Short temporary rainfall forecasting method and device, electronic equipment and storage medium

Abstract

The invention discloses a short temporary rainfall forecasting method and device, electronic equipment and a storage medium. The forecasting method comprises the steps that a forecasting model with station short temporary rainfall as a learning target is built; short temporary rainfall data are obtained, a data set used for training a prediction model is established, and the short temporary rainfall data comprise radar forecast reflectivity, site observation rainfall and site positions; and inputting the data set and the time characteristic data into a prediction model, and carrying out model training based on different learning branches on the prediction model so as to output a prediction result of short temporary rainfall for expected time based on the prediction model, wherein the learning branch comprises a first branch for radar forecast reflectivity learning and a second branch for site observation precipitation learning based on site position control. According to the method, site observation precipitation is integrated, and different branch processing is carried out on radar forecast reflectivity, so that forecast errors caused by radar extrapolation are reduced, and the precipitation forecast effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of weather forecasting, and particularly to a method and device for forecasting short-term and imminent precipitation, an electronic device, and a storage medium. Background Art

[0002] The forecasting of short-term and imminent precipitation usually refers to the forecasting of precipitation within the next few hours, such as the weather forecast of precipitation within the next six hours. The forecasting of short-term and imminent precipitation provides important information and strong data support for human production and life, travel, disaster prevention and mitigation, early warning decision-making, etc., and also provides historical data reference for more accurate numerical model simulation and weather forecasting.

[0003] Currently, the forecasting of short-term and imminent precipitation mainly includes numerical forecasting models and radar extrapolation models, etc. Although numerical forecasting can target the global and local regions to forecast the hourly precipitation at future times. However, the calculation process of this model is complex, the operation cost is high, and its forecasting performance depends to a large extent on the accuracy of the initial field and boundary conditions. The radar extrapolation model uses the ZR relationship (the relationship between radar reflectivity Z and rainfall intensity R) on the basis of extrapolating the radar to obtain the precipitation. However, for the prediction of hourly precipitation, this model needs to use the ZR relationship and other methods to calculate. And for calculating precipitation using the ZR relationship, it will be affected by the raindrop size distribution and thus cannot adapt to the variability of factors such as terrain and season, resulting in relatively large forecasting errors.

[0004] Therefore, there is an urgent need for a forecasting method suitable for short-term and imminent precipitation to overcome the defects of the existing technology. Summary of the Invention

[0005] The purpose of the present invention aims to solve at least one of the above technical problems to a certain extent.

[0006] To this end, the present invention proposes a method for forecasting short-term and imminent precipitation. The method obtains station observed precipitation data and radar forecast reflectivity, combines data such as time encoding, builds a network model with station short-term and imminent precipitation as the learning target, and learns the prediction of station precipitation based on different training branches, so as to obtain a prediction model capable of predicting precipitation at a preset time interval, and then realizes the forecasting of short-term and imminent precipitation. This method does not require complex calculations, effectively alleviates the problem of the applicability of the same ZR relationship in different regions and seasons, and at the same time uses the observed precipitation data of adjacent stations to reduce the precipitation prediction error to a certain extent, thereby improving the accuracy of short-term and imminent precipitation forecasting.

[0007] To achieve the above object, an embodiment of the first aspect of the present invention provides a short-term and imminent precipitation forecasting method, which includes: building a prediction model with the short-term and imminent precipitation at stations as the learning target; obtaining the short-term and imminent precipitation data at stations and establishing a data set for training the prediction model, where the short-term and imminent precipitation data includes radar forecast reflectivity, station observed precipitation, and station location; inputting the data set and time interval feature data into the prediction model, and performing model training on the prediction model based on different learning branches to output a forecasting result of the short-term and imminent precipitation for the expected time based on the prediction model, where the learning branches include a first branch for learning the radar forecast reflectivity controlled by the station location and a second branch for learning the station observed precipitation.

[0008] According to an embodiment of the present invention, the short-term and imminent precipitation data further includes surface elevation data and longitude and latitude data.

[0009] According to an embodiment of the present invention, the prediction model includes: an input layer, a position encoding layer, a convolutional layer, a fusion layer, a mapping layer, and an output layer. Among them, the input layer is used to input the short-term and imminent precipitation data; the position encoding layer is used to perform position encoding on the input feature and learn the feature corresponding to the input feature position; the convolutional layer is used to perform a convolutional operation on the short-term and imminent precipitation data and map it to a latent space; the fusion layer is used to transform and fuse the features of the short-term and imminent precipitation data; the mapping layer is used to map the latent space data output by the fusion layer to the data size of the original input; the output layer is used to output a forecasting result of the short-term and imminent precipitation at the station.

