A power mode constrained drought prediction method and system

Abstract

The application discloses a kind of power mode constrained drought prediction method and system, the method is inputed into multi-branch drought prediction model with the dynamic mode data of target area, historical meteorological data, climate index data and static geographic environment data, and the SPI index of predicted forecast period is obtained;Multi-branch drought prediction model includes first to fourth feature extraction module, fusion mapping module and prediction module;First to fourth feature extraction module carries out feature extraction to dynamic mode data, historical meteorological data, climate index data and static geographic environment data respectively by neural network, fusion mapping module is spliced to the feature of first to fourth feature extraction module output and obtains fusion feature, and prediction module is predicted to obtain the SPI index of forecast period according to the fusion feature.The application comprehensively utilizes dynamic climate model prediction information, historical climate state information and regional environment information, realizes the spatiotemporal variation characteristic prediction of regional drought under different forecast period conditions.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an adaptive distribution method and system for multi-source data. Background Technology

[0002] Drought refers to a persistent lack of water resources in a region compared to normal conditions. It is characterized by its long duration, wide impact, and severe damage, making it one of the most destructive natural disasters. Accurate and reliable drought forecasts provide crucial preliminary information for developing scientific and effective drought response strategies and reducing disaster losses. Currently, regional drought prediction methods mainly include prediction methods based on dynamic climate models and data-driven prediction methods based on historical data statistical relationships.

[0003] Existing dynamical forecasting methods typically utilize global or regional climate models to simulate future meteorological elements such as precipitation and temperature, and further calculate drought indices such as the Standardized Precipitation Index (SPI) to achieve drought prediction. While these methods have a clear physical basis, their accuracy is significantly limited at regional scales, especially in the medium to long term, due to the complexity of atmospheric systems, initial field errors, model structure biases, and uncertainties in regional precipitation processes.

[0004] In recent years, with the development of artificial intelligence technology, data-driven methods based on machine learning and deep learning have been gradually applied to regional drought prediction. These methods can effectively uncover nonlinear relationships in historical meteorological data, improving prediction accuracy. However, existing deep learning-based data-driven methods primarily rely on historical observation data for feature learning. Due to the "data-driven" nature of deep learning, the model's predictive ability is limited by the distribution of training data. When future climate conditions deviate from historical climate conditions, the model's generalization and extrapolation capabilities significantly decrease. Furthermore, existing deep learning drought prediction models typically use conventional loss functions such as mean squared error for training, focusing primarily on the overall prediction error and applying uniform weights to all samples. In actual climate events, the number of normal climate state samples is usually significantly greater than the number of drought event samples, leading to normal state samples dominating the model training process while drought event samples contribute insufficiently, resulting in insufficient learning ability for drought events. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a drought prediction method and system constrained by dynamic models, which solves the problems of existing deep learning models that are highly dependent on historical samples, have insufficient generalization and extrapolation capabilities when facing future climate changes, have insufficient contribution of drought event samples under skewed sample distribution conditions, and have limited learning ability of models for drought processes, thereby improving the accuracy of regional drought prediction under different forecast periods.

[0006] Technical solution: The drought prediction method constrained by dynamic models according to the present invention includes the following steps:

[0007] The input factors for the target area are obtained, including dynamic model data, historical meteorological data, climate index data, and static geographic environment data; the dynamic model data includes forecast precipitation and temperature data, and the historical meteorological data includes historical meteorological observation data and historical SPI index data.

[0008] The input factors are input into the multi-branch drought prediction model to obtain the predicted lead time SPI index;

[0009] The multi-branch drought prediction model includes a first to fourth feature extraction module, a fusion mapping module, and a prediction module. The first to fourth feature extraction modules extract features from dynamic model data, historical meteorological data, climate index data, and static geographic environment data through neural networks, respectively. The fusion mapping module concatenates the features output by the first to fourth feature extraction modules to obtain fused features. The prediction module predicts the SPI index for the foreseeability period based on the fused features.

[0010] Furthermore, the first feature extraction module extracts features from the input dynamic pattern data using a convolutional neural network to obtain the first feature; the second feature extraction module extracts features from the input historical meteorological data using a convolutional neural network to obtain the second feature; the third feature extraction module extracts features from the input climate index data using a fully connected network to obtain the third feature; and the fourth feature extraction module extracts features from the input static geographic environment data using a convolutional neural network to obtain the fourth feature.

