Plum rain prediction method based on physical constraint-hybrid power machine learning

Abstract

The invention discloses a plum rain prediction method based on physical constraint-hybrid power machine learning. The method comprises the steps of obtaining historical plum rain amount and key physical system observation data through historical physical system indexes; based on the key physical system index prediction factor, taking the historical plum rain amount as a prediction object, respectively adopting a plurality of machine learning methods, and respectively training and constructing a plurality of corresponding plum rain prediction models; outputting a key physical system index as an atmospheric bridge based on a power numerical mode, substituting the atmospheric bridge into the machine learning model, and outputting plum rainfall predicted values of various machine learning methods; according to the method, the weights are dynamically distributed through a multi-model weighted integration strategy, the over-fitting risk of a single model to a specific meteorological scene is reduced, and data support is provided for intelligent forecasting research based on a meteorological mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of weather forecasting, and in particular to a plum rain forecasting method based on physical constraint-hybrid power machine learning. Background Art

[0002] The Meiyu season is a unique summer weather and climate phenomenon in East Asia, characterized by significant interannual variability. Its accurate prediction is crucial for agricultural production, flood control and disaster reduction, and traffic scheduling. Existing Meiyu season prediction methods fall into two main categories: traditional physical models and dynamic numerical models. Traditional physical models construct synoptic or statistical models based on physical principles such as fluid mechanics and thermodynamics. These models describe the physical mechanisms of Meiyu formation through analytical models and combine them with real-time observations or historical data to predict Meiyu. Dynamic numerical models, centered on numerical weather prediction, use a system of atmospheric fluid dynamics equations and high-performance computers to solve them, predicting atmospheric circulation and precipitation during the Meiyu season.

[0003] Traditional physical models for plum rain prediction mainly rely on idealized physical correlations, such as short-term forecasts of temperature, humidity, and precipitation in the plum rain area, or predict the arrival of the plum rain season by observing and analyzing the influence of subtropical high pressure and southwest monsoon. These models are difficult to truly reflect actual processes such as complex atmospheric systems and terrain forcing, and their ability to predict the total amount of precipitation during the plum rain season is insufficient.

[0004] Dynamic numerical models rely solely on initial field data to drive simulations, failing to effectively integrate long-term statistical patterns from historical observations. This results in unstable total precipitation forecasts during the plum rain season, with large errors in precipitation predictions during the late plum rain season and at its end. Therefore, a plum rain forecasting method based on physical constraints and hybrid machine learning is needed. Summary of the Invention

[0005] The purpose of the present invention is to provide a plum rain prediction method based on physical constraint-hybrid machine learning.

[0006] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0007] A first aspect of the present invention provides a plum rain prediction method based on physical constraint-hybrid machine learning, comprising the following steps:

[0008] Obtain historical plum rain amounts and key physical system observation data through historical physical system indices;

[0009] Based on the key physical system index prediction factor, the historical plum rain amount is used as the prediction object, and multiple machine learning methods are used to train and construct multiple corresponding plum rain prediction models;

[0010] Based on the output of key physical system indices from the dynamic numerical model as atmospheric bridges, the machine learning model is fed with the output of plum rain amount prediction values ​​using various machine learning methods.

[0011] Furthermore, the key physical system index uses the meridional wind in the lower troposphere of the Bay of Bengal and nearby areas to characterize the influence of the South Asian summer monsoon; uses the meridional wind in the lower troposphere on the east side of the Qinghai-Tibet Plateau to characterize the influence of the plateau summer monsoon; uses the zonal wind in the lower troposphere of the Jiangnan region to characterize the influence of the subtropical summer monsoon; uses the mid-tropospheric geopotential height in the northeastern cold vortex activity area to characterize the cold air activity affecting the plum rain area; and uses the mid-tropospheric geopotential height in the South China Sea-western Pacific region to characterize the influence of the subtropical high pressure in the western Pacific:

[0012] Furthermore, global atmospheric reanalysis data with a spatial resolution of 2.5°×2.5° and a temporal resolution of 1 day were used to calculate the historical data of the key physical system index affecting the plum rain, with a time length of 33 years.

[0013] Furthermore, the machine learning method includes Linear machine learning method, Gbdt machine learning method, Lgbm machine learning method, and Xgb machine learning method to train and construct four plum rain prediction models F respectively. M .

