CN114548595A - Strong convection weather physical characteristic quantity prediction method and system based on attention mechanism - Google Patents
Strong convection weather physical characteristic quantity prediction method and system based on attention mechanism Download PDFInfo
- Publication number
- CN114548595A CN114548595A CN202210202985.0A CN202210202985A CN114548595A CN 114548595 A CN114548595 A CN 114548595A CN 202210202985 A CN202210202985 A CN 202210202985A CN 114548595 A CN114548595 A CN 114548595A
- Authority
- CN
- China
- Prior art keywords
- weather
- data
- strong convection
- model
- longitude
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000007246 mechanism Effects 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 40
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 29
- 238000012549 training Methods 0.000 claims description 22
- 239000007788 liquid Substances 0.000 claims description 13
- 238000009792 diffusion process Methods 0.000 claims description 5
- 230000007774 longterm Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000010200 validation analysis Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 15
- 238000012545 processing Methods 0.000 description 9
- 241000282414 Homo sapiens Species 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000012800 visualization Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 238000001556 precipitation Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000002123 temporal effect Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000000739 chaotic effect Effects 0.000 description 2
- 239000013078 crystal Substances 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 230000001066 destructive effect Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 230000015654 memory Effects 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012950 reanalysis Methods 0.000 description 2
- 238000005096 rolling process Methods 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 239000005436 troposphere Substances 0.000 description 2
- 108091026890 Coding region Proteins 0.000 description 1
- 235000019013 Viburnum opulus Nutrition 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000003920 cognitive function Effects 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000013079 data visualisation Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 208000013409 limited attention Diseases 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- JTJMJGYZQZDUJJ-UHFFFAOYSA-N phencyclidine Chemical class C1CCCCN1C1(C=2C=CC=CC=2)CCCCC1 JTJMJGYZQZDUJJ-UHFFFAOYSA-N 0.000 description 1
- 238000005381 potential energy Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000012887 quadratic function Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
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
- G01—MEASURING; TESTING
- G01W—METEOROLOGY
- G01W1/00—Meteorology
- G01W1/10—Devices for predicting weather conditions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06Q10/067—Enterprise or organisation modelling
-
- 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
- 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)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Marketing (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Development Economics (AREA)
- Computational Linguistics (AREA)
- General Business, Economics & Management (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Entrepreneurship & Innovation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Educational Administration (AREA)
- Environmental & Geological Engineering (AREA)
- Primary Health Care (AREA)
- Atmospheric Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Ecology (AREA)
- Environmental Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a strong convection weather physical characteristic quantity prediction method and a strong convection weather physical characteristic quantity prediction system based on an attention mechanism, wherein the method is mainly embodied in the establishment of a model, the model comprises an attention mechanism module, a coding module and a decoding module, when input data enter the model, the input data pass through the attention mechanism module, a current input result is predicted according to a previous hidden state, the current input result and the current input data are overlapped together and are coded through the decoding and coding module to form a current hidden state, and after the historical hidden state and the current hidden state are added, the historical hidden state and the current hidden state are decoded through the decoding module and then are convolved to generate a prediction result. The invention effectively solves the problem of multi-dimension and multi-input space-time sequence prediction.
Description
Technical Field
The invention relates to the technical field of meteorology, in particular to a strong convection weather physical characteristic quantity prediction method and system based on an attention mechanism.
Background
Since the birth of human beings, weather affects the daily life of people. Weather shapes various atmospheric events and produces a "snowball effect" for real-life production and life systems such as agriculture, transportation, tourism, aviation, etc., in which data scientists attempt to predict the next state of the weather system, and with the success of space-time series (spatio-temporalisris) models, the industry has created new solutions for weather numerical prediction (NWP), which have significant advantages over traditional methods. Compared with the traditional physical model of numerical weather forecast, the deep learning model can provide results within minutes of receiving data, utilizes a large amount of historical data collected for many years, and uses a model with lower cost for accurate prediction. Strong Convection Weather (SCW), which is a weather phenomenon caused by strong vertical movement of air, generally includes storms, lightning, hailstones, convection gusts, short-term heavy rainfall, tornadoes, and poses serious threats to lives and properties in most regions of the world. Due to the rapid change of the small-scale convection system and the complex interaction with the environment, the prediction of the physical parameters of strong convection weather is still a challenging and significant problem under the task of weather numerical prediction in the current business meteorological field. Physical characteristics of strong convection include potential height (geotropic), Temperature (Temperature), U-wind component of wind (U-wind), V-wind component of wind (V-wind), relative humidity (relative humidity), specific humidity (specificity), vertical velocity (Verticalvelocity), Vorticity (Vorticity), potential Vorticity (polytronticity), specific liquid water content (specific liquidwater content), and the like.
The prediction problem of the strong convection weather physical characteristic quantity is a multi-time and multi-space sequence prediction problem (hereinafter referred to as space-time sequence prediction). In the field of multi-time series, classical machine learning algorithms have enjoyed great success in time series applications. In the prior art, Neural Networks (NNs) have been successfully trained to predict the energy of each region. Also, Recurrent Neural Networks (RNN) and Convolutional Neural Networks (CNN) have been built in a multi-step and recursive fashion, which has shown great success compared to statistical models. In a more challenging task of predicting non-stationary time series data, the M4 dataset may also be predicted in conjunction with statistical methods and deep learning models. Furthermore, as new methods of time series prediction have been developed, one can combine multiple time series with attention mechanisms, or perform temporally extended random convolutions, to produce accurate long-term predictions. However, all of these methods focus on the prediction of the next value of a single target sequence. For example, covariates of multiple input sequences are used to predict multiple time sequences. However, in the spatio-temporal sequence prediction, there are a plurality of spatio-temporal covariances having spatio-temporal covariances to be predicted.
