CN118114546B - Numerical mode driven deep learning lightning early warning system and method - Google Patents
Numerical mode driven deep learning lightning early warning system and method Download PDFInfo
- Publication number
- CN118114546B CN118114546B CN202311800415.2A CN202311800415A CN118114546B CN 118114546 B CN118114546 B CN 118114546B CN 202311800415 A CN202311800415 A CN 202311800415A CN 118114546 B CN118114546 B CN 118114546B
- Authority
- CN
- China
- Prior art keywords
- time
- data
- lightning
- deep learning
- thunder
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a numerical mode-driven deep learning lightning early warning system and a numerical mode-driven deep learning lightning early warning method, wherein the numerical mode-driven deep learning lightning early warning system comprises a data fusion processing module, a sample construction module, a deep learning model construction module and a parameter optimization verification module, and a space-time fusion processing method is adopted to perform data fusion processing on lightning monitoring data and EC weather forecast data; training the thunder and lightning time-by-time prediction deep learning model by using thunder and lightning monitoring data and EC weather forecast data, optimizing super parameters by using a Bayesian optimization strategy, and evaluating the prediction precision of the thunder and lightning time-by-time prediction deep learning model by using a ten-fold layered cross validation method to obtain the optimized thunder and lightning time-by-time prediction deep learning model. The EC data adopted by the invention has no space coverage blind area and full meteorological elements, is relatively easy to acquire, and is beneficial to the business floor application of the lightning prediction model.
Description
Technical Field
The invention belongs to the field of lightning early warning, and particularly relates to a numerical mode driven deep learning lightning early warning system and method.
Background
Lightning disasters are one of the most serious natural disasters, and form a great threat to the industries such as electric power, traffic and the like and the production and living of people. Predicting the occurrence of lightning helps to improve the lightning protection capacity. However, the occurrence of lightning is strong in randomness, and the occurrence of the lightning is indistinguishable from the atmospheric conditions, and the occurrence or non-occurrence of the lightning under different conditions and the characteristics of activities are quite different, so that the occurrence time and the place of the lightning are not easy to accurately predict.
Currently, common lightning prediction methods include extrapolation of live lightning monitoring data and forecasting of lightning potential using numerical patterns. However, due to the complex evolution of lightning generation and extinction, prediction accuracy after 2 hours is rapidly reduced by the extrapolation-based prediction method, and the numerical weather prediction has certain deviation in space and time. Therefore, the historical lightning monitoring data and the numerical weather forecast are required to be fused, the development trend of recent lightning is reflected by utilizing the historical lightning monitoring data, and the atmospheric state information corresponding to the target forecast period is provided by combining the numerical weather forecast. In addition, the simulated space-time deviation of the digital weather forecast can be corrected by utilizing the latest historical lightning monitoring data.
Disclosure of Invention
The invention aims to provide a numerical mode-driven deep learning lightning early warning system and a numerical mode-driven deep learning lightning early warning method, which effectively integrate lightning monitoring data and EC (European Centre for Medium-RANGE WEATHER Forecasts, european middle weather forecast center) weather forecast data, can predict lightning for 2 hours in the future in a rolling manner time by time according to the EC weather forecast data with 3-hour resolution, have higher practicability and flexibility, and can provide powerful support for lightning prediction and lightning stroke early protection.
In order to achieve the purpose, the numerical mode-driven deep learning lightning early warning system comprises a data fusion processing module, a sample construction module, a deep learning model construction module and a parameter optimization verification module; the data fusion processing module is used for carrying out data fusion processing on lightning monitoring data and EC weather forecast data by adopting a space-time fusion processing method to obtain lightning monitoring rasterization data and EC weather forecast interpolation data with the same space-time resolution, the sample construction module is used for constructing data samples based on the lightning monitoring rasterization data and the EC weather forecast interpolation data obtained by the data fusion processing module, the data samples are divided into training set sample data and verification set sample data as lightning prediction factors and corresponding labels, the deep learning model construction module is used for constructing a lightning time-by-time prediction deep learning model, the training set sample data obtained by the sample construction module is input into the lightning time-by-time prediction deep learning model, mapping relation between the lightning prediction factors and the corresponding labels is mined by the lightning time-by-time prediction deep learning model to obtain a trained lightning time-by-time prediction deep learning model, the parameter optimization verification module is used for optimizing the super-by-fold cross-fold prediction deep learning model after training, and the lightning time-by-time prediction deep learning model is obtained by adopting a cross-fold verification method.
