44 results about "Nowcasting" patented technology

The invention relates to a rainfall nowcasting method based on U-KAN grading loss weighting and frequency self-adaption, which comprises the following steps: (1) carrying out quality control and screening on a radar puzzle, and establishing a data set; (2) dividing a training set, a verification set and a test set, and standardizing; (3) constructing a U-KAN model, selecting training parameters and inputting data: combining a traditional Unet structure with a KAN network to construct the U-KAN model; then performing model training to obtain a prediction result; the prediction result is restored to the original magnitude through destandardization; (4) introducing a loss function based on a root-mean-square error and grade weighting in a model training stage, and performing post-processing on model output by adopting a frequency deviation correction method; (5) integrating and averaging the forecast products processed by the two complementary strategies, and recording the forecast products as U-KANE; and (6) predicting a rainfall result in the next three hours by using radar echo data in the past one hour, and outputting a rainfall short-term and imminent forecast result by the U-KANE.

The invention discloses a rainfall nowcasting method based on multi-scale feature fusion of a lightweight generative adversarial network, and the method comprises the steps: carrying out the down-sampling and up-sampling of a radar echo sequence through a classic U-net structure, and generating a prediction result; a prediction result and a true value are respectively combined with original input and are input into a discriminator for discrimination, in this way, the generator and the discriminator continuously carry out confrontation training, and finally, the generator can generate radar prediction data which are vivid enough. According to the method, the loss of a discriminator is calculated by using a mixed loss function in which a heavy rainfall area mask is introduced, so that the model pays more attention to the generation quality of the heavy rainfall area. In addition, the parameter quantity of the model is reduced by adopting grouping convolution, so that the requirements on calculation power and hardware resources are reduced.

The invention relates to an improved weather radar echo short-time nowcasting method based on a diffusion model, and the method comprises the steps: obtaining a to-be-measured radar echo image, combining the to-be-measured radar echo image with time sequence information, inputting the to-be-measured radar echo image into a radar echo extrapolation model, and obtaining a denoised radar echo image; the radar echo extrapolation model is obtained by using a training set to train a diffusion model; the training set is a combined reflectivity data set; and processing the image blocks through a core denoising module in the radar echo extrapolation model, and predicting noise. According to the invention, the precision and anti-interference capability of short temporary rainfall prediction can be improved.

The invention discloses a thunderstorm and gale dynamic extrapolation forecasting method based on multi-task MSTA-ConvLSTM, and relates to the technical field of atmospheric sciences, and the method comprises the steps: constructing a multi-source fusion spatio-temporal data set and a gale process record file, and obtaining a data set; a thunderstorm gale nowcasting model based on an MSTA-ConvLSTM model is constructed; according to the method, a dynamic gating mechanism of lightning data is introduced, whether a lightning probability prediction task is started or not is intelligently judged according to the lightning activity intensity, thunderstorm, gale and co-evolution characteristics of the lightning process are considered, 0-3-hour high-temporal-spatial-resolution nowcasting is achieved, the limitation of traditional single-task forecasting is broken through, a multi-task learning structure is adopted, and the prediction efficiency is improved. The model is guided to recognize the lightning occurrence probability during main task gale prediction, and the overall perception capability of a severe convection system is improved; extreme sample learning is enhanced through a weighted loss function, and output is optimized in combination with a gust coefficient model, so that the forecasting precision and practicability are effectively improved.

The invention discloses a precipitation nowcasting method based on deep learning, and the method comprises the following steps: collecting historical radar echo data, carrying out the preprocessing of a collected radar echo image sequence, and constructing a time-space sequence data set; constructing a dynamic flow space-time generative adversarial network model which comprises a generator and a discriminator; designing a discriminator based on a Transform architecture, performing authenticity evaluation on the generated rainfall forecast sequence by using a self-attention mechanism, and evaluating the time-space consistency and visual quality of the generated sequence; a two-stage training strategy is adopted to optimize the network, in the first stage, reconstruction loss and decoupling loss are used to pre-train a generator, in the second stage, adversarial loss and feature matching loss are introduced to carry out joint training, and high-quality rainfall nowcasting is achieved; according to the invention, the modeling capability of the model for complex meteorological phenomena is improved.