Furthermore, previous studies on spatiotemporal sequences, including ConvLSTM, did not take advantage of weather digital data sets such as satellite images, numerical values, and weather station observations. These efforts have all focused on precipitation prediction. Precipitation prediction aims to accurately and timely predict the rainfall intensity in a relatively short period of time in the area. The input and output are a series of radar images, which essentially solve an image transformation problem, rather than a spatio-temporal sequence prediction problem, which focuses on a data source. However, weather is a chaotic system that is affected by many parameters, such as vegetation, geographical contours, and artifacts. Therefore, a smaller amount of information is not sufficient to model such chaotic behavior.
The weather numerical prediction problem belongs to the multi-space-time sequence prediction problem. The ConvLSTM model was introduced in a predictive network consisting of an encoder and a decoder built from a stack of ConvLSTMs. The model can capture a space-time covariance matrix and show high performance in Moving-MNIST data sets and predicting rainfall intensity of an area. Furthermore, hurricanes have been tracked using this architecture in the prior art. Later, a trajectory-gated recursion unit (Traj-GRU) was introduced in the prior art to learn the position-invariant motion between the input radar images, showing higher performance. However, the network architecture proposed in the prior art must employ a one-dimensional tensor to predict the one-dimensional output. In addition, the prediction architecture is improved by the downsampling and the upsampling layer between the TrajGRU unit and the output convolution in the prior art, and the multidimensional input problem is ignored.
Attention mechanism (attentionchannels) can be developed rapidly in deep learning, and especially in recent years, along with its wide application in fields of natural language processing, speech recognition, computer vision and the like, the attention mechanism attracts high attention of people. The attention mechanism is derived from research on human vision, and is an indispensable complex cognitive function of human beings, and people have the selective ability to focus on some information and ignore other information. The attention mechanism conforms to the logic of human beings looking at pictures, and when we see a picture, the whole content of the picture is not always seen, but the attention is focused on a certain important part of the picture. The focus part is generally called a focus part, and more attention resources are put into the focus part to acquire more target detail information needing to be focused and ignore other useless information. In the deep learning application of the means for screening out high-value information from a large amount of information by using limited attention resources, the efficiency and performance of natural language processing and image processing can be greatly improved.
To enable human action recognition, the prior art has developed spatiotemporal attention mechanisms with fully connected NNs and LSTMs. For example, a dual attention mechanism is implemented in time series prediction. Still other techniques implement a multi-tiered attention mechanism that fuses ancillary information with multiple geo-sensory time series data to predict air quality. For traffic prediction, one has implemented a mechanism of interest in LSTM outputs that accept inputs from different periods in the past. While these efforts successfully implement the attention mechanism, the input is a single or multiple time series, rather than a three-dimensional tensor. One has implemented an attention mechanism to further refine ConvLSTM, which takes a three-dimensional tensor as input; however, the performance is hardly improved. Recently, CNN is concerned with the architecture and temporal connectivity architecture of the LSTM (Conv + LSTM) and ConvLSTM spread in time. They observe the change in loss from different inputs, e.g., features, locations where model decisions are interpreted. ConvLSTM exhibits high performance compared to Conv + LSTM. However, they only obtained interpretable results for the Conv + LSTM model, their interpretation being at the feature and location level, not at the activation of the model.
Therefore, in the prior art, a method for improving prediction accuracy and fusing natural flow in the atmosphere is needed to solve the problem of predicting the physical characteristic quantity of strong convection weather, and a method for realizing weather space-time sequence prediction by fully fusing various weather parameters and better learning spatial correlation is needed.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a strong convection weather physical characteristic quantity prediction method and system based on an attention mechanism. The invention introduces a recursion network structure based on attention to a space-time sequence by adding more channel numbers (atmospheric parameters are used as auxiliary information) to optimize a prediction result, and solves the problem of multi-dimensional and multi-input space-time sequence prediction.
The specific technical scheme of the invention is as follows:
according to a first technical scheme of the invention, a strong convection weather physical characteristic quantity prediction method based on an attention mechanism is provided, and comprises the following steps: acquiring a meteorological parameter file; acquiring a history record of screened strong convection weather; extracting and fusing data to produce a data set; building a model, wherein the model comprises an attention mechanism module, a coding module and a decoding module, when input data enter the model, the input data pass through the attention mechanism module, predict a current input result according to a previous hidden state, overlap the current input result and the current input data together, and carry out coding through the coding module to form a current hidden state, add a historical hidden state and the current hidden state, carry out convolution after decoding through the decoding module to generate a prediction result; based on the data set, training the model by using a configured parameter file, and stopping training when the loss value does not decrease for a plurality of times; and predicting the next observation value of each weather factor of the strong convection based on the model.
According to a second aspect of the present invention, there is provided an attention-based strong convection weather physical characteristic quantity prediction system, including a processor configured to: acquiring meteorological parameter texts to obtain a history record of screened out strong convection weather; extracting and fusing data to produce a data set; building a model, wherein the built model comprises an attention mechanism module, a coding module and a decoding module, when input data enter the model, the input data pass through the attention mechanism module, a current input result is predicted according to a previous hidden state, the current input result and the current input data are overlapped together and are coded by the coding module to form a current hidden state, and after the historical hidden state and the current hidden state are added, the historical hidden state and the current hidden state are decoded by the decoding module and then are convolved to generate a prediction result; based on the data set, training the model by using a configured parameter file, and stopping training when the loss value does not decrease for a plurality of times; and predicting the next observation value of each weather factor of the strong convection based on the model.
