CN109472396B - Mountain fire prediction method based on deep network learning - Google Patents
Mountain fire prediction method based on deep network learning Download PDFInfo
- Publication number
- CN109472396B CN109472396B CN201811209832.9A CN201811209832A CN109472396B CN 109472396 B CN109472396 B CN 109472396B CN 201811209832 A CN201811209832 A CN 201811209832A CN 109472396 B CN109472396 B CN 109472396B
- Authority
- CN
- China
- Prior art keywords
- data
- matrix
- convolution
- mountain fire
- layer
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000012545 processing Methods 0.000 claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 13
- 239000011159 matrix material Substances 0.000 claims description 105
- 238000007500 overflow downdraw method Methods 0.000 claims description 24
- 238000013058 risk prediction model Methods 0.000 claims description 16
- 230000004927 fusion Effects 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 9
- 230000002776 aggregation Effects 0.000 claims description 6
- 238000004220 aggregation Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 125000004057 biotinyl group Chemical group [H]N1C(=O)N([H])[C@]2([H])[C@@]([H])(SC([H])([H])[C@]12[H])C([H])([H])C([H])([H])C([H])([H])C([H])([H])C(*)=O 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 5
- 238000007499 fusion processing Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 2
- 238000000605 extraction Methods 0.000 abstract description 6
- 238000004458 analytical method Methods 0.000 abstract description 5
- 238000005516 engineering process Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 241001465754 Metazoa Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 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
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000002123 temporal effect 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 mountain fire prediction method based on deep network learning, which can improve prediction efficiency and prediction accuracy. According to the mountain fire prediction method based on deep network learning, the mountain fire prediction early warning analysis is achieved by constructing an automatic feature extraction network in time and space dimensions and learning massive historical data. The bottleneck of the traditional learning mode on mass data is overcome, and the GPU and the distributed type are effectively utilized for rapid training; the traditional manual feature mode is abandoned, and feature automation is realized; the depth network is used for flexibly and effectively fusing the features with different dimensions, so that the prediction efficiency and the prediction accuracy of mountain fire risk prediction are improved. Is suitable for popularization and application in the technical field of data processing.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a mountain fire prediction method based on deep network learning.
Background
In natural disasters, mountain fires are a very special and damaging disaster; not only can have a great influence on ecology, but also can have great losses on people, animals and some 'assets' in places; towers, cables, substations as in electrical equipment; signal towers in communications, and the like. The generation mechanism of the mountain fire is very complex, and the mountain fire can only occur under the interaction of a plurality of factors.
With the development of technologies such as the Internet of things, the digitalization and the Internet of things, a large amount of data can be obtained in a reasonable mode, and natural disasters can be analyzed, predicted, early-warned and the like in a non-experimental mode by combining with the improvement of the computing capability. Second, the current data is not only big data, but also diversified and complicated. Similar to the problem of natural disasters, the method not only comprises the geospatial temperature, but also comprises the time dimension, and the effective combination modeling of the two dimensions is also a difficulty of many traditional mathematical models at present. Therefore, the invention mainly utilizes the deep learning of the current hot spot technology to solve the problem of time and space modeling, and combines the characteristics of mountain fire to apply in mountain fire prediction and early warning.
The traditional feature extraction mode is to extract in space and time dimensions respectively, and then to carry out subsequent analysis and modeling, so that the feature extraction mode has certain subjectivity, and effective features cannot be easily found to support a subsequent model, so that a model effect is poor.
Traditional machine learning models are based on patterns of artificial features, cannot make efficient use of spatial features, model in the time dimension, traditional time series cannot build long-term time dependencies, and time series are mainly also statistical-based. Such conventional machine learning algorithms are designed without spatial or temporal concepts, and are feature-based learning processes.
In feature fusion, each feature is typically added to each sample, and even if there is a global feature, this approach is not flexible enough and problems can occur when different types of features or inconsistent features are fused.
The traditional learning mode is based on a certain data volume, and problems can occur when the data volume is too small or too large. When the data amount is too small, the reliability of the result is low. However, the traditional learning method is very difficult to process mass data and has a certain bottleneck.
