Reference crop evapotranspiration prediction method and device, medium and equipment

Abstract

The invention discloses a reference crop evapotranspiration prediction method and device, a medium and equipment, and relates to the field of agrometeorology, water resource management and intelligent computation.The reference crop evapotranspiration prediction method comprises the steps that a historical reference crop evapotranspiration ET0 sequence and historical day-by-day meteorological factor data are obtained, and day-by-day meteorological data are determined according to the day-by-day meteorological factor data; empirical mode EMD decomposition is carried out on the historical ET0 sequence, and historical ET0 multi-scale time-frequency features are determined; serializing the day-by-day meteorological data, and carrying out weight dynamic allocation and fusion on the day-by-day meteorological data processed by adopting a multi-head self-attention mechanism to generate meteorological factor data dynamic characteristics; performing time sequence feature analysis and fusion on the meteorological factor data dynamic features and the historical ET0 multi-scale time-frequency features to generate meteorological factor features and historical ET0 features; and carrying out dynamic weighted fusion on the meteorological factor features and the historical ET0 features, determining fused features, carrying out time sequence feature analysis on the fused features, and obtaining a reference crop evapotranspiration ET0 prediction result.

Description

Technical Field

[0001] The present invention relates to the intersecting fields of agricultural meteorology, water resources management and intelligent computing, and in particular to a reference crop evapotranspiration prediction method, device, medium and equipment. Background Art

[0002] Reference Crop Evapotranspiration (ET0) is an important indicator for measuring crop water demand. Accurate prediction of ET0 is of great significance for optimizing irrigation schedules and improving water resource utilization efficiency.

[0003] The FAO-56 Penman-Monteith (PM) formula is the standard method for calculating ET0 (ET0) recommended by the Food and Agriculture Organization (FAO). However, its practical application in ET0 prediction places extremely stringent demands on meteorological factors, requiring forecasts of key meteorological factors such as solar radiation, relative humidity, wind speed, sunshine hours, and temperature. This makes ET0 predictions based on the FAO-56 PM formula difficult to widely apply. In recent years, deep learning models such as long short-term memory (LSTM) networks and convolutional neural networks have been introduced to the field of ET0 prediction. These models use a limited number of meteorological factors to predict ET0, alleviating the limitations of ET0 prediction based on the FAO-56 PM formula. However, these methods still have the following drawbacks: existing studies employ a stacked architecture of multiple deep learning models and fail to design feature extraction strategies tailored to meteorological factors and historical ET0 data, resulting in forecasts that fail to accurately reflect actual conditions. Summary of the Invention

[0004] The present invention provides a reference crop evapotranspiration prediction method, device, medium, and equipment to solve the above-mentioned problem existing in the prior art, namely, how to improve the accuracy of crop evapotranspiration prediction in the prior art. The present invention provides a reference crop evapotranspiration prediction method, which includes: Obtain the historical reference crop evapotranspiration ET0 series and historical daily meteorological factor data, and determine the daily meteorological data based on the daily meteorological factor data; The empirical mode decomposition (EMD) of the historical reference crop evapotranspiration (ET0) series was performed to determine the multi-scale time-frequency characteristics of the historical ET0. The daily meteorological data is serialized and the multi-head self-attention mechanism is used to dynamically assign weights to the serialized daily meteorological data and fuse them to generate dynamic features of meteorological factor data. The dynamic characteristics of meteorological factor data and the multi-scale time-frequency characteristics of historical ET0 were analyzed and fused in the forward and reverse time series to generate the corresponding meteorological factor characteristics and historical ET0 characteristics. The gating mechanism was used to dynamically weight the meteorological factor characteristics and historical ET0 characteristics to determine the fused characteristics. The fused characteristics were then analyzed in the forward and reverse time series to obtain the reference crop evapotranspiration ET0 prediction results.

[0005] Optionally, performing empirical mode decomposition (EMD) on the historical reference crop evapotranspiration (ET0) series to determine the multi-scale time-frequency characteristics of the historical ET0 specifically includes: Obtain all local maximum and minimum points in the historical reference crop evapotranspiration ET0 series, interpolate the maximum and minimum points respectively using the interpolation method to generate the upper and lower envelopes, and determine the mean of the upper and lower envelopes; Determine the mean envelope according to the mean of the upper envelope and the lower envelope; Subtract the mean envelope from the original ET0 sequence to determine the intermediate signal, and extract the IMF component based on the intermediate signal; The residual signal is determined by subtracting the IMF component from the original ET0 sequence; Based on the IMF components and residual signals, the multi-scale time-frequency characteristics of historical ET0 are determined.

[0006] Optionally, the reference crop evapotranspiration ET0 prediction result is obtained using a feature learning network, and the feature learning network specifically includes: The first BiLSTM module is used to perform forward and reverse time series feature analysis and fusion on the dynamic features of meteorological factor data to generate corresponding meteorological factor features; The second BiLSTM module is used to perform forward and reverse time series feature analysis and fusion on the multi-scale time-frequency features of historical ET0 to generate corresponding historical ET0 features; The gated network is used to dynamically weight the fusion of meteorological factor features and historical ET0 features to determine the fused features; The third BiLSTM module is used to perform forward and reverse time series feature analysis on the fused features to obtain the reference crop evapotranspiration ET0 prediction results.