The numerical mode-driven deep learning lightning early warning method comprises the following steps of carrying out data fusion processing on lightning monitoring data and EC weather forecast data by adopting a space-time fusion processing method to obtain lightning monitoring rasterized data and EC weather forecast interpolation data with the same space-time resolution;
Constructing a data sample based on the lightning monitoring rasterization data and the EC weather forecast interpolation data, and dividing the data sample into training set sample data and verification set sample data as lightning prediction factors and corresponding labels;
The training set sample data is input into the thunder and lightning time-by-time prediction deep learning model, and the thunder and lightning prediction factors and corresponding labels are mapped when the thunder and lightning happens by the thunder and lightning time-by-time prediction deep learning model to obtain a trained thunder and lightning time-by-time prediction deep learning model;
And inputting the sample data of the verification set into the trained thunder and lightning time-by-time prediction deep learning model by using a Bayesian optimization strategy, optimizing the super-parameters, and evaluating the prediction accuracy of the thunder and lightning time-by-time prediction deep learning model by using a ten-fold layered cross verification method to obtain the optimized thunder and lightning time-by-time prediction deep learning model.
The lightning monitoring system has the beneficial effects that the historical lightning monitoring data is fused with EC value weather forecast data, so that the development trend of recent lightning is reflected by the historical lightning monitoring data, the atmospheric state information corresponding to a target forecast period is provided by utilizing the value weather forecast, and a more accurate lightning 2-hour forecast result can be obtained. The EC numerical weather forecast data adopted by the invention has no space coverage blind area and full meteorological elements, and the EC data is relatively easy to obtain, thereby being beneficial to the business floor application of the lightning prediction model.
Drawings
FIG. 1 is a block diagram of a system of the present invention;
FIG. 2 is a flow chart of the method of the present invention;
FIG. 3 is a chart of EC value weather forecast data forecast times;
FIG. 4 is a graph comparing the model predictive result of a primary thunderstorm process with live lightning monitoring;
the system comprises a 1-data fusion processing module, a 2-sample construction module, a 3-deep learning model construction module, a 4-parameter optimization verification module, a 5-thunder and lightning prediction module and a 6-result display application module.
Detailed Description
The invention is described in further detail below with reference to the attached drawings and specific examples:
A numerical mode-driven deep learning lightning early warning system is shown in fig. 1, and comprises a data fusion processing module 1, a sample construction module 2, a deep learning model construction module 3 and a parameter optimization verification module 4;
The data fusion processing module 1 is used for carrying out data fusion processing on lightning monitoring data and EC weather forecast data by adopting a space-time fusion processing method to obtain lightning monitoring rasterized data and EC weather forecast interpolation data with the same space-time resolution;
The sample construction module 2 constructs a data sample based on the lightning monitoring rasterization data and the EC weather forecast interpolation data obtained by the data fusion processing module 1, and the data sample is divided into training set sample data and verification set sample data as lightning prediction factors and corresponding labels;
The training set sample data obtained by the sample construction module 2 is input into the thunder time-by-time prediction deep learning model, the thunder prediction factors (the thunder prediction factors are input into the trained thunder time-by-time prediction deep learning model to obtain thunder prediction results) and corresponding labels when the thunder occurs are mined by the thunder time-by-time prediction deep learning model, and the trained thunder time-by-time prediction deep learning model is obtained by inputting the prediction factors into the trained thunder time-by-time prediction deep learning model;
The parameter optimization verification module 4 inputs the verification set sample data obtained by the sample construction module 2 into the trained thunder and lightning time-by-time prediction deep learning model obtained by the deep learning model construction module 3 by using a Bayesian optimization strategy, optimizes super parameters, and evaluates the prediction precision of the thunder and lightning time-by-time prediction deep learning model by using a ten-fold layered cross verification method to obtain the optimized thunder and lightning time-by-time prediction deep learning model.