[0014] Furthermore, the dynamic numerical model is a key physical system predicted by the dynamic numerical model CFS-v2.

[0015] Furthermore, the explained variance of a model is calculated as follows:

[0016]

[0017] where R i represents the actual observed plum rain amount in the i-th year before the predicted year; R' i is the output value of plum rain in the i-th year before the current year from an independent test report of a certain model; Represents the residual sum of squares divided by the number of samples; Var(R) is the variance of the plum rain amount in the previous 10 years of the predicted year:

[0018]

[0019] Finally, we get the plum rain amount forecast for year N after weighted integration

[0020]

[0021] in, represents the weight of model M in the i-th year before the forecast year N, Represents the plum rain amount of model M for year N.

[0022] In a second aspect, an embodiment of the present application further provides an electronic device, comprising: a processor; and a memory arranged to store computer-executable instructions, wherein the executable instructions, when executed, cause the processor to perform the method steps described in the first aspect.

[0023] In a third aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple applications, the electronic device executes the method steps described in the first aspect.

[0024] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0025] (1) The present invention introduces the key circulation system index of the Asian summer monsoon system to construct a complex correlation model of multiple physical elements, thereby solving the problems of insufficient utilization of dynamic numerical model data and poor prediction stability.

[0026] (2) The present invention dynamically allocates weights through a multi-model weighted integration strategy, reducing the risk of overfitting of a single model to a specific meteorological scenario, especially improving the prediction robustness under abnormal climate conditions such as El Niño / La Niña. By explicitly introducing key physical system indices and analyzing the importance of machine learning features (such as the feature gain value of GBDT), the influence weight of each circulation system on the plum rain amount can be quantified, solving the "black box" problem of traditional machine learning methods and providing data support for intelligent prediction research based on meteorological mechanisms. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flowchart of the steps of the plum rain prediction method based on physical constraints-hybrid machine learning of the present invention;

[0028] Figure 2 Schematic diagram of the regional range of five key physical system indices based on physical constraint-hybrid machine learning of the present invention;

[0029] Figure 3 This is a schematic diagram of the structure of an electronic device in an embodiment of this specification. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0031] Reference Figure 1 As shown, the present invention provides a plum rain prediction method based on physical constraint-hybrid machine learning, including:

[0032] 1. Identify the key physical systems that influence the Meiyu season, including five key circulation system indices of the Asian summer monsoon;

[0033] 2. Based on historical plum rain amounts and observational data of key physical systems, a plum rain model was established using four machine learning methods (Linear, Gbdt, Lgbm, and Xgb);

[0034] 3. The key physical systems predicted by the dynamic numerical model CFS-v2 are used as atmospheric bridges and substituted into the machine learning model to obtain the Meiyu forecast value;

[0035] 4. Based on the results of independent trial reports, the plum rain prediction values ​​of the four machine learning models are weighted and integrated to obtain the final plum rain amount forecast.

[0036] 1. Identify the key physical systems that influence plum rains

[0037] The key system index of the Asian summer monsoon was constructed, and the different periods of Meiyu precipitation anomalies affected by different systems were diagnosed and analyzed. A total of five key circulation systems affecting the entire Meiyu period were identified, which respectively represent: South Asian summer monsoon, plateau summer monsoon, subtropical summer monsoon, cold air influence, and subtropical high pressure influence over the South China Sea.

[0038] 2. Obtain historical plum rain amounts and key physical system observation data

[0039] Referencing the national standard "Meiyu Monitoring Indicators" (GB / T 33671-2017), meteorological station observation data are used to determine the start and end of the Meiyu rainy season, and the Meiyu amount (R) is calculated based on this:

[0040]

[0041] Among them, S represents the beginning of the plum rain season, E represents the end of the plum rain season, and P i It represents the precipitation on the i-th day during the plum rain period, and then obtains the historical observation data of plum rain amount, providing a historical prediction object data basis for constructing the plum rain prediction method.