[0010] According to an embodiment of the present invention, establishing a data set for training the prediction model with the short-term and imminent precipitation data includes: spatiotemporally matching the radar forecast reflectivity, the station observed precipitation, and the station location to obtain sample data of the short-term and imminent precipitation; performing a balancing operation on the sample data based on the data distribution characteristics of the sample data of the short-term and imminent precipitation; dividing the balanced sample data into a training set and a test set according to a preset ratio.

[0011] According to an embodiment of the present invention, performing model training on the prediction model based on different learning branches includes: coupling the training results of the first branch and the training results of the second branch, and iteratively learning the coupled results to train the prediction model to learn the short-term and imminent precipitation forecasting based on the radar forecast reflectivity and the station observed precipitation.

[0012] According to an embodiment of the present invention, coupling the training results of the first branch and the training results of the second branch, and iteratively learning the coupled results to update the training results of the first branch includes: using the site location as a control condition to perform feature learning on the radar forecast reflectivity to obtain the training results of the first branch; performing feature learning on the site observed precipitation to obtain the training results of the second branch; coupling the training results of the first branch and the training results of the second branch to obtain a coupled result for the first branch and the second branch; re-inputting the coupled result into the first branch, and repeating the above steps until the prediction model meets the training requirements for the fusion feature learning based on the radar reflectivity and the site observed precipitation.

[0013] According to an embodiment of the present invention, the forecasting method further includes: comparing the forecast result output by the prediction model with the true result, calculating the corresponding loss value and / or calculating the scoring value of the forecasting performance of the prediction model for nowcasting precipitation by using an evaluation model; optimizing the prediction model based on the loss value and / or the scoring value.

[0014] According to an embodiment of the present invention, the forecasting method further includes: cutting the entire area to be forecast into multiple local areas, and recording the position of each local area in the entire area; inputting the nowcasting precipitation data of each local area into the prediction model to obtain the forecast result for each local area; merging the forecast results of each local area according to its position in the entire area to obtain the forecast result for the entire area.

[0015] To achieve the above object, an embodiment of the second aspect of the present invention provides a nowcasting precipitation forecasting device, which includes: a building unit for building a prediction model with site nowcasting precipitation as the learning target; an establishing unit for obtaining nowcasting precipitation data and establishing a data set for training the prediction model, where the nowcasting precipitation data includes radar forecast reflectivity, site observed precipitation, and site location; a predicting unit for inputting the data set and the time interval feature data into the prediction model, and performing model training on the prediction model based on different learning branches to output the forecast result of nowcasting precipitation for the expected time based on the prediction model, where the learning branches include a first branch for learning the radar forecast reflectivity controlled by the site location and a second branch for learning the site observed precipitation.

[0016] To achieve the above object, an embodiment of the third aspect of the present invention provides an electronic device, which includes: a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the short-term precipitation forecasting method according to the embodiment of the first aspect of the present invention is implemented.

[0017] To achieve the above object, a computer-readable storage medium provided by an embodiment of the fourth aspect of the present invention, when the computer program is executed by a processor, implements the short-term precipitation forecasting method according to the embodiment of the first aspect of the present invention.

[0018] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and / or additional aspects and advantages of the present invention will become apparent and understandable from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0020] Figure 1 is a schematic flowchart of a short-term precipitation forecasting method shown according to an exemplary embodiment;

[0021] Figure 2 is a schematic diagram of the network structure of a prediction model shown according to an exemplary embodiment;

[0022] Figure 3 is a schematic diagram of the comparison between the prediction result and the true result based on the prediction model shown according to an exemplary embodiment;

[0023] Figure 4 is a schematic block diagram of a short-term precipitation forecasting device shown according to an exemplary embodiment; and

[0024] Figure 5 is a schematic diagram of the structure of an electronic device shown according to an exemplary embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present invention, and should not be construed as limiting the present invention.

[0026] Specifically, a short-term precipitation amount forecasting method, a forecasting device, an electronic device, and a storage medium according to embodiments of the present invention will be described below with reference to the accompanying drawings.

[0027] Figure 1It is a schematic flowchart of a short-term and imminent precipitation forecasting method shown according to an exemplary embodiment. It should be noted that the short-term and imminent precipitation forecasting method of the embodiments of the present invention can be applied to the short-term and imminent precipitation forecasting device of the embodiments of the present invention. The short-term and imminent precipitation forecasting device can be configured on an electronic device or can be configured in a server. Among them, the electronic device can be a PC or a mobile terminal. The embodiments of the present invention do not make too many limitations on this.

[0028] As Figure 1 shown, the short-term and imminent precipitation forecasting method includes:

[0029] Step S110, build a prediction model with the short-term and imminent precipitation of a station as the learning target.