In the above technical solution, the lightning monitoring data in the data fusion processing module 1 includes longitude information and latitude information of a lightning landing zone;
The EC weather forecast data in the data fusion processing module 1 comprises atmospheric environment variables with a large relation with thunder, wherein the atmospheric environment variables comprise convection precipitation, large-scale precipitation, temperature at air pressure 500hPa, temperature at air pressure 925hPa, zero-degree high-level, convection effective potential energy and vertical relative vorticity at air pressure 950hPa, the convection precipitation is marked as CP, the large-scale precipitation is marked as LSP, the temperature at air pressure 500hPa is marked as T500, the temperature at air pressure 925hPa is marked as T925, the zero-degree high-level is marked as DEG0l, the convection effective potential energy is marked as CAPE, and the vertical relative vorticity at air pressure 950hPa is marked as Vor950. Large scale precipitation is mainly affected by large scale circulation (e.g., low pressure, fronts, etc.), which can lead to large area, sustained precipitation. The large scale precipitation in EC is the precipitation volume predicted by cloud regime. Convective precipitation is primarily affected by local regional temperature and humidity differences that can lead to convective cloud formation and precipitation. Convection precipitation in EC is the amount of precipitation predicted by the convection regime.
In the above technical solution, the specific implementation method of the data fusion processing module 1 is as follows:
The method comprises the steps of spatially rasterizing discrete lightning monitoring data with the spatial precision of hundreds of meters, wherein the spatial resolution of the spatially rasterized lightning monitoring data is the same as that of the EC weather forecast data, temporally rasterizing the discrete lightning monitoring data with the time precision of microseconds to generate time rasterized lightning monitoring data, transforming the time resolution of the EC weather forecast data by adopting a nearest-neighbor interpolation method to obtain interpolated EC weather forecast data, the time resolution of the time rasterized lightning monitoring data is the same as that of the interpolated EC weather forecast data, the time resolution of the time rasterized lightning monitoring data is 1 hour, the original time resolution of the EC weather forecast data is 3 hours, and the time resolution of the interpolated EC weather forecast data is 1 hour.
In the above technical solution, the specific implementation method of the nearest neighbor interpolation method is as follows:
The method for processing the EC grid data with the time resolution of 3 hours into the time resolution of 1 hour by adopting the nearest neighbor interpolation method comprises the following steps of outputting EC weather forecast data of 1 time every 3 hours of EC according to the 3-hour resolution characteristic of the EC data, copying backwards for 2 hours, and obtaining the EC weather forecast data time by time in 3 hour periods.
In the above technical solution, the specific implementation method of the sample construction module 2 is as follows:
the data samples constructed by the sample construction module 2 comprise a historical lightning monitoring rasterization data set, a future lightning monitoring rasterization data set and a historical EC weather forecast interpolation data set;
dividing the data sample into training set sample data and verification set sample data according to a set proportion, wherein the set proportion is preferably 7:3 in the embodiment;
The historical lightning monitoring data set is the lightning monitoring rasterization data of the model from time to time in the past m hours at the time of starting, wherein m is more than or equal to 3, and the embodiment is preferably 3;
Selecting a future lightning monitoring data set as a label, wherein the future lightning monitoring data set is lightning monitoring rasterization data of n hours in the future at the time of model starting, n is determined according to the time range to be forecasted, the time range to be forecasted in the embodiment is 2 hours, and n is preferably 2;
And selecting a historical EC weather forecast data set as a lightning prediction factor, wherein the historical EC weather forecast data set is EC weather forecast interpolation data of the last k times nearest to the starting time of the model, and k is preferably 2 in the embodiment. For example, when the reporting time is world time 3, the EC data used are the EC data at time 0 and time 3 times. According to the characteristic that the EC time resolution is 3 hours, three models are required to be established for realizing rolling prediction of lightning for 2 hours in the future time by time.
In the above technical solution, the time-by-time thunder and lightning prediction deep learning model trained by the deep learning model construction module 3 includes a time-by-time thunder and lightning prediction deep learning model a, a time-by-time thunder and lightning prediction deep learning model B, and a time-by-time thunder and lightning prediction deep learning model C, and the specific implementation method is as follows:
According to the characteristic that the time resolution of the EC data is 3 hours, as shown in figure 3, a strategy of multiplexing the EC weather forecast data three times and rolling and updating the lightning data time by time is provided;
Dividing the data samples constructed by the sample construction module 2 into first time data samples from the EC weather forecast data updating time, dividing the data samples at the corresponding time into second time data samples at 1 time after the corresponding time, and dividing the data samples at 2 time after the corresponding time into third time data samples;
The first time data sample is input into the thunder time-by-time prediction deep learning model to obtain a thunder time-by-time prediction deep learning model A, the second time data sample is input into the thunder time-by-time prediction deep learning model to obtain a thunder time-by-time prediction deep learning model B, and the third time data sample is input into the thunder time-by-time prediction deep learning model to obtain a thunder time-by-time prediction deep learning model C.
