batch data input mountain fire risk forecast model, and the same day is extracted from batch data
Meteorological data inputs the full articulamentum of mountain fire risk forecast model, and final mountain fire Occurrence forecast probability can be obtained.
This is mentioned based on the mountain fire prediction technique of depth e-learning by the automated characterization on building time and Spatial Dimension
Network is taken, and mass historical data is learnt, reaches the analysis of mountain fire prediction and warning.Traditional mode of learning is overcome in magnanimity
Bottleneck in data can efficiently use GPU and distribution quickly to be trained;The mode of traditional artificial feature has been abandoned, it is real
Existing feature automation;Depth network has been used flexibly to carry out effective integration to different dimensions feature, to improve mountain fire
The forecasting efficiency and forecasting accuracy of risk profile.
In the above-described embodiment, in 1) step, mountain fire risk forecast model is established using depth online learning methods
The following steps are included:
A, for the needs of mountain fire risk profile task, acquisition is distant in the phase of history time forward since current time
Feel data and meteorological data;The remotely-sensed data includes: Fuel loads FMC, combustible load FL, fuel type FT;
Elevation, the gradient, slope aspect;The meteorological data includes temperature, humidity, rainfall, wind speed, wind direction;
B, handle to obtain the meteorology as unit of day by temporal resolution fusion method to the meteorological data of step A acquisition
Data;The temporal resolution fusion method is described in detail below:
By the d days meteorological index m in observation data m hourlyd,i(i ∈ 0 ..., 23) daily carry out polymerization generation accordingly
Statistical indicator, comprising: same day maximum value max_md, minimum value min_md, very poor value range_md, as daily amount total_md;
Wherein, max_md=max (md,i)(i∈0,…,23)
min_md=min (md,i)(i∈0,…,23)
range_md=max (md,i)-min(md,i)(i∈0,…,23)
total_md=sum (md,i)(i∈0,…,23)
Maximum value max_md, minimum value min_mdFor temperature, humidity, rainfall, wind speed, very poor value range_mdSystem is used for
Temperature, humidity, as daily amount total_mdFor rainfall level;
C, the meteorological data as unit of day for obtaining remotely-sensed data and step B is real by spatial data fusion method
The fusion matching of existing spatial data;The spatial data fusion method is used close to matching method or krige interpolation method;
D, the remotely-sensed data, the meteorological data that handle by step C are all converted to space rate is 500m* respectively
500m;Temporal resolution is 1 day, and format is the data of tiff format;
E, the data that step D is obtained are obtained into batch data by following processing, concrete processing procedure is as described below:
A, tiff data are resolved to matrix form by the parsing of tiff data;Missing Data Filling, filling are carried out to data again
Mode be to be filled by adjacent point: it is specific as follows shown in:
x′I, j=xI-1, j|xI, j-1;
B, the remotely-sensed data of matrix form is divided into multiple remotely-sensed data matrixes, and records the longitude and latitude in each point
Position;
C, meteorological data is corresponded to according to longitude and latitude position, is generated identical as remotely-sensed data matrix size in b step
Meteorological data matrix;
D, meteorological data matrix and remotely-sensed data matrix are merged and ultimately generates large-scale matrix, the format of matrix are as follows:
[time step number, matrix line number, matrix columns, characteristic];
E, target sample matrix is generated by history fire point data, is generated in the size and step d of target sample matrix
Large-scale matrix is consistent;
F, the data of the large-scale matrix generated in the data of target sample matrix and step d are mapped, and will be entire
Data are divided into different size of data set, can form batch data;During batch data assembling, positive example data are less,
In order to solve proportional imbalance, all positive example data be assembled into each batch in each use;
F, based on batch data construct depth network model, the depth network model include convolutional layer, fused layer,
LSTM layers, full articulamentum;
The convolutional layer is stacked by multiple convolution units, and each convolution unit obtains with the following method: first
Batch data are subjected to convolutional calculation as input matrix and obtain convolution results, the convolutional calculation formula are as follows: C=A*B,
Middle * indicates that convolution algorithm, A indicate input matrix and a B convolution nuclear matrix, and the convolution algorithm uses following discrete calculation
Formula:Wherein m, n are the sizes values of convolution kernel.wm,nFor m in convolution kernel, the position n
Value, b be bigoted item;Then obtained convolution results are subjected to maximum pond and obtain convolution unit
The fused layer merges obtained convolution unit using ADD method, specifically, the volume that convolutional layer is obtained
Product unit carries out flatten expansion, and new feature is added, and the shape of the new feature is consistent with the matrix size of input,
Stack is carried out according still further to time dimension, obtains fusion matrix [batch_size, time_steps, depth], wherein depth=
w*h;
Described LSTM layers includes LSTMCell layers, dropper layers and full-connect layers, will merge matrix [batch_
Size, time_steps, depth] it inputs after LSTMCell layers and successively passes through at dropper layers and full-connect layers again
It manages and is exported to obtain LSTM data using result function, the result function is using dimension tanh function output, the dimension
Tanh function is as described below:
The full articulamentum merges the same day meteorological data for obtaining LSTM data and extracting from batch data
Processing;
G, depth network model is trained using more epoch, and carries out the instruction of depth network model using more GPU
Practice, CPU is responsible for the update of parameter, and GPU carries out the training of network, obtains final mountain fire prediction model.