[0030] For example, the embodiments of the present invention can build a prediction model with the short-term and imminent precipitation of a station as the learning target based on different neural network models. For example, the embodiments of the present invention preferably build it based on the Adaptive Fourier Neural Operator (AFNO). The AFNO is a neural network operator based on the Fourier transform. It is designed as an effective feature fusion device that can transform features from the time domain to the frequency domain for fusion. It can be regarded as a continuous global convolution and does not depend on the input resolution, so it can reduce the video memory occupation, reduce the computational complexity, and improve the computational efficiency while ensuring accuracy. Specifically, the present invention can build a prediction model with the short-term and imminent precipitation learning of a station as the target based on the AFNO. This prediction model can also be called the Station Conditioned Adaptive Fourier Neural Operator (StationCondAFNO) model. It is a Fourier neural operator that adaptively adjusts according to the station conditions. By fusing the precipitation characteristics of the station and introducing an adaptive mechanism, the model can dynamically adjust according to the precipitation characteristics of the station and better capture and predict the precipitation meteorological conditions at the station location.

[0031] It should be noted that building a prediction model based on the AFNO in the embodiments of the present invention is only a preferred method. Those skilled in the art can build it using other neural network models according to actual needs, and no too many limitations are made on this.

[0032] Step S120, obtain short-term and imminent precipitation data, and establish a data set for training the prediction model. The short-term and imminent precipitation data includes radar forecast reflectivity, station observed precipitation, and station location.

[0033] For example, the embodiments of the present invention are characterized by radar short-term prediction reflectivity and precipitation observed at stations, and calculations and model learning are performed at the station locations. Therefore, the embodiments of the present invention can obtain radar short-term prediction data, precipitation data observed at each station, and the location information of each corresponding station, and process these data to construct a data set for model training.

[0034] Step S130, input the data set and time characteristics into the prediction model, and perform model training on the prediction model to output a prediction result of short-term precipitation for the expected time based on the prediction model, where the model training is performed with the station location as the control condition and the precipitation observed at the station as a separate training branch.

[0035] For example, in order to enable the prediction model to predict the precipitation for a preset time interval, such as predicting the precipitation for the next two hours. The embodiments of the present invention need to input time characteristics during the model training stage to distinguish the prediction of precipitation for different future time intervals. For example, distinguish the prediction of precipitation for the first hour and the second hour in the future. And the present invention uses the precipitation observed at the station as an independent branch for training and learning, and uses the station location as the control condition to train and learn the radar prediction reflectivity, so as to predict the precipitation for a certain future time interval.

[0036] It should be noted that the specific implementation details of the above steps will be described in detail in the following embodiments and will not be elaborated here.

[0037] In the embodiments of the present invention, a network model is built with the station precipitation data as the control condition, by integrating the precipitation observed at the station and using the precipitation observed at the station as an independent branch for learning and training, so that the model can more effectively combine the station precipitation data and the radar short-term prediction data for short-term precipitation prediction. Compared with the prediction based on the ZR relationship, the embodiments of the present invention can reduce the prediction error caused by radar extrapolation and improve the prediction accuracy.

[0038] Next, the present invention will be further described in detail by combining the drawings and examples.

[0039] In a preferred embodiment, the short-term precipitation data further includes surface elevation data and longitude and latitude data.

[0040] For example, considering that some regions have rich surface information, the embodiments of the present invention introduce data such as surface elevation data and longitude and latitude data, so that the model can comprehensively train and learn various factor data, thereby improving the performance of the prediction model in different regions and at different observation distances.

[0041] It should be noted that the embodiments of the present invention are not limited to this. Those skilled in the art can introduce other data related to short-term and imminent precipitation according to actual needs, so that the prediction model can be adapted to the precipitation prediction for certain specific conditions. The present invention will not elaborate on this too much.

[0042] In a preferred embodiment, the prediction model includes an input layer, a position encoding layer, a convolutional layer, a fusion layer, a mapping layer, and an output layer. Among them, the input layer is used to input the short-term and imminent precipitation data and the time features; the position encoding layer is used to perform position encoding on the input features and learn the features corresponding to the positions of the input features; the convolutional layer is used to perform convolutional operations on the short-term and imminent precipitation data and map it to the hidden space; the fusion layer is used to transform and fuse the features of the short-term and imminent precipitation data; the mapping layer is used to map the hidden space data output by the fusion layer to the data size of the original input; the output layer is used to output the short-term and imminent precipitation forecast results at a preset time interval.

[0043] For example, as mentioned above, the prediction model of the embodiments of the present invention is trained with the station location as the control condition and the station observed rainfall as an independent branch. Therefore, the input layer of the embodiments of the present invention includes at least two parts of input. The first part of the input includes radar forecast reflectivity and time features, etc., and the second part of the input includes the station observed precipitation, and the station observed precipitation is processed as an independent branch.