Has the advantages that:
the invention can solve the problem of multi-dimension and multi-input space-time sequence prediction, optimizes the prediction result by adding more channels (atmospheric parameters are used as auxiliary information), introduces a recursive model based on attention for a space-time sequence, and takes a convolution operation and decoding coding module of an attention mechanism module as a memory unit. Based on the method, the device and the system can perform long-term prediction, can better learn spatial correlation, simultaneously allow a plurality of auxiliary information to be used, generate fine prediction with high interpretability, and further effectively solve prediction of strong convection parameters.
Drawings
In order to more clearly illustrate the detailed description of the invention or the technical solutions in the prior art, the drawings that are needed in the detailed description of the invention or the prior art will be briefly described below. Throughout the drawings, like elements or portions are generally identified by like reference numerals. In the drawings, elements or portions are not necessarily drawn to scale.
Fig. 1 is a flowchart of a strong convection weather physical characteristic quantity prediction method based on an attention mechanism according to an embodiment of the invention;
FIG. 2 is a diagram of physical characteristics suitable for heavy convection weather according to an embodiment of the invention;
fig. 3a is a schematic diagram of temperature visualization according to an embodiment of the present invention, in which the horizontal axis represents longitude and the vertical axis represents latitude.
FIG. 3b is a schematic view of a divergence visualization in accordance with an embodiment of the present invention; wherein the horizontal axis represents longitude and the vertical axis represents latitude; .
FIG. 3c is a schematic diagram illustrating the visualization of potential vorticity according to an embodiment of the present invention; wherein the horizontal axis represents longitude and the vertical axis represents latitude; .
FIG. 3d is a schematic diagram of a potential visualization according to an embodiment of the present invention; wherein the horizontal axis represents longitude and the vertical axis represents latitude; .
FIG. 3e is a schematic view of a barosphere temperature visualization according to an embodiment of the invention; wherein the horizontal axis represents longitude and the vertical axis represents latitude; .
Fig. 4 is a diagram illustrating distribution of different strong convection weathers according to an embodiment of the present invention.
Fig. 5 is a schematic diagram illustrating a principle of a context pairing mechanism according to an embodiment of the present invention.
FIG. 6 is a diagram of spatio-temporal sequences according to an embodiment of the present invention.
FIG. 7 is a schematic diagram of an attention mechanism according to an embodiment of the present invention.
Fig. 8 is a schematic view of the atmospheric flow predicted by a strong convection weather physical characteristic quantity prediction method based on an attention mechanism according to an embodiment of the present invention.
Fig. 9 is a hardware diagram of a strong convection weather physical characteristic quantity prediction system based on an attention mechanism according to an embodiment of the invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely below, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that all the directional indicators (such as upper, lower, left, right, front and rear … …) in the embodiment of the present invention are only used to explain the relative position relationship between the components, the movement situation, etc. in a specific posture (as shown in the drawing), and if the specific posture is changed, the directional indicator is changed accordingly.
In addition, the descriptions related to "first", "second", etc. in the present invention are only for descriptive purposes and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention will now be further described with reference to the accompanying drawings.
Fig. 1 shows a flowchart of a strong convection weather physical characteristic quantity prediction method based on an attention mechanism according to an embodiment of the invention. As shown in fig. 1, an embodiment of the present invention provides a strong convection weather physical characteristic quantity prediction method based on an attention mechanism, which starts with step S100 to obtain a weather parameter file. It will be appreciated that the meteorological parameter file is made up of a number of meteorological parameters. Note that the meteorological reference described herein may be in any format, chosen specifically for the actual needs. For example, it may be netCDF format or GRIB2 format or MICAPS4 format, etc. The embodiment of the present invention is not particularly limited thereto.
In some embodiments, based on a historical database, a plurality of meteorological parameters are acquired in a preset period and used as meteorological parameter files; the meteorological parameters comprise potential height, temperature, wind speed components in the horizontal direction, wind speed components in the vertical direction, relative humidity, specific humidity, vertical speed, vorticity, potential vorticity and specific cloud liquid water content under different atmospheric pressure heights.
For example, ERA5 data provided for ECMWF, ERA5 was the fifth generation ECMWF reanalysis data for the last 4 to 7 10 years of global climate and weather, used by embodiments of the present invention. The current data is from 1950 and is divided into the entries of climate data storage after 1950-. ERA5 replaced ERA-Interim reanalysis data. ERA5 provides hourly estimates of large amounts of atmospheric, ocean and surface quantities. For convenience, the overall mean and spread have been pre-calculated. The basic product comprises: potential height (geotropic), Temperature (Temperature), U-wind component of wind (U-wind), V-wind component of wind (V-wind), relative humidity (relative humidity), specific humidity (specific), vertical velocity (vertical), Vorticity (Vorticity), potential Vorticity (polytertialvorticity), specific cloud liquid water content (specific liquid water content), and the like.
Wherein the potential height (Geopotential) parameter is the gravitational potential energy of a unit mass at a specific location relative to the mean sea level. This is also the amount of work that must be done to lift the unit mass from average sea level to that position against gravity. The potential height can be calculated by dividing the potential by the gravitational acceleration g (=9.80665m/s-2) of the earth. The potential height plays an important role in weather meteorology (weather pattern analysis). Potential altitude maps plotted at constant pressure levels (e.g., 300, 500, or 850 hPa) can be used to identify weather systems such as cyclones, anticyclones, troughs, and ridges. At the earth's surface, this parameter shows changes in the potential (height) of the earth's surface, commonly referred to as the terrain.