[0011] Furthermore, the loss function of the multi-branch drought prediction model consists of a drought index regression error term and a drought event identification loss term; wherein, the drought index regression error term is calculated based on the difference between historical SPI index data and predicted SPI index data, and the drought event identification loss term is calculated based on the probability of drought occurrence.

[0012] Furthermore, the loss function of the multi-branch drought prediction model ;

[0013] Drought Index Regression Error Term ;

[0014] Identifying Loss Items in Drought Events ;

[0015] in, Let Ω be the spatial mask and Ω be the target region. This is historical SPI index data. To predict SPI index data, It is an adaptive weighting for drought level, based on Determine the weight values; This is the drought probability mapping function. It is the sigmoid function; Labeling drought events The threshold for determining the occurrence of drought; This is a smoothing control function.

[0016] Furthermore, the calculation method for the adaptive weight of drought level is as follows:

[0017] .

[0018] Furthermore, before inputting the input factors into the multi-branch drought prediction model, the method further includes preprocessing the input factors, including:

[0019] The dynamic model data is downscaled, and the statistical relationship between historical dynamic model simulation results and observational data is used to correct the bias in the dynamic model data for the forecast period.

[0020] The present invention discloses a drought prediction system constrained by dynamic patterns, comprising:

[0021] The input factor acquisition unit acquires input factors for the target area, including dynamic model data, historical meteorological data, climate index data, and static geographic environment data; the dynamic model data includes forecast precipitation and temperature data, and the historical meteorological data includes historical meteorological observation data and historical SPI index data.

[0022] The drought index prediction unit is used to input the input factors into a multi-branch drought prediction model to obtain the predicted SPI index for the foreseeable period. The multi-branch drought prediction model includes a first to fourth feature extraction module, a fusion mapping module, and a prediction module. The first to fourth feature extraction modules extract features from dynamic model data, historical meteorological data, climate index data, and static geographic environment data through neural networks, respectively. The fusion mapping module concatenates the features output by the first to fourth feature extraction modules to obtain fused features. The prediction module predicts the SPI index for the foreseeable period based on the fused features.

[0023] The computer-readable storage medium of the present invention stores a computer program, which, when executed by a processor, implements the drought prediction method constrained by dynamic patterns.

[0024] The electronic device of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the drought prediction method constrained by dynamic patterns.

[0025] The computer program product of the present invention includes a computer program that, when executed by a processor, implements the drought prediction method constrained by dynamic patterns.

[0026] Beneficial Effects: Compared with existing technologies, the advantages of this invention lie in utilizing the extrapolation advantages of dynamic climate models for future states, combining them with the nonlinear feature learning capabilities of deep learning, and simultaneously improving the model's response to drought events through an optimization mechanism designed for differences in drought samples. This intelligent drought prediction method enhances the accuracy of regional drought prediction under different forecast periods. Specifically, this invention has the following advantages:

[0027] (1) This invention introduces the information on the evolution of future climate state provided by dynamic climate models into the deep learning prediction framework, and realizes the synergistic effect of dynamic model information and data-driven model through soft constraints. It makes full use of the simulation and extrapolation capabilities of dynamic climate models for future climate processes, makes up for the problem that traditional deep learning models rely on historical samples and have insufficient generalization ability when facing changes in future climate state, and improves the stability and prediction reliability of the model under different forecast conditions.

[0028] (2) In view of the problem that the drought index sample distribution has obvious skewness, the present invention designs an adaptive weighted joint loss function based on drought level. By dynamically adjusting the contribution weight of samples of different drought levels in the model training process, the dominant role of normal climate state samples in the model optimization process is reduced, the model's learning ability for low-frequency and moderate to severe drought events is enhanced, and the model's ability to characterize drought processes, especially severe drought states, is improved.

[0029] (3) This invention directly uses the standardized drought index (SPI) as the prediction target to achieve end-to-end prediction from climate information input to future drought state output, avoiding the error accumulation that may occur in the traditional method of first predicting meteorological variables and then calculating the drought index, and improving the efficiency and applicability of the regional scale drought prediction process.

