A method, device, equipment and medium for revising precipitation product monitoring values

Abstract

The application relates to the technical field of data fusion, and provides a precipitation product monitoring value revision method and device, equipment and a medium, the method comprising the following steps: acquiring a plurality of initial precipitation monitoring values corresponding to a plurality of time periods output by a target precipitation product, and an observation station coordinate observation precipitation value; fitting the plurality of initial precipitation monitoring values, and performing first iteration training according to the fitted initial precipitation monitoring values to obtain an initial feature vector and a coordinate bias value; fitting the observation station coordinate observation precipitation value according to the initial feature vector and the coordinate bias value, and performing second iteration training according to the fitted observation station coordinate observation precipitation value to obtain a target feature vector and a target transfer coefficient; and obtaining a revised precipitation monitoring value according to the target feature vector, the target transfer coefficient and the coordinate bias value. In the application, the observation station coordinate observation precipitation value can improve the precipitation monitoring precision of each grid of the precipitation product.

Description

Technical Field

[0001] This application relates to the field of data fusion technology, specifically to a method, apparatus, equipment, and medium for correcting precipitation product monitoring values. Background Technology

[0002] With the advancement of global warming and data analysis technologies, precipitation products are used to monitor and predict the spatiotemporal characteristics of precipitation, such as intensity and frequency. Fusion precipitation products, such as CMPAS (China Meteorological Administration Multi-Source Precipitation Product) and CMORPH (Climate Prediction Center Morphing technique, CPC deformation technology, a global high spatiotemporal resolution precipitation data developed based on various microwave and infrared data), collect precipitation data from ground observation stations, radar, and satellites to produce real-time precipitation products gridded with geographic latitude and longitude coordinates. These products provide comprehensive precipitation monitoring data and reliable data support for precipitation-related tasks (such as flood monitoring).

[0003] However, numerous studies have shown that the accuracy of precipitation monitoring values ​​in fused precipitation products is somewhat flawed compared to the most reliable station precipitation data. Furthermore, station precipitation data is limited by station deployment conditions and cannot provide comprehensive regional coverage. Therefore, corrections based on station precipitation data are necessary for fused precipitation products. Currently, the widely used correction method in meteorology, based on probability density matching (PDF), has limited effectiveness in improving fused precipitation products and cannot precisely enhance the accuracy of precipitation monitoring values ​​in each grid.

[0004] Therefore, how to improve the accuracy of precipitation monitoring values ​​in each grid of precipitation products is an urgent problem to be solved. Summary of the Invention

[0005] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a method, apparatus, equipment and medium for correcting precipitation product monitoring values, so as to solve the problem of how to improve the precipitation monitoring accuracy of each grid of precipitation products in the prior art.

[0006] To achieve the above and other related objectives, this application provides a method for correcting precipitation product monitoring values, the method comprising:

[0007] Acquire multiple initial precipitation monitoring values ​​corresponding to multiple time periods of the target precipitation product output, as well as the precipitation values ​​observed at the coordinates of the observation station;

[0008] The multiple initial precipitation monitoring values ​​are fitted, and the first iteration training is performed based on the fitted initial precipitation monitoring values ​​to obtain the initial representation vector and coordinate bias value;

[0009] Based on the initial representation vector and coordinate bias value, the observed precipitation value at the observation station coordinates is fitted, and a second iteration training is performed based on the fitted observed precipitation value at the observation station coordinates to obtain the target representation vector and target migration coefficient.

[0010] Based on the target representation vector, target migration coefficient, and coordinate offset value, the corrected precipitation monitoring value is obtained.

[0011] In one embodiment of this application, the initial precipitation monitoring value includes longitude information and latitude information.

[0012] In one embodiment of this application, fitting the plurality of initial precipitation monitoring values ​​includes:

[0013] The multiple initial precipitation monitoring values ​​are fitted into the sum of the dot product of the longitude representation vector and the latitude representation vector and the coordinate offset value;

[0014] The methods for representing the fitted initial precipitation monitoring values ​​include:

[0015] P ij ≈lat i ·lon j +bias ij

[0016] Among them, P ij For each of the initial precipitation monitoring values ​​after fitting, lat i Let lon be the representation vector of the latitude. j Let bias be the representation vector of the longitude. ij The coordinate offset value is given.