In the above technical solution, the model super-parameters optimized by the parameter optimization verification module 4 include learning rate, optimizer, batch sample number, training round, hidden layer number, hidden layer node number, and activation function.
In this embodiment, the number of hidden layer nodes is preferably 4, and the number of nodes in each layer is 256, 512 and 128 in sequence, so that the model is prevented from being overfitted by adopting an early stop strategy. The early-stop strategy core idea is to stop training once the validation loss stops decreasing by checking the performance of the deep learning model on the validation set data. This approach aims to prevent model overfitting by ending training early.
The numerical mode-driven deep learning lightning early warning system also comprises a lightning prediction module 5 and a result display application module 6;
The lightning prediction module 5 inputs real-time lightning monitoring rasterization data and real-time EC weather forecast interpolation data of a target test time period into the optimized lightning time-by-time prediction deep learning model obtained by the parameter optimization verification module 4, and a lightning early warning result, namely lightning occurrence probability, of n hours in the future is obtained;
The result display application module 6 is used for performing visual drawing on the lightning early warning result obtained by the lightning prediction module 5 according to the probability of occurrence of lightning, storing the result into a result data table and synchronizing the result data to a lightning early warning data application platform.
In the above technical solution, the implementation method of the lightning prediction module 5 inputting the lightning monitoring data and the EC weather forecast data of the target test period into the optimized lightning time-by-time prediction deep learning model obtained by using the parameter optimization verification module 4 includes:
Dividing the lightning monitoring data and the EC weather forecast data at corresponding moments into first moment real-time data samples from the moment of updating the EC weather forecast data, dividing the lightning monitoring data and the EC weather forecast data 1 time after the corresponding moments into second moment real-time data samples, and dividing the lightning monitoring data and the EC weather forecast data 2 time after the corresponding moments into third moment real-time data samples;
The method comprises the steps of inputting the first time real-time data sample into the optimized thunder and lightning time-by-time prediction deep learning model A to obtain a thunder and lightning time-by-time prediction result n hours after the first time, inputting the second time real-time data sample into the optimized thunder and lightning time-by-time prediction deep learning model B to obtain a thunder and lightning time-by-time prediction result n hours after the second time, and inputting the third time real-time data sample into the optimized thunder and lightning time-by-time prediction deep learning model C to obtain a thunder and lightning time-by-time prediction result n hours after the third time.
In the actual prediction process, the thunder time-by-time prediction deep learning model A, the thunder time-by-time prediction deep learning model B and the thunder time-by-time prediction deep learning model C are used in a crossing mode, and the thunder prediction effect is improved through time-by-time iteration rolling correction.
A numerical mode-driven deep learning lightning early warning method is shown in figure 2, and comprises the following steps of carrying out data fusion processing on lightning monitoring data and EC weather forecast data by adopting a space-time fusion processing method to obtain lightning monitoring rasterized data and EC weather forecast interpolation data with the same space-time resolution;
Constructing a data sample based on the lightning monitoring rasterization data and the EC weather forecast interpolation data, and dividing the data sample into training set sample data and verification set sample data as lightning prediction factors and corresponding labels;
The training set sample data is input into the thunder and lightning time-by-time prediction deep learning model, and the thunder and lightning prediction factors and corresponding labels are mapped when the thunder and lightning happens by the thunder and lightning time-by-time prediction deep learning model to obtain a trained thunder and lightning time-by-time prediction deep learning model;
And inputting the sample data of the verification set into the trained thunder and lightning time-by-time prediction deep learning model by using a Bayesian optimization strategy, optimizing the super-parameters, and evaluating the prediction accuracy of the thunder and lightning time-by-time prediction deep learning model by using a ten-fold layered cross verification method to obtain the optimized thunder and lightning time-by-time prediction deep learning model.