The invention discloses a short-time approaching rainfall prediction method based on multi-source fusion data and an MIM network, and the method comprises the steps: carrying out the space-time matching of radar observation data, Chinese land surface data assimilation system data and site observation data, carrying out the preprocessing of the matched data, including feature factor extraction, missing value processing and the like, building a deep learning model, and carrying out the prediction of the short-time approaching rainfall. The model is trained, and a multi-source fusion rainfall inversion result is output based on real-time radar data; furthermore, errors of a falling area and rainfall intensity of intelligent grid rainfall forecast are identified, a phase correction technology is utilized to correct a position error of an intelligent grid forecast rainfall zone, meanwhile, precision evaluation is performed on a prediction effect of the deep learning model according to the rainfall amount of short-time nowcasting, and intelligent prediction of short-time nowcasting is realized.

The invention relates to the technical field of rainfall prediction, in particular to a high-frequency second-length time sequence short-imminent forecasting method and system based on generative AI, and the method comprises the following steps: compressing an original radar reflectivity sequence into a compact hidden variable Token sequence through employing a pre-trained variational auto-encoder; in the compressed hidden space, with the hidden variable Token sequence as a condition, a prediction hidden variable Token sequence in a future time period is generated step by step in a probability generation mode; and reconstructing the predicted hidden variable Token sequence into a predicted radar reflectivity sequence in a future time period by using a decoder of the variational auto-encoder. According to the method, high-resolution and long-time-efficiency radar nowcasting is realized, the accuracy of rainfall forecasting in a 2-6-hour short-term and imminent forecasting grey area time window is improved, and a solid support is provided for constructing early warning and accurate decision-making of extreme weather.

The invention relates to the technical field of meteorology and artificial intelligence, in particular to a short-time rainfall nowcasting method based on multi-source meteorological data and a neural network, and the method comprises the steps: collecting the multi-source meteorological data of a target, and building a historical data set; respectively expanding the training set and the verification set based on a sliding window; fusing the radar data and the multi-source meteorological variable data by using a multi-head self-attention mechanism; training a rainfall nowcasting neural network driven by the multi-source meteorological data by using the historical data set; and on the basis of the trained rainfall nowcasting neural network, selecting parameters meeting a preset optimal condition, performing quantitative evaluation on the test set, generating an evaluation result of short-time rainfall nowcasting, and forecasting the short-time rainfall amount of the target period according to the real-time radar rainfall and the multi-meteorological variable data on the basis of the evaluation result. Therefore, the problems that an existing rainfall nowcasting method is single in driving data source, low in rainfall forecasting precision, difficult to forecast a complex rainfall process and the like are solved.

The invention provides a short-time rainfall nowcasting method fusing quantum calculation and a deep learning model, and belongs to the field of short-time rainfall nowcasting, and the method comprises the following steps: firstly, extracting the precipitation amount of a target region and the long-term historical data of meteorological elements, then taking the precipitation amount at a to-be-predicted moment as a prediction target amount, and carrying out the prediction of the prediction target amount; the meteorological elements and the precipitation amount M hours before the to-be-forecasted moment are used as model input characteristic quantities. And then changing the meteorological elements and the precipitation amount M hours before the to-be-forecasted moment into high-dimensional quantum probability characteristics related to the precipitation amount at the to-be-forecasted moment by adopting a quantum calculation method. Finally, the quantum line precipitation probability information serves as input, the precipitation in the next one hour serves as output, and a short-time precipitation nowcasting model is constructed and trained. The quantum line and the quantum superposition state are utilized, classic data are mapped to a high-dimensional quantum Hilbert space, local optimal solution traps can be avoided, the calculation burden is remarkably reduced, and the calculation efficiency is improved. And short-time rainfall forecasting under the condition of massive high-dimensional meteorological feature input is realized.

The invention discloses a high-resolution low-altitude meteorological real-time and short-time nowcasting method, system and device, and relates to the technical field of low-altitude meteorological evaluation.The method comprises the steps that on the basis of the similarity of meteorological physical processes, according to underlying surface features of a target area and a long-term climate background field used for representing the long-term state of climate elements, the long-term climate background field is used for representing the long-term state of the climate elements; dividing the target area to obtain a plurality of sub-areas and corresponding meteorological physical characteristics; for each sub-region, matching and selecting an adaptive micro-scale meteorological model configuration scheme according to the meteorological physical characteristics of the sub-region; and respectively loading the multi-source real-time meteorological observation data of each sub-region and a real-time climate background field used for representing the recent state of climate elements into the called micro-scale meteorological model configuration scheme, and generating a high-resolution meteorological live field of each sub-region through rapid update cycle assimilation and parallel calculation. The method and the device have the effect of improving the forecasting accuracy.