[0017] In one embodiment of this application, the first iteration training is performed based on the fitted initial precipitation monitoring values, including:

[0018] By using stochastic gradient descent, the longitude representation vector, latitude representation vector, and coordinate offset value are iteratively trained based on the fitted initial precipitation monitoring values.

[0019] The representation methods for iteratively training the representation vector of the aforementioned dimension include:

[0020] lat i ←lat i +λ lat ·(P ij -lat i ·lon j -bias ij )·lon j

[0021] Where, λlat The learning rate of the iterative training model corresponding to the representation vector of the stated dimension;

[0022] The representation methods for iteratively training the longitude representation vector include:

[0023] lon j ←lon j +λ lon ·(P ij -lat i ·lon j -bias ij )·lat i

[0024] Where, λ lon The learning rate of the iterative training model corresponding to the representation vector of the longitude;

[0025] The representation methods for iterative training using the multiple coordinate offset values ​​include:

[0026] bias ij ←bias ij +λ bias ·(P ij -lat i ·lon j -bias ij )

[0027] Where, λ bias The learning rate of the iterative training model corresponding to the coordinate offset value.

[0028] In one embodiment of this application, fitting the observed precipitation values ​​at the observation station coordinates based on the initial characterization vector and coordinate offset values ​​includes:

[0029] The initial representation vector and coordinate bias value are assigned to a pre-built transfer learning model to fit the observed precipitation values ​​at the observation station coordinates.

[0030] The fitted representation of the observed precipitation values ​​at the observation station coordinates includes:

[0031] G ij ≈lat i ·lon j +τ·bias ij

[0032] Wherein, τ is the transfer coefficient corresponding to the transfer learning model.

[0033] In one embodiment of this application, the second iteration training is performed based on the observed precipitation values ​​from the fitted observation station coordinates, including:

[0034] By using stochastic gradient descent, precipitation values ​​are observed based on the fitted station coordinates, and the longitude representation vector, latitude representation vector, and migration coefficient are iteratively trained.

[0035] The representation methods for iteratively training the representation vector of the aforementioned dimension include:

[0036] lat i ←lat i +λ lat ·(G ij -lat i ·lon j -τ·bias ij )·lon j

[0037] The representation methods for iteratively training the longitude representation vector include:

[0038] lon j ←lon j +λ lon ·(G ij -lat i ·lon j -τ·bias ij )·lat i

[0039] The representation methods for iterative training using the multiple coordinate offset values ​​include:

[0040] τ←τ+λ τ ·(G ij -lat i ·lon j -τ·bias ij )·bias ij

[0041] Where, λ τ The learning rate of the iterative training model corresponding to the transfer coefficient.

[0042] In one embodiment of this application, the corrected precipitation monitoring values ​​are represented in the following ways:

[0043] P ui =max(lat i ·lon j +τ·bias ij ,0.0)

[0044] in, The corrected precipitation monitoring value is referred to here.

[0045] In one embodiment of this application, a precipitation product monitoring value correction device is also provided, the device comprising:

[0046] The data acquisition module is used to acquire multiple initial precipitation monitoring values ​​corresponding to multiple time periods of the target precipitation product output, as well as the precipitation values ​​observed at the coordinates of the observation station;

[0047] The first iteration training module is used to fit the multiple initial precipitation monitoring values ​​and perform the first iteration training based on the fitted initial precipitation monitoring values ​​to obtain the initial representation vector and coordinate bias value;

[0048] The second iteration training module is used to fit the observed precipitation values ​​at the observation station coordinates based on the initial representation vector and coordinate bias values, and to perform a second iteration training based on the fitted observed precipitation values ​​at the observation station coordinates to obtain the target representation vector and the target migration coefficient.

[0049] The correction module is used to obtain the corrected precipitation monitoring value based on the target representation vector, target migration coefficient and coordinate offset value.

[0050] In one embodiment of this application, an electronic device is also provided, the electronic device comprising:

[0051] One or more processors;

[0052] A storage device for storing one or more programs that, when executed by one or more processors, cause the electronic device to implement the precipitation product monitoring value correction method as described above.

[0053] In one embodiment of this application, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a computer processor, causes the computer to perform the precipitation product monitoring value correction method as described above.