[0044] Correspondingly, the convolutional layer in the present invention also includes two parts. The first part of the convolution corresponds to the input of data such as radar reflectivity and time features, while the second part of the convolution corresponds to the input of the station observed precipitation.

[0045] The convolutional layer divides the input data into small blocks (patches) of a fixed size. Each small block can be regarded as an independent unit for convolutional operations to extract local features within the small block. After the convolutional operation, each small block is converted into a corresponding feature vector, and this feature vector contains the key information within the small block. Thus, these feature vectors are combined together to form a feature map, and this feature map represents the representation of the original input data in the hidden space. Mapping the original input to the hidden space can enable the model to learn the essential laws of the data, which is helpful for the learning and optimization of the model, and improves the efficiency and performance of model training. Since the training and learning of the radar forecast reflectivity are based on the station location as the control condition, the output of the position encoding layer (station position encoding) is used as the input of the first part of the convolutional layer in the embodiments of the present invention.

[0046] The fusion layer is the main component of this prediction model, and it performs feature processing such as Fourier transform, frequency domain feature fusion, inverse Fourier transform, and channel feature mixing on the data output by the convolutional layer.

[0047] The mapping layer maps the latent space data processed and output by the fusion layer to the original data size through upsampling. Since the data in the latent space is usually low-dimensional data obtained after a series of transformations and feature extractions, it is necessary to restore the low-dimensional data to the original data size or resolution through upsampling. At the same time, details can be enhanced and lost data information can be restored, etc. It should be noted that the present invention does not limit the specific method used for upsampling, and will not elaborate on this too much.

[0048] After the data is mapped and restored, the output layer can output the predicted precipitation according to the time characteristics preset by the model. For example, if the model is set to predict the precipitation within the next two hours, the output layer outputs the precipitation prediction results hour by hour, that is, the precipitation in the next hour and the next two hours.

[0049] In the embodiment of the present invention, the prediction model established in the embodiment of the present invention is trained and learned based on different branches for radar forecast reflectivity and station observed precipitation, so as to fuse the characteristics of the two to predict the precipitation at future times. At the same time, taking the station location as a control condition improves the prediction results output by the model and reduces the forecast error brought by radar extrapolation.

[0050] In a preferred embodiment, the model training of the prediction model based on different learning branches includes: coupling the training results of the first branch and the training results of the second branch, and iteratively learning the coupled results to train the prediction model to learn short-term precipitation forecasting based on the radar forecast reflectivity and the station observed precipitation.

[0051] For example, as mentioned above, the model training in the present invention is based on two learning branches. Therefore, in order to better enable the prediction model to fuse the data characteristics of the radar forecast reflectivity and the station observed precipitation for short-term precipitation forecasting, the embodiment of the present invention couples the training results of the first branch and the training results of the second branch in the middle stage of training, and then continues to learn the coupled features, so as to continuously fuse the data characteristics of the station observed precipitation during training.

[0052] In a more preferred embodiment, coupling the training result of the first branch with the training result of the second branch and iteratively learning the coupled result to update the training result of the first branch includes: using the site location as a control condition to perform feature learning on the radar forecast reflectivity to obtain the training result of the first branch; performing feature learning on the site observed precipitation to obtain the training result of the second branch; coupling the training result of the first branch with the training result of the second branch to obtain a coupled result for the first branch and the second branch; re-inputting the coupled result into the first branch, and repeating the above steps until the prediction model meets the training requirements for the fusion feature learning based on the radar reflectivity and the site observed precipitation.

[0053] For example, in an embodiment of the present invention, the site location can be input into a position encoding layer as a control condition, and the radar forecast reflectivity can be trained through a convolutional layer and a first branch of multiple fusion layers to obtain the training result of the first branch. Furthermore, in an embodiment of the present invention, the site observed precipitation is trained through a second branch of another convolutional layer to obtain the training result of the second branch. In this way, in an embodiment of the present invention, the output of the second branch training can be coupled with the output of each fusion layer in the first branch training respectively, and the coupled output result can be used as the input of the next fusion layer in the first branch to continue to perform further fusion learning on the coupled output result in the next fusion layer. After the next fusion layer outputs the next training result of the first branch, it is continuously coupled with the training result of the second branch, and so on iteratively, continuously fusing and learning the feature data of the two in the training.

[0054] Further, in order to better understand this preferred embodiment, the following will be described in conjunction with Figure 2 the network structure of the exemplary implementation prediction model shown. As Figure 2 shown, the prediction model is built based on AFNO as the main component, where Figure 2 the fusion layer includes multiple AFNO modules (AFNO block). Among them, the AFNO block includes Fourier transform (FFT), frequency fusion (multiply frequency), inverse Fourier transform (IFFT), and channel mixing. FFT converts data from the spatial domain to the frequency domain, multiply frequency is used for the fusion of different spatial features, IFFT converts data from the frequency domain to the spatial domain, and channel mixing is used for the fusion between different channel features.