In this embodiment, to implement lightning 2 hours prediction for a certain rectangular space region, the following operations are adopted:
and (3) data fusion, namely adopting 2020-2021 lightning monitoring and EC weather forecast data. Discrete lightning monitoring data (as shown in the example data of table 1) are converted into grid data with a space-time accuracy of 1 hour and 0.5 ° respectively. According to the characteristic that the time resolution of the EC data is 3 hours (see figure 2), the nearest neighbor interpolation method is adopted for EC grid data with the time resolution of 3 hours, namely, the pre-reported time weather data output by EC are copied backwards for 2 times and processed into 1 hour time resolution, so that the effective space-time fusion of lightning monitoring and EC weather forecast data is realized;
table 1 example of lightning monitoring data (including time, latitude and longitude information of lightning occurrence) in Sichuan province at 7.1 and 15.7.1
| Time of | Microsecond | Latitude of latitude | Longitude and latitude | Electric current |
| 21:20:21 | 2804791 | 29.103894 | 105.059658 | 10.3 |
| 21:20:23 | 221116 | 29.432289 | 91.774612 | 17.7 |
| 21:20:23 | 221293 | 29.303452 | 91.62205 | 13.2 |
| 21:20:25 | 1430403 | 29.151595 | 105.10164 | -46.1 |
| 21:20:25 | 3190804 | 29.166534 | 105.087603 | -34.2 |
| 21:20:25 | 3747653 | 29.162258 | 105.085361 | -8.7 |
| 21:20:25 | 6138372 | 29.132312 | 105.113735 | -24 |
| 21:20:25 | 6590998 | 29.133157 | 105.112463 | -8.3 |
| 21:20:26 | 2964764 | 30.344877 | 92.540879 | -38.3 |
| 21:20:26 | 2991063 | 29.199563 | 105.162618 | -12.2 |
| 21:20:27 | 8449831 | 29.384269 | 91.606833 | 18.8 |
| 21:20:28 | 619837 | 29.39498 | 91.625346 | -8.1 |
| 21:20:28 | 927344 | 29.395301 | 91.625251 | -7.4 |
| 21:20:28 | 5781953 | 28.398339 | 100.852021 | 28.4 |
| 21:20:29 | 4824370 | 29.091413 | 105.164682 | 13.5 |
| 21:20:30 | 1497839 | 29.650652 | 105.263556 | -43.3 |
Sample construction, namely taking the previous 3 times of lightning and the previous 2 times of EC as lightning predictors, and taking the future 2 hours of lightning as a tag. According to the 3-hour resolution characteristic of the EC data, the lightning data is updated in a rolling way time by time and the EC data is multiplexed for 2 time by time, so that the lightning and EC sample data are updated time by time.
Model training, namely constructing three deep learning thunder and lightning prediction models by adopting a historical sample data set updated time by time. The model framework is a convolutional neural network, 4 hidden layers are arranged, the number of nodes of each layer is 256, 512 and 128 in sequence, and effective mining of data information is ensured. Meanwhile, in the training process, an early stop strategy is adopted to prevent the model from being fitted excessively. And optimizing the super parameters of the deep learning model by using a Bayes optimization strategy, and evaluating the precision of the deep learning model by using ten-fold hierarchical cross verification. And outputting and drawing the loss value, the accuracy and the ROC curve change of the training set and the verification set along with the training turn change, finally obtaining the overall evaluation index of the model, wherein the hit rate of the model is 85.31%, the false alarm rate is 45.84%, the risk score is 0.45, and the overall performance of the model is better.
And (3) model prediction, namely calling the constructed lightning early warning model, multiplexing the EC data for 2 hours according to the characteristic that the time resolution of the EC data is 3 hours, and rolling and updating the lightning data time by time to realize rolling and predicting the lightning time by time for 2 hours in the future. The three models are used in a crossing mode, and the prediction effect is improved through iterative rolling correction time by time.
And (3) visualizing the result, namely performing visual drawing on the lightning early warning result according to the occurrence probability of the lightning (see figure 4), wherein the result of model reasoning shows that the model prediction result basically reflects the generation and elimination and movement trend of the lightning. And storing the early warning result into a result data table, and synchronizing the early warning result to a corresponding lightning early warning data application platform.
The above examples illustrate how lightning monitoring data and EC numerical weather forecast data can be used to achieve a modified future 2-hour lightning forecast on a time-by-time basis. According to the invention, the lightning prediction factors of the past three hours and the past 2 hours are adopted as the lightning prediction factors, so that the development trend of the atmospheric background field along with time is considered, and the early warning timeliness and accuracy are improved. The method and the system can help the user to better obtain the lightning early warning information and provide powerful support for the early protection of the lightning.