[0054] The beneficial effects of this invention are:

[0055] First, multiple initial precipitation monitoring values ​​corresponding to multiple time periods of the target precipitation product output are obtained, along with the precipitation values ​​observed at the station coordinates. Then, these initial precipitation monitoring values ​​are fitted, and a first iteration of training is performed based on the fitted initial precipitation monitoring values ​​to obtain an initial representation vector and coordinate bias values. Next, based on the initial representation vector and coordinate bias values, the precipitation values ​​observed at the station coordinates are fitted, and a second iteration of training is performed based on the fitted precipitation values ​​observed at the station coordinates to obtain a target representation vector and a target migration coefficient. Finally, based on the target representation vector, target migration coefficient, and coordinate bias values, the corrected precipitation monitoring values ​​are obtained. In this invention, based on the idea of ​​matrix factorization, multiple initial precipitation monitoring values ​​corresponding to multiple time periods of the target precipitation product are fitted. Then, through the first iteration of training, the obtained initial representation vector can improve the accuracy of the spatiotemporally continuous monitoring values ​​of the precipitation product output. Fitting the precipitation values ​​observed at the station coordinates based on the initial representation vector, and then performing a second iteration of training based on the idea of ​​transfer learning, can further improve the precipitation monitoring accuracy of each grid of the precipitation product.

[0056] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0057] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings:

[0058] Figure 1 This is a schematic diagram illustrating the implementation environment of a precipitation product monitoring value correction method, as shown in an exemplary embodiment of this application.

[0059] Figure 2 This is a schematic flowchart illustrating a precipitation product monitoring value correction method according to an exemplary embodiment of this application;

[0060] Figure 3 This is a flowchart illustrating a method for correcting precipitation product monitoring values, as shown in another exemplary embodiment of this application.

[0061] Figure 4 This is a block diagram illustrating a precipitation product monitoring value correction device according to an exemplary embodiment of this application;

[0062] Figure 5 A schematic diagram of the structure of a computer system suitable for an electronic device according to an embodiment of this application is shown. Detailed Implementation

[0063] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0064] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0065] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0066] First, it should be noted that with the rapid development of meteorological observation systems, the amount of observational data acquired using automatic weather stations, radar, and satellites is increasing, and the quality of simulation data from various numerical models is also continuously improving. At the same time, various industries have increasingly higher requirements for gridded, spatiotemporally continuous meteorological data products. Utilizing data fusion and assimilation technologies to integrate observational data from multiple sources and multi-model simulation data to obtain high-precision, high-quality, spatiotemporally continuous multi-source data fusion meteorological gridded products is an effective approach.

[0067] In existing data fusion technologies, various methods exist to improve the quality and resolution of precipitation products. For example, one approach uses automatic weather station precipitation data as a baseline, employing probability density function matching to correct systematic biases in radar and satellite-estimated precipitation products. Then, a Bayesian model averaging method is used to fuse radar and satellite precipitation products, forming a comprehensive and optimal joint precipitation background field. Another approach uses satellite-radar Bayesian model-averaged joint precipitation products as the background, employing spatial structure information from 1km radar-estimated precipitation for downscaling, which effectively improves the quality of the resolution background field. Then, using satellite-radar joint precipitation products at different resolutions as backgrounds, statistical methods are used to quantify error estimates, followed by optimal interpolation to integrate ground observations. While these two fusion methods offer higher accuracy than single-source precipitation products, the improvement in accuracy is limited and cannot finely enhance the accuracy of precipitation monitoring values ​​in each grid.

[0068] Therefore, to achieve rapid and smooth power response during load increases and decreases under varying load amplitudes without overshoot or undershoot, and to maintain stable and accurate steady-state power across different values, the control parameters of the power closed-loop control strategy need to be adaptive. Specifically, the parameters of the power closed-loop control strategy need to be determined based on different power load states (load increases, load decreases, or steady-state), the varying amplitudes of load increases or decreases, and the magnitude of the steady-state power. This enables adaptive net power control of the fuel cell engine based on load conditions, improving the power control response speed, accuracy, and stability.

[0069] Figure 1 This is a schematic diagram illustrating the implementation environment of a precipitation product monitoring value correction method, as shown in an exemplary embodiment of this application.