[0042] Using global atmospheric reanalysis data (spatial resolution of 2.5°×2.5° and temporal resolution of 1 day), we calculated historical data of five key physical system indices (VS / VN / UE / H8 / H9) that affect the Meiyu season:

[0043] a. The influence of the South Asian summer monsoon is characterized by the lower tropospheric meridional wind (VS) over the Bay of Bengal and nearby areas:

[0044] VS= <v 700 >(10°-20°N,85°-100°E)

[0045] Among them, v 700 It represents the 700hPa meridional wind, and (10°-25°N, 85°-100°E) represents the selected area.

[0046] b. The influence of the plateau summer monsoon is characterized by the lower tropospheric meridional wind (VN) east of the Qinghai-Tibet Plateau:

[0047] VN= <v 700 > (25°-40°N,100°-105°E)

[0048] Among them, v 700 It represents the 700hPa meridional wind, and (25°-40°N, 100°-105°E) represents the selected area.

[0049] c. The influence of the subtropical summer monsoon is characterized by the lower tropospheric zonal wind (UE) over the Jiangnan region:

[0050] UE= 700 > (25°-35°N,105°-120°E)

[0051] Among them, u 700 It represents the 700hPa zonal wind, and (25°-35°N, 105°-120°E) represents the selected area.

[0052] d. The mid-tropospheric geopotential height (H8) of the Northeast Cold Vortex activity area is used to characterize the cold air activity affecting the Meiyu area:

[0053] H8= <Z 500 > (25°-35°N,105°-120°E)

[0054] Among them, Z 500 represents the 500hPa geopotential height, and (35°-60°N, 115°-145°E) represents the selected area.

[0055] e. The influence of the western Pacific subtropical high pressure is characterized by the mid-tropospheric geopotential height (H9) over the South China Sea-western Pacific region:

[0056] H9= <Z 500 > (10°-20°N,110°-125°E)

[0057] Among them, Z 500 represents the 500hPa geopotential height, and (10°-20°N, 110°-125°E) represents the selected area.

[0058] 3. Build a machine learning plum rain prediction model based on physical constraints

[0059] like Figure 2 ​As shown in the figure, the key physical systems such as the South Asian summer monsoon, plateau summer monsoon, subtropical summer monsoon, cold air influence, and subtropical high pressure over the South China Sea from historical observations in June and July are used as prediction factors (VS / VN / UE / H8 / H9), and the historical plum rain amount (R) is used as the prediction object. Four machine learning methods (including Linear / Gbdt / Lgbm / Xgb) are used to train and construct four plum rain prediction models.

[0060]

[0061] in, It represents the plum rain forecast value in year "N" (e.g., 2023) obtained by using the machine learning method Gbdt to train the historical observed plum rain sequences and key physical systems before that year.

[0062] 4. Hybrid Power Mode Prediction

[0063] In real-time prediction, the above key physical system indices predicted by the dynamic numerical model CFS-v2 are used as atmospheric bridges and substituted into the machine learning model. The output of the plum rain amount prediction values ​​of the four machine learning methods is R M N .

[0064] For the predicted value of plum rain amount R M N The explanation is as follows: R represents the amount of plum rain, M represents a machine learning method (specifically corresponding to Linear / Gbdt / Lgbm / Xgb; N represents the year). For example, R Gbdt 2023 It represents the Meiyu amount in 2023 outputted by the Gbdt method.

[0065] The steps for entering the dynamic numerical model are as follows:

[0066] Calculate the predicted values ​​(VS) of the above five key physical system indices in the year "N" (such as 2023) by the dynamic numerical model (such as CFS-v2). N / VN N / UE N / H8 N / H9 N ), VS N / VN N / UE N / H8 N / H9 N Input the four plum rain machine learning models trained in 3) and obtain the plum rain amount prediction value R output by the model M N ,Right now:

[0067]

[0068] 5. Dynamic Weighted Ensemble Prediction

[0069] In real-time prediction, for each machine learning method, the variance Q of the actual plum rain amount explained by the prediction of the 10 years before year N is calculated. M (As Q linear The four machine learning methods were weighted and the real-time plum rain prediction values ​​of the four machine learning models were further weighted integrated to obtain the final plum rain amount prediction R. ens N .

[0070] The proportion of explained variance corresponding to each model Q M (As Q linear etc.) are calculated as follows:

[0071]

[0072] Among them, Con M The explained variances of the four models are: Con linear / Con Gbdt / Con Lgbm / Con Xgb .