Disclosure of Invention
The invention aims to provide a forest fire prediction method based on deep network learning, which can improve prediction efficiency and prediction accuracy.
The technical scheme adopted for solving the technical problems is as follows: the mountain fire prediction method based on deep network learning comprises the following steps:
1) Establishing a forest fire risk prediction model by using a deep network learning method;
the mountain fire risk prediction model comprises a convolution layer, a fusion layer, an LSTM layer and a full connection layer;
2) Aiming at the requirements of mountain fire risk prediction tasks, remote sensing data and meteorological data in a historical time from the current time to the front are collected;
3) Processing the meteorological data acquired in the step 2) by a time resolution fusion method to obtain meteorological data taking a day as a unit;
4) The remote sensing data and the weather data which is obtained in the step 3) and takes the day as a unit are fused and matched with each other through a spatial data fusion method;
5) All the remote sensing data and the meteorological data obtained through the processing of the step 4) are converted into the space division ratio 500m by 500m; time resolution is 1 day, and data with a tiff format is formed;
6) And 5) processing the data obtained in the step 5) to obtain batch data, wherein the specific processing process is as follows:
a. analyzing tiff data, namely analyzing the tiff data into a matrix form; and filling the missing values of the data in a manner of filling through adjacent points: the method is specifically as follows:
x′ i,j =x i-1,j |x i,j-1 ;
b. dividing the remote sensing data in a matrix form into a plurality of remote sensing data matrixes, and recording longitude and latitude positions in each point;
c. the meteorological data are corresponding according to the longitude and latitude positions, and a meteorological data matrix with the same size as the remote sensing data matrix in the step b is generated;
d. combining the meteorological data matrix and the remote sensing data matrix to finally generate a large matrix, wherein the format of the matrix is as follows: time step number, matrix line number, matrix column number, feature number ];
e. generating a target sample matrix through historical fire data, wherein the size of the target sample matrix is consistent with that of the large matrix generated in the step d;
f. mapping the data of the target sample matrix and the data of the large matrix generated in the step d, and dividing the whole data into data sets with different sizes to form batch data;
7) And inputting the batch data into a convolution layer of the mountain fire risk prediction model, extracting the weather data of the current day from the batch data, and inputting the weather data into a full-connection layer of the mountain fire risk prediction model, so that the final mountain fire occurrence prediction probability can be obtained.
Further, in the step 3), the time resolution fusion method is specifically as follows:
will be on day dObservation data m of meteorological index m per hour d,i (i.epsilon.0, …, 23) to generate corresponding statistical indexes by aggregation according to days, including: maximum value of current day max_m d Minimum value min_m d Range_m d Total of current day total_m d ;
Wherein max_m d =max(m d,i )(i∈0,…,23)
min_m d =min(m d,i )(i∈0,…,23)
range_m d =max(m d,i )-min(m d,i )(i∈0,…,23)
total_m d =sum(m d,i )(i∈0,…,23)
Maximum max_m d Minimum value min_m d For temperature, humidity, rainfall, wind speed, range_m d The total amount of the current day total_m is used for temperature and humidity d For rainfall levels.
Further, in the step 4), the spatial data fusion method adopts a close-proximity matching method or a krige interpolation method.