The mountain fire prediction model method for building up based on depth e-learning passes through oneself on building time and Spatial Dimension
Dynamic feature extraction network, and mass historical data is learnt, reach the analysis of mountain fire prediction and warning, overcomes traditional study side
Bottleneck of the formula in mass data can efficiently use GPU and distribution quickly to be trained;Traditional artificial feature is abandoned
Mode realizes feature automation;Depth network has been used flexibly to carry out effective integration to different dimensions feature, to be promoted
The forecasting efficiency and forecasting accuracy of mountain fire risk forecast model.
Claims (6)
1. the mountain fire prediction technique based on depth e-learning, it is characterised in that the following steps are included:
1), mountain fire risk forecast model is established using depth online learning methods;
The mountain fire risk forecast model includes convolutional layer, fused layer, LSTM layers, full articulamentum;
2), for the needs of mountain fire risk profile task, the remote sensing in the phase of history time forward is acquired since current time
Data and meteorological data;
3), handle to obtain the meteorological number as unit of day by temporal resolution fusion method to the meteorological data of 2) step acquisition
According to;
4), the meteorological data as unit of day for obtaining remotely-sensed data and 3) step is realized by spatial data fusion method
The fusion of spatial data matches;
5), by space is all converted to by remotely-sensed data that 4) step process obtains, meteorological data, rate is 500m* respectively
500m;Temporal resolution is 1 day, and format is the data of tiff format;
6) data that step 5) obtains, are obtained into batch data by following processing, concrete processing procedure is as described below:
A, tiff data are resolved to matrix form by the parsing of tiff data;Missing Data Filling, the side of filling are carried out to data again
Formula is filled by adjacent point: shown in specific as follows:
x′I, j=xI-1, j|xI, j-1;
B, the remotely-sensed data of matrix form is divided into multiple remotely-sensed data matrixes, and records the longitude and latitude position in each point
It sets;
C, meteorological data is corresponded to according to longitude and latitude position, generates gas identical with remotely-sensed data matrix size in b step
Image data matrix;
D, meteorological data matrix and remotely-sensed data matrix are merged and ultimately generates large-scale matrix, the format of matrix are as follows: [time
Step number, matrix line number, matrix columns, characteristic];
E, target sample matrix, the large size generated in the size and step d of target sample matrix are generated by history fire point data
Matrix is consistent;
F, the data of the large-scale matrix generated in the data of target sample matrix and step d are mapped, and by entire data
It is divided into different size of data set, batch data can be formed;
7), batch data are inputted to the convolutional layer of mountain fire risk forecast model, and extract the meteorology on the same day from batch data
Data input the full articulamentum of mountain fire risk forecast model, and final mountain fire Occurrence forecast probability can be obtained.
2. as described in claim 1 based on the mountain fire prediction technique of depth e-learning, it is characterised in that: in 3) step,
The temporal resolution fusion method is described in detail below:
By the d days meteorological index m in observation data m hourlyd,i(i ∈ 0 ..., 23) daily carry out the corresponding statistics of polymerization generation
Index, comprising: same day maximum value max_md, minimum value min_md, very poor value range_md, as daily amount total_md;
Wherein, max_md=max (md,i)(i∈0,…,23)
min_md=min (md,i)(i∈0,…,23)
range_md=max (md,i)-min(md,i)(i∈0,…,23)
total_md=sum (md,i)(i∈0,…,23)
Maximum value max_md, minimum value min_mdFor temperature, humidity, rainfall, wind speed, very poor value range_mdSystem for temperature,
Humidity, as daily amount total_mdFor rainfall level.
3. as claimed in claim 2 based on the mountain fire prediction technique of depth e-learning, it is characterised in that: in 4) step,
The spatial data fusion method is used close to matching method or krige interpolation method.
