CN-121559636-B - Temperature medium-term prediction method based on physical constraint

CN121559636BCN 121559636 BCN121559636 BCN 121559636BCN-121559636-B

Abstract

The application discloses a temperature metaphase prediction method based on physical constraint, which comprises the steps of preprocessing data in a historical data set containing a plurality of variables to obtain preprocessed data, inputting the preprocessed data into a trained temperature metaphase prediction model to conduct temperature metaphase prediction to obtain a temperature metaphase prediction result, dividing the preprocessed data by adopting a sliding window, inputting input tensors corresponding to each variable divided by the sliding window into ConvLSTM D layers through independent channels to extract space-time characteristics corresponding to each variable, removing redundant information in the space-time characteristics corresponding to each variable to obtain space-time characteristics of each variable after removing redundant information, carrying out feature fusion on the space-time characteristics of all variables after removing redundant information to obtain fusion characteristics, and carrying out temperature metaphase prediction based on the fusion characteristics to obtain the temperature metaphase prediction result. The method can improve the prediction accuracy of the 2m temperature medium-term prediction.

Inventors

LI XIAOYONG
LI QIXIAO
GU MINGHAO
WU SONG
TAN MINGKUN
ZHANG KAIYU
ZHAO HAOTIAN
SHAO CHENGCHENG
REN XIAOLI
ZHU XIANG

Assignees

中国人民解放军国防科技大学

Dates

Publication Date: 20260512
Application Date: 20260122

Claims (8)