Further, in the step 1), the step of establishing the mountain fire risk prediction model by using the deep network learning method comprises the following steps:
A. aiming at the requirements of mountain fire risk prediction tasks, remote sensing data and meteorological data in a historical time from the current time to the front are collected;
B. c, processing the meteorological data acquired in the step A by a time resolution fusion method to obtain meteorological data taking a day as a unit;
C. the remote sensing data and the weather data which is obtained in the step B and takes the day as a unit are fused and matched with each other through a space data fusion method;
D. all the remote sensing data and the meteorological data obtained through the processing in the step C are converted into the space division ratio 500 m-500 m; time resolution is 1 day, and data with a tiff format is formed;
E. and D, processing the data obtained in the step D to obtain batch data, wherein the specific processing process is as follows:
a. analyzing tiff data, namely analyzing the tiff data into a matrix form; and filling the missing values of the data in a manner of filling through adjacent points: the method is specifically as follows:
x′ i,j =x i-1,j |x i,j-1 ;
b. dividing the remote sensing data in a matrix form into a plurality of remote sensing data matrixes, and recording longitude and latitude positions in each point;
c. the meteorological data are corresponding according to the longitude and latitude positions, and a meteorological data matrix with the same size as the remote sensing data matrix in the step b is generated;
d. combining the meteorological data matrix and the remote sensing data matrix to finally generate a large matrix, wherein the format of the matrix is as follows: time step number, matrix line number, matrix column number, feature number ];
e. generating a target sample matrix through historical fire data, wherein the size of the target sample matrix is consistent with that of the large matrix generated in the step d;
f. mapping the data of the target sample matrix and the data of the large matrix generated in the step d, and dividing the whole data into data sets with different sizes to form batch data;
F. constructing a depth network model based on batch data, wherein the depth network model comprises a convolution layer, a fusion layer, an LSTM layer and a full connection layer;
the convolution layer is formed by stacking a plurality of convolution units, and each convolution unit is obtained by adopting the following method: firstly, taking batch data as an input matrix to carry out convolution calculation to obtain a convolution result, wherein the convolution calculation formula is as follows: c=a×b, where x represents a convolution operation, a represents an input matrix and a B convolution kernel matrix, the convolution operation using the following discrete calculation formula:where m, n are the size values of the convolution kernel. w (w) m,n The value of m and n positions in the convolution kernel, and b is a paranoid item; and then carrying out maximum pooling on the obtained convolution result to obtain a convolution unit
The fusion layer fuses the obtained convolution units by adopting an ADD method, specifically, the convolution units obtained by the convolution layer are subjected to flat expansion, new features are added, the shape of the new features is consistent with the size of an input matrix, and stack is carried out according to a time dimension to obtain a fusion matrix [ batch_size, time_steps, depth ], wherein depth=w×h;
the LSTM layer comprises an LSTMCell layer, a drope layer and a full-connect layer, and is used for fusing matrixes (batch_size, time_steps, depth)]After being input into the LSTMCell layer, the LSTM data is processed by the dropper layer and the full-connect layer in sequence and is output by using a result function, wherein the result function is output by adopting a vitamin H function, and the vitamin H function is as follows:
the full connection layer carries out fusion processing on the obtained LSTM data and the current day meteorological data extracted from the batch data;
G. and training the depth network model by adopting multiple epochs, training the depth network model by adopting multiple GPUs, updating parameters by the CPU, and training the network by adopting the GPUs to obtain a final mountain fire prediction model.
Further, in the step B, the time resolution fusion method is specifically as follows:
observation data m of d weather image index m in each hour d,i (i.epsilon.0, …, 23) to generate corresponding statistical indexes by aggregation according to days, including: maximum value of current day max_m d Minimum value min_m d Range_m d Total of current day total_m d ;
Wherein max_m d =max(m d,i )(i∈0,…,23)
min_m d =min(m d,i )(i∈0,…,23)
range_m d =max(m d,i )-min(m d,i )(i∈0,…,23)
total_m d =sum(m d,i )(i∈0,…,23)
Maximum max_m d Minimum value min_m d For temperature, humidity, rainfall, wind speed, range_m d The total amount of the current day total_m is used for temperature and humidity d For rainfall levels.
Further, in the step C, the spatial data fusion method adopts a close-proximity matching method or a krige interpolation method.
The invention has the beneficial effects that: according to the mountain fire prediction method based on deep network learning, the mountain fire prediction early warning analysis is achieved by constructing an automatic feature extraction network in time and space dimensions and learning massive historical data. The bottleneck of the traditional learning mode on mass data is overcome, and the GPU and the distributed type are effectively utilized for rapid training; the traditional manual feature mode is abandoned, and feature automation is realized; the depth network is used for flexibly and effectively fusing the features with different dimensions, so that the prediction efficiency and the prediction accuracy of mountain fire risk prediction are improved.