[0070] Reference Figure 1 As shown, the implementation environment may include a data acquisition terminal 101, a monitoring value correction terminal 102, and a server 103. The technical solution provided in this application embodiment can be applied to the monitoring value correction terminal 102. The monitoring value correction terminal 102 is used to acquire multiple initial precipitation monitoring values ​​corresponding to multiple time periods of the target precipitation product output by the data acquisition terminal 101, as well as the precipitation values ​​observed at the observation station coordinates. Then, the precipitation product monitoring value correction method in this invention is implemented through the initial precipitation monitoring values ​​and the precipitation values ​​observed at the observation station coordinates. After obtaining the corrected precipitation monitoring values, they are transmitted to the server 103. When precipitation-related tasks such as water supply detection are required, the corrected precipitation monitoring values ​​can be retrieved from the server 103 through a pre-configured data interface.

[0071] In one embodiment of this application, the monitoring value correction terminal 102 acquires multiple initial precipitation monitoring values ​​corresponding to multiple time periods of the target precipitation product output, as well as the observation station coordinate observation precipitation values; it fits the multiple initial precipitation monitoring values, and performs a first iteration training based on the fitted initial precipitation monitoring values ​​to obtain an initial representation vector and coordinate bias value; it fits the observation station coordinate observation precipitation values ​​based on the initial representation vector and coordinate bias value, and performs a second iteration training based on the fitted observation station coordinate observation precipitation values ​​to obtain a target representation vector and a target migration coefficient; and it obtains the corrected precipitation monitoring values ​​based on the target representation vector, target migration coefficient, and coordinate bias value. In this embodiment, based on the idea of ​​matrix factorization, multiple initial precipitation monitoring values ​​corresponding to multiple time periods of the target precipitation product can be fitted. Then, through the first iteration training, the obtained initial representation vector can improve the accuracy of the continuous data of the precipitation product output monitoring values ​​in time and space. Based on the initial representation vector, the precipitation values ​​observed at the station coordinates are fitted, and then a second iteration training is carried out based on the idea of ​​transfer learning, which can further improve the precipitation monitoring accuracy of each grid of the precipitation product.

[0072] The above section describes an exemplary implementation environment for applying the technical solution of this application. Next, we will continue to describe the method for correcting precipitation product monitoring values ​​of this application.

[0073] To address the problem of improving the accuracy of precipitation monitoring values ​​for each grid in the prior art, embodiments of this application propose a precipitation monitoring value correction method, a precipitation monitoring value correction device, an electronic device, a computer-readable storage medium, and a computer program product. These embodiments will be described in detail below.

[0074] Please see Figure 2 , Figure 2 This is a schematic flowchart illustrating a method for correcting precipitation product monitoring values, as shown in an exemplary embodiment of this application. This method can be applied to... Figure 1 The implementation environment is shown. It should be understood that this method can also be applied to other exemplary implementation environments and specifically executed by devices in other implementation environments. This embodiment does not limit the implementation environment to which the method is applicable.

[0075] like Figure 2 As shown, in an exemplary embodiment, the precipitation product monitoring value correction method includes at least steps S210 to S240, which are described in detail below:

[0076] In step S210, multiple initial precipitation monitoring values ​​corresponding to multiple time periods of the target precipitation product output are obtained, as well as the precipitation values ​​observed at the observation station coordinates.

[0077] It should be noted that since the fused precipitation products are usually generated hourly precipitation monitoring values, the precipitation product monitoring value correction method in this application occurs after each monitoring value is generated.

[0078] In step S220, multiple initial precipitation monitoring values ​​are fitted, and the first iteration training is performed based on the fitted initial precipitation monitoring values ​​to obtain the initial representation vector and coordinate bias value.

[0079] For example, this application uses the Transferred Matrix Factorization (TMF) algorithm to correct precipitation product monitoring values. The correction process can be divided into a pre-training stage and a fine-tuning stage. In the pre-training stage, based on the matrix factorization approach, for the precipitation values ​​at latitude i and longitude j in the precipitation fusion product P, they are fitted as the sum of the dot product of the latitude and longitude representation vectors and the coordinate bias value. Then, stochastic gradient descent is performed on each fitted initial precipitation monitoring value to conduct multiple iterations of training, resulting in the initial representation vector and coordinate bias value after multiple iterations of training.

[0080] It should be noted that, in the embodiments of this application, the design significance of the coordinate offset value is to reduce the number of differences between precipitation coordinate points and non-precipitation coordinate points at the same latitude (longitude), so as to make the training process of latitude and longitude representation vectors smoother.