Potential vorticity (momentum) is a measure of the ability of air to rotate in the atmosphere. If we ignore the effects of heating and friction, the potential vorticity is conserved behind the air mass. It is used to find places where large storms can occur and develop. The potential vorticity increases strongly above the top of the troposphere, so it can also be used for studies related to stratospheric and stratospheric-troposphere exchanges. When the column of air in the atmosphere begins to rotate, a large storm is created. The potential vorticity is calculated according to the wind, the temperature and the pressure of an air column in the atmosphere.
Relative humidity (relative humidity), which is the percentage of the water vapor pressure duty gas saturation value (the point at which water vapor begins to condense into liquid water or deposit into ice). For temperatures above 0 ℃ (273.15K), the saturation of water is calculated. At temperatures below-23 ℃, the saturation on ice is calculated. Between-23 ℃ and 0 ℃, this parameter is calculated by interpolating between ice and water values using a quadratic function.
Specific humidity (specific humidity), the parameter being the mass of water vapour per kg of humid air. The total mass of humid air is the sum of dry air, water vapor, cloud liquid, cloud ice, rain, and snowfall.
Vertical velocity (vertical velocity), which is the velocity of air movement in the upward or downward direction. The ECMWF Integrated Forecast System (IFS) uses a pressure-based vertical coordinate system, with pressure decreasing with altitude, so that negative values of vertical velocity indicate upward motion. Vertical velocity is very useful for understanding the large scale dynamics of the atmosphere, including regions of up/up (negative) and down/sink (positive).
Vorticity (Vorticity), a parameter that is a measure of the rotation of air in a horizontal direction about a vertical axis relative to a fixed point on the earth's surface. On the scale of the weather system, valleys (which may include weather features of rainfall) are associated with counterclockwise rotation (in the northern hemisphere), while ridges (weather features that bring about a light or stationary wind) are associated with clockwise rotation. Adding the effect of earth rotation, i.e., the coriolis parameter, to the relative vorticity produces absolute vorticity.
The specific cloud liquid water content (specific cloud liquid water content), the large amount of water produced by a large scale cloud of raindrop size, and thus may fall to the surface as precipitation. The cloud scenario in the ECMWF Integrated Forecast System (IFS) generates large-scale clouds. The cloud plots represent the formation and dissipation of clouds and large scale precipitation due to the variation of the large volumes (e.g., pressure, temperature and humidity) that IFS directly predicts on a grid box or larger spatial scale. This amount is expressed in kilograms per kilogram of total mass of humid air. The "total mass of humid air" is the sum of dry air, water vapor, cloud liquid, cloud ice, rain and snow. This parameter represents the average of the grid box. Clouds contain a continuum of water droplets and ice particles of varying sizes. The IFS cloud scheme reduces it to represent a number of discrete cloud droplets/particles, including cloud water droplets, rain droplets, ice crystals, and snow (polymerized ice crystals). The processes of droplet formation, phase transition and aggregation are also highly simplified in IFS.
FIG. 2 shows a graph of physical characteristics suitable for heavy convection weather, according to an embodiment of the invention. As shown in fig. 2, after data stored in the netCDF format is acquired, reading is performed, data visualization is performed, a timestamp is converted into a representation mode of a metric year, a physical parameter of a certain dimension in a specific time axis corresponding to a geographic range is read, and a partial weather parameter display diagram is displayed by a pyplot. Partial meteorological parameter displays are shown in fig. 3 a-3 e.
In step S200, a history of screened out strong convective weather is taken.
In some embodiments, the screened history of strong convective weather includes ID number, event occurrence time, revision time, accuracy, latitude, and weather category of strong convection.
Wherein the weather category of strong convection includes strong wind, destructive lightning, hail, typhoon, heavy snow.
Illustratively, the history selected by the embodiment of the present invention is a detailed history of weather with strong convection occurring in france provided by ESDW, and specifically includes the following basic data items, ID number, event occurrence time, revision time, accuracy, latitude, and specific weather category of strong convection. As shown in fig. 4, the present embodiment screened 2016 for strong convection weather that frequently occurred in france in 2018: strong wind, rainfall, hail, typhoon and lightning, and counting the occurrence frequency of the strong wind, the rainfall, the hail, the typhoon and the lightning.
In step S300, data is extracted and fused to create a data set.
In order to ensure that the positive samples and the negative samples of the acquired data set are uniformly distributed, the embodiment of the invention respectively extracts strong convection weather data and common weather data.
Specifically, strong convection weather data and common weather data are extracted; the method comprises the steps that strong convection weather data are used for locating all weather parameters of a time period corresponding to a strong convection event in a weather parameter file according to historical records of the strong convection weather, a point diffusion method is applied according to longitude and latitude coordinates of the event, the point is diffused outwards to form a first longitude and latitude range, and corresponding weather parameters are cut out from the weather parameter file based on the first longitude and latitude range and serve as positive samples of a data set;
the method comprises the steps that common weather data locate all weather parameters of a corresponding occurrence time period of a common event in a weather parameter file according to the history of the common weather, a second longitude and latitude range which is as large as the first longitude and latitude range is randomly selected according to the common event, and the corresponding weather parameters are cut out from the weather parameter file based on the second longitude and latitude range and serve as negative samples of a data set.
In some embodiments, the first warp range and the second warp range may be 1.75 ° x 1.75 ° in size. Of course, the embodiments of the present invention are only examples of longitude and latitude ranges, and the specific range size is not limited.