Detailed Description
The mountain fire prediction method based on deep network learning comprises the following steps:
1) Establishing a forest fire risk prediction model by using a deep network learning method;
the mountain fire risk prediction model comprises a convolution layer, a fusion layer, an LSTM layer and a full connection layer;
2) Aiming at the requirements of mountain fire risk prediction tasks, remote sensing data and meteorological data in a historical time from the current time to the front are collected; the remote sensing data comprises: the water content FMC of the combustible, the load FL of the combustible and the type FT of the combustible; elevation, gradient, slope direction; the meteorological data comprise temperature, humidity, rainfall, wind speed and wind direction;
3) Processing the meteorological data acquired in the step 2) by a time resolution fusion method to obtain meteorological data taking a day as a unit; the time resolution fusion method is specifically as follows:
observation data m of d weather image index m in each hour d,i (i.epsilon.0, …, 23) to generate corresponding statistical indexes by aggregation according to days, including: maximum value of current day max_m d Minimum value min_m d Range_m d Total of current day total_m d ;
Wherein max_m d =max(m d,i )(i∈0,…,23)
min_m d =min(m d,i )(i∈0,…,23)
range_m d =max(m d,i )-min(m d,i )(i∈0,…,23)
total_m d =sum(m d,i )(i∈0,…,23)
Maximum max_m d Minimum value min_m d For temperature, humidity, rainfall, wind speed, range_m d The total amount of the current day total_m is used for temperature and humidity d For rainfall levels;
4) The remote sensing data and the weather data which is obtained in the step 3) and takes the day as a unit are fused and matched with each other through a spatial data fusion method; the spatial data fusion method adopts an adjacent matching method or a krige interpolation method;
5) All the remote sensing data and the meteorological data obtained through the processing of the step 4) are converted into the space division ratio 500m by 500m; time resolution is 1 day, and data with a tiff format is formed;
6) And 5) processing the data obtained in the step 5) to obtain batch data, wherein the specific processing process is as follows:
a. analyzing tiff data, namely analyzing the tiff data into a matrix form; and filling the missing values of the data in a manner of filling through adjacent points: the method is specifically as follows:
x′ i,j =x i-1,j |x i,j-1 ;
b. dividing the remote sensing data in a matrix form into a plurality of remote sensing data matrixes, and recording longitude and latitude positions in each point;
c. the meteorological data are corresponding according to the longitude and latitude positions, and a meteorological data matrix with the same size as the remote sensing data matrix in the step b is generated;
d. combining the meteorological data matrix and the remote sensing data matrix to finally generate a large matrix, wherein the format of the matrix is as follows: time step number, matrix line number, matrix column number, feature number ];
e. generating a target sample matrix through historical fire data, wherein the size of the target sample matrix is consistent with that of the large matrix generated in the step d;
f. mapping the data of the target sample matrix and the data of the large matrix generated in the step d, and dividing the whole data into data sets with different sizes to form batch data; in the batch data assembling process, the positive example data are fewer, and in order to solve the proportion imbalance, the positive example data are assembled into each batch when each time of use;
7) And inputting the batch data into a convolution layer of the mountain fire risk prediction model, extracting the weather data of the current day from the batch data, and inputting the weather data into a full-connection layer of the mountain fire risk prediction model, so that the final mountain fire occurrence prediction probability can be obtained.
According to the mountain fire prediction method based on deep network learning, the mountain fire prediction early warning analysis is achieved by constructing an automatic feature extraction network in time and space dimensions and learning massive historical data. The bottleneck of the traditional learning mode on mass data is overcome, and the GPU and the distributed type are effectively utilized for rapid training; the traditional manual feature mode is abandoned, and feature automation is realized; the depth network is used for flexibly and effectively fusing the features with different dimensions, so that the prediction efficiency and the prediction accuracy of mountain fire risk prediction are improved.