[0030] (4) This invention can comprehensively utilize dynamic climate model prediction information, historical climate status information and regional environmental information to predict the spatiotemporal variation characteristics of regional drought under different forecast periods, and provide more reliable technical support for regional drought monitoring, early warning and water resource management. Attached Figure Description

[0031] Figure 1 This is a flowchart of the drought prediction method in Embodiment 1 of the present invention.

[0032] Figure 2 This is a diagram of the multi-branch prediction model architecture of Embodiment 2 of the present invention.

[0033] Figure 3 This is a comparison diagram of the model effects in Embodiment 2 of the present invention. Detailed Implementation

[0034] This invention introduces information on future climate state evolution provided by dynamic climate models into a deep learning prediction framework. It leverages the simulation capabilities of dynamic models to provide soft constraints for data-driven models, enhancing their adaptability to future climate states that exceed the distribution range of historical samples. Simultaneously, addressing the issue of significant skewness in the distribution of drought index samples, an adaptive weighted loss optimization mechanism based on drought samples is constructed. By dynamically adjusting the contribution weights of samples with different drought levels during model training, the model pays more attention to the characteristics of drought samples, thereby improving its ability to learn and predict drought events.

[0035] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0036] Example 1

[0037] like Figure 1 As shown, the drought prediction method constrained by dynamic patterns according to the present invention includes the following steps.

[0038] Step 1: Acquisition and preprocessing of multi-source drought prediction data.

[0039] We acquire dynamic model data, historical meteorological observation data, large-scale climate index data, and regional static geographic environment data for the target area, and construct a multi-source information dataset for regional drought prediction.

[0040] Dynamical model data includes meteorological variables such as precipitation and temperature for the foreseeable future, providing information on future climate evolution. To address the issues of low spatial resolution and systematic bias in dynamical climate model output data, a statistical downscaling method is first employed to transform the spatial scale of the dynamical model output, mapping the coarse-resolution dynamical model data to a high-resolution grid scale for the target region, such as bilinear interpolation. Subsequently, a bias correction method is used, utilizing the statistical relationship between historical dynamical model simulation results and observed data, to correct the future dynamical model output results, thereby reducing systematic bias and improving the applicability of dynamical prediction information. This bias correction method includes, but is not limited to, linear correction and quantile mapping (QM), ensuring that the corrected dynamical model data maintains statistical consistency with regional observational data.

[0041] Among them, historical meteorological observation data includes historical precipitation, temperature and historical drought index information, which are used to characterize the region's previous water status and drought evolution memory characteristics.

[0042] Among them, large-scale climate index data includes ocean-atmosphere coupled climate anomaly indices, which are used to describe global and regional climate background changes.

[0043] Among them, static geographic environment data includes spatial attribute data such as digital elevation models (DEM), which are used to characterize the spatial differences of the underlying surface in the region.

[0044] Spatial matching was performed on data with different spatial resolutions, and spatial interpolation was used to unify the downscaled and bias-corrected data of the dynamic model to the grid scale of the target prediction region. Subsequently, a spatial mask was constructed based on the study area to filter effective grids in the target region, avoiding interference from data outside the study area for model training. Finally, all input variables were standardized to eliminate dimensional differences between variables and improve model training stability. Simultaneously, based on the Standardized Precipitation Index (SPI) calculation method, SPI indices at different time scales were calculated using historical precipitation data, and the SPI index for the future target period was used as the target variable for model prediction, enabling direct prediction of future drought conditions.

[0045] Step 2: Construct a multi-branch drought prediction model under dynamic mode constraints.

[0046] The multi-branch drought prediction model includes the first to fourth feature extraction modules, namely the dynamic model prediction information feature extraction module, the historical meteorological information feature extraction module, the climate index feature extraction module, and the static environmental information feature extraction module; it also includes a fusion mapping module and a prediction module.

[0047] Among them, the dynamic model prediction information feature extraction module takes the precipitation and temperature prediction data of different forecast periods output by the dynamic climate model as input, and extracts future meteorological driving features through a convolutional neural network to reflect the impact of future climate conditions on drought development.

[0048] Among them, the historical observation information feature extraction module takes precipitation, temperature and historical SPI index sequences of the target area at several time scales in the past as input, and extracts historical meteorological state features through convolutional neural networks to capture the water memory effect in the early stage of drought development in the region.