The invention provides a rainfall nowcasting method and a rainfall nowcasting system based on a space-time attention diffusion model, which are used for simulating the space-time evolution process of rainfall so as to realize accurate rainfall nowcasting. The method comprises the following steps: decoupling historical rainfall data into a motion field feature and an intensity residual feature; based on an STA UNet architecture, extrapolating future evolution features through a noise iteration removal process by taking motion field features and intensity residual features as conditions; and reconstructing the future evolution characteristics into future rainfall prediction by utilizing Warp operation guided by a motion field and intensity residual fusion. According to the method, the uncertainty of the rainfall evolution process is considered, the modeling capability of the space and time dependency relationship in the rainfall process is enhanced, and an accurate and stable short-term rainfall nowcasting result is generated.

The invention discloses a low-altitude three-dimensional wind field nowcasting method and device, and the method comprises the steps: firstly collecting and preprocessing wind field data through an API, constructing a spatial-temporal feature project, and dividing a data set according to a time sequence; thirdly, a LightGBM framework is adopted, and a static optimal forecasting model is trained by means of Bayesian optimization; the method is characterized in that the static optimal forecasting model and an ensemble Kalman filtering framework are deeply fused, a dynamic assimilation forecasting system is constructed, forecasting output is continuously corrected and optimized, and error accumulation is effectively restrained. And finally, after performance verification, the system is operated in a business mode, and a final wind field forecasting result containing the NaN identifier is output.

The application provides a short-time precipitation nowcasting method fusing quantum computation and a deep learning model, and belongs to the field of short-time precipitation nowcasting, and steps are as follows: firstly, long-term historical data of precipitation and meteorological elements of a target area are extracted; then, the precipitation at a time of tentative prediction is taken as a prediction target quantity, and meteorological elements and precipitation of M hours before the time of tentative prediction are taken as model input characteristic quantities; then, a quantum computation method is used to change the meteorological elements and precipitation of M hours before the time of tentative prediction into high-dimensional quantum probability characteristics about the precipitation at the time of tentative prediction; finally, quantum circuit precipitation probability information is taken as input, and the precipitation of 1 hour in the future is taken as output, a short-time precipitation nowcasting model is constructed and trained, and the application uses quantum circuits and quantum superposition states to map classical data to a high-dimensional quantum Hilbert space, can avoid a local optimal solution trap, significantly reduces a calculation burden, and realizes short-time precipitation prediction under the condition of massive high-dimensional meteorological element inputs.

The application discloses a radar echo extrapolation method based on a space-time attention mechanism, first obtains radar echo data of a target area, and carries out pretreatment on the obtained radar echo data; the radar echo data after the pretreatment is subjected to sliding grouping and data augmentation, so that a radar echo sequence data set is obtained, and then the radar echo sequence data set is divided into a training set and a test set; a radar echo prediction network model based on a SimVP architecture is constructed and trained and tested, a multi-target loss function is used to supervise model training; and a future radar echo image is predicted in real time. Through the method, complex weather phenomena such as storms can be more accurately captured in dynamic evolution, the precision and interpretability of short-term nowcasting can be significantly improved, more reliable technical support is provided for timely and accurate early warning of meteorological disasters, and the method has important practical application value.

The invention relates to a local surface layer wind field fusion and nowcasting method, and belongs to the technical field of meteorological wind field fusion and forecasting. According to the method, the wind shear index calculated by a 90-meter anemometer tower is utilized to fill up a wind profile at the height of 90 meters above a surrounding automatic station, Barnes step-by-step corrected wind field fusion is carried out in combination with WRF-LES large eddy simulation, a convolutional neural network model is established between a large eddy simulation data set and a fused wind field data set, model parameters are obtained, and the wind field fusion is realized. And the parameters are used in WRF-LES real-time forecast to obtain proximity correction forecast of the near-earth wind field. According to the method, the local anemometer tower and encrypted automatic station data are fully applied, WRF-LES modeling is integrated, the method can be used for forecasting a local wind field, and the corresponding algorithm and process are suitable for wind field forecasting of wind power, low-altitude economy and the like.