In the above embodiment, in step 1), the establishing the forest fire risk prediction model by using the deep network learning method includes the steps of:
A. aiming at the requirements of mountain fire risk prediction tasks, remote sensing data and meteorological data in a historical time from the current time to the front are collected; the remote sensing data comprises: the water content FMC of the combustible, the load FL of the combustible and the type FT of the combustible; elevation, gradient, slope direction; the meteorological data comprise temperature, humidity, rainfall, wind speed and wind direction;
B. c, processing the meteorological data acquired in the step A by a time resolution fusion method to obtain meteorological data taking a day as a unit; the time resolution fusion method is specifically as follows:
observation data m of d weather image index m in each hour d,i (i.epsilon.0, …, 23) to generate corresponding statistical indexes by aggregation according to days, including: maximum value of current day max_m d Minimum value min_m d Range_m d Total of current day total_m d ;
Wherein max_m d =max(m d,i )(i∈0,…,23)
min_m d =min(m d,i )(i∈0,…,23)
range_m d =max(m d,i )-min(m d,i )(i∈0,…,23)
total_m d =sum(m d,i )(i∈0,…,23)
Maximum max_m d Minimum value min_m d For temperature, humidity, rainfall, wind speed, range_m d The total amount of the current day total_m is used for temperature and humidity d For rainfall levels;
C. the remote sensing data and the weather data which is obtained in the step B and takes the day as a unit are fused and matched with each other through a space data fusion method; the spatial data fusion method adopts an adjacent matching method or a krige interpolation method;
D. all the remote sensing data and the meteorological data obtained through the processing in the step C are converted into the space division ratio 500 m-500 m; time resolution is 1 day, and data with a tiff format is formed;
E. and D, processing the data obtained in the step D to obtain batch data, wherein the specific processing process is as follows:
a. analyzing tiff data, namely analyzing the tiff data into a matrix form; and filling the missing values of the data in a manner of filling through adjacent points: the method is specifically as follows:
x′ i,j =x i-1,j |x i,j-1 ;
b. dividing the remote sensing data in a matrix form into a plurality of remote sensing data matrixes, and recording longitude and latitude positions in each point;
c. the meteorological data are corresponding according to the longitude and latitude positions, and a meteorological data matrix with the same size as the remote sensing data matrix in the step b is generated;
d. combining the meteorological data matrix and the remote sensing data matrix to finally generate a large matrix, wherein the format of the matrix is as follows: time step number, matrix line number, matrix column number, feature number ];
e. generating a target sample matrix through historical fire data, wherein the size of the target sample matrix is consistent with that of the large matrix generated in the step d;
f. mapping the data of the target sample matrix and the data of the large matrix generated in the step d, and dividing the whole data into data sets with different sizes to form batch data; in the batch data assembling process, the positive example data are fewer, and in order to solve the proportion imbalance, the positive example data are assembled into each batch when each time of use;
F. constructing a depth network model based on batch data, wherein the depth network model comprises a convolution layer, a fusion layer, an LSTM layer and a full connection layer;
the convolution layer is formed by stacking a plurality of convolution units, and each convolution unit is obtained by adopting the following method: firstly, taking batch data as an input matrix to carry out convolution calculation to obtain a convolution result, wherein the convolution calculation formula is as follows: c=a×b, where x represents a convolution operation, a represents an input matrix and a B convolution kernel matrix, the convolution operation using the following discrete calculation formula:where m, n are the size values of the convolution kernel. w (w) m,n The value of m and n positions in the convolution kernel, and b is a paranoid item; then willThe obtained convolution result is maximally pooled to obtain a convolution unit
The fusion layer fuses the obtained convolution units by adopting an ADD method, specifically, the convolution units obtained by the convolution layer are subjected to flat expansion, new features are added, the shape of the new features is consistent with the size of an input matrix, and stack is carried out according to a time dimension to obtain a fusion matrix [ batch_size, time_steps, depth ], wherein depth=w×h;
the LSTM layer comprises an LSTMCell layer, a drope layer and a full-connect layer, and is used for fusing matrixes (batch_size, time_steps, depth)]After being input into the LSTMCell layer, the LSTM data is processed by the dropper layer and the full-connect layer in sequence and is output by using a result function, wherein the result function is output by adopting a vitamin H function, and the vitamin H function is as follows:
the full connection layer carries out fusion processing on the obtained LSTM data and the current day meteorological data extracted from the batch data;
G. and training the depth network model by adopting multiple epochs, training the depth network model by adopting multiple GPUs, updating parameters by the CPU, and training the network by adopting the GPUs to obtain a final mountain fire prediction model.