4. as described in claim 1 based on the mountain fire prediction technique of depth e-learning, it is characterised in that: in 1) step,
Establish mountain fire risk forecast model using depth online learning methods the following steps are included:
A, for the needs of mountain fire risk profile task, the remote sensing number in the phase of history time forward is acquired since current time
According to and meteorological data;
B, handle to obtain the meteorological number as unit of day by temporal resolution fusion method to the meteorological data of step A acquisition
According to;
C, the meteorological data as unit of day for obtaining remotely-sensed data and step B is realized empty by spatial data fusion method
Between data fusion matching;
D, the remotely-sensed data, the meteorological data that handle by step C are all converted to space rate is 500m*500m respectively;
Temporal resolution is 1 day, and format is the data of tiff format;
E, the data that step D is obtained are obtained into batch data by following processing, concrete processing procedure is as described below:
A, tiff data are resolved to matrix form by the parsing of tiff data;Missing Data Filling, the side of filling are carried out to data again
Formula is filled by adjacent point: shown in specific as follows:
x′I, j=xI-1, j|xI, j-1;
B, the remotely-sensed data of matrix form is divided into multiple remotely-sensed data matrixes, and records the longitude and latitude position in each point
It sets;
C, meteorological data is corresponded to according to longitude and latitude position, generates gas identical with remotely-sensed data matrix size in b step
Image data matrix;
D, meteorological data matrix and remotely-sensed data matrix are merged and ultimately generates large-scale matrix, the format of matrix are as follows: [time
Step number, matrix line number, matrix columns, characteristic];
E, target sample matrix, the large size generated in the size and step d of target sample matrix are generated by history fire point data
Matrix is consistent;
F, the data of the large-scale matrix generated in the data of target sample matrix and step d are mapped, and by entire data
It is divided into different size of data set, batch data can be formed;
F, based on batch data construct depth network model, the depth network model include convolutional layer, fused layer, LSTM layers,
Full articulamentum;
The convolutional layer is stacked by multiple convolution units, and each convolution unit obtains with the following method: first will
Batch data carry out convolutional calculation as input matrix and obtain convolution results, the convolutional calculation formula are as follows: C=A*B, wherein *
Indicate that convolution algorithm, A indicate input matrix and a B convolution nuclear matrix, the convolution algorithm is public using following discrete calculation
Formula:Wherein m, n are the sizes values of convolution kernel.wm,nFor m in convolution kernel, the position n
Value, b are bigoted item;Then obtained convolution results are subjected to maximum pond and obtain convolution unit
The fused layer merges obtained convolution unit using ADD method, specifically, the convolution list that convolutional layer is obtained
Member carries out flatten expansion, and new feature is added, and the shape of the new feature is consistent with the matrix size of input, then presses
Stack is carried out according to time dimension, obtains fusion matrix [batch_size, time_steps, depth], wherein depth=w*h;
Described LSTM layers include LSTMCell layer, dropper layers and full-connect layers, will merge matrix [batch_size,
Time_steps, depth] it inputs after LSTMCell layers and successively passes through dropper layers and full-connect layers processing again and make
It is exported to obtain LSTM data with result function, the result function is using dimension tanh function output, the dimension tanh function
It is as described below:
The same day meteorological data for obtaining LSTM data and extracting from batch data is carried out fusion place by the full articulamentum
Reason;
G, depth network model is trained using more epoch, and carries out the training of depth network model using more GPU,
CPU is responsible for the update of parameter, and GPU carries out the training of network, obtains final mountain fire prediction model.
5. as claimed in claim 4 based on the mountain fire prediction technique of depth e-learning, it is characterised in that: in step B, institute
It is described in detail below to state temporal resolution fusion method:
By the d days meteorological index m in observation data m hourlyd,i(i ∈ 0 ..., 23) daily carry out the corresponding statistics of polymerization generation
Index, comprising: same day maximum value max_md, minimum value min_md, very poor value range_md, as daily amount total_md;
Wherein, max_md=max (md,i)(i∈0,…,23)
min_md=min (md,i)(i∈0,…,23)
range_md=max (md,i)-min(md,i)(i∈0,…,23)
total_md=sum (md,i)(i∈0,…,23)
Maximum value max_md, minimum value min_mdFor temperature, humidity, rainfall, wind speed, very poor value range_mdSystem for temperature,
Humidity, as daily amount total_mdFor rainfall level.