[0049] Among them, the climate index feature extraction module takes large-scale climate teleconnection index as input and extracts climate background features through a fully connected network to characterize the regulatory role of ocean-atmosphere anomalies in the formation and development of regional drought.

[0050] Among them, the static geographic information feature extraction module takes the regional digital elevation model and other static environmental factors as input, and extracts spatial background features through a convolutional neural network to describe the impact of regional spatial heterogeneity on drought distribution.

[0051] The fusion mapping module encodes the high-dimensional features output by the multiple feature extraction modules and performs feature concatenation and nonlinear mapping in the fusion layer to realize the input of future GCM information and the fusion of historical information.

[0052] The prediction module inputs the fused features into the prediction network, further extracts comprehensive features through multiple convolutional layers and fully connected layers, and outputs the standardized precipitation index (SPI) prediction results for the target area under different forecast periods.

[0053] Specifically, considering the significant skewness in the drought index sample distribution, this invention constructs a drought adaptive weighted joint loss function. The loss function (L) of the multi-branch drought prediction model is composed of the regression error term of the drought index (SPI3). ) and drought event identification loss items ( )composition:

[0054] ;

[0055] The regression error term for the drought index is as follows:

[0056] ;

[0057] in, This represents a spatial mask, and Ω represents the target region. These are measured values, i.e., historical SPI index data. These are simulated values, i.e., predicted SPI index data;

[0058] ω is the adaptive weighting function for drought level, and ω is used as a model hyperparameter.

[0059] The loss items identified in the drought event are as follows:

[0060] ;

[0061] in, The drought probability mapping function is constructed using the sigmoid function. It is the sigmoid function; Labeling drought events The threshold for determining the occurrence of drought (usually -1); For smoothing control constants, As a model hyperparameter, it is manually tuned to select the optimal one.

[0062] Step 3, Model Training and Optimization.

[0063] A training dataset consisting of historical samples is constructed, and the multi-source input factors are used as model inputs, while the measured SPI index of the corresponding period is used as model outputs to train the multi-source information fusion prediction model.

[0064] The backpropagation algorithm is used to optimize the model parameters, and the Adam optimization algorithm, learning rate adjustment strategy and early stopping strategy are combined to improve the model training stability and generalization ability.

[0065] To address the issues of limited drought sample size and difficulty in predicting extreme drought events, low SPI value samples are weighted during model training to improve the model's ability to identify drought conditions, especially severe drought events.

[0066] Step 4, Regional drought prediction application.

[0067] By using the trained model to input real-time updated multi-source meteorological information of the target area, the spatial distribution results of the regional drought index for different forecast periods are output, enabling the prediction of the occurrence, development trend and impact range of regional drought, and providing a predictive basis for drought monitoring and early warning, water resource allocation and agricultural disaster prevention and mitigation.

[0068] Example 2

[0069] This embodiment verifies the effectiveness of the method described in this invention through a specific case.

[0070] Taking drought prediction in a certain region as an example, the drought prediction method in this embodiment includes the following steps.

[0071] S1, multi-source data acquisition.

[0072] The dynamic climate model predicts data using output products from the Climate Forecast System version 2 (CFSv2). Precipitation and temperature forecasts for different lead times from January 1981 to December 2020 are selected to describe future climate condition changes. This embodiment sets multiple lead times, including drought prediction tasks one month, two months, and three months in advance.

[0073] Historical meteorological observation data includes monthly precipitation and temperature data from January 1961 to December 2020, which are used to calculate the historical SPI index and characterize the historical moisture change status of the region.

[0074] Large-scale climate index data includes climate index data that can reflect the anomaly state of ocean-atmosphere coupling, and are used to characterize the impact of global and regional climate background changes on drought formation processes.

[0075] Static geographic environment data includes spatial attribute data such as the target area digital elevation model (DEM), which is used to characterize regional topographic differences and spatial heterogeneity.

[0076] S2, Data Preprocessing.

[0077] The dynamic climate model prediction data obtained in step S1 is rescaled using bilinear interpolation. The gridded data is then subjected to quantile mapping for bias correction.

[0078] The data obtained in step S1 is rescaled using bilinear interpolation. Grid dataset.