1. A method for mid-temperature prediction based on physical constraints, the method comprising: Preprocessing data in a historical data set containing a plurality of variables to obtain preprocessed data, wherein the variables comprise the acquired 2m height temperature, total cloud cover, east-west component wind speed, north-south component wind speed and net radiation flux; Inputting the preprocessed data into a trained temperature metaphase prediction model for temperature metaphase prediction to obtain a temperature metaphase prediction result, wherein the trained temperature metaphase prediction model is obtained by training a target loss function with physical constraint and an enhanced training data set, and comprises the following steps: calculating a temperature theoretical value based on a base temperature, the total cloud amount, and the net radiant flux; Constructing a radiation balance constraint according to the temperature theoretical value and the 2m height temperature, wherein the radiation balance constraint comprises the following steps: ; Constructing a soil atmosphere thermal coupling constraint based on a current 2m altitude temperature, a previous time step 2m altitude temperature, the east-west direction component wind speed, and the north-south direction component wind speed, comprising: ; Wherein, the Representing the radiation balance constraint of the beam, The temperature at the height of 2m is indicated, The base temperature is indicated as such, Representing the radiation influence coefficient(s), Indicating the net radiant flux that is to be emitted, The total amount of cloud is represented by the total amount of cloud, The density of the soil is indicated and, Represents the specific heat capacity of the material, Representing the thermal coupling constraints of the soil atmosphere, Indicating the current 2m altitude temperature, A 2m height temperature representing the previous time step, The east-west component wind speed representing a 10m altitude, Representing the current 2m height temperature pair The partial derivative of the shaft, Representing the component wind speed in the north-south direction of 10m altitude, Representing the current 2m height temperature pair The partial derivative of the shaft, The time step is represented by a time step, Taking a positive integer, wherein the positive integer is taken, Representing the norm square; Carrying out weighted summation on the radiation balance constraint and the soil atmosphere thermal coupling constraint to construct a total physical loss function; Adding the total physical loss function and the mean square error loss function to construct a target loss function; Training a temperature metaphase prediction model through the target loss function to obtain a trained temperature metaphase prediction model; inputting the preprocessed data into a trained temperature metaphase prediction model for temperature metaphase prediction to obtain a temperature metaphase prediction result, wherein, Dividing the preprocessed data by adopting a sliding window, inputting an input tensor corresponding to each variable divided by the sliding window into a ConvLSTM D layer through an independent channel, and extracting space-time characteristics corresponding to each variable; Removing redundant information in the space-time characteristics corresponding to each variable to obtain space-time characteristics corresponding to each variable after removing the redundant information; performing feature fusion on the space-time features of all variables after redundant information is removed, so as to obtain fusion features; And carrying out mid-temperature prediction based on the fusion characteristics to obtain mid-temperature prediction results.
2. The method for medium-term temperature prediction based on physical constraints according to claim 1, wherein preprocessing the data in the historical dataset including a plurality of variables to obtain preprocessed data comprises: Filling the missing value of the data in the historical data set to obtain filled data; normalizing the filled data to obtain normalized data; and splicing and fusing the normalized data and the static land-sea mask data to obtain preprocessed data.
3. The physical constraint-based mid-temperature prediction method of claim 1, wherein the trained mid-temperature prediction model is trained from an objective loss function with physical constraints and an enhanced training data set, comprising: preprocessing the data in the acquired training data set to obtain a preprocessed training data set; Carrying out random horizontal overturn on the input tensor corresponding to each variable in the preprocessed training data set to obtain a new tensor after random horizontal overturn; randomly and vertically overturning the input tensor corresponding to each variable in the preprocessed training data set to obtain a new tensor after random and vertical overturning; carrying out random time migration on the input tensor corresponding to each variable in the preprocessed training data set to obtain a new tensor after random time migration; adding the new tensor after random horizontal overturn, the new tensor after random vertical overturn and the new tensor after random time shift into the training data set after preprocessing to obtain an enhanced training data set; And training the middle temperature prediction model through a target loss function with physical constraint and the enhanced training data set to obtain a trained middle temperature prediction model.
4. The physical constraint-based mid-temperature prediction method of claim 1, wherein the calculating a temperature theoretical value based on a base temperature, the total cloud cover, and the net radiant flux comprises: Acquiring soil density and specific heat capacity; calculating a temperature theoretical value based on the base temperature, the total cloud cover, the net radiant flux, the soil density, and the specific heat capacity.
5. The physical constraint-based mid-temperature prediction method of claim 1, wherein said constructing a radiation balance constraint from said temperature theoretical value and said 2 m-height temperature comprises: Performing difference calculation on the temperature theoretical value and the 2m height temperature to obtain a difference result; and carrying out norm square solving on the difference result to construct radiation balance constraint.
6. A medium temperature prediction system based on physical constraints, the system comprising: The data preprocessing unit is used for preprocessing data in a historical data set containing a plurality of variables to obtain preprocessed data, wherein the variables comprise the acquired 2m height temperature, total cloud cover, east-west direction component wind speed, north-south direction component wind speed and net radiation flux; The temperature metaphase prediction unit is used for inputting the preprocessed data into a trained temperature metaphase prediction model for temperature metaphase prediction to obtain a temperature metaphase prediction result, wherein the trained temperature metaphase prediction model is obtained by training a target loss function with physical constraint and an enhanced training data set, and comprises the following steps: calculating a temperature theoretical value based on a base temperature, the total cloud amount, and the net radiant flux; Constructing a radiation balance constraint according to the temperature theoretical value and the 2m height temperature, wherein the radiation balance constraint comprises the following steps: ; Constructing a soil atmosphere thermal coupling constraint based on a current 2m altitude temperature, a previous time step 2m altitude temperature, the east-west direction component wind speed, and the north-south direction component wind speed, comprising: ; Wherein, the Representing the radiation balance constraint of the beam, The temperature at the height of 2m is indicated, The base temperature is indicated as such, Representing the radiation influence coefficient(s), Indicating the net radiant flux that is to be emitted, The total amount of cloud is represented by the total amount of cloud, The density of the soil is indicated and, Represents the specific heat capacity of the material, Representing the thermal coupling constraints of the soil atmosphere, Indicating the current 2m altitude temperature, A 2m height temperature representing the previous time step, The east-west component wind speed representing a 10m altitude, Representing the current 2m height temperature pair The partial derivative of the shaft, Representing the component wind speed in the north-south direction of 10m altitude, Representing the current 2m height temperature pair The partial derivative of the shaft, The time step is represented by a time step, Taking a positive integer, wherein the positive integer is taken, Representing the norm square; Carrying out weighted summation on the radiation balance constraint and the soil atmosphere thermal coupling constraint to construct a total physical loss function; Adding the total physical loss function and the mean square error loss function to construct a target loss function; Training a temperature metaphase prediction model through the target loss function to obtain a trained temperature metaphase prediction model; inputting the preprocessed data into a trained temperature metaphase prediction model for temperature metaphase prediction to obtain a temperature metaphase prediction result, wherein, Dividing the preprocessed data by adopting a sliding window, inputting an input tensor corresponding to each variable divided by the sliding window into a ConvLSTM D layer through an independent channel, and extracting space-time characteristics corresponding to each variable; Removing redundant information in the space-time characteristics corresponding to each variable to obtain space-time characteristics corresponding to each variable after removing the redundant information; performing feature fusion on the space-time features of all variables after redundant information is removed, so as to obtain fusion features; And carrying out mid-temperature prediction based on the fusion characteristics to obtain mid-temperature prediction results.
7. An electronic device comprising at least one control processor and a memory communicatively coupled to the at least one control processor, the memory storing instructions executable by the at least one control processor to enable the at least one control processor to perform the physical constraint based mid-temperature prediction method of any one of claims 1 to 5.
8. A computer-readable storage medium storing computer-executable instructions for causing a computer to perform the medium-term temperature prediction method based on physical constraints according to any one of claims 1 to 5.