[0073] As an important indicator for evaluating how well the model's predictions compare to the actual observed values, the explained variance of a model is calculated as follows:

[0074]

[0075] where R i represents the actual observed plum rain amount in the i-th year before the predicted year; R' i is the output value of plum rain in the i-th year before the current year of the forecast of a certain model in an independent trial report; n is the number of years of independent trial report, which is 10, and the 10 years before the forecast year N are taken; Represents the residual sum of squares divided by the number of samples; Var(R) is the variance of the plum rain amount in the previous 10 years of the predicted year:

[0076]

[0077] Finally, we get the plum rain amount forecast for year N after weighted integration

[0078]

[0079] in, represents the weight of model M in the i-th year before the forecast year N, Represents the plum rain amount of model M for year N.

[0080] In this implementation example, the prediction of plum rain amount in the middle and lower reaches of the Yangtze River in 2024 is taken as an example:

[0081] 1. Obtain historical plum rain amounts and key physical system observation data

[0082] With reference to the national standard of "Meiyu Monitoring Indicators", the start and end of the Meiyu rainy season are determined using meteorological station observation data, and the historical observation data R of Meiyu in the middle and lower reaches of the Yangtze River from 1991 to 2023 are calculated.

[0083] Using global atmospheric reanalysis data with a spatial resolution of 2.5°×2.5° and a temporal resolution of 1 day (such as the China Meteorological Administration's atmospheric reanalysis data CRA40), historical data of five key physical system indices affecting the Meiyu season (South Asian summer monsoon VS, Plateau summer monsoon VN, subtropical summer monsoon UE, cold air impact H8, and subtropical high pressure index over the South China Sea H9) were calculated, with a time span of 1991-2023.

[0084] 2. Build a machine learning plum rain prediction model based on physical constraints

[0085] The five key physical system indices (VS / VN / UE / H8 / H9) from historical observations in June and July were used as prediction factors (i.e., model input values), and the historical plum rain amount in the middle and lower reaches of the Yangtze River was used as the prediction object (i.e., model output value). Four machine learning methods (including Linear / Gbdt / Lgbm / Xgb) were used to train and construct four plum rain prediction models F. M (VS, VN, UE, H8, H9), and use the trained model to output the predicted value R of the plum rain amount from 2014 to 2023 M N , where M is the model name, which can be Linear / Gbdt / Lgbm / Xgb, and N is the year, ranging from 2014 to 2023. For example, the prediction result of the Linear model for the plum rain amount in 2014 is R Linear 2014 .

[0086] 3. Hybrid Power Mode Prediction

[0087] The predicted values ​​of the above five key physical system indices in 2024 (VS 2024 / VN 2024 / UE 2024 / H8 2024 / H9 2024 ), VS 2024 / VN 2024 / UE 2024 / H8 2024 / H9 2024 Input the four plum rain machine learning models trained in 2) and obtain the plum rain amount prediction value R output by the model M 2024 , taking Gbdt as an example:

[0088]

[0089] 4. Dynamic Weighted Ensemble Prediction

[0090] Calculate the predicted value R of Meiyu amount for each prediction model from 2014 to 2023 M N (such as R Linear 2014 ,...R Linear 2023 )'s variance V M 2014-2023 With observation R 2014 ,...R 2023 The ratio Q of the variance V M 2014-2023 (As Q Linear 2014-2023 ), according to Q M 2014-2023 The weights of the four machine learning methods were assigned, and the real-time plum rain prediction values ​​R of the four machine learning models were further calculated. M 2024 Perform weighted integration to obtain the final plum rain amount forecast R ens 2024 .

[0091]

[0092] The present invention includes determining key physical systems, acquiring historical data, constructing a machine learning model, substituting the dynamic numerical model into the weighted integrated data of multiple models, calculating the proportion of explained variance of the independent test reports of each machine learning model to the actual situation in the past 10 years, and determining the corresponding weight of each model in the multi-model integration. The above weights are the results of the past 10 years sliding each year, so it is a dynamic weighted integration strategy based on explained variance.

[0093] Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. Figure 3 At the hardware level, the electronic device includes a processor and, optionally, an internal bus, a network interface, and memory. The memory may include internal memory, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk drive. Of course, the electronic device may also include other hardware required for its services.