[0081] In step S230, the observation precipitation values ​​at the station coordinates are fitted based on the initial representation vector and coordinate offset values, and a second iteration training is performed based on the fitted observation precipitation values ​​at the station coordinates to obtain the target representation vector and target migration coefficient.

[0082] For example, in the fine-tuning phase, the idea of ​​transfer learning is used. The pre-trained representation vector and coordinate bias values ​​are used as the initial values ​​of the model for this phase. The corresponding representation vector is adjusted according to the observed precipitation values ​​at the observation station coordinates. At the same time, since different coordinate bias values ​​are independent of each other, it is meaningless to adjust the individual coordinate bias values. Therefore, a global transfer coefficient is trained in this phase, while the coordinate bias values ​​remain unchanged in this phase.

[0083] In step S240, the corrected precipitation monitoring value is obtained based on the target characterization vector, the target migration coefficient, and the coordinate offset value.

[0084] For example, after the model training in the fine-tuning phase is completed, the corrected precipitation monitoring values ​​can be calculated. To prevent negative precipitation values, a maximum filtering function is used to avoid them.

[0085] As can be seen from steps S210 to S240 above, the scheme proposed in this embodiment can fit multiple initial precipitation monitoring values ​​corresponding to multiple time periods of the target precipitation product based on the idea of ​​matrix factorization. Then, through the first iteration training, the obtained initial representation vector can improve the accuracy of the continuous data of the precipitation product output monitoring values ​​in time and space. Based on the initial representation vector, the precipitation values ​​observed at the station coordinates are fitted, and then the second iteration training is carried out based on the idea of ​​transfer learning, which can further improve the precipitation monitoring accuracy of each grid of the precipitation product.

[0086] In one embodiment of this application, the initial precipitation monitoring values ​​include longitude information and latitude information.

[0087] For example, the precipitation monitoring values ​​output by the target precipitation product are real-time data in a grid format, using geographical latitude and longitude coordinates.

[0088] In one embodiment of this application, fitting the plurality of initial precipitation monitoring values ​​includes:

[0089] The multiple initial precipitation monitoring values ​​are fitted into the sum of the dot product of the longitude representation vector and the latitude representation vector and the coordinate offset value;

[0090] The methods for representing the fitted initial precipitation monitoring values ​​include:

[0091] P ij ≈lat i ·lon j +bias ij

[0092] Among them, P ij For each of the initial precipitation monitoring values ​​after fitting, lat i Let lon be the representation vector of the latitude. j Let bias be the representation vector of the longitude. ij The coordinate offset value is given.

[0093] It should be noted that in this embodiment, matrix factorization is used to fit multiple initial precipitation monitoring values ​​into the sum of the dot product of the longitude and latitude representation vectors and the coordinate offset values. Matrix factorization is a method to simplify a matrix to its constituent parts. This method can simplify more complex matrix operations, which can be performed on the decomposed matrix rather than on the original matrix itself. Matrix factorization is mainly used in the field of recommendation systems. Through matrix factorization, on the one hand, the amount of computation can be reduced, and on the other hand, it can effectively solve the problem of sparsity of the behavior matrix caused by an excessive number of users and items.

[0094] In one embodiment of this application, the first iterative training is performed based on the fitted initial precipitation monitoring values, including:

[0095] By using stochastic gradient descent, the longitude representation vector, latitude representation vector, and coordinate offset value are iteratively trained based on the fitted initial precipitation monitoring values.

[0096] The representation methods for iteratively training the representation vector of the aforementioned dimension include:

[0097] lat i ←lat i +λ lat ·(P ij -lat i ·lon j -bias ij )·lon j

[0098] Where, λ lat The learning rate of the iterative training model corresponding to the representation vector of the stated dimension;

[0099] The representation methods for iteratively training the longitude representation vector include:

[0100] lon j ←lon j +λ lon ·(P ij -lat i ·lon j -bias ij )·lat i

[0101] Where, λ lon The learning rate of the iterative training model corresponding to the representation vector of the longitude;

[0102] The representation methods for iterative training using the multiple coordinate offset values ​​include:

[0103] bias ij ←bias ij +λ bias ·(P ij -lat i ·lon j -bias ij )

[0104] Where, λ bias The learning rate of the iterative training model corresponding to the coordinate offset value.

[0105] For example, before iterative training, the number of iterations can be preset. If the preset number of iterations is reached during training, the iterative training will stop.

[0106] It's important to note that most deep learning models currently employ batch stochastic gradient descent (BDD) for optimization. BDD optimization involves batch size and learning rate. These two factors, besides the gradient itself, directly determine the model's weight updates, and from an optimization perspective, they are the most crucial parameters affecting model performance and convergence. The learning rate determines the extent (or speed, and step size) of parameter adjustment in each update step. The learning rate directly impacts the model's convergence state. It's a key hyperparameter for training neural networks, representing the step size by which the gradient moves towards the optimal solution of the loss function in each iteration. Its magnitude determines the network's learning speed. During network training, the model provides predictions based on sample data, calculates the cost function, and adjusts parameters through backpropagation. This process is repeated until the model parameters gradually approach the optimal solution, resulting in the optimal model.

[0107] In one embodiment of this application, fitting the observed precipitation values ​​at the observation station coordinates based on the initial characterization vector and coordinate offset values ​​includes:

[0108] The initial representation vector and coordinate bias value are assigned to a pre-built transfer learning model to fit the observed precipitation values ​​at the observation station coordinates.

[0109] The fitted representation of the observed precipitation values ​​at the observation station coordinates includes:

[0110] G ij ≈lat i ·lon j +τ·bias ij

[0111] Wherein, τ is the transfer coefficient corresponding to the transfer learning model.

[0112] It's important to note that transfer learning utilizes existing knowledge to learn new knowledge. Its core is finding similarities between existing and new knowledge, and achieving the goal of transfer learning through the transfer of these similarities. Everything in the world shares commonalities; the core issue of transfer learning is how to reasonably find these similarities and then use this bridge to help learn new knowledge.

[0113] In one embodiment of this application, a second iteration of training is performed based on the fitted observation station coordinates and observed precipitation values, including:

[0114] By using stochastic gradient descent, precipitation values ​​are observed based on the fitted station coordinates, and the longitude representation vector, latitude representation vector, and migration coefficient are iteratively trained.

[0115] The representation methods for iteratively training the representation vector of the aforementioned dimension include:

[0116] lat i ←lat i +λ lat ·(G ij -lat i ·lon j -τ·bias ij )·lon j

[0117] The representation methods for iteratively training the longitude representation vector include:

[0118] lon j ←lon j +λ lon ·(G ij -lat i ·lon j -τ·bias ij )·lat i

[0119] The representation methods for iterative training using the multiple coordinate offset values ​​include:

[0120] τ←τ+λ τ ·(G ij -lat i ·lon j -τ·bias ij )·bias ij

[0121] Where, λ τ The learning rate of the iterative training model corresponding to the transfer coefficient.

[0122] For example,

[0123] In one embodiment of this application, the corrected precipitation monitoring values ​​are represented in the following ways:

[0124]

[0125] in, The corrected precipitation monitoring value is referred to here.

[0126] In one embodiment of this application, see Figure 3 , Figure 3This is a schematic flowchart illustrating a method for correcting precipitation product monitoring values, as shown in another exemplary embodiment of this application. Figure 3 As can be seen, the TMF algorithm diagram is as follows: Figure 3 As shown, the process mainly consists of two stages: pre-training and fine-tuning. In the pre-training stage, for the precipitation values ​​in the precipitation fusion product based on latitude and longitude, they are fitted as the sum of the dot product of the latitude and longitude representation vectors and the coordinate bias value. Through stochastic gradient descent, an iterative formula for the representation vector and coordinate bias value is obtained. In the fine-tuning stage, using the idea of ​​transfer learning, the pre-trained representation vector and coordinate bias value are used as the initial values ​​for this stage of the model. The corresponding representation vector is then adjusted based on the observed precipitation values ​​at the observation station coordinates.

[0127] In one embodiment of this application, taking the China Meteorological Administration's Multi-Source Precipitation Product (CMPAS) as an example, hourly precipitation data from Chongqing area from May 1 to July 31, 2021, are used as both training and test datasets (with a ratio ranging from 1:9 to 9:1). Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are used as evaluation metrics, and PDF, Support Vector Regression (SVR), and Matrix Factorization (MF) algorithms are used as comparison methods. See Tables 1 and 2. Table 1 shows the errors of multiple correction methods when the training dataset ratio is 10%, 20%, 30%, 40%, and 50%, and Table 2 shows the errors of multiple correction methods when the training dataset ratio is 60%, 70%, 80%, and 90%.