The mountain fire prediction model establishment method based on deep network learning achieves mountain fire prediction early warning analysis by constructing an automatic feature extraction network in time and space dimensions and learning massive historical data, overcomes the bottleneck of a traditional learning mode on massive data, and can effectively utilize a GPU and a distributed mode to perform rapid training; the traditional manual feature mode is abandoned, and feature automation is realized; the depth network is used for flexibly and effectively fusing the features with different dimensions, so that the prediction efficiency and the prediction accuracy of the forest fire risk prediction model are improved.
Claims (6)
1. The mountain fire prediction method based on deep network learning is characterized by comprising the following steps of:
1) Establishing a forest fire risk prediction model by using a deep network learning method;
the mountain fire risk prediction model comprises a convolution layer, a fusion layer, an LSTM layer and a full connection layer;
2) Aiming at the requirements of mountain fire risk prediction tasks, remote sensing data and meteorological data in a historical time from the current time to the front are collected;
3) Processing the meteorological data acquired in the step 2) by a time resolution fusion method to obtain meteorological data taking a day as a unit;
4) The remote sensing data and the weather data which is obtained in the step 3) and takes the day as a unit are fused and matched with each other through a spatial data fusion method;
5) All the remote sensing data and the meteorological data obtained through the processing of the step 4) are converted into the space division ratio 500m by 500m; time resolution is 1 day, and data with a tiff format is formed;
6) And 5) processing the data obtained in the step 5) to obtain batch data, wherein the specific processing process is as follows:
a. analyzing tiff data, namely analyzing the tiff data into a matrix form; and filling the missing values of the data in a manner of filling through adjacent points: the method is specifically as follows:
x′ i,j =x i-1,j ||x i,j-1 ;
b. dividing the remote sensing data in a matrix form into a plurality of remote sensing data matrixes, and recording longitude and latitude positions in each point;
c. the meteorological data are corresponding according to the longitude and latitude positions, and a meteorological data matrix with the same size as the remote sensing data matrix in the step b is generated;
d. combining the meteorological data matrix and the remote sensing data matrix to finally generate a large matrix, wherein the format of the matrix is as follows: time step number, matrix line number, matrix column number, feature number ];
e. generating a target sample matrix through historical fire data, wherein the size of the target sample matrix is consistent with that of the large matrix generated in the step d;
f. mapping the data of the target sample matrix and the data of the large matrix generated in the step d, and dividing the whole data into data sets with different sizes to form batch data;
7) And inputting the batch data into a convolution layer of the mountain fire risk prediction model, extracting the weather data of the current day from the batch data, and inputting the weather data into a full-connection layer of the mountain fire risk prediction model, so that the final mountain fire occurrence prediction probability can be obtained.
2. The mountain fire prediction method based on deep network learning as claimed in claim 1, wherein: in the step 3), the time resolution fusion method is specifically as follows:
observation data m of d weather image index m in each hour d,i (i.epsilon.0, …, 23) to generate corresponding statistical indexes by aggregation according to days, including: maximum value of current day max_m d Minimum value min_m d Range_m d Total of current day total_m d ;
Wherein max_m d =max(m d,i )(i∈0,…,23)
min_m d =min(m d,i )(i∈0,…,23)
range_m d =max(m d,i )-min(m d,i )(i∈0,…,23)
total_m d =sum(m d,i )(i∈0,…,23)
Maximum max_m d Minimum value min_m d For temperature, humidity, rainfall, wind speed, range_m d The total amount of the current day total_m is used for temperature and humidity d For rainfall levels.
3. The mountain fire prediction method based on deep network learning as claimed in claim 2, wherein: in the step 4), the spatial data fusion method adopts a close-proximity matching method or a krige interpolation method.