[0079] S3, Model Architecture Design.

[0080] like Figure 2 As shown, the historical meteorological information feature extraction module in this embodiment uses spatial data such as precipitation, temperature, and SPI index from past periods as input, and employs a convolutional neural network (CNN) to extract spatial features of historical meteorological conditions. This module includes two two-dimensional convolutional layers (Conv2D), followed by a batch normalization layer and a non-linear activation function layer, and further reduces the feature dimensionality through a max pooling layer.

[0081] The convolutional structures of the dynamic pattern prediction information feature extraction module and the static environment information feature extraction module are the same as those of the historical meteorological information feature extraction module, both including two two-dimensional convolutional layers, normalization layers, activation layers, and max pooling layers.

[0082] The dynamic model prediction information feature extraction module takes future precipitation and temperature prediction data output by the dynamic climate model as input, and uses a convolutional neural network to extract future climate driving features to characterize future changes in meteorological conditions. The static environment information feature extraction module takes digital elevation model and regional spatial attribute information as input, extracts regional spatial background features through convolutional layers, and obtains static environment features through batch normalization, activation function and pooling operations.

[0083] The climate index feature extraction module takes large-scale climate indices and their lagged information as input, extracts climate background features through a fully connected network (Dense), and transforms them into spatial feature representations through feature transformation.

[0084] After the four feature extraction modules complete their respective feature extraction, they enter the fusion mapping module. The features are then fused through a concatenation layer to form a comprehensive feature set containing historical state, future dynamic predictions, climate background, and spatial environment information. The fused features are further input into a convolutional fusion module, which learns the nonlinear relationships between different information sources through two convolutional operations (including one convolutional layer (Conv2D), one batch normalization layer (BN), and a ReLU activation function) and Dropout regularization. Subsequently, an upsampling layer (UpSampling2D) is used to restore the feature spatial resolution, and two more convolutional operations (including one convolutional layer (Conv2D), one batch normalization layer (BN), and a ReLU activation function) and one output layer generate the future spatial distribution of the SPI3 drought index for the target region. The output layer uses a 1×1 convolutional layer to achieve grid-by-grid SPI3 index prediction.

[0085] S4, Model Training.

[0086] The completed training samples are input into the multi-branch drought prediction model, and the measured SPI index is used as the supervision target to train the model parameters.

[0087] During model training, the Adam optimization algorithm is used to update network parameters, and an early stopping strategy is adopted to avoid model overfitting, taking into account the changes in validation set error.

[0088] To address the issue of the low frequency of drought events in historical samples, the weights of samples with different drought levels are adjusted to improve the model's ability to identify moderate to severe drought conditions.

[0089] After training, a multi-branch drought prediction model capable of predicting future SPI indices is obtained.

[0090] S5, Model Application and Evaluation.

[0091] The root mean square error and relative error were used to evaluate the model performance. Specific results are shown in Table 1. Figure 3 As shown.

[0092] Table 1. Prediction performance of the dual-branch drought prediction model of the present invention in a certain region.

[0093]

[0094] The results show that the method proposed in this invention can effectively utilize future information from dynamic models and take into account sample imbalance to achieve accurate forecasting of regional drought.

[0095] Example 3

[0096] The present invention discloses a drought prediction system constrained by dynamic patterns, comprising:

[0097] The input factor acquisition unit acquires input factors for the target area, including dynamic model data, historical meteorological data, climate index data, and static geographic environment data; the dynamic model data includes forecast precipitation and temperature data, and the historical meteorological data includes historical meteorological observation data and historical SPI index data.

[0098] The drought index prediction unit is used to input the input factors into a multi-branch drought prediction model to obtain the predicted SPI index for the foreseeable period. The multi-branch drought prediction model includes a first to fourth feature extraction module, a fusion mapping module, and a prediction module. The first to fourth feature extraction modules extract features from dynamic model data, historical meteorological data, climate index data, and static geographic environment data through neural networks, respectively. The fusion mapping module concatenates the features output by the first to fourth feature extraction modules to obtain fused features. The prediction module predicts the SPI index for the foreseeable period based on the fused features.

[0099] Example 4

[0100] The computer-readable storage medium of the present invention stores a computer program, which, when executed by a processor, implements the drought prediction method constrained by dynamic patterns.