[0094] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0095] The memory is used to store programs. Specifically, the program may include program code, which includes computer operating instructions. The memory may include internal memory and non-volatile memory, and provides instructions and data to the processor.

[0096] The processor reads the corresponding computer program from the non-volatile memory into the internal memory and then runs it, logically forming a device for constructing a virtual reality-based intelligent product simulation test scenario. The processor executes the program stored in the memory and is specifically configured to execute any of the aforementioned plum rain prediction methods based on physical constraints and hybrid machine learning.

[0097] The present invention can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, the steps of the above method can be completed by hardware integrated logic circuits in the processor or by software instructions. The above processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The various methods, steps, and logic block diagrams disclosed in the embodiments of this application can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0098] An embodiment of the present application also proposes a computer-readable storage medium, which stores one or more programs, and the one or more programs include instructions. When the instructions are executed by an electronic device including multiple applications, they execute any of the aforementioned plum rain prediction methods based on physical constraints and hybrid machine learning.

[0099] The above content is merely an example and explanation of the structure of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.

Claims

1. A plum rain prediction method based on physical constraints and hybrid machine learning, characterized by: The following steps are involved: Obtain historical plum rain amounts and key physical system observation data through historical physical system indices; Based on the key physical system index prediction factor, the historical plum rain amount is used as the prediction object, and multiple machine learning methods are used to train and construct multiple corresponding plum rain prediction models; Based on the output of key physical system indices from the dynamic numerical model as atmospheric bridges, the machine learning model is fed with the output of plum rain amount prediction values ​​using various machine learning methods. Based on the results of independent trial reports in the 10 years before the forecast year, the explained variance percentage of each machine learning model was calculated, weights were assigned to multiple machine learning methods, and the real-time plum rain prediction values ​​of the corresponding machine learning models were weighted integrated to obtain the final plum rain amount forecast.

2. The plum rain prediction method based on physical constraint-hybrid machine learning according to claim 1 is characterized in that: The key physical system index uses the lower tropospheric meridional wind in the Bay of Bengal and nearby areas to represent the influence of the South Asian summer monsoon; uses the lower tropospheric meridional wind on the east side of the Qinghai-Tibet Plateau to represent the influence of the plateau summer monsoon; uses the lower tropospheric zonal wind in the Jiangnan region to represent the influence of the subtropical summer monsoon; uses the mid-tropospheric potential height in the northeastern cold vortex activity area to represent the cold air activity affecting the plum rain area; and uses the mid-tropospheric potential height in the South China Sea-western Pacific region to represent the influence of the western Pacific subtropical high pressure.

3. The plum rain prediction method based on physical constraint-hybrid machine learning according to claim 1 is characterized in that: The historical data of the key physical system indices affecting the Meiyu region were calculated using global atmospheric reanalysis data with a spatial resolution of 2.5°×2.5° and a temporal resolution of 1 day, with a time length of 33 years.

4. The plum rain prediction method based on physical constraint-hybrid machine learning according to claim 1 is characterized in that: The machine learning method includes Linear machine learning method, Gbdt machine learning method, Lgbm machine learning method, and Xgb machine learning method to train and construct four plum rain prediction models F respectively. M .

5. The plum rain prediction method based on physical constraint-hybrid machine learning according to claim 1 is characterized in that: The dynamic numerical model is the key physical system predicted by the dynamic numerical model CFS-v2.

6. The plum rain prediction method based on physical constraint-hybrid machine learning according to claim 1 is characterized in that: The explained variance of a model is calculated as follows: where R i represents the actual observed plum rain amount in the i-th year before the predicted year; R' i is the output value of plum rain in the i-th year before the current year from an independent test report of a certain model; Represents the residual sum of squares divided by the number of samples; Var(R) is the variance of the plum rain amount in the previous 10 years of the predicted year: Finally, we get the weighted integrated plum rain amount forecast R for year N. ens N ; in, represents the weight of model M in the i-th year before the forecast year N, Represents the plum rain amount of model M for year N.

7. An electronic device comprising: processor; and a memory arranged to store computer-executable instructions, which, when executed, cause the processor to perform the method according to any one of claims 1 to 6.

8. A computer-readable storage medium storing one or more programs, wherein when the one or more programs are executed by an electronic device including a plurality of application programs, the electronic device executes the method according to any one of claims 1 to 6.

Images