[0128] Table 1 shows the errors of multiple correction methods when the training dataset proportions are 10%, 20%, 30%, 40%, and 50%.

[0129]

[0130]

[0131] Table 2 shows the errors of multiple correction methods when the training dataset proportions are 60%, 70%, 80%, and 90%.

[0132]

[0133] As shown in Tables 1 and 2, TMF significantly improves the accuracy of CMPAS precipitation monitoring values ​​compared to the other three correction methods. Specifically, the mean absolute error is reduced by at least 20.49%, and the root mean square error is reduced by at least 15.17%.

[0134] Figure 4 This is a block diagram illustrating a precipitation product monitoring value correction device according to an exemplary embodiment of this application. The device can be applied to... Figure 1 The implementation environment shown is not limited to this embodiment. This device can also be applied to other exemplary implementation environments and specifically configured in other devices. This embodiment does not limit the implementation environment to which the device is applicable.

[0135] like Figure 4 As shown, this exemplary precipitation product monitoring value correction device includes:

[0136] The data acquisition module 401 is used to acquire multiple initial precipitation monitoring values ​​corresponding to multiple time periods of the target precipitation product output, as well as the precipitation values ​​observed at the coordinates of the observation station.

[0137] The first iteration training module 402 is used to fit the plurality of initial precipitation monitoring values ​​and perform the first iteration training based on the fitted initial precipitation monitoring values ​​to obtain the initial representation vector and coordinate bias value.

[0138] The second iteration training module 403 is used to fit the observed precipitation values ​​at the observation station coordinates based on the initial representation vector and coordinate bias value, and to perform a second iteration training based on the fitted observed precipitation values ​​at the observation station coordinates to obtain the target representation vector and the target migration coefficient.

[0139] The correction module 404 is used to obtain the corrected precipitation monitoring value based on the target representation vector, target migration coefficient and coordinate offset value.

[0140] In this exemplary precipitation product monitoring value correction device, multiple initial precipitation monitoring values ​​corresponding to multiple time periods of the target precipitation product can be fitted based on the idea of ​​matrix factorization. Then, through the first iteration training, the obtained initial representation vector can improve the accuracy of the continuous data of the precipitation product output monitoring values ​​in time and space. Based on the initial representation vector, the precipitation values ​​observed at the station coordinates are fitted, and then a second iteration training is carried out based on the idea of ​​transfer learning, which can further improve the precipitation monitoring accuracy of each grid of the precipitation product.

[0141] It should be noted that the precipitation product monitoring value correction device and the precipitation product monitoring value correction method provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the precipitation product monitoring value correction device provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0142] Embodiments of this application also provide an electronic device, including: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the electronic device to implement the precipitation product monitoring value correction method provided in the above embodiments.

[0143] Figure 5 A schematic diagram of a computer system suitable for an electronic device according to an embodiment of this application is shown. It should be noted that... Figure 5 The computer system 500 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0144] like Figure 5 As shown, the computer system 500 includes a Central Processing Unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 502 or programs loaded from Storage Unit 508 into Random Access Memory (RAM) 503, such as performing the methods described in the above embodiments. The RAM 503 also stores various programs and data required for system operation. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An Input / Output (I / O) interface 505 is also connected to the bus 504.

[0145] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to I / O interface 505 as needed. Removable media 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 510 as needed so that computer programs read from them can be installed into storage section 508 as needed.

[0146] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs various functions defined in the system of this application.

[0147] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0148] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0149] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0150] Another aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a computer's processor, causes the computer to perform the precipitation product monitoring value correction method as described above. This computer-readable storage medium may be included in the electronic device described in the above embodiments, or it may exist independently and not incorporated into the electronic device.

[0151] Another aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the precipitation product monitoring value correction method provided in the various embodiments described above.