Description

Temperature medium-term prediction method based on physical constraint Technical Field The application relates to the technical field of temperature prediction, in particular to a temperature medium-term prediction method based on physical constraint. Background The 2m temperature is the atmospheric temperature at the height of 2m from the ground surface in the weather science, is one of the reference indexes of global weather observation, and the medium-term prediction of the 2m temperature is also the key research direction of the weather science and the application field, and is widely applied to the fields of weather research, agriculture, traffic and the like. The 2m temperature short-term forecasting method commonly used at present mainly comprises a traditional numerical forecasting method and a pure data driving method based on a physical equation. However, the existing 2m temperature mid-term forecasting method cannot meet the requirements of the modern weather service on quick response and physical interpretability, and the forecasting accuracy of the 2m temperature mid-term forecasting is low. Disclosure of Invention The application aims to provide a temperature mid-term prediction method based on physical constraint, which can meet the requirements of modern weather service on quick response and physical interpretability and can improve the prediction accuracy of 2m temperature mid-term prediction. In a first aspect, an embodiment of the present application provides a method for mid-temperature prediction based on physical constraints, where the method includes: Preprocessing data in a historical data set containing a plurality of variables to obtain preprocessed data, wherein the variables comprise the acquired 2m height temperature, total cloud cover, east-west component wind speed, north-south component wind speed and net radiation flux; Inputting the preprocessed data into a trained temperature metaphase prediction model for temperature metaphase prediction to obtain a temperature metaphase prediction result, wherein the trained temperature metaphase prediction model is obtained by training a target loss function with physical constraint and an enhanced training data set, Dividing the preprocessed data by adopting a sliding window, inputting an input tensor corresponding to each variable divided by the sliding window into a ConvLSTM D layer through an independent channel, and extracting space-time characteristics corresponding to each variable; Removing redundant information in the space-time characteristics corresponding to each variable to obtain space-time characteristics corresponding to each variable after removing the redundant information; performing feature fusion on the space-time features of all variables after redundant information is removed, so as to obtain fusion features; And carrying out mid-temperature prediction based on the fusion characteristics to obtain mid-temperature prediction results. Compared with the prior art, the first aspect of the application has the following beneficial effects: the method comprises the steps of preprocessing data in a historical dataset comprising a plurality of variables, obtaining preprocessed data, wherein the variables comprise acquired 2m height temperature, total cloud quantity, east-west direction component wind speed, north-south direction component wind speed and net radiation flux, inputting the preprocessed data into a trained temperature metaphase prediction model for performing temperature metaphase prediction to obtain a temperature metaphase prediction result, training the trained temperature metaphase prediction model through a target loss function with physical constraints and an enhanced training dataset, dividing the preprocessed data by adopting a sliding window, inputting input tensor corresponding to each variable divided by the sliding window into a ConvLSTM D layer through an independent channel, extracting space-time characteristics corresponding to each variable, removing redundant information in the space-time characteristics corresponding to each variable, obtaining the space-time characteristics corresponding to each variable after removing the redundant information, performing characteristic fusion on the space-time characteristics corresponding to all the variables, obtaining the temperature metaphase prediction result, and performing temperature metaphase prediction based on the fusion characteristics. In this way, the variables with physical significance are comprehensively considered as input tensors, then the trained middle temperature prediction model is obtained through training of the target loss function with physical constraint and the enhanced training data set, the prediction accuracy of the trained middle temperature prediction model is improved, finally the temperature prediction is carried out through the trained middle temperature prediction model, the obtained middle temperature prediction resul