4. The mountain fire prediction method based on deep network learning as claimed in claim 1, wherein: in the step 1), the step of establishing a mountain fire risk prediction model by using a deep network learning method comprises the following steps of:
A. aiming at the requirements of mountain fire risk prediction tasks, remote sensing data and meteorological data in a historical time from the current time to the front are collected;
B. c, processing the meteorological data acquired in the step A by a time resolution fusion method to obtain meteorological data taking a day as a unit;
C. the remote sensing data and the weather data which is obtained in the step B and takes the day as a unit are fused and matched with each other through a space data fusion method;
D. all the remote sensing data and the meteorological data obtained through the processing in the step C are converted into the space division ratio 500 m-500 m; time resolution is 1 day, and data with a tiff format is formed;
E. and D, processing the data obtained in the step D to obtain batch data, wherein the specific processing process is as follows:
a. analyzing tiff data, namely analyzing the tiff data into a matrix form; and filling the missing values of the data in a manner of filling through adjacent points: the method is specifically as follows:
x′ i,j =x i-1,j |x i,j-1 ;
b. dividing the remote sensing data in a matrix form into a plurality of remote sensing data matrixes, and recording longitude and latitude positions in each point;
c. the meteorological data are corresponding according to the longitude and latitude positions, and a meteorological data matrix with the same size as the remote sensing data matrix in the step b is generated;
d. combining the meteorological data matrix and the remote sensing data matrix to finally generate a large matrix, wherein the format of the matrix is as follows: time step number, matrix line number, matrix column number, feature number ];
e. generating a target sample matrix through historical fire data, wherein the size of the target sample matrix is consistent with that of the large matrix generated in the step d;
f. mapping the data of the target sample matrix and the data of the large matrix generated in the step d, and dividing the whole data into data sets with different sizes to form batch data;
F. constructing a depth network model based on batch data, wherein the depth network model comprises a convolution layer, a fusion layer, an LSTM layer and a full connection layer;
the convolution layer is formed by stacking a plurality of convolution units, and each convolution unit is obtained by adopting the following method: firstly, taking batch data as an input matrix to carry out convolution calculation to obtain a convolution result, wherein the convolution calculation formula is as follows: c=a×b, where x represents a convolution operation, a represents an input matrix and a B convolution kernel matrix, the convolution operation using the following discrete calculation formula:where m, n are the size values of the convolution kernel, w m,n The value of m and n positions in the convolution kernel, and b is a paranoid item; and then carrying out maximum pooling on the obtained convolution result to obtain a convolution unit
The fusion layer fuses the obtained convolution units by adopting an ADD method, specifically, the convolution units obtained by the convolution layer are subjected to flat expansion, new features are added, the shape of the new features is consistent with the size of an input matrix, and stack is carried out according to a time dimension to obtain a fusion matrix [ batch_size, time_steps, depth ], wherein depth=w×h;
the LSTM layer comprises an LSTMCell layer, a drope layer and a full-connect layer, and is used for fusing matrixes (batch_size, time_steps, depth)]Inputting LSTMCell layer, sequentially passing through dropperThe layers and full-connect layers process and output LSTM data using a result function that is output using a vitamin H function, which is described below:
the full connection layer carries out fusion processing on the obtained LSTM data and the current day meteorological data extracted from the batch data;
G. and training the depth network model by adopting multiple epochs, training the depth network model by adopting multiple GPUs, updating parameters by the CPU, and training the network by adopting the GPUs to obtain a final mountain fire prediction model.
5. The mountain fire prediction method based on deep network learning as claimed in claim 4, wherein: in the step B, the time resolution fusion method is specifically as follows:
observation data m of d weather image index m in each hour d,i (i.epsilon.0, …, 23) to generate corresponding statistical indexes by aggregation according to days, including: maximum value of current day max_m d Minimum value min_m d Range_m d Total of current day total_m d ;
Wherein max_m d =max(m d,i )(i∈0,…,23)
min_m d =min(m d,i )(i∈0,…,23)
range_m d =max(m d,i )-min(m d,i )(i∈0,…,23)
total_m d =sum(m d,i )(i∈0,…,23)
Maximum max_m d Minimum value min_m d For temperature, humidity, rainfall, wind speed, range_m d Is used for temperature, humidity and dayTotal total_m d For rainfall levels.