[0101] Example 5

[0102] The electronic device of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the drought prediction method constrained by dynamic patterns.

[0103] The processor is used to execute a computer program stored in memory to implement the various steps in the methods described in the above embodiments.

[0104] Example 6

[0105] The computer program product of the present invention includes a computer program that, when executed by a processor, implements the drought prediction method constrained by dynamic patterns.

[0106] The processor is used to execute a computer program stored in memory to implement the various steps in the methods described in the above embodiments.

Claims

1. A drought prediction method constrained by dynamic models, characterized in that, Includes the following steps: The input factors for the target area are obtained, including dynamic model data, historical meteorological data, climate index data, and static geographic environment data; the dynamic model data includes forecast precipitation and temperature data, and the historical meteorological data includes historical meteorological observation data and historical SPI index data. The input factors are input into the multi-branch drought prediction model to obtain the predicted lead time SPI index; The multi-branch drought prediction model includes a first to fourth feature extraction module, a fusion mapping module, and a prediction module. The first to fourth feature extraction modules extract features from dynamic model data, historical meteorological data, climate index data, and static geographic environment data through neural networks, respectively. The fusion mapping module concatenates the features output by the first to fourth feature extraction modules to obtain fused features. The prediction module predicts the SPI index for the foreseeability period based on the fused features.

2. The drought prediction method constrained by dynamic models according to claim 1, characterized in that, The first feature extraction module extracts features from the input dynamic pattern data using a convolutional neural network to obtain the first feature; the second feature extraction module extracts features from the input historical meteorological data using a convolutional neural network to obtain the second feature; and the third feature extraction module extracts features from the input climate index data using a fully connected network to obtain the third feature. The fourth feature extraction module extracts the fourth feature from the input static geographic environment data using a convolutional neural network.

3. The drought prediction method constrained by dynamic models according to claim 1, characterized in that, The loss function of the multi-branch drought prediction model consists of a drought index regression error term and a drought event identification loss term; wherein, the drought index regression error term is calculated based on the difference between historical SPI index data and predicted SPI index data, and the drought event identification loss term is calculated based on the probability of drought occurrence.

4. The drought prediction method constrained by dynamic models according to claim 3, characterized in that, The loss function of the multi-branch drought prediction model ; Drought Index Regression Error Term ; Identifying Loss Items in Drought Events ; in, Let Ω be the spatial mask and Ω be the target region. This is historical SPI index data. To predict SPI index data, It is an adaptive weighting for drought level, based on Determine the weight values; This is the drought probability mapping function. It is the sigmoid function; Labeling drought events The threshold for determining the occurrence of drought; This is a smoothing control function.

5. The drought prediction method constrained by dynamic patterns according to claim 4, characterized in that, The method for calculating the adaptive weight of drought level is as follows: 。 6. The drought prediction method constrained by dynamic models according to claim 1, characterized in that, Before inputting the input factors into the multi-branch drought prediction model, the method further includes preprocessing the input factors, including: The dynamic model data is downscaled, and the statistical relationship between historical dynamic model simulation results and observational data is used to correct the bias in the dynamic model data for the forecast period.

7. A drought prediction system constrained by dynamic models, characterized in that, include: The input factor acquisition unit acquires input factors for the target area, including dynamic model data, historical meteorological data, climate index data, and static geographic environment data; the dynamic model data includes forecast precipitation and temperature data, and the historical meteorological data includes historical meteorological observation data and historical SPI index data. The drought index prediction unit is used to input the input factors into a multi-branch drought prediction model to obtain the predicted SPI index for the foreseeable period. The multi-branch drought prediction model includes a first to fourth feature extraction module, a fusion mapping module, and a prediction module. The first to fourth feature extraction modules extract features from dynamic model data, historical meteorological data, climate index data, and static geographic environment data through neural networks, respectively. The fusion mapping module concatenates the features output by the first to fourth feature extraction modules to obtain fused features. The prediction module predicts the SPI index for the foreseeable period based on the fused features.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the drought prediction method constrained by dynamic patterns according to any one of claims 1-7.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is loaded into the processor, it implements the drought prediction method constrained by dynamic patterns according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the drought prediction method constrained by dynamic patterns according to any one of claims 1-7.

Images