A computer readable storage medium storing a computer program which, when executed by a processor, implements the steps of a method as described above.
What is not described in detail in this specification is prior art known to those skilled in the art. It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It should be finally understood that the foregoing embodiments are merely illustrative of the technical solutions of the present invention and not limiting the scope of protection thereof, and although the present invention has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that various changes, modifications or equivalents may be made to the specific embodiments of the invention, and these changes, modifications or equivalents are within the scope of protection of the claims appended hereto.
Claims (11)
1. The numerical mode-driven deep learning lightning early warning system is characterized by comprising a data fusion processing module (1), a sample construction module (2), a deep learning model construction module (3) and a parameter optimization verification module (4);
The data fusion processing module (1) is used for carrying out data fusion processing on lightning monitoring data and EC weather forecast data by adopting a space-time fusion processing method to obtain lightning monitoring rasterized data and EC weather forecast interpolation data with the same space-time resolution;
The sample construction module (2) constructs a data sample based on the lightning monitoring rasterization data and the EC weather forecast interpolation data obtained by the data fusion processing module (1), and the data sample is divided into training set sample data and verification set sample data as lightning prediction factors and corresponding labels;
the training set sample data obtained by the sample construction module (2) is input into the thunder and lightning time-by-time prediction deep learning model, and the thunder and lightning prediction factors and corresponding labels are used for mining the mapping relation between the thunder and lightning time-by-time prediction deep learning model and the thunder and lightning occurrence time, so as to obtain a trained thunder and lightning time-by-time prediction deep learning model;
The parameter optimization verification module (4) utilizes a Bayesian optimization strategy to input verification set sample data obtained by the sample construction module (2) into the trained thunder time-by-time prediction deep learning model obtained by the deep learning model construction module (3), and optimizes super parameters to obtain an optimized thunder time-by-time prediction deep learning model;
the specific implementation method of the data fusion processing module (1) comprises the following steps:
The method comprises the steps of spatially rasterizing lightning monitoring data, temporally rasterizing the lightning monitoring data to generate time-rasterized lightning monitoring data, transforming the time resolution of the EC weather forecast data by adopting a nearest-neighbor interpolation method to obtain interpolated EC weather forecast data, wherein the time resolution of the time-rasterized lightning monitoring data is the same as that of the interpolated EC weather forecast data.
2. The numerical mode-driven deep learning lightning early warning system is characterized in that a ten-fold layered cross validation method is adopted to evaluate the prediction accuracy of the lightning time-by-time prediction deep learning model.
3. The numerical mode-driven deep learning lightning early warning system according to claim 1, wherein the nearest interpolation method is specifically implemented by the following steps:
The method for processing the EC grid data with the time resolution of 3 hours into the time resolution of 1 hour by adopting the nearest neighbor interpolation method comprises the following steps of outputting EC weather forecast data of 1 time every 3 hours by EC, copying backwards for 2 hours, and obtaining the EC weather forecast data time by time in 3 hour periods.
4. The numerical mode-driven deep learning lightning early warning system according to claim 1, wherein the sample construction module (2) is specifically implemented by the following steps:
the data samples constructed by the sample construction module (2) comprise a historical lightning monitoring rasterization data set, a future lightning monitoring rasterization data set and a historical EC weather forecast interpolation data set;
dividing the data sample into training set sample data and verification set sample data according to a set proportion;
The historical lightning monitoring data set is the lightning monitoring rasterization data of the model from time to time in the past m hours at the time of starting, and m is more than or equal to 3;
selecting a future lightning monitoring data set as a label, wherein the future lightning monitoring data set is lightning monitoring rasterization data time by time for n hours in the future at the time of model starting, and n is determined according to a time range to be forecasted;
And selecting a historical EC weather forecast data set as a lightning prediction factor, wherein the historical EC weather forecast data set is EC weather forecast interpolation data of the last k times nearest to the starting time of the model.