6. as claimed in claim 5 based on the mountain fire prediction technique of depth e-learning, it is characterised in that: in step C, institute
Spatial data fusion method is stated to use close to matching method or krige interpolation method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811209832.9A CN109472396B (en) | 2018-10-17 | 2018-10-17 | Mountain fire prediction method based on deep network learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811209832.9A CN109472396B (en) | 2018-10-17 | 2018-10-17 | Mountain fire prediction method based on deep network learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109472396A true CN109472396A (en) | 2019-03-15 |
CN109472396B CN109472396B (en) | 2023-06-20 |
Family
ID=65665116
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811209832.9A Active CN109472396B (en) | 2018-10-17 | 2018-10-17 | Mountain fire prediction method based on deep network learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109472396B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110210676A (en) * | 2019-06-06 | 2019-09-06 | 国网湖南省电力有限公司 | The long-term prediction of situation method and system of transmission line forest fire |
CN111445011A (en) * | 2020-04-01 | 2020-07-24 | 成都思晗科技股份有限公司 | Mountain fire early warning method based on meteorological and remote sensing data |
CN111460733A (en) * | 2020-04-01 | 2020-07-28 | 成都卡普数据服务有限责任公司 | Mountain fire early warning method based on deep learning |
CN111882128A (en) * | 2020-07-28 | 2020-11-03 | 中原工学院 | TCN-based flood season climate trend prediction method |
CN111931645A (en) * | 2020-08-10 | 2020-11-13 | 成都思晗科技股份有限公司 | Real-time mountain fire risk monitoring method based on remote sensing data |
CN112395924A (en) * | 2019-08-16 | 2021-02-23 | 阿里巴巴集团控股有限公司 | Remote sensing monitoring method and device |
CN113486286A (en) * | 2021-06-08 | 2021-10-08 | 电子科技大学 | Method for estimating water content of 10-h dead combustible by combining deep learning and physical model |
CN113553764A (en) * | 2021-07-13 | 2021-10-26 | 广东工业大学 | Mountain fire prediction method based on deep learning network |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105678419A (en) * | 2016-01-05 | 2016-06-15 | 天津大学 | Fine grit-based forest fire hazard probability forecasting system |
CN107104978A (en) * | 2017-05-24 | 2017-08-29 | 赖洪昌 | A kind of network risks method for early warning based on deep learning |
US20180096253A1 (en) * | 2016-10-04 | 2018-04-05 | Civicscape, LLC | Rare event forecasting system and method |
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 |
CN108510132A (en) * | 2018-07-03 | 2018-09-07 | 华际科工(北京)卫星通信科技有限公司 | A kind of sea-surface temperature prediction technique based on LSTM |
CN108520363A (en) * | 2018-04-18 | 2018-09-11 | 电子科技大学 | A kind of appraisal procedure for predicting the following phase forest fire occurrence risk |
-
2018
- 2018-10-17 CN CN201811209832.9A patent/CN109472396B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105678419A (en) * | 2016-01-05 | 2016-06-15 | 天津大学 | Fine grit-based forest fire hazard probability forecasting system |
US20180096253A1 (en) * | 2016-10-04 | 2018-04-05 | Civicscape, LLC | Rare event forecasting system and method |
CN107104978A (en) * | 2017-05-24 | 2017-08-29 | 赖洪昌 | A kind of network risks method for early warning based on deep learning |
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 |
CN108520363A (en) * | 2018-04-18 | 2018-09-11 | 电子科技大学 | A kind of appraisal procedure for predicting the following phase forest fire occurrence risk |
CN108510132A (en) * | 2018-07-03 | 2018-09-07 | 华际科工(北京)卫星通信科技有限公司 | A kind of sea-surface temperature prediction technique based on LSTM |
Non-Patent Citations (2)
Title |
---|
孔亚奇: "基于弱监督深度网络的火灾检测技术研究与实现", 《工程科技Ⅰ辑》 * |
林作永等: "基于深度卷积神经网络的火灾预警算法研究", 《信息通信》 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110210676A (en) * | 2019-06-06 | 2019-09-06 | 国网湖南省电力有限公司 | The long-term prediction of situation method and system of transmission line forest fire |
CN112395924B (en) * | 2019-08-16 | 2024-02-20 | 阿里巴巴集团控股有限公司 | Remote sensing monitoring method and device |
CN112395924A (en) * | 2019-08-16 | 2021-02-23 | 阿里巴巴集团控股有限公司 | Remote sensing monitoring method and device |
CN111445011B (en) * | 2020-04-01 | 2023-07-28 | 成都思晗科技股份有限公司 | Mountain fire early warning method based on meteorological and remote sensing data |
CN111445011A (en) * | 2020-04-01 | 2020-07-24 | 成都思晗科技股份有限公司 | Mountain fire early warning method based on meteorological and remote sensing data |