[0152] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method of revising a monitoring value of a precipitation product, characterized by, The method comprises: obtaining a plurality of initial precipitation monitoring values corresponding to a plurality of time periods of target precipitation product output, and precipitation values observed by observation station coordinates; fitting the plurality of initial precipitation monitoring values, and performing first iteration training according to each of the fitted initial precipitation monitoring values to obtain an initial feature vector and a coordinate bias value; fitting the precipitation values observed by the observation station coordinates according to the initial feature vector and the coordinate bias value, and performing second iteration training according to the fitted precipitation values observed by the observation station coordinates to obtain a target feature vector, a target transfer coefficient; obtaining a revised precipitation monitoring value according to the target feature vector, the target transfer coefficient and the coordinate bias value; wherein fitting the plurality of initial precipitation monitoring values comprises: fitting the plurality of initial precipitation monitoring values into the sum of the dot product of the longitude feature vector and the latitude feature vector and the coordinate bias value.

2. The precipitation product monitoring value revision method according to claim 1, characterized by, The initial precipitation monitoring value comprises longitude information and latitude information.

3. The precipitation product monitoring value revision method according to claim 1, characterized by, The representation of each of the fitted initial precipitation monitoring values comprises: wherein, is the fitted initial precipitation monitoring value, is the representative vector of the latitude, is the representative vector of the longitude, is the coordinate bias value.

4. The precipitation product monitoring value revision method according to claim 3, characterized by, performing first iteration training according to each of the fitted initial precipitation monitoring values comprises: performing iteration training on the longitude feature vector, the latitude feature vector and the coordinate bias value according to each of the fitted initial precipitation monitoring values by random gradient descent; the representation of performing iteration training on the latitude feature vector comprises: wherein, is the learning rate of the iterative trained model corresponding to the representation vector of the latitude. the representation of performing iteration training on the longitude feature vector comprises: wherein, is the learning rate of the iterative trained model corresponding to the representation vector of the longitude. the representation of performing iteration training on the plurality of coordinate bias values comprises: wherein, is the iteration training model learning rate corresponding to the coordinate bias value.

5. The method for revising the precipitation product monitoring value according to claim 3, characterized in that, fitting the precipitation values observed by the observation station coordinates according to the initial feature vector and the coordinate bias value comprises: fitting the initial feature vector and the coordinate bias value into a pre-constructed transfer learning model to fit the precipitation values observed by the observation station coordinates; the representation of the fitted precipitation values observed by the observation station coordinates comprises: wherein, is a transfer coefficient corresponding to the transfer learning model.

6. The precipitation product monitoring value revision method according to claim 5, characterized by, performing second iteration training according to the fitted precipitation values observed by the observation station coordinates comprises: performing iteration training on the longitude feature vector, the latitude feature vector and the transfer coefficient according to the fitted precipitation values observed by the observation station coordinates by random gradient descent; the representation of performing iteration training on the latitude feature vector comprises: the representation of performing iteration training on the longitude feature vector comprises: the representation of performing iteration training on the plurality of coordinate bias values comprises: wherein, is the learning rate of the iterative training model corresponding to the migration coefficient.

7. The precipitation product monitoring value revision method according to claim 6, characterized by, the representation of the revised precipitation monitoring value comprises: wherein, is the revised precipitation monitoring value.

8. A precipitation product monitoring value revision device characterized by comprising: The device comprises: a data acquisition module configured to obtain a plurality of initial precipitation monitoring values corresponding to a plurality of time periods of target precipitation product output, and precipitation values observed by observation station coordinates; a first iteration training module configured to fit the plurality of initial precipitation monitoring values, and perform first iteration training according to each of the fitted initial precipitation monitoring values to obtain an initial feature vector and a coordinate bias value; a second iteration training module configured to fit the precipitation values observed by the observation station coordinates according to the initial feature vector and the coordinate bias value, and perform second iteration training according to the fitted precipitation values observed by the observation station coordinates to obtain a target feature vector, a target transfer coefficient; A correction module is configured to obtain a corrected precipitation monitoring value according to the target characteristic vector, a target migration coefficient, and a coordinate bias value. The fitting of the plurality of initial precipitation monitoring values includes: The plurality of initial precipitation monitoring values are fitted into a sum of a dot product of a longitude characteristic vector and a latitude characteristic vector and a coordinate bias value.

9. An electronic device, comprising: The electronic device includes: one or more processors; a storage device for storing one or more programs that, when executed by the one or more processors, cause the electronic device to implement the precipitation product monitoring value correction method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to perform the precipitation product monitoring value correction method of any one of claims 1 to 7.

Images