5. The numerical mode-driven deep learning lightning early warning system is characterized in that the lightning time-by-time prediction deep learning model trained by the deep learning model building module (3) comprises a lightning time-by-time prediction deep learning model A, a lightning time-by-time prediction deep learning model B and a lightning time-by-time prediction deep learning model C, and the specific implementation method is as follows:
Dividing the data samples constructed by the sample construction module (2) into first time data samples from the EC weather forecast data updating time, dividing the data samples at the corresponding time into second time data samples at 1 time after the corresponding time, and dividing the data samples at 2 time after the corresponding time into third time data samples;
The first time data sample is input into the thunder time-by-time prediction deep learning model to obtain a thunder time-by-time prediction deep learning model A, the second time data sample is input into the thunder time-by-time prediction deep learning model to obtain a thunder time-by-time prediction deep learning model B, and the third time data sample is input into the thunder time-by-time prediction deep learning model to obtain a thunder time-by-time prediction deep learning model C.
6. The numerical mode-driven deep learning lightning early warning system is characterized in that the model super parameters optimized by the parameter optimization verification module (4) comprise learning rate, optimizers, batch sample number, training rounds, hidden layer number, hidden layer node number and activation functions.
7. The numerical mode-driven deep learning lightning early warning system is characterized by further comprising a lightning prediction module (5) and a result display application module (6);
The lightning prediction module (5) inputs real-time lightning monitoring grid data and real-time EC weather forecast interpolation data of a target test time period into the optimized lightning time-by-time prediction deep learning model obtained by the parameter optimization verification module (4) to obtain lightning early warning results of n hours in the future;
The result display application module (6) is used for carrying out visual drawing on the lightning early warning result obtained by the lightning prediction module (5), storing the result data into a result data table and synchronizing the result data to the lightning early warning data application platform.
8. The numerical mode-driven deep learning lightning early warning system is characterized in that the lightning prediction module (5) inputs lightning monitoring data and EC weather forecast data of a target test time period into the optimized lightning time-by-time prediction deep learning model obtained by the parameter optimization verification module (4), and the implementation method comprises the following steps:
Dividing the lightning monitoring data and the EC weather forecast data at corresponding moments into first moment real-time data samples from the moment of updating the EC weather forecast data, dividing the lightning monitoring data and the EC weather forecast data 1 time after the corresponding moments into second moment real-time data samples, and dividing the lightning monitoring data and the EC weather forecast data 2 time after the corresponding moments into third moment real-time data samples;
The method comprises the steps of inputting the first time real-time data sample into the optimized thunder and lightning time-by-time prediction deep learning model A to obtain a thunder and lightning time-by-time prediction result n hours after the first time, inputting the second time real-time data sample into the optimized thunder and lightning time-by-time prediction deep learning model B to obtain a thunder and lightning time-by-time prediction result n hours after the second time, and inputting the third time real-time data sample into the optimized thunder and lightning time-by-time prediction deep learning model C to obtain a thunder and lightning time-by-time prediction result n hours after the third time.
9. A method for deep learning lightning early warning driven by a numerical mode of the system of claim 1 is characterized by comprising the following steps:
Carrying out data fusion processing on the lightning monitoring data and the EC weather forecast data by adopting a space-time fusion processing method to obtain lightning monitoring rasterization data and EC weather forecast interpolation data with the same space-time resolution;
Constructing a data sample based on the lightning monitoring rasterization data and the EC weather forecast interpolation data, and dividing the data sample into training set sample data and verification set sample data as lightning prediction factors and corresponding labels;
The training set sample data is input into the thunder and lightning time-by-time prediction deep learning model, and the thunder and lightning prediction factors and corresponding labels are mapped when the thunder and lightning happens by the thunder and lightning time-by-time prediction deep learning model to obtain a trained thunder and lightning time-by-time prediction deep learning model;
And inputting the sample data of the verification set into the trained thunder and lightning time-by-time prediction deep learning model by using a Bayesian optimization strategy, optimizing the super-parameters, and evaluating the prediction accuracy of the thunder and lightning time-by-time prediction deep learning model by using a ten-fold layered cross verification method to obtain the optimized thunder and lightning time-by-time prediction deep learning model.
10. An electronic device comprising a memory and a processor, said memory and said processor being communicatively coupled to each other, said memory storing computer instructions, said processor executing said computer instructions to perform the numerical mode-driven deep learning lightning early warning method of claim 9.