[0055] The input layer consists of two parts. The first part includes time features (timedelta), radar reflectivity, longitude, latitude, and surface elevation data (lon, lat, dem). The second part is the station observed precipitation, which is input as another separate part. Figure 2 The middle convolutional layer (parch embedding) corresponding to the above input also consists of two parts. Among them, Figure 2 the convolutional layer on the left combines the time features, radar reflectivity, longitude, latitude, and surface elevation data with the station location encoding output by the position embedding layer as input, so as to be able to learn and train the radar reflectivity data based on the station location as a control condition. And ​ the convolutional layer on the right takes the station observed precipitation as input and processes the station observed precipitation as an independent branch separately. ​ There are multiple ANFO blocks in it. It performs feature fusion processing on the data output by the left convolutional layer in the first branch. At the same time, in the second branch, the output of the right convolutional layer is respectively coupled with the features output by each ANFO block in the middle of the network, and the result of the coupling process is used as the input of the next fusion layer, and so on for iterative processing. In addition, the coupling in the embodiments of the present invention can adopt summation processing.

[0056] After coupling the training results of multiple branches, the output of the last AFNO block is processed subsequently to obtain the precipitation forecast result.

[0057] It should be noted that the embodiments of the present invention are only described in combination with the network structure of a preferred prediction model, and the present invention is not limited thereto, and no excessive limitation is made in this regard.

[0058] It can be seen that on the one hand, the prediction model of the embodiments of the present invention can train and learn the features of radar reflectivity, and on the other hand, it iteratively couples the features of station observed precipitation on the basis of learning the radar forecast reflectivity, so that the trained prediction model can effectively fuse the station observed precipitation on the basis of radar reflectivity for prediction, reducing the precipitation error of radar extrapolation to a certain extent and improving the accuracy of precipitation forecast.

[0059] In the embodiments of the present invention, based on the network structure mainly composed of AFNO, combined with short-term precipitation data as input, there is no need for various parameterized complex calculations, the amount of calculation is small, and the error caused by relying only on radar extrapolation data is reduced.

[0060] In a preferred embodiment, for the short-term precipitation data, to establish a data set for training the prediction model, it includes: spatiotemporally matching the radar forecast reflectivity, the precipitation observed at the station, and the station location to obtain sample data of short-term precipitation; based on the data distribution characteristics of the sample data of short-term precipitation, performing a balancing operation on the sample data; and dividing the balanced sample data into a training set and a test set according to a preset ratio.

[0061] For example, embodiments of the present invention can obtain radar forecast reflectivity, precipitation data observed at national stations at adjacent times, and the corresponding station locations, and perform spatiotemporal matching on them to ensure the consistency of these data in time and space. Moreover, embodiments of the present invention can also cut these original data to produce sample data. Due to precipitation uncertainty and a large number of small precipitation data, the sample data will have the characteristics that the data with precipitation is less than the data without precipitation, and the data with large precipitation is less than the data with small precipitation, resulting in an unbalanced distribution of sample data. In view of this, embodiments of the present invention can also perform a balancing operation on the samples. Specifically, for example, samples with less than 100 precipitation observation points at the station or the sum of hourly radar reflectivities less than 1000 can be removed to balance the distribution of sample data. It should be noted that other methods can also be adopted for the sample balancing operation according to actual needs, and the present invention does not limit this too much.

[0062] In addition, it should also be noted that since the precipitation data observed at the station may have missing measurements, abnormal values, etc. due to reasons such as rain gauge failures, etc., before mapping the precipitation data, it is necessary to first perform outlier detection and removal to ensure the normality of the finally output prediction results.

[0063] In addition, to test the performance and generalization performance of the model, embodiments of the present invention also divide the sample data into a training set and a test set according to a ratio of, for example, 4:1. The sample data in the training set is used for model training and learning, and the sample data in the test set is used to verify the prediction performance of the model, etc.

[0064] In the embodiments of the present invention, establishing a data set based on radar short-term forecast reflectivity data and precipitation data observed at the station can enable the model to more effectively integrate the precipitation data observed at the station for short-term precipitation forecasting. At the same time, balancing the sample data can improve the accuracy of model prediction and prevent model bias.

[0065] In a preferred embodiment, the forecasting method further includes: comparing the forecast result output by the prediction model with the real result, calculating the corresponding loss value and / or calculating the score value of the forecasting performance of the prediction model for short-term precipitation by using an evaluation model; and optimizing the prediction model based on the loss value and / or the score value.

[0066] For example, during the model training process, embodiments of the present invention can measure the difference between the output of the prediction model and the true result based on a preset loss function. For example, the loss function can be a weighted mean squared error (MSE), which can effectively increase the loss weight for large rainfall amounts. Furthermore, embodiments of the present invention can select a suitable optimizer to update the model parameters according to the model architecture and the loss function. On the other hand, after the model is trained, in order to view the performance and generalization of the model, embodiments of the present invention can also evaluate the predicted results of the model output using, for example, TS and POD evaluation models. For example, taking the TS evaluation model as an example, embodiments of the present invention can calculate the TS scores for different rainfall magnitudes and can view the texture clarity of the rainfall amount plan view. Embodiments of the present invention compare the TS scores for different forecasting methods at different rainfall levels, as shown in the following table:

[0067]

[0068] Among them, in the above table, ZR refers to the precipitation forecasting method based on the ZR relationship, Resnet 18 (with no station precip) refers to the precipitation forecasting method based on the Resnet 18 network model (without adding station observed precipitation), AFNONet (no station skip) refers to the precipitation prediction method based on the AFNO network model (without adding station observed precipitation), and AFNONet (with station skip) refers to the precipitation prediction method based on the AFNO network model (adding station observed precipitation), that is, the prediction method adopted by the present invention. It can be seen from the above table that the method adopted by the present invention is relatively prominent in the scoring performance of the precipitation prediction for the next one hour and two hours at precipitation levels of 0.1 mm, 2 mm, 5 mm, 10 mm, and 20 mm. And in terms of the overall precipitation prediction score, the present invention has a higher overall score and better effect compared with other methods.

[0069] Therefore, it can be seen from the comparison of the TS scores that by adding station observed precipitation data as a control condition, the present invention can more effectively integrate station observed precipitation data for hourly precipitation forecasting at future times.

[0070] ​ It is a comparison plan view of rainfall amounts ​On the left side is the real data of precipitation observed at the station, and on the right side is the predicted precipitation result based on the prediction model. It can be seen that the predicted precipitation area and intensity are very close to the actual precipitation. Furthermore, in the embodiments of the present invention, by calculating the loss and various scores of the model, the performance of the model is analyzed, and the advantages and disadvantages of the model are identified, so that operations such as adjusting the model network architecture, increasing or decreasing the number of layers, changing the activation function, and adjusting the learning rate can be performed to optimize the model.

[0071] In the embodiments of the present invention, by calculating and evaluating the loss function for the output of the prediction model, the performance of the prediction model can be effectively analyzed, the model can be optimized in a timely manner, so as to improve the performance of the prediction model, and further improve the accuracy of the prediction model for short-term and impending precipitation forecasting.

[0072] In a preferred embodiment, the prediction method further includes: the forecasting method further includes: cutting the entire area to be forecast into multiple local areas, and recording the position of each local area in the entire area; inputting the short-term and impending precipitation data of each local area into the prediction model to obtain the forecasting result for each local area; and merging the forecasting results of each local area according to their positions in the entire area to obtain the forecasting result for the entire area.

[0073] For example, considering that the model has a fixed requirement for the size of the input image data, in the embodiments of the present invention, before inputting the data into the prediction model, the large image of the entire area needs to be cut into multiple small images of local areas first, so that each small image matches the input requirements of the prediction model. Then, the cut small images are input into the prediction model, and the prediction results of the small images are re-stitched into the large image of the entire area. Among them, when cutting and stitching before and after inputting the model, in order to ensure the continuity of the result image, the embodiments of the present invention can adopt a cross-cutting method, which not only retains the spatial continuity between adjacent two small images, makes the results obtained by the model more continuous in space, but also has a higher accuracy. In addition, the cut small images can be input into the prediction model in parallel, so that the prediction model can process the data of multiple small images in parallel and output the prediction results of the small images in parallel.

[0074] In the embodiments of the present invention, by cutting the data to be input into the prediction model, not only can it better meet the input requirements of the prediction model, but also can more effectively utilize the computing power of the prediction model, improve the prediction efficiency, and ensure the integrity and accuracy of the prediction results.

[0075] After the model training is completed, the embodiments of the present invention save the model, so as to obtain a prediction model suitable for short-term and impending precipitation prediction. Furthermore, the embodiments of the present invention can perform precipitation forecasting based on real-time data. The following gives a short-term and impending precipitation forecasting process, and the specific process is as follows:

[0076] 1) Obtain radar reflectivity data at the predicted future time according to the current time.

[0077] 2) Obtain precipitation data observed at nearby stations according to the current time, perform quality control on the precipitation data, and map it to a grid matching the radar reflectivity data.

[0078] 3) Obtain surface elevation data and longitude and latitude data, and preprocess the obtained data.

[0079] 3) Cut the large national map so that the cut small maps match the input size of the model, and record the position of each small map in the large map.

[0080] 4) Jointly input the station observation quantity data, radar forecast reflectivity data, surface elevation data, longitude and latitude data, etc. into the trained prediction model to perform precipitation forecasting for the future time.

[0081] 5) Reassemble the output results of the prediction model back into the large map according to the position of the small map in the large map.

[0082] 6) Save the result data and pictures, and publish the pictures to the web page.

[0083] In summary, the short-term and imminent precipitation forecasting method provided by the present invention has the following advantages:

[0084] 1) Introduce station precipitation control conditions to build a prediction model, train the station precipitation as a separate network branch, so that the prediction model can more effectively integrate station observed precipitation data for hourly precipitation forecasting at the expected time;

[0085] 2) Compared with the traditional precipitation forecasting based on the ZR relationship, the present invention performs short-term and imminent precipitation forecasting by fusing station precipitation data with extrapolated radar forecast reflectivity data, which has stronger applicability, reduces the precipitation forecasting error caused by radar extrapolation, and improves the precipitation forecasting effect;

[0086] 3) By adding surface elevation and longitude and latitude data, the applicability of the model in different regions is further enhanced;

[0087] 4) By using the method of cross-cutting the large map, the spatial continuity between adjacent two small maps is retained, making the prediction results output by the prediction model more continuous in space and having higher accuracy at the same time.

[0088] Based on the same inventive concept as the above method, an embodiment of the present invention provides a short-term and imminent precipitation forecasting device, as ​As shown, the forecasting device 400 includes: The short-term and imminent precipitation forecasting device includes: A building unit 410, configured to build a prediction model with the short-term and imminent precipitation at stations as the learning objective; A establishing unit 420, configured to obtain short-term and imminent precipitation data and establish a data set for training the prediction model, where the short-term and imminent precipitation data includes radar forecast reflectivity, station observed precipitation, and station location; A predicting unit 430, configured to input the data set and time interval feature data into the prediction model, and perform model training on the prediction model based on different learning branches, so as to output a forecast result of short-term and imminent precipitation for the expected time based on the prediction model, where the learning branches include a first branch for learning the radar forecast reflectivity controlled based on the station location and a second branch for learning the station observed precipitation.

[0089] For the specific implementation content and advantages of the weather forecasting device of the present invention, reference can be made to the embodiments of the above weather forecasting method, and details are not described herein again.

[0090] Correspondingly, an embodiment of the present invention further provides an electronic device, such as ​ As shown, it shows a schematic structural diagram of an electronic device suitable for implementing the embodiments of the present invention. The electronic device in the embodiments of the present invention may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), PMPs (Portable Multimedia Players), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. ​ The electronic device shown is only an example and should not impose any limitation on the functions and usage scope of the embodiments of the present invention.

[0091] Such as ​ As shown, the electronic device may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device are also stored. The processing device 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0092] Typically, the following devices can be connected to the I / O interface 505: input devices 506 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 508 including, for example, magnetic tapes, hard disks, etc.; and a communication device 509. The communication device 509 can allow the electronic device to communicate with other devices wirelessly or wireline to exchange data. Although ​ an electronic device with various devices is shown, it should be understood that it is not required to implement or have all the shown devices. Instead, more or fewer devices can be implemented or had.

[0093] Specifically, according to an embodiment of the present invention, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, an embodiment of the present invention provides a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above functions defined in the methods of the embodiments of the present invention are executed.

[0094] The above computer-readable medium can be included in the above electronic device; or can exist separately without being assembled into the electronic device.

[0095] The above computer-readable medium carries one or more programs, and when the one or more programs are executed by the electronic device, the electronic device is caused to: build a prediction model with short-term precipitation at a site as the learning target; obtain short-term precipitation data, establish a data set for training the prediction model, where the short-term precipitation data includes radar forecast reflectivity, site observed precipitation, and site location; input the data set and time feature data into the prediction model, and perform model training on the prediction model based on different learning branches, so as to output a forecast result of short-term precipitation for an expected time based on the prediction model, where the learning branches include a first branch for learning the radar forecast reflectivity controlled based on the site location and a second branch for learning the site observed precipitation.

[0096] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a portion of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that, in some alternative implementations, the functions noted in the blocks may occur in a different order than that noted in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or by a combination of dedicated hardware and computer instructions.

[0097] The units described in the embodiments of the present invention can be implemented in software or in hardware. Among them, the name of the unit does not constitute a limitation on the unit itself in some cases. For example, the first acquisition unit can also be described as "the unit for acquiring at least two Internet protocol addresses".

[0098] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that can be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), and so on.

[0099] In the context of the present invention, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0100] The above description is only a preferred embodiment of the present invention and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of disclosure involved in the present invention is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosure concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the present invention.

[0101] In addition, although the operations are depicted in a particular order, this should not be construed as requiring that the operations be performed in the particular order shown or in sequential order. In certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the present invention. Certain features described in the context of separate embodiments may also be implemented in combination in a single embodiment. Conversely, the various features described in the context of a single embodiment may also be implemented separately or in any suitable sub-combination in multiple embodiments.

[0102] Although the subject matter has been described in language specific to structural features and / or methodological logical acts, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. On the contrary, the specific features and acts described above are merely example forms for implementing the claims.

Claims

1. A short-term and impending precipitation forecasting method, characterized in that, The prediction method includes: Building a prediction model with short-term and nowcasting precipitation at stations as the learning target; Obtaining short-term and nowcasting precipitation data, and establishing a data set for training the prediction model, where the short-term and nowcasting precipitation data includes radar forecast reflectivity, station observed precipitation, and station location; Inputting the data set and time feature data into the prediction model, and performing model training on the prediction model based on different learning branches, so as to output a forecast result of short-term and nowcasting precipitation for the expected time based on the prediction model, where the learning branches include a first branch for learning the radar forecast reflectivity controlled by the station location and a second branch for learning the station observed precipitation.

2. The prediction method according to claim 1, wherein The short-term and nowcasting precipitation data further includes surface elevation data and longitude and latitude data.

3. The prediction method according to claim 1, characterized in that, The prediction model includes an input layer, a position encoding layer, a convolutional layer, a fusion layer, a mapping layer, and an output layer, where The input layer is used to input the short-term and nowcasting precipitation data; The position encoding layer is used to perform position encoding on the input features and learn the features corresponding to the input feature positions; The convolutional layer is used to perform a convolutional operation on the short-term and nowcasting precipitation data and map it to a latent space; The fusion layer is used to transform and fuse the features of the short-term and nowcasting precipitation data; The mapping layer is used to map the latent space data output by the fusion layer to the data size of the original input; The output layer is used to output a short-term and nowcasting precipitation forecast result at a preset time interval.

4. The prediction method according to claim 1, wherein Establishing a data set for training the prediction model from the short-term and nowcasting precipitation data includes: Spatiotemporally matching the radar forecast reflectivity, the station observed precipitation, and the station location to obtain sample data of short-term and nowcasting precipitation; Performing a balancing operation on the sample data based on the data distribution characteristics of the sample data of short-term and nowcasting precipitation; Dividing the balanced sample data into a training set and a test set according to a preset ratio.

5. The prediction method according to claim 1, wherein The model training on the prediction model based on different learning branches includes: Coupling the training results of the first branch and the training results of the second branch, and iteratively learning the coupled results to train the prediction model to learn short-term and nowcasting precipitation forecasting based on the radar forecast reflectivity and the station observed precipitation.

6. The forecasting method according to claim 5, characterized in that: Coupling the training results of the first branch and the training results of the second branch, and iteratively learning the coupled results to update the training results of the first branch, including: Using the station location as a control condition to perform feature learning on the radar forecast reflectivity to obtain the training results of the first branch; Performing feature learning on the station observed precipitation to obtain the training results of the second branch; Coupling the training results of the first branch and the training results of the second branch to obtain a coupled result for the first branch and the second branch; Inputting the coupled result into the first branch again, and repeating the above steps until the prediction model satisfies the training requirements for learning the fusion features based on the radar reflectivity and the station observed precipitation.

7. The prediction method according to claim 1, wherein The prediction method further includes: Compare the forecast result output by the prediction model with the true result, calculate the corresponding loss value and / or calculate the score value of the prediction performance of the prediction model for short-term and imminent precipitation by using an evaluation model; Optimize the prediction model based on the loss value and / or score value.

8. The prediction method according to claim 1, characterized in that, The forecast method further includes: Cut the entire area to be forecast into multiple local areas, and record the position of each local area in the entire area; Input the short-term and imminent precipitation data of each local area into the prediction model to obtain the forecast result for each local area; Merge the forecast results of each local area according to their positions in the entire area to obtain the forecast result for the entire area.

9. A short-term and imminent precipitation forecasting device, characterized in that, The short-term and imminent precipitation forecasting device includes: A building unit for building a prediction model with the short-term and imminent precipitation at stations as the learning target; An establishment unit for obtaining short-term and imminent precipitation data and establishing a data set for training the prediction model, where the short-term and imminent precipitation data includes radar forecast reflectivity, station observed precipitation, and station location; A prediction unit for inputting the data set and time interval feature data into the prediction model, and performing model training on the prediction model based on different learning branches, so as to output the forecast result of short-term and imminent precipitation for the expected time based on the prediction model, where the learning branches include a first branch for learning the radar forecast reflectivity controlled by the station location and a second branch for learning the station observed precipitation.

10. An electronic device, comprising: At least one processor; And A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the short-term and imminent precipitation forecasting method according to any one of claims 1-8.

11. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to execute the short-term and imminent precipitation forecasting method according to any one of claims 1-8.

Images