6. The mountain fire prediction method based on deep network learning as claimed in claim 5, wherein: in the step C, the spatial data fusion method adopts a close-proximity matching method or a 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 CN109472396A (en) | 2019-03-15 |
CN109472396B true 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) |
Families Citing this family (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 |
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 |
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 |
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 |
CN111882128B (en) * | 2020-07-28 | 2021-09-28 | 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 |
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 |
CN113553764B (en) * | 2021-07-13 | 2023-08-04 | 广东工业大学 | Mountain fire prediction method based on deep learning network |
Citations (5)
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 |
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 |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180096253A1 (en) * | 2016-10-04 | 2018-04-05 | Civicscape, LLC | Rare event forecasting system and method |
-
2018
- 2018-10-17 CN CN201811209832.9A patent/CN109472396B/en active Active
Patent Citations (5)
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 |
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 |
---|
基于弱监督深度网络的火灾检测技术研究与实现;孔亚奇;《工程科技Ⅰ辑》;20180615;68 * |
基于深度卷积神经网络的火灾预警算法研究;林作永等;《信息通信》;20180515;38-42 * |
Also Published As
Publication number | Publication date |
---|---|
CN109472396A (en) | 2019-03-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109472396B (en) | Mountain fire prediction method based on deep network learning | |
CN109508476B (en) | Mountain fire prediction modeling method based on deep network learning | |
Altintas et al. | Towards an integrated cyberinfrastructure for scalable data-driven monitoring, dynamic prediction and resilience of wildfires | |
CN111445011B (en) | Mountain fire early warning method based on meteorological and remote sensing data | |
Saha et al. | Autoencoder-based identification of predictors of Indian monsoon | |
CN115113301B (en) | Emergency short-term forecasting method and system based on multi-source data fusion | |
CN112434663B (en) | Power transmission line forest fire detection method, system and medium based on deep learning | |
Linhui et al. | Extracting the forest type from remote sensing images by random forest | |
McCandless et al. | Enhancing wildfire spread modelling by building a gridded fuel moisture content product with machine learning | |
Nguyen | Hybrid models based on deep learning neural network and optimization algorithms for the spatial prediction of tropical forest fire susceptibility in Nghe An province, Vietnam | |
CN111460733B (en) | Mountain fire early warning method based on deep learning | |
CN114444757A (en) | Combined prediction method for plateau mountain multi-model multi-scale new energy power station output | |
Zheng et al. | Integrating gate and attention modules for high-resolution image semantic segmentation | |
Premkumar et al. | IoT-enabled edge computing model for smart irrigation system | |
Man et al. | W-mae: Pre-trained weather model with masked autoencoder for multi-variable weather forecasting | |
Dong et al. | Tropical cyclone track prediction with an encoding-to-forecasting deep learning model | |
Ullerud et al. | Distribution modelling of vegetation types in the boreal–alpine ecotone | |
Nguyen | Land cover change detection in northwestern Vietnam using Landsat images and Google Earth Engine | |
Feng et al. | National-scale mapping of building footprints using feature super-resolution semantic segmentation of Sentinel-2 images | |
Qiannan et al. | High-resolution cropland extraction in Shandong province using MPSPNet and UNet network | |
Yang et al. | Extraction of land covers from remote sensing images based on a deep learning model of NDVI-RSU-Net | |
CN116665040A (en) | Building change detection method based on attention mechanism and with multi-scale input and output | |
Liu et al. | Integration transformer for ground-based cloud image segmentation | |
CN114418243B (en) | Distributed new energy cloud grid prediction method and system | |
CN115527113A (en) | Bare land classification method and device for remote sensing image |
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 | ||
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. |
|
GR01 | Patent grant | ||
GR01 | Patent grant |