In the case of selecting 1.75 ° × 1.75 °, a piece of data is taken from the CSV data provided in step S200, two time stamps adjacent to each other are located according to the occurrence event, that is, a strong convection event is represented by two observation data spaced one hour before and after the occurrence event, and then a longitude and latitude range of 1.75 ° × 1.75 ° is formed by outward diffusion with the longitude and latitude coordinates of the occurrence event as the center, and a feature map whose tensor is 7 × 7 is converted, and a multidimensional feature map is superimposed (stack) in a parameter file according to the selected prediction data to form a tensor of 7 × 7 × 144.
In step S400, a model is built, the model comprises an attention mechanism module, a coding module and a decoding module, when input data enter the model, the input data pass through the attention mechanism module, a current input result is predicted according to a previous hidden state, the current input result and the current input data are overlapped together and coded by the coding module to form a current hidden state, a historical hidden state and the current hidden state are added, and then the historical hidden state and the current hidden state are decoded by the decoding module and then convolved to generate a prediction result.
Illustratively, the input tensor of the attention mechanism module is of a form of B × T × M × N, the hidden state is a tensor of B × hidden × M × N, and the size of the attention weight matrix output through fusion is B × 1 × M × N. The decoding and encoding modules may refer to the ConvLSTM implementation where the forward propagated input tensor is b x t x m x n x d and the output is b x t x m x n x d.
In some embodiments, a model built according to embodiments of the present invention includes:
the context pairing mechanism: the method provided by the invention is used for disordering the coding sequence and inputting the disordering sequence into the decoding layer, and experiments verify that the mechanism can effectively improve the accuracy of the prediction result and accelerate the process of model fitting.
X: training data (meteorological parameters) of the input model, t: subscript, which represents data corresponding to t time period.
Y: tag data (meteorological parameters) of the input model, t: subscript, which represents data corresponding to t time period.
H: historical data output by ConvLSTM in the coding layer is called hidden state, subscript t-1 represents data corresponding to a previous time period, and superscript k represents the kth dimension of the data.
D: and historical data output by ConvLSTM in the decoding layer, wherein a subscript t-1 represents data corresponding to a previous time period, and a superscript k represents the kth dimension of the data.
ConvLSTM: the basic processing unit of the model mainly adopts convolution and LSTM algorithm to obtain short-time memory.
Conv 2D: and performing a rolling machine filtering operation on the two-dimensional data so as to reduce the data.
ReLU: linear rectification function, activation function commonly used in artificial neural networks.
In the model, training data acquired in the last step is input, assuming that the training data is meteorological parameters Xt corresponding to a t time period, then using the attention mechanism method proposed by the method, obtaining an attention matrix A, multiplying A by Xt, it is understood that the training data is weighted, flow information therein is highlighted, and then the multiplication result is input to the encoding layer, ConvLSTM encodes each dimension of data and generates a hidden state for the next calculation, then inputs the tag data of t time period and the encoded history data calculated by the context pairing mechanism proposed by the invention into the decoding layer, obtaining coded data input this time, obtaining predictive label data input this time through subsequent Conv2D and ReLU operations, the data is a primary prediction, can be input into a CNN network for classification, and can also be directly combined with dimensions to generate a radar emission prediction graph.
Fig. 5 shows a schematic diagram of a context pairing mechanism according to an embodiment of the present invention. It is understood that the up-down pairing mechanism can be carriedAnd (5) memorizing in the upper stage. As shown in fig. 5, H in fig. 5 represents the historical data output by ConvLSTM in the coding layer, which is also called hidden state, where ConvLSTM specifically refers to the basic processing unit of the model, and mainly adopts convolution and LSTM algorithms to obtain short-term memory. The superscript of parameter H represents the dimension of the data. For example H 11 st dimension, H, representing the datakRepresents the k dimension, H, of the dataKRepresenting the K-th dimension of the data. The parameter D represents the historical data output by ConvLSTM in the decoding layer, the subscript t-1 represents the data corresponding to the previous time period, and the superscript k represents the kth dimension of the data. For example,
historical data representing dimension 1 of the ConvLSTM output in the previous time period decoding layer,
historical data representing the k-th dimension of the ConvLSTM output in the previous time period decoding layer,
history data representing the K-th dimension of the ConvLSTM output in the last time period decoding layer. Therefore, the embodiment of the invention can arrange all the previous hidden states based on a context pairing mechanism and add the hidden states to carry long-term reliability information, a decoding module (decoder) further decodes the hidden states generated by the encoder according to the context pairing mechanism, the hidden states are further convolved after being input into the decoder to generate a prediction result, and the prediction of strong convection weather characteristic parameters and the drawing of weather parameter prediction graphs can be realized by inputting a pre-designed parameter extraction program based on predicted physical characteristic forecast data.
Illustratively, the input data may be, for example, extracting potential height (geodetial), Temperature (Temperature), U-component (U-wind) of wind, V-component (V-wind) of wind, Relative humidity (Relative humidity), Specific humidity (Specific humidity), Vertical velocity (Vertical velocity), Vorticity (Vorticity), potential (e.g., potential) from multiple historical dataPhysical quantities such as vorticity (vorticity), Specific liquid water content (Specific liquid water content), etc., and the physical quantities are organized into
The embodiment of the invention defines input data at different time points into a plurality of space-time sequences in space-time
Each of which
M, N represents the spatial dimension and T represents the temporal dimension. FIG. 6 illustrates a spatio-temporal sequence diagram according to an embodiment of the present invention. The predictor tensor of the final model output can be expressed as
And relevant data extraction operation is carried out according to the prediction result of the future meteorological factors output by the model, so that the prediction of the physical characteristics of the strong convection weather can be realized.