11. A computer readable medium having stored thereon a computer program which, when run, performs the numerical mode driven deep learning lightning early warning method of claim 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311800415.2A CN118114546B (en) | 2023-12-26 | 2023-12-26 | Numerical mode driven deep learning lightning early warning system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311800415.2A CN118114546B (en) | 2023-12-26 | 2023-12-26 | Numerical mode driven deep learning lightning early warning system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118114546A CN118114546A (en) | 2024-05-31 |
| CN118114546B true CN118114546B (en) | 2024-12-03 |
Family
ID=91220112
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311800415.2A Active CN118114546B (en) | 2023-12-26 | 2023-12-26 | Numerical mode driven deep learning lightning early warning system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118114546B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109447315A (en) * | 2018-09-18 | 2019-03-08 | 中国电力科学研究院有限公司 | A kind of electric power meteorology numerical weather forecast method and apparatus based on multiple space and time scales |
| CN113064222A (en) * | 2021-03-09 | 2021-07-02 | 中国气象科学研究院 | Lightning warning and forecasting method and system |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114118598A (en) * | 2021-12-01 | 2022-03-01 | 国网湖南省电力有限公司 | Power grid lightning disaster prediction method and system based on deep learning prediction model |
| CN114966233B (en) * | 2022-05-16 | 2024-08-13 | 国网电力科学研究院武汉南瑞有限责任公司 | Lightning forecasting system and method based on deep neural network |
| CN117148360B (en) * | 2023-10-31 | 2024-01-19 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Lightning approach prediction method and device, electronic equipment and computer storage medium |
-
2023
- 2023-12-26 CN CN202311800415.2A patent/CN118114546B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109447315A (en) * | 2018-09-18 | 2019-03-08 | 中国电力科学研究院有限公司 | A kind of electric power meteorology numerical weather forecast method and apparatus based on multiple space and time scales |
| CN113064222A (en) * | 2021-03-09 | 2021-07-02 | 中国气象科学研究院 | Lightning warning and forecasting method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118114546A (en) | 2024-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20230128989A1 (en) | Systems and methods of data preprocessing and augmentation for neural network climate forecasting models | |
| Li et al. | Urban growth models: progress and perspective | |
| Wang et al. | Wind power forecasting in the presence of data scarcity: A very short-term conditional probabilistic modeling framework | |
| CN117909888B (en) | Intelligent artificial intelligence climate prediction method | |
| JP2020064446A (en) | Prediction system and prediction method | |
| Sun et al. | Accident prediction and emergency management for expressways using big data and advanced intelligent algorithms | |
| CN120494472B (en) | Low-voltage distribution network variation risk assessment and intelligent early warning methods and systems | |
| CN119918750B (en) | A schedule management system and method for power engineering | |
| Willems et al. | Probabilistic Restoration Modeling of Wide-Area Power Outage | |
| Zou et al. | A data-driven framework for fast building energy demand estimation across future climate conditions | |
| CN121529517A (en) | Regional Electricity Load Forecasting Method and Related Equipment Based on Periodic Characteristics | |
| CN118114546B (en) | Numerical mode driven deep learning lightning early warning system and method | |
| Shen et al. | Hazard-responsive digital twin for climate-driven urban resilience and equity | |
| Sajadifar et al. | A probabilistic Markov chain model for short-term water demand forecasting | |
| CN113761806A (en) | Atmospheric visibility prediction method based on neural network | |
| Li et al. | Dynamic risk assessment model for urban power lifeline based on machine learning | |
| CN120974121B (en) | Bird diversity prediction method and device | |
| Rowley et al. | From smart homes to smart communities: Advanced data acquisition and analysis for improved sustainability and decision making | |
| CN121352434B (en) | A method, apparatus, and equipment for dynamic water distribution in wetland communities based on an adaptive optimization model. | |
| Zheng et al. | A Deep Reinforcement Learning–Based Urban Traffic Control Model for Vehicle‐to‐Everything Ecosystem | |
| Hu et al. | Urban AQI prediction using an inverse variance weighted Gcn and bilstm combined model | |
| Liu et al. | An intrinsic dynamics capture network for long-term airspace traffic prediction | |
| Sherif et al. | Diverse occupancy simulation and presence sensing viability for residential thermal energy regulation: Review and false positive modeling initial findings | |
| Chakraborty et al. | Machine Learning for Building Energy Modeling | |
| Zheng | Digital twin for renewable energy landscape planning and design |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |