Machine learning based near real-time satellite retrieval precipitation correction method, device and system

Abstract

This invention discloses a near-real-time satellite inversion precipitation correction method, apparatus, and system based on machine learning. The method includes: acquiring near-real-time satellite inversion precipitation data in real time; sending the near-real-time satellite inversion precipitation data to a pre-trained machine learning model to obtain precipitation correction data; the machine learning model is obtained through the following steps: acquiring a training set and a validation set pre-constructed based on historical near-real-time satellite inversion precipitation data and observed precipitation data; comprehensively considering regression accuracy, time complexity, and input data volume, selecting the optimal machine learning model from multiple candidate models for the current training set; and training the selected machine learning model using a preset algorithm based on the training set and validation set to obtain the pre-trained machine learning model. This invention employs machine optimization technology to extract features from indirect simulation and observational data, making the correction results more reliable.

Description

Technical Field

[0001] This invention belongs to the field of satellite precipitation data processing technology, specifically relating to a near real-time satellite inversion precipitation correction method, device, and system based on machine learning. Background Technology

[0002] Precipitation is a crucial component of the global water cycle system. Obtaining high spatiotemporal resolution precipitation information, especially near-real-time precipitation data, plays a vital role in addressing critical scientific issues concerning the safety of people's lives and property, such as runoff forecasting, flood warnings, and reservoir management.

[0003] Satellite precipitation measurement has the advantages of being unrestricted by the underlying surface, rapidly acquiring large-scale precipitation information, and having relatively high spatial distribution accuracy. The Fengyun-4 series satellites are a new generation of geostationary meteorological satellites independently developed by China. In 2016, Fengyun-4A (FY4A), carrying a geostationary radiation imager with internationally leading performance, was successfully launched and put into use.

[0004] Since 2018, the National Meteorological Center has been releasing official precipitation products retrieved from Fengyun satellites (FY4A-REGC), achieving initial success in the localization of near-real-time satellite precipitation measurement. However, due to limitations in sensor system errors, spatiotemporal sampling, and inversion algorithms, the measurement accuracy of real-time satellite precipitation products remains low. The accuracy of precipitation retrieved from Fengyun satellites still lags behind that of the international benchmark product IMERG-Early.

[0005] With the rapid development of machine learning algorithms, the correction methods for satellite-measured precipitation products have been profoundly affected, but there is still no data correction method for near real-time precipitation inversion by domestic satellites. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a near real-time satellite inversion precipitation correction method, device, and system based on machine learning. It employs machine optimization technology to extract features from indirect simulation and observation data, thereby making the correction results more reliable.

[0007] To achieve the above-mentioned technical objectives and effects, the present invention is implemented through the following technical solution:

[0008] In a first aspect, the present invention provides a near real-time satellite inversion precipitation correction method based on machine learning, comprising:

[0009] Real-time acquisition of near-real-time satellite inversion precipitation data;

[0010] The near-real-time satellite inversion precipitation data is sent to a pre-trained machine learning model to obtain precipitation correction data.

[0011] The machine learning model is obtained through the following steps:

[0012] Obtain a training set and a validation set pre-constructed based on historical near-real-time satellite inversion precipitation data and observed precipitation data;

[0013] Taking into account regression accuracy, time complexity, and input data volume, the optimal machine learning model for the current training set is selected from multiple candidate machine learning models.

[0014] Based on the training set and validation set, the selected machine learning model is trained using a preset algorithm to obtain a pre-trained machine learning model.

[0015] Optionally, the method for constructing the training set includes:

[0016] The precipitation events are selected from the first near-real-time satellite inversion precipitation data, the second near-real-time satellite inversion precipitation data, and the automatic station observation data to generate the first precipitation data, the second precipitation data, and the observed precipitation data.

[0017] The spatiotemporal resolution of the first and second precipitation data is resampled to match the observed precipitation data. Furthermore, a preset proportion of the first precipitation data is selected as input data, and the second precipitation data is used as calibration data to generate a training set.

[0018] Optionally, the first near-real-time satellite inversion precipitation data is the FY4A-REGC near-real-time satellite inversion precipitation data from the domestic satellite Fengyun-4A; the second near-real-time satellite inversion precipitation data is the IMERG-Early near-real-time satellite inversion precipitation data from the Global Precipitation Measurement Project (GPM).

[0019] Optionally, the step of training the selected machine learning model using a preset algorithm based on the training set and validation set to obtain a pre-trained machine learning model includes:

[0020] Using a pre-built training set as input, a pre-defined algorithm is used to train the selected machine learning model to obtain the first machine learning model.

[0021] The first machine learning model is validated using a pre-built validation set. If the prediction results of the first machine learning model meet the set conditions, the first machine learning model is used as the final machine learning model.

[0022] Optionally, the step of validating the first machine learning model using a pre-built validation set, and if the prediction result of the first machine learning model meets the set conditions, then using the first machine learning model as the final machine learning model, includes:

[0023] A first precipitation data of a preset proportion is selected as input data, and the observed precipitation data is used as the true value of the validation set to generate a validation set.

[0024] The first precipitation data at the preset ratio is input into the first machine learning model to obtain corrected data;

[0025] Based on the corrected data and observed precipitation data, a preset index value is calculated;

[0026] If the preset index values ​​meet the requirements, the first machine learning model will be used as the final machine learning model.

[0027] Optionally, the calculation formula for the preset index value includes:

[0028]

[0029]

[0030]

[0031] Where CC represents the correlation coefficient, RMSE represents the root mean square coefficient, Bias represents the relative error, n represents the total amount of the corrected first precipitation data, and G represents the observed precipitation data. S represents the mean of observed precipitation data; i This represents the i-th value in the first set of corrected precipitation data. This represents the mean of the first precipitation data after correction.

[0032] Optionally, the alternative machine learning models include the XGBoost model, the LightGBM model, and the RandomForest model;

[0033] When selecting the optimal machine learning model for the current training set from multiple candidate machine learning models, each machine learning model uses default parameters. Once the optimal machine learning model is selected, the best parameters for that machine learning model are obtained through a grid search method.

[0034] Optionally, considering regression accuracy, time complexity, and input data volume, the optimal machine learning model for the current training set is selected from multiple candidate machine learning models, including:

[0035] Calculate the regression accuracy and time required for each candidate machine learning model under different input data volumes;

[0036] Based on the regression accuracy and time required under different input data volumes, the optimal machine learning model is selected.

[0037] Secondly, the present invention provides a near real-time satellite inversion precipitation correction device based on machine learning, comprising:

[0038] The acquisition module is used to acquire near-real-time satellite inversion precipitation data in real time.

[0039] The correction module is used to send the near real-time satellite inversion precipitation data to a pre-trained machine learning model to obtain precipitation correction data;

[0040] The machine learning model is obtained through the following steps:

[0041] Obtain a training set and a validation set pre-constructed based on historical near-real-time satellite inversion precipitation data and observed precipitation data;

[0042] Taking into account regression accuracy, time complexity, and input data volume, the optimal machine learning model for the current training set is selected from multiple candidate machine learning models.

[0043] Based on the training set and validation set, the selected machine learning model is trained using a preset algorithm to obtain a pre-trained machine learning model.

[0044] Thirdly, the present invention provides a near real-time satellite inversion precipitation correction system based on machine learning, including a storage medium and a processor;

[0045] The storage medium is used to store instructions;

[0046] The processor is configured to operate according to the instructions to perform the method according to any one of the first aspects.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] This invention comprehensively considers multiple machine learning models, selecting the optimal model based on a combination of regression accuracy, time complexity, and input data volume. This optimal model is then trained to obtain the best parameter model, rather than using a fixed machine learning model. System evaluation shows that the method described in this invention can effectively and quickly improve the precipitation estimation quality of FY4A-REGC, achieving accuracy close to that of IMERG-Early. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein:

[0050] Figure 1 This is a schematic diagram of the basic process of the precipitation correction method in one embodiment of the present invention;

[0051] Figure 2(a) is a scatter plot of the relationship between FY4A-Adj and CMPA automatic station data in one embodiment of the present invention;

[0052] Figure 2(b) is a scatter plot of the relationship between FY4A-REGC and CMPA automatic station data in one embodiment of the present invention;

[0053] Figure 2(c) is a scatter plot of the relationship between IMERG-Early and CMPA automatic station data in one embodiment of the present invention;

[0054] Figure 3 This is a heatmap showing the regression accuracy and time required to train a machine learning model after running input data of various data volumes for each model in one embodiment of the present invention.

[0055] Figure 4 This refers to the regression accuracy (RMSE) calculated from the training and validation truth values ​​of each dataset in one embodiment of the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0057] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0058] Example 1

[0059] This invention provides a near real-time satellite inversion precipitation correction method based on machine learning, comprising the following steps:

[0060] (1) Real-time acquisition of near-real-time satellite inversion precipitation data;

[0061] (2) Send the near-real-time satellite inversion precipitation data to a pre-trained machine learning model to obtain precipitation correction data;

[0062] The machine learning model is obtained through the following steps:

[0063] Obtain a training set and a validation set pre-constructed based on historical near-real-time satellite inversion precipitation data and observed precipitation data;

[0064] Taking into account regression accuracy, time complexity, and input data volume, the optimal machine learning model for the current training set is selected from multiple candidate machine learning models.

[0065] Based on the training set and validation set, the selected machine learning model is trained using a preset algorithm to obtain a pre-trained machine learning model.

[0066] This invention comprehensively considers multiple machine learning models, selects the best machine learning model based on regression accuracy, time complexity, and input data volume, and trains the best machine learning model to obtain the optimal parameter model, rather than using a fixed machine learning model.

[0067] In one specific embodiment of the present invention, the method for constructing the training set includes:

[0068] The precipitation events are selected from the first near-real-time satellite inversion precipitation data, the second near-real-time satellite inversion precipitation data, and the automatic station observation data to generate the first precipitation data, the second precipitation data, and the observed precipitation data.

[0069] The spatiotemporal resolution of the first and second precipitation data is resampled to match the observed precipitation data. Furthermore, a preset proportion of the first precipitation data is selected as input data, and the second precipitation data is used as calibration data to generate a training set.

[0070] In one specific embodiment of the present invention, the first near-real-time satellite inversion precipitation data is the FY4A-REGC near-real-time satellite inversion precipitation data of the domestic satellite Fengyun-4A; the second near-real-time satellite inversion precipitation data is the IMERG-Early near-real-time satellite inversion precipitation data of the Global Precipitation Measurement Project (GPM).

[0071] In one specific embodiment of the present invention, the step of training the selected machine learning model using a preset algorithm based on the training set and validation set to obtain a pre-trained machine learning model includes:

[0072] Using a pre-built training set as input, a pre-defined algorithm is used to train the selected machine learning model to obtain the first machine learning model.

[0073] The first machine learning model is validated using a pre-built validation set. If the prediction results of the first machine learning model meet the set conditions, the first machine learning model is used as the final machine learning model.

[0074] Specifically, the step of validating the first machine learning model using a pre-built validation set, and if the prediction error of the first machine learning model meets a preset threshold, then using the first machine learning model as the final machine learning model, includes:

[0075] A first precipitation data of a preset proportion is selected as input data, and the observed precipitation data is used as the true value of the validation set to generate a validation set.

[0076] The first precipitation data at the preset ratio is input into the first machine learning model to obtain corrected data;

[0077] Based on the corrected data and observed precipitation data, a preset index value is calculated;

[0078] If the preset index values ​​meet the requirements, the first machine learning model will be used as the final machine learning model.

[0079] In one specific embodiment of the present invention, the calculation formula for the preset index value includes:

[0080]

[0081]

[0082]

[0083] Where CC represents the correlation coefficient, RMSE represents the root mean square coefficient, Bias represents the relative error, n represents the total amount of the corrected first precipitation data, and G represents the observed precipitation data. S represents the mean of observed precipitation data; i This represents the i-th value in the first set of corrected precipitation data. This represents the mean of the first precipitation data after correction.

[0084] In one specific embodiment of this invention, the candidate machine learning models include the XGBoost model, the LightGBM model, and the Random Forest model. When selecting the optimal machine learning model for the current training set from multiple candidate models, each machine learning model uses default parameters. Once the optimal machine learning model is selected, its optimal parameters are obtained through a grid optimization method. Different data and different spatiotemporal conditions have different characteristics, making it necessary to select a machine learning model more suitable for the corresponding data characteristics. The grid parameter tuning method in this invention requires a large number of calculations. For models with high computational time complexity, finding the optimal parameters takes a very long time, which is difficult to meet in reality. The hyperparameter selection in this invention is an adaptation of data-driven models to data characteristics, which may involve coincidence; therefore, default parameters are generally used to judge model performance. Preferably, during the experiment, it was found that the quality of the near-real-time inversion precipitation data FY4A-REGC from the domestic satellite Fengyun-4A changes over time and improves. Therefore, in the specific implementation, the LightGBM model, with its faster computation speed, is selected as the optimal machine learning model. Experiments have shown that the machine learning model performs best when the amount of input data in the training set is 2 to the power of 23. Therefore, whenever new input data is acquired, the old data is replaced with the new input data to keep the amount of input data constant at 2 to the power of 23. In a specific embodiment of this invention, considering regression accuracy, time complexity, and the amount of input data, the optimal machine learning model for the current training set is selected from multiple candidate machine learning models, including:

[0085] The regression accuracy and time required for each candidate machine learning model under different input data volumes were calculated. During the experiment, it was found that the machine learning model performed best when the data volume was kept to the power of 2^23.

[0086] Based on the regression accuracy and time required under different input data volumes, the optimal machine learning model is selected.

[0087] FY4A-REGC is a near-real-time precipitation retrieval product, and its timeliness must be considered during calibration. Therefore, this invention requires selecting a model that can still balance runtime and calibration accuracy under conditions of progressively increasing data volume. Figure 3 The heatmaps present the regression accuracy and training time required for each machine learning model after running on input data of various magnitudes (the running platforms are shown in Table 1).

[0088] Table 1

[0089]

[0090] As the amount of training input data increases, the training time for all models increases linearly, but the training accuracy depends on the selected data segment itself and is not significantly affected by the order of magnitude of the training input. While there is no significant difference in training accuracy among the three models, LightGBM requires significantly less training time than the other two. In terms of training accuracy, before the training data volume increases to 2^23, LightGBM's accuracy is slightly better than XGBoost and Random Forest. XGBoost and LightGBM, however, slightly outperform Random Forest when given 2^23 data input. In terms of training time, although all three achieve similar training results, XGBoost requires approximately 5 times the training time of LightGBM, while Random Forest requires approximately 25 times the training time. After the data volume increases, the training time further increases to about 40 times that of LightGBM.

[0091] Therefore, LightGBM was selected as the main method for rapidly correcting the near-real-time precipitation inversion product FY4A-REGC.

[0092] Figure 4 The RMSE calculated from the training and validation ground truth values ​​for each dataset is presented to avoid overfitting (i.e., the model overfits the data, resulting in weakened generalization ability) during training. Therefore, this invention focuses on the FY4A-Adj dataset (FY4A near real-time correction product) generated by inputting validation set data (not used in training) into a LightGBM model tuned via grid search. The parameters adjusted using the grid search method are shown in Table 2. Different models were generated for different orders of magnitude of data input to determine the appropriate sequence length for the application.

[0093] Table 2

[0094]

[0095] The method in the embodiments of the present invention will be described in detail below with reference to a specific implementation method.

[0096] The machine learning-based near-real-time satellite inversion precipitation correction method for domestically produced satellites includes the following steps:

[0097] Step 1, Data Acquisition:

[0098] Acquire near-real-time satellite inversion precipitation data FY4A-REGC from the domestic satellite Fengyun-4A, near-real-time satellite inversion precipitation data IMERG-Early from the Global Precipitation Measurement Project (GPM), and automatic station observation data CMPA. Extract the portion of the above data that has experienced precipitation for use in subsequent steps two to four.

[0099] The precipitation data retrieved from the domestically produced Fengyun-4A near-real-time satellite (FY4A-REGC) and the precipitation data retrieved from the Global Precipitation Measurement Project (GPM) near-real-time satellite (IMERG-Early) are used as input data for model training. The automatic weather station observation data (CMPA) includes observation methods such as satellite remote sensing inversion and rain gauge observations, with a time span including but not limited to 2018 to 2019.

[0100] Step 2, Data Preprocessing:

[0101] The spatiotemporal resolution of FY4A-REGC and IMERG-Early was resampled to 0.1° / 1 hour to match the automatic station observation data.

[0102] Step 3, Dataset Construction:

[0103] 80% of FY4A-REGC was selected as the input training set, and IMERG-Early was used as the training dataset for calibration; the remaining 20% ​​of FY4A-REGC was used as the input validation set, and automatic station observation data was used as the validation ground truth for the validation set.

[0104] The method for dividing the training dataset and the test dataset is specifically designed for this invention. The information characteristics obtained by satellite inversion precipitation, automatic station observation and other means are different. Therefore, it is necessary to divide the training dataset and the test dataset according to the data characteristics.

[0105] For example, satellite-retrieved precipitation information is relatively accurate and has strong temporal and spatial coverage, making it suitable as a basic training dataset; automatic station observation data is the most accurate, but the amount of data is small and it is point information, making it suitable as a dataset for optimization and verification.

[0106] Step 4: Pre-select three machine learning models—XGBoost, LightGBM, and Random Forest—and comprehensively consider the relationship between regression accuracy, time complexity, and input data volume to obtain the optimal machine learning model for the current task with default parameters. After determining the optimal machine learning model, further optimize the model's parameters using a grid search method to obtain the model and its corresponding parameters.

[0107] The machine learning methods used are all improvements on decision tree models. The optimal machine learning model for the current task is determined by considering three main factors: regression accuracy, time complexity, and input data volume. The machine learning methods used include, but are not limited to, XGBoost, LightGBM, and Random Forest. Default parameters are used when comparing models, and once the optimal model is selected, its best parameters are obtained through grid search.

[0108] Step 5, Model Run:

[0109] Input the validation set into the model obtained in the previous process, and output the correction dataset FY4A-Adj (spatial-temporal resolution 1 hour / 0.1°).

[0110] Step 6, Result Verification: The effect is evaluated using indicators such as correlation coefficient (CC), root mean square coefficient (RMSE), and relative error (Bias).

[0111] The preset index value calculation formula includes:

[0112]

[0113]

[0114]

[0115] Where CC represents the correlation coefficient, RMSE represents the root mean square coefficient, Bias represents the relative error, n represents the total amount of the corrected first precipitation data, and G represents the observed precipitation data. S represents the mean of observed precipitation data; i This represents the i-th value in the first set of corrected precipitation data. This represents the mean of the first precipitation data after correction.

[0116] Evaluation results as follows Figures 2(a)-2(c) As shown:

[0117] Figures 2(a)-2(c) Scatter plots of the relationships between FY4A-Adj, FY4A-REGC, IMERG-Early, and CMPA automatic weather station data are presented. Figures 2(b) and 2(c) show that during the study period, the precipitation estimated by IMERG-Early and FY4A-REGC is mainly concentrated in the range of 0–5 mm, and can be relatively evenly distributed near the 45° line. Figure 2(a) shows the scatter plot of the relationships between FY4A-Adj and CMPA automatic weather station data. It can be found that after correction, more data points are concentrated on the 45° line (especially in the range of 0–2 mm precipitation). This indicates that the method in the embodiments of the present invention is very effective in correcting FY4A-REGC.

[0118] Example 2

[0119] Based on the same inventive concept as Embodiment 1, the present invention provides a near real-time satellite inversion precipitation correction device based on machine learning, comprising:

[0120] The acquisition module is used to acquire near-real-time satellite inversion precipitation data in real time.

[0121] The correction module is used to send the near real-time satellite inversion precipitation data to a pre-trained machine learning model to obtain precipitation correction data;

[0122] The machine learning model is obtained through the following steps:

[0123] Obtain training and validation sets based on historical near-real-time satellite inversion precipitation data and observed precipitation data.

[0124] Taking into account regression accuracy, time complexity, and input data volume, the optimal machine learning model for the current training set is selected from multiple candidate machine learning models.

[0125] Based on the training set and validation set, the selected machine learning model is trained using a preset algorithm to obtain a pre-trained machine learning model.

[0126] The rest are the same as in Example 1.

[0127] Example 3

[0128] This invention provides a near real-time satellite inversion precipitation correction system based on machine learning, including a storage medium and a processor;

[0129] The storage medium is used to store instructions;

[0130] The processor is configured to operate according to the instructions to execute the method according to any one of Embodiment 1.

[0131] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0132] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0133] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0134] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0135] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

[0136] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A near real-time satellite inversion precipitation correction method based on machine learning, characterized in that, include: Real-time acquisition of near-real-time satellite inversion precipitation data; The near-real-time satellite inversion precipitation data is sent to a pre-trained machine learning model to obtain precipitation correction data. The machine learning model is obtained through the following steps: Obtain a training set and a validation set pre-constructed based on historical near-real-time satellite inversion precipitation data and observed precipitation data; Taking into account regression accuracy, time complexity, and input data volume, the optimal machine learning model for the current training set is selected from multiple candidate machine learning models. Based on the training set and validation set, the selected machine learning model is trained using a preset algorithm to obtain a pre-trained machine learning model. The methods for constructing the training set include: The precipitation events are selected from the first near-real-time satellite inversion precipitation data, the second near-real-time satellite inversion precipitation data, and the automatic station observation data to generate the first precipitation data, the second precipitation data, and the observed precipitation data. The spatiotemporal resolution of the first precipitation data and the second precipitation data is resampled to match the observed precipitation data. Furthermore, a preset proportion of the first precipitation data is selected as input data, and the second precipitation data is used as calibration data to generate a training set. Taking into account regression accuracy, time complexity, and input data volume, the optimal machine learning model for the current training set is selected from multiple candidate models, including: Calculate the regression accuracy and time required for each candidate machine learning model under different input data volumes; based on the regression accuracy and time required under different input data volumes, select the optimal machine learning model.

2. The near real-time satellite inversion precipitation correction method based on machine learning according to claim 1, characterized in that: The first near-real-time satellite inversion precipitation data is the FY4A-REGC near-real-time satellite inversion precipitation data from the domestic satellite Fengyun-4A; the second near-real-time satellite inversion precipitation data is the IMERG-Early near-real-time satellite inversion precipitation data from the Global Precipitation Measurement Project (GPM).

3. The near real-time satellite inversion precipitation correction method based on machine learning according to claim 1, characterized in that: The step of training the selected machine learning model using a preset algorithm based on the training set and validation set to obtain a pre-trained machine learning model includes: Using a pre-built training set as input, a pre-defined algorithm is used to train the selected machine learning model to obtain the first machine learning model. The first machine learning model is validated using a pre-built validation set. If the prediction results of the first machine learning model meet the set conditions, the first machine learning model is used as the final machine learning model.

4. The near real-time satellite inversion precipitation correction method based on machine learning according to claim 3, characterized in that: The step of validating the first machine learning model using a pre-built validation set, and if the prediction result of the first machine learning model meets the set conditions, then using the first machine learning model as the final machine learning model, includes: A first precipitation data of a preset proportion is selected as input data, and the observed precipitation data is used as the true value of the validation set to generate a validation set. The first precipitation data at the preset ratio is input into the first machine learning model to obtain corrected data; Based on the corrected data and observed precipitation data, a preset index value is calculated; If the preset index values ​​meet the requirements, the first machine learning model will be used as the final machine learning model.

5. The near real-time satellite inversion precipitation correction method based on machine learning according to claim 4, characterized in that: The preset index value calculation formula includes: Where CC represents the correlation coefficient, RMSE represents the root mean square coefficient, Bias represents the relative error, and n represents the total amount of the corrected first precipitation data; G i This represents the i-th value of the observed precipitation data. This represents the mean of the observed precipitation data; S i This represents the i-th value in the first corrected precipitation data. This represents the mean of the first precipitation data after correction.

6. The near real-time satellite inversion precipitation correction method based on machine learning according to claim 1, characterized in that: The alternative machine learning models include the XGBoost model, the LightGBM model, and the RandomForest model; When selecting the optimal machine learning model for the current training set from multiple candidate machine learning models, each machine learning model uses default parameters. Once the optimal machine learning model is selected, the best parameters for that machine learning model are obtained through a grid search method.

7. A near real-time satellite inversion precipitation correction device based on machine learning, characterized in that, include: The acquisition module is used to acquire near-real-time satellite inversion precipitation data in real time. The correction module is used to send the near real-time satellite inversion precipitation data to a pre-trained machine learning model to obtain precipitation correction data; The machine learning model is obtained through the following steps: Obtain a training set and a validation set pre-constructed based on historical near-real-time satellite inversion precipitation data and observed precipitation data; Taking into account regression accuracy, time complexity, and input data volume, the optimal machine learning model for the current training set is selected from multiple candidate machine learning models. Based on the training set and validation set, the selected machine learning model is trained using a preset algorithm to obtain a pre-trained machine learning model. The methods for constructing the training set include: The precipitation events are selected from the first near-real-time satellite inversion precipitation data, the second near-real-time satellite inversion precipitation data, and the automatic station observation data to generate the first precipitation data, the second precipitation data, and the observed precipitation data. The spatiotemporal resolution of the first precipitation data and the second precipitation data is resampled to match the observed precipitation data. Furthermore, a preset proportion of the first precipitation data is selected as input data, and the second precipitation data is used as calibration data to generate a training set. Taking into account regression accuracy, time complexity, and input data volume, the optimal machine learning model for the current training set is selected from multiple candidate models, including: Calculate the regression accuracy and time required for each candidate machine learning model under different input data volumes; based on the regression accuracy and time required under different input data volumes, select the optimal machine learning model.

8. A near real-time satellite inversion precipitation correction system based on machine learning, characterized in that, Including storage media and processor; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the method according to any one of claims 1-6.

Images