FIG. 7 illustrates a schematic diagram of an attention mechanism according to an embodiment of the present invention. As shown in fig. 7, the core of the model in the embodiment of the present invention is the attention mechanism module. With regard to Attention mechanism module (Attention), wherein X: training data (meteorological parameters) of the input model, t: subscript, which represents data corresponding to t time period. X in FIG. 71Training data, X, representing a first dimension i Is shown asiTraining data of dimension, X l Is shown aslTraining data for the dimensions. H: historical data output by ConvLSTM in the coding layer is called hidden state, subscript t-1 represents data corresponding to a previous time period, and superscript k represents the kth dimension of the data. ConvLSTM specifically refers to the basic processing unit of the model, and mainly adopts convolution and LSTM algorithms to obtain short-term memory. In FIG. 7
The ConvLSTM in the coding layer representing the last time period outputs history data in a first dimension. tan h: a hyperbolic tangent function; conv 2D: performing a rolling machine filtering operation in the two-dimensional data, thereby simplifying the data; softmax: the softmax logistic regression model is a generalization of the logistic regression model to the multi-classification problem. A: the attention matrix is calculated through the step. Based on the attention mechanism module, an attention weight A can be calculatedt. After the attention mechanism principle is applied, the accuracy of the overall prediction can be improved. FIG. 8 shows a predicted atmospheric flow diagram according to an embodiment of the invention. As shown in fig. 8, the horizontal and vertical axes in fig. 8 represent positions (unit: km) for visualizing the flow of the atmosphere, so that the embodiment of the present invention can well simulate the flow of the atmosphere in nature.
Therefore, in step S500, the model is trained using the configured parameter file based on the data set, and the training is stopped and the weight file is saved when the loss value does not decrease any more continuously. The parameter file of the configured model comprises a global start date (global _ start _ date), a global end date (global _ end _ date), a data step size, a data interval duration, a proportion of a training set verification set test set, a down-sampling mode (selectable, average), physical features contained in the data set, physical features needing to be predicted, the size of batch-size and the like.
Finally, in step S600, based on the model, the next observed value of each weather factor of the strong convection is predicted. Specifically, the next observed value of each weather factor is predicted. The method includes the steps that latest observed data and previous hidden states are calculated through an encoder-decoder structure of a proposed model to obtain prediction results of weather factors, the prediction results are divided into two parts, Geopontinial, temporal, U-wind, V-wind, Relative hub, Specific hub and the like are used as one part to be input into six layers of cnn which are trained in advance to carry out strong-convection weather type classification, and data corresponding to dimensions to be predicted are visualized to form a prediction graph to assist judgment of meteorologists.
Fig. 9 shows a hardware diagram of a strong convection weather physical characteristic quantity prediction system based on an attention mechanism according to an embodiment of the invention. As shown in fig. 9, the embodiment of the present invention further provides a strong convection weather physical characteristic quantity prediction system based on the attention mechanism. The system 700 comprises a processor 701, the processor 701 being configured to: acquiring a meteorological parameter file; acquiring a history record of screened strong convection weather; extracting and fusing data to produce a data set; building a model, wherein the model comprises an attention mechanism module, an encoding module and a decoding module, when input data enter the model, the input data pass through the attention mechanism module, predict a current input result according to a previous hidden state, overlap the current input result and the current input data together, encode the current input result and the current input data through the decoding and encoding module to form a current hidden state, add a historical hidden state and the current hidden state, and perform convolution to generate a prediction result after decoding through the decoding module; based on the data set, training the model by using a configured parameter file, and stopping training when the loss value does not decrease for a plurality of times; and predicting the next observation value of each weather factor of the strong convection based on the model.
The processor 701 may be a processing device, such as a microprocessor, Central Processing Unit (CPU), Graphics Processing Unit (GPU), etc., including one or more general-purpose processing devices. More specifically, the processor 701 may be a Complex Instruction Set Computing (CISC) microprocessor, Reduced Instruction Set Computing (RISC) microprocessor, Very Long Instruction Word (VLIW) microprocessor, processor running other instruction sets, or processors running a combination of instruction sets. The processor 701 may also be one or more special-purpose processing devices such as an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), a Digital Signal Processor (DSP), a system on a chip (SoC), or the like.
In some embodiments, the processor 701 may be further configured to: extracting strong convection weather data and extracting common weather data; the method comprises the steps that strong convection weather data are used for locating all weather parameters of a time period corresponding to a strong convection event in a weather parameter file according to historical records of the strong convection weather, a point diffusion method is applied according to longitude and latitude coordinates of the event, the point is diffused outwards to form a first longitude and latitude range, and corresponding weather parameters are cut out from the weather parameter file based on the first longitude and latitude range and serve as positive samples of a data set; the method comprises the steps that common weather data locate all weather parameters of a corresponding occurrence time period of a common event in a weather parameter file according to the history of the common weather, a second longitude and latitude range which is as large as the first longitude and latitude range is randomly selected according to the common event, and the corresponding weather parameters are cut out from the weather parameter file based on the second longitude and latitude range and serve as negative samples of a data set.
In some embodiments, the first warp range has a size of 1.75 ° x 1.75 °.
In some embodiments, the processor 701 may be further configured to: and inputting the data of the newly observed strong convection weather as input data into the model to obtain a prediction result.
In some embodiments, the configured parameter file includes a global start date, a global end date, a data step size, a data interval duration, a proportion of a training set validation set test set, a down-sampling mode, a physical feature contained in the data set, and a physical feature to be predicted.
In some embodiments, based on a historical database, a plurality of meteorological parameters are acquired in a preset period and used as meteorological parameter files; the meteorological parameters comprise potential height, temperature, wind speed components in the horizontal direction, wind speed components in the vertical direction, relative humidity, specific humidity, vertical speed, vorticity, potential vorticity and specific cloud liquid water content under different atmospheric pressure heights.
In some embodiments, the screened history of strong convective weather includes ID number, event occurrence time, revision time, accuracy, latitude, and weather category of strong convection.
In some embodiments, the weather categories of strong convection include strong wind, destructive lightning, hail, typhoon, heavy snow.
The above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit and scope of the present invention, and they should be construed as being included in the following claims and description.
Claims (10)
1. The strong convection weather physical characteristic quantity prediction method based on the attention mechanism is characterized by comprising the following steps of:
acquiring a meteorological parameter file;
acquiring a history record of screened strong convection weather;
extracting and fusing data to produce a data set;
building a model, wherein the model comprises an attention mechanism module, a coding module and a decoding module, when input data enter the model, the input data pass through the attention mechanism module, predict a current input result according to a previous hidden state, overlap the current input result and the current input data together, and carry out coding through the coding module to form a current hidden state, add a historical hidden state and the current hidden state, carry out convolution after decoding through the decoding module to generate a prediction result;
based on the data set, training the model by using a configured parameter file, and stopping training when the loss value does not decrease for a plurality of times;
and predicting the next observation value of each weather factor of the strong convection based on the model.
2. The method of claim 1, wherein said extracting and fusing data to produce a data set comprises the steps of:
extracting strong convection weather data and extracting common weather data; the method comprises the steps that strong convection weather data are located to all weather parameters of an occurrence time period corresponding to a strong convection event in a weather parameter file according to historical records of the strong convection weather, a point diffusion method is applied according to longitude and latitude coordinates of the event, longitude and latitude coordinate points of the event are diffused outwards to form a first longitude and latitude range, and corresponding weather parameters are cut out from the weather parameter file based on the first longitude and latitude range to serve as a positive sample of a data set;
the method comprises the steps that common weather data locate all weather parameters of a corresponding occurrence time period of a common event in a weather parameter file according to the history of the common weather, a second longitude and latitude range which is as large as the first longitude and latitude range is randomly selected according to the common event, and the corresponding weather parameters are cut out from the weather parameter file based on the second longitude and latitude range and serve as negative samples of a data set.
3. A method according to claim 2, characterized in that said first warp range has a size of 1.75 ° x 1.75 °.
4. The method of claim 1, wherein predicting next observed values for strong convection weather factors based on the model comprises the steps of:
and inputting the data of the newly observed strong convection weather as input data into the model to obtain a prediction result.
5. The method of claim 1, wherein the configured parameter file comprises a global start date, a global end date, a data step size, a data interval duration, a training set validation set test set proportion, a down-sampling mode, a physical feature contained in the data set, and a physical feature to be predicted.
6. The method of claim 1, wherein said obtaining a meteorological parameter file comprises the steps of:
based on a historical database, acquiring a plurality of meteorological parameters in a preset period to serve as meteorological parameter files;
the meteorological parameters comprise potential height, temperature, wind speed components in the horizontal direction, wind speed components in the vertical direction, relative humidity, specific humidity, vertical speed, vorticity, potential vorticity and specific cloud liquid water content under different atmospheric pressure heights.
7. The method of claim 1, wherein the screened history of strongly convective weather includes ID number, time of occurrence of event, revision time, accuracy, latitude, and strongly convective weather category.
8. The method of claim 7, wherein after adding the historical concealment state and the current concealment state, decoding the current concealment state by the decoding module and performing convolution to generate the prediction result, comprising the steps of:
and based on a context pairing mechanism, sorting all previous hidden states, adding the hidden states and the current hidden states to carry long-term trust information, further decoding the hidden states generated by the encoding module by the decoding module according to the context pairing mechanism, and further performing convolution to generate a prediction result after the hidden states are input into a decoder.
9. The strong convection weather physical characteristic quantity prediction system based on the attention mechanism is characterized by comprising a processor, wherein the processor is configured to:
acquiring a meteorological parameter file;
acquiring a history record of screened strong convection weather;
extracting and fusing data to produce a data set;
building a model, wherein the built model comprises an attention mechanism module, a coding module and a decoding module, when input data enter the model, the input data pass through the attention mechanism module, a current input result is predicted according to a previous hidden state, the current input result and the current input data are overlapped together and are coded by the coding module to form a current hidden state, and after the historical hidden state and the current hidden state are added, the historical hidden state and the current hidden state are decoded by the decoding module and then are convolved to generate a prediction result;
based on the data set, training the model by using a configured parameter file, and stopping training when the loss value does not decrease for a plurality of times;
and predicting the next observation value of each weather factor of the strong convection based on the model.
10. The system of claim 9, wherein the processor is configured to:
extracting strong convection weather data and extracting common weather data; the method comprises the steps that strong convection weather data are used for locating all weather parameters of a time period corresponding to a strong convection event in a weather parameter file according to historical records of the strong convection weather, a point diffusion method is applied according to longitude and latitude coordinates of the event, the point is diffused outwards to form a first longitude and latitude range, and corresponding weather parameters are cut out from the weather parameter file based on the first longitude and latitude range and serve as positive samples of a data set;
the method comprises the steps that common weather data locate all weather parameters of a corresponding occurrence time period of a common event in a weather parameter file according to the history of the common weather, a second longitude and latitude range which is as large as the first longitude and latitude range is randomly selected according to the common event, and the corresponding weather parameters are cut out from the weather parameter file based on the second longitude and latitude range and serve as negative samples of a data set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210202985.0A CN114548595A (en) | 2022-03-03 | 2022-03-03 | Strong convection weather physical characteristic quantity prediction method and system based on attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210202985.0A CN114548595A (en) | 2022-03-03 | 2022-03-03 | Strong convection weather physical characteristic quantity prediction method and system based on attention mechanism |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114548595A true CN114548595A (en) | 2022-05-27 |
Family
ID=81661100
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210202985.0A Pending CN114548595A (en) | 2022-03-03 | 2022-03-03 | Strong convection weather physical characteristic quantity prediction method and system based on attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114548595A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116629142A (en) * | 2023-07-24 | 2023-08-22 | 合肥工业大学 | Lightning positioning track prediction method, system and storage medium based on transformer mechanism |
| CN117650834A (en) * | 2024-01-29 | 2024-03-05 | 南京理工大学 | Space-time flow prediction method of space-time integrated network based on knowledge distillation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200026760A1 (en) * | 2018-07-23 | 2020-01-23 | Google Llc | Enhanced attention mechanisms |
| CN111612245A (en) * | 2020-05-18 | 2020-09-01 | 北京中科三清环境技术有限公司 | Atmospheric pollution condition prediction method and device, electronic equipment and storage medium |
| CN111652437A (en) * | 2020-06-04 | 2020-09-11 | 上海眼控科技股份有限公司 | Runway visual range predicted value obtaining method, computer equipment and storage medium |
| CN113779892A (en) * | 2021-09-27 | 2021-12-10 | 中国人民解放军国防科技大学 | Wind speed and wind direction prediction method |
-
2022
- 2022-03-03 CN CN202210202985.0A patent/CN114548595A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200026760A1 (en) * | 2018-07-23 | 2020-01-23 | Google Llc | Enhanced attention mechanisms |
| CN111612245A (en) * | 2020-05-18 | 2020-09-01 | 北京中科三清环境技术有限公司 | Atmospheric pollution condition prediction method and device, electronic equipment and storage medium |
| CN111652437A (en) * | 2020-06-04 | 2020-09-11 | 上海眼控科技股份有限公司 | Runway visual range predicted value obtaining method, computer equipment and storage medium |
| CN113779892A (en) * | 2021-09-27 | 2021-12-10 | 中国人民解放军国防科技大学 | Wind speed and wind direction prediction method |
Non-Patent Citations (1)
| Title |
|---|
| 相峰等: "关于强对流天气预报的信息化分析", 《科学与信息化》, 30 November 2017 (2017-11-30) * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116629142A (en) * | 2023-07-24 | 2023-08-22 | 合肥工业大学 | Lightning positioning track prediction method, system and storage medium based on transformer mechanism |
| CN116629142B (en) * | 2023-07-24 | 2023-09-29 | 合肥工业大学 | Lightning positioning track prediction method and system based on transformer mechanism |
| CN117650834A (en) * | 2024-01-29 | 2024-03-05 | 南京理工大学 | Space-time flow prediction method of space-time integrated network based on knowledge distillation |
| CN117650834B (en) * | 2024-01-29 | 2024-04-12 | 南京理工大学 | Space-time flow prediction method of space-time integrated network based on knowledge distillation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Giffard-Roisin et al. | Tropical cyclone track forecasting using fused deep learning from aligned reanalysis data | |
| Ghimire et al. | Self-adaptive differential evolutionary extreme learning machines for long-term solar radiation prediction with remotely-sensed MODIS satellite and Reanalysis atmospheric products in solar-rich cities | |
| Jonnalagadda et al. | Forecasting atmospheric visibility using auto regressive recurrent neural network | |
| Dupuy et al. | ARPEGE cloud cover forecast postprocessing with convolutional neural network | |
| CN114548595A (en) | Strong convection weather physical characteristic quantity prediction method and system based on attention mechanism | |
| Zhou et al. | Quantitative precipitation forecast experiment based on basic NWP variables using deep learning | |
| Ali et al. | Sea ice forecasting using attention-based ensemble LSTM | |
| Liu et al. | Short-term daily prediction of sea ice concentration based on deep learning of gradient loss function | |
| CN117008217A (en) | Short-time strong precipitation prediction method integrating self-attention module and Unet model | |
| CN113496310A (en) | Atmospheric pollutant prediction method and system based on deep learning model | |
| Jena et al. | Deep convolutional network based machine intelligence model for satellite cloud image classification | |
| Pegion et al. | Understanding predictability of daily southeast US precipitation using explainable machine learning | |
| Larraondo et al. | Automating weather forecasts based on convolutional networks | |
| Verma et al. | Deep learning techniques in extreme weather events: A review | |
| Kim et al. | Improvement in forecasting short-term tropical cyclone intensity change and their rapid intensification using deep learning | |
| Chen et al. | Aviation visibility forecasting by integrating Convolutional Neural Network and long short-term memory network | |
| Socaci et al. | XNow: A deep learning technique for nowcasting based on radar products’ values prediction | |
| CN114355482B (en) | Strong convection weather identification method and system based on optical flow and meteorological numerical prediction | |
| Karpatne et al. | Earth science applications of sensor data | |
| Zhu et al. | Tropical Cyclones Intensity Prediction in the Western North Pacific Using Gradient Boosted Regression Tree Model | |
| Bose et al. | Simulation of atlantic hurricane tracks and features: A deep learning approach | |
| CN115877386A (en) | Radar echo extrapolation method introducing three-dimensional environment field data | |
| Bose et al. | Forecasting the evolution of north atlantic hurricanes: A deep learning approach | |
| Rae | A modified supercell composite parameter for supercell thunderstorms over the Gauteng Province, South Africa | |
| Wang et al. | The prediction method of tropical cyclone intensity change based on deep learning |
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 |