CN111460733A (en) * | 2020-04-01 | 2020-07-28 | 成都卡普数据服务有限责任公司 | Mountain fire early warning method based on deep learning |
CN111460733B (en) * | 2020-04-01 | 2023-10-03 | 成都卡普数据服务有限责任公司 | Mountain fire early warning method based on deep learning |
CN111882128B (en) * | 2020-07-28 | 2021-09-28 | 中原工学院 | TCN-based flood season climate trend prediction method |
CN111882128A (en) * | 2020-07-28 | 2020-11-03 | 中原工学院 | TCN-based flood season climate trend prediction method |
CN111931645B (en) * | 2020-08-10 | 2023-05-23 | 成都思晗科技股份有限公司 | Real-time mountain fire risk monitoring method based on remote sensing data |
CN111931645A (en) * | 2020-08-10 | 2020-11-13 | 成都思晗科技股份有限公司 | Real-time mountain fire risk monitoring method based on remote sensing data |
CN113486286B (en) * | 2021-06-08 | 2023-04-07 | 电子科技大学 | Method for estimating water content of 10-h dead combustible by combining deep learning and physical model |
CN113486286A (en) * | 2021-06-08 | 2021-10-08 | 电子科技大学 | Method for estimating water content of 10-h dead combustible by combining deep learning and physical model |
CN113553764A (en) * | 2021-07-13 | 2021-10-26 | 广东工业大学 | Mountain fire prediction method based on deep learning network |
CN113553764B (en) * | 2021-07-13 | 2023-08-04 | 广东工业大学 | Mountain fire prediction method based on deep learning network |
Also Published As
Publication number | Publication date |
---|---|
CN109472396B (en) | 2023-06-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109472396A (en) | Mountain fire prediction technique based on depth e-learning | |
CN109508476A (en) | Mountain fire based on depth e-learning predicts modeling method | |
CN104298841B (en) | A kind of Flood Forecasting Method and system based on historical data | |
CN114493050B (en) | Multi-dimensional fusion new energy power parallel prediction method and device | |
CN104217384A (en) | A nuclear accident emergency processing and auxiliary decision support system | |
CN114493052B (en) | Multi-model fusion self-adaptive new energy power prediction method and system | |
CN111968019A (en) | Landslide hazard susceptibility prediction method based on CNN and RNN integration | |
CN116882211B (en) | Reservoir water condition forecasting simulation method and system based on digital twin | |
CN109299208A (en) | Transmission tower intelligent visual methods of risk assessment under a kind of typhoon disaster | |
CN117351521A (en) | Digital twinning-based power transmission line bird detection method, system, medium and equipment | |
CN116722544A (en) | Distributed photovoltaic short-term prediction method and device, electronic equipment and storage medium | |
CN106446546A (en) | Meteorological data complement method based on automatic convolutional encoding and decoding algorithm | |
CN106771847A (en) | A kind of 35kV power distribution networks transmission line lightning stroke Risk Forecast Method | |
TWI587222B (en) | Neural networks based water level predicting system and method for reservoirs | |
CN112381332A (en) | Population spatial distribution prediction method based on settlement object | |
CN114418243B (en) | Distributed new energy cloud grid prediction method and system | |
CN106295175A (en) | Station meteorological data missing value fill method based on svd algorithm | |
CN106202920A (en) | A kind of application and interpretation method of single sea level pressure of standing | |
CN113869804B (en) | Power grid equipment risk early warning method and system under flood disaster | |
CN115830807A (en) | Mine safety early warning method, system, equipment and medium | |
CN113779892B (en) | Method for predicting wind speed and wind direction | |
Lin et al. | Integrating ANFIS and Qt Framework to Develop a Mobile‐Based Typhoon Rainfall Forecasting System | |
CN114552570A (en) | Offshore wind power prediction management system | |
Gao et al. | Research on integrated management platform of smart park based on CIM | |
CN112784477A (en) | WRF-LES and BP-PSO-Bagging combined wind power prediction method |
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 | ||
TA01 | Transfer of patent application right |
Effective date of registration: 20230112 Address after: 610000 1, 3, 1, 366 north section of lakeside road, Tianfu New District, Chengdu, Sichuan Applicant after: Chengdu Cap Data Service Co.,Ltd. Address before: Room 503, 5th floor, unit 1, building 12, 333 Taihe 2nd Street, high tech Zone, Chengdu, Sichuan 610041 Applicant before: SICHUAN JIALIAN ZHONGHE ENTERPRISE MANAGEMENT CONSULTATION Co.,Ltd. |
|
TA01 | Transfer of patent application right | ||
GR01 | Patent grant | ||
GR01 | Patent grant |