This invention discloses a method for minute-level rolling spatiotemporal correction of precipitation based on multi-band weather radar fusion, relating to the fields of meteorological observation and short-term nowcasting. The method includes: acquiring multi-band weather radar base data and automatic rain gauge observation data, and preprocessing them; generating a minute-level quantitative precipitation estimation field based on the preprocessed multi-band weather radar base data; extrapolating to generate a minute-level quantitative precipitation forecast field; constructing station residuals and generating a spatial error field; using the spatial error field to spatially correct the minute-level quantitative precipitation forecast field to obtain an intermediate forecast field; extracting precipitation dynamic evolution characteristics to determine time-rolling correction coefficients, and performing time correction on the intermediate forecast field to obtain a corrected minute-level quantitative precipitation forecast field. By generating a precipitation estimation field through multi-band radar fusion, and then combining this with station observations to construct a spatial error field and time correction coefficients, it achieves accurate spatial and temporal correction of minute-level extrapolated forecasts.

The application provides a precipitation near extrapolation intelligent prediction method and device and electronic equipment, and the method comprises the following steps: obtaining precipitation data to be predicted; performing clipping processing on the precipitation data, and inputting the clipped precipitation data into an intelligent prediction model; the intelligent prediction model adopts a specific model structure, which comprises: adding a hierarchical upsampling fusion module HUFM and a CA attention mechanism on the basis of a U2-net model; through the intelligent prediction model, feature extraction and deep learning precipitation extrapolation are performed on the clipped real-time data, an extrapolation prediction result is obtained, the extrapolation prediction result is adaptively weighted with the prediction data, and a precipitation prediction result in a specified period in the future is output. The application is a new type of deep learning nowcasting method which simultaneously considers spatial accuracy, prediction timeliness and precipitation intensity description capability, and can meet the urgent needs of modern meteorological services for high-resolution and intelligent precipitation prediction.

The invention provides a rainfall nowcasting method based on a multi-scale adaptive fusion model, and provides a multi-scale space-time fusion attention framework (MSTFA), and through explicit modeling and adaptive fusion of local to global features, the ability to maintain the complex form and strength of a storm is well improved; a learnable space-time wavelet enhancement module (LSTWE) is designed, high-frequency detail perception is enhanced in a frequency domain through adaptive wavelet transform, and the problems of prediction result fuzziness and edge distortion are effectively relieved; an adaptive dynamic convolutional network (ADCN) is constructed, dynamically generated parameters are used for replacing a fixed convolution kernel, and the fitting capability of a nonlinear dynamic process in a precipitation system is enhanced; the three modules are organically embedded into a Transform backbone network to form an end-to-end forecasting framework, leading performance is achieved on a plurality of standard data sets, and effectiveness and complementarity of the modules are verified through detailed ablation experiments.

The application discloses a radar near precipitation prediction method based on frequency domain perception and conditional diffusion, and belongs to the technical field of image data processing. VAE The method comprises the following steps: obtaining a radar echo data set, pre-training a variational autoencoder M VAE ; constructing a convolution enhanced frequency domain Transformer; constructing an improved denoising network and M VAE based on the multi-layer convolution enhanced frequency domain Transformer, generating an improved conditional diffusion model and training a near precipitation prediction model; and being used for near precipitation prediction. The application overcomes the core defects of poor fusion of local and global features, prediction ambiguity caused by signal-noise confusion and the like in the prior art through frequency domain-space collaborative modeling and efficient conditional diffusion architecture design, enhances the adaptation capability to complex weather scenes, can more accurately capture high-frequency details and edge features of a convection system, significantly improves the prediction accuracy of a weak echo area, and improves the accuracy of short-term and near-term precipitation prediction.

The application discloses a radar echo space-time extrapolation method based on asymmetric space modeling, belongs to the cross field of meteorological radar data processing and artificial intelligence technology, and first constructs an asymmetric input space sample containing a target region and a surrounding extended region by preprocessing original meteorological radar observation data, builds an asymmetric space modeling prediction network based on a 3D U-Net architecture, adopts a regional mask mean square error loss function to complete model training, and finally realizes accurate space-time extrapolation of the target region radar echo. The application solves the technical problems of information fragmentation and significant long-time prediction error accumulation of the traditional symmetric modeling method, improves the precision and stability of radar echo extrapolation under complex weather systems, and is suitable for 0-1 hour nowcasting and meteorological disaster accurate early warning scenes.

The invention provides a physical constraint enhanced type encoder-decoder network model. The model adopts an enhanced encoder-decoder architecture, and the architecture combines a normalized space-time convolutional memory network and a space channel dual attention module. The normalized spatio-temporal convolutional memory network enhances the multi-scale temporal precipitation and cloud top temperature features extracted from the main encoder. The space channel double attention module further guides the network to pay attention to important features in space and channel dimensions. In addition, in order to better capture the space-time evolution trend of rainfall and the learning process of the constraint model, the model designs a new physical constraint loss function. For coastal areas of China and Southeast Asia, the model carries out rainfall forecasting hour by hour by using global rainfall measurement (GPM) rainfall data and Himawari satellite observation data.

The invention relates to the technical field of radar signal processing, in particular to a weather radar echo nowcasting method fusing a mesoscale numerical mode. Comprising the following steps: S1, acquiring radar live echo signal intensity data and radar monitoring terrain data; carrying out binarization processing on the radar real-time echo signal intensity to obtain the radar real-time echo signal intensity after binarization processing; based on the radar monitoring terrain data, matching the radar live echo signal intensity after binarization processing with the radar monitoring terrain to obtain a matching result; and S2, on the basis of a result after matching, radar live echo data are obtained, and CMA-meso original data are collected. According to the method, the Fi-radar and Fusion models are provided, multi-source data are fused to eliminate the space-time difference, the extrapolation definition is optimized, and the disaster weather early warning precision is improved through adaptive adjustment of CSI, FAR, POD and MSE indexes.

The application discloses a precipitation nowcasting method based on a multi-scale feature fusion of a lightweight generative adversarial network, a generator generates a prediction result by performing down-sampling and up-sampling on a radar echo sequence through a classical U-net structure, and the prediction result is combined with a true value and an original input respectively and input into a discriminator for discrimination, in this way, the generator and the discriminator continuously perform adversarial training, and finally, the generator can generate radar prediction data which is realistic enough. The method uses a hybrid loss function with a strong precipitation area mask introduced, calculates the loss of the discriminator, and makes the model pay more attention to the generation quality of the strong precipitation area. In addition, grouping convolution is adopted to reduce the parameter quantity of the model, so as to reduce the requirement for computing power and hardware resources.

The invention discloses a convection inception intelligent forecasting system and method based on multi-source data. The system comprises a data fusion module, a feature extraction module, a trend discrimination module, a fusion deduction module and a forecasting output module which are connected in sequence. The data fusion module is used for constructing a unified spatial-temporal feature data set; the feature extraction module adopts a deep neural network, combines time sequence modeling and an attention mechanism, and extracts weak signal features of a convection newborn stage; the trend discrimination module identifies the vertical development trend of the cloud body and the spatial evolution characteristics of the convection embryo through a characteristic fusion model; the fusion deduction module outputs a convection inception probability field based on the features; the forecast output module generates a short-time nowcasting result and realizes visual display and threshold alarm of the service terminal; by introducing a multi-source meteorological observation and intelligent feature fusion model, potential convection units can be found earlier, the development trend of the potential convection units can be dynamically tracked, and the advance and precision of short-time severe convection forecasting are greatly improved.

The invention discloses a lightning potential and proximity forecasting method based on a diffusion model, and the method comprises the steps: obtaining a data set comprising time sequence weather state data, and carrying out the training of a diffusion model, and obtaining a trained diffusion model; acquiring the weather state data Xt of the current time step t and the weather state data Xt-1 of the historical time step, and predicting the weather state Xt + 1 of the future t + 1 moment based on the trained diffusion model; carrying out autoregression iteration on the prediction process, and predicting weather state data of T time steps in the future; the weather state data comprises lightning potential forecast or lightning nowcasting. According to the method, uncertainty in lightning forecasting is quantified through the diffusion model, the lightning forecasting based on multi-source data feature driving uncertainty quantification is researched, unification of lightning occurrence probability and evolution process forecasting is further achieved, and therefore dynamic lightning forecasting based on multi-source data and diffusion model driving is achieved.

The present invention relates to a method and system for estimating rainfall timing based on cloud base height variations, belonging to the field of radio. The method comprises: inputting THI data, making a judgment based on a reflectivity threshold, eliminating interference data, and calculating the cloud base height in each radial direction, where THI stands for vertical to top scanning; determining whether rainfall is likely to form based on cloud base height variations and cloud base velocity; and performing data fitting if the conditions are met; estimating the rainfall time based on the fitting results and outputting the results. The present invention solves the problem that prior art techniques fail to fully utilize the strong correlation between dynamic changes in cloud base height and rainfall formation. It can provide meteorological service guarantees for major events, offer technical support for weather modification operations, and provide prerequisite services for business forecasters to issue nowcast information.

A high-resolution spatiotemporal nowcasting method and storage medium for urban flooding based on a U-RNN (U-RNN) is proposed. Input features are derived from the current rainfall sequence and urban spatial factors, and these features are fed into the U-RNN to produce real-time urban flooding forecasts. The U-RNN uses a sliding window warm-up training paradigm during training. The U-RNN and the shallow water equation solver form an inverse problem training framework. The U-RNN and the shallow water equation solver are alternately trained based on the inverse problem training framework, thereby alternately optimizing network parameters and physical parameters to obtain the trained U-RNN. The U-RNN's prediction accuracy is improved by optimizing the physical parameters. Based on the higher-precision prediction results, the physical parameters are then inverted and optimized to achieve global coordinated optimization, thereby improving the overall accuracy of the system and enabling faster, more accurate, and high-spatiotemporal resolution urban flooding nowcasting.

The invention discloses a radar echo extrapolation method integrating physical guidance and a diffusion coupling mechanism, and relates to the technical field of meteorological short-term and imminent forecast. The method comprises the following steps: firstly, coding radar echo data by using a physical guidance network, capturing long-distance space-time dependence through a multi-scale gating axial block, and generating a preliminary prediction sequence with macroscopic physical consistency in combination with a multi-domain space-time evolution module and an explicit physical advection guidance mechanism; and then, adopting a channel decoupling and cascade diffusion strategy, taking the preliminary prediction sequence as a physical prior skeleton, extracting guide features by using a double-flow spatio-temporal context network, and executing reverse iteration denoising and refined texture repair through a diffusion model. According to the method, the physical guidance composite loss function is introduced, the problems of image blurring and physical evolution logic deficiency existing in a traditional method are effectively relieved, and the forecasting precision and image fidelity of severe convective weather are improved.

This application proposes a convective nowcasting system and method based on multimodal data, belonging to the field of convective nowcasting technology. This invention fundamentally solves the problem of fusion distortion caused by physical conflicts in multi-source observation data, achieving end-to-end optimization from "raw multimodal input" to "high-quality forecast output." Compared with existing technologies, this invention significantly improves the ability to capture the initial stage of convection, enhances the predictive stability of strong echo structures in the mature stage, and effectively eliminates false signals in the dissipation stage, thereby greatly improving the accuracy, physical rationality, and timeliness of the 0–2 hour radar reflectivity factor prediction field. This not only enhances the lead time and hit rate of short-term severe convective warnings but also provides more reliable technical support for disaster prevention and mitigation decision-making, possessing outstanding substantive features and significant progress.

The invention discloses a thunderstorm gale short-time nowcasting method based on an ENTwo evolution network and Uform network hybrid architecture, and the method comprises the steps: obtaining historical live data, which comprises a radar echo sample data set, a maximum wind speed data set and a lightning data set; generating a grid thunderstorm gale label data set based on the gale wind speed and the lightning data; an ENTwo evolution network and a Uform hybrid model are constructed, the evolution network performs space-time extrapolation on the input data to generate a forecast field, and the Uform fuses an extrapolation result and historical data to output a thunderstorm and gale forecast field; training the model through a loss function; and inputting real-time data into the training model, and outputting a lattice forecasting field. According to the method, the accuracy and timeliness of 0-2-hour thunderstorm gale forecasting are effectively improved, the spatial resolution reaches 1 kilometer, the strong echo and extreme wind speed forecasting performance is superior to that of a traditional method, and high-precision technical support is provided for disaster prevention and reduction.