[0007] Optionally, the gating mechanism is used to dynamically weight the meteorological factor features and the historical ET0 features to determine the fused features, specifically including: Based on the Softmax function, the following formula is used to generate normalized weight distribution: ; in, Z 1 is the meteorological factor characteristic,Z 2 is the historical ET0 characteristic, g =[ g 1, g 2], and g 1+ g 2=1; The fused features are obtained using the following formula: ; in, Z fuse is the fused feature.

[0008] Optionally, before performing forward and reverse time series feature analysis on the fused features, random perturbations are added to the fused features through the Dropout layer to perform a zeroing operation on the feature values.

[0009] Optionally, before serializing the daily meteorological data, preprocessing the daily meteorological data may be performed, specifically including: Missing value handling, outlier detection and normalization.

[0010] Optionally, the daily meteorological factor data specifically includes: Maximum temperature, minimum temperature, wind speed, relative humidity and sunshine hours.

[0011] The present invention provides a reference crop evapotranspiration prediction device, comprising: An acquisition module is used to obtain the historical reference crop evapotranspiration ET0 series and historical daily meteorological factor data, and determine the daily meteorological data based on the daily meteorological factor data; The decomposition module is used to perform empirical mode EMD decomposition on the historical reference crop evapotranspiration ET0 series to determine the multi-scale time-frequency characteristics of historical ET0; The fusion module is used to serialize the daily meteorological data and use the multi-head self-attention mechanism to dynamically assign weights to the serialized daily meteorological data and fuse them to generate dynamic features of meteorological factor data; The prediction module is used to perform forward and reverse time series feature analysis and fusion on the dynamic features of meteorological factor data and the multi-scale time-frequency features of historical ET0, respectively, to generate corresponding meteorological factor features and historical ET0 features; a gating mechanism is used to perform dynamic weighted fusion of meteorological factor features and historical ET0 features, determine the fused features, perform forward and reverse time series feature analysis on the fused features, and obtain the reference crop evapotranspiration ET0 prediction results.

[0012] The present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the reference crop evapotranspiration prediction method is implemented.

[0013] The present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned reference crop evapotranspiration prediction method is implemented.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides a reference crop evapotranspiration prediction method, which breaks the limitation of traditional models relying only on a single data source by utilizing effective information in meteorological data and historical ET0 data, and realizes data complementarity and fusion; by inputting the dynamic characteristics of meteorological factor data and the multi-scale time-frequency characteristics of historical ET0 into two independent BiLSTM modules for forward and reverse time series feature analysis and fusion, the information loss caused by premature feature fusion is avoided, the characteristics of each data modality are fully utilized, and the various factors affecting the reference crop evapotranspiration and their interaction mechanisms can be more comprehensively grasped, so that the actual situation can be more accurately reflected during prediction, and the prediction accuracy is effectively improved; in addition, the present invention realizes the dynamic fusion of meteorological factor characteristics and historical ET0 characteristics by introducing a gating mechanism, and adaptively allocates weights, which effectively solves the problem that the feature fusion method in the traditional model is fixed and cannot adapt to dynamic changes in data, so that the crop evapotranspiration prediction is more in line with the actual situation, thereby effectively improving the crop evapotranspiration prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0016] Figure 1 1 is a flow chart of a reference crop evapotranspiration prediction method provided by an embodiment of the present invention; Figure 2 The IMFi(t) component and the residual term after EMD decomposition provided by the embodiment of the present invention are R n (t) graph; Figure 3 Schematic diagram of the multi-head self-attention mechanism MHSA module provided by an embodiment of the present invention; Figure 4 Schematic diagram of a BiLSTM module provided by an embodiment of the present invention; Figure 5 Schematic diagram of an ETO prediction method using a feature learning network according to an embodiment of the present invention; Figure 6 This is a comparison chart of the prediction errors between the model of the present invention and the traditional model; Figure 7Schematic diagram of computer equipment for the reference crop evapotranspiration prediction method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0018] The following describes in detail the technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems using specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The following embodiments of the present invention are described in conjunction with the accompanying drawings.

[0019] Figure 1 FIG. 1 is a flow chart of a reference crop evapotranspiration prediction method provided by an embodiment of the present invention. Figure 1 As shown, this embodiment shows a reference crop evapotranspiration prediction method, including: S1: Obtain the historical reference crop evapotranspiration ET0 series and historical daily meteorological factor data, and determine the daily meteorological data based on the daily meteorological factor data.

[0020] For example, daily meteorological factor data for N days in the test area (the first meteorological station, the second meteorological station, and the third meteorological station) are collected, including meteorological factor data such as the maximum temperature (Tmax), the minimum temperature (Tmin), the wind speed (u2), the relative humidity (RH), and the number of sunshine hours (n). The data are organized into daily meteorological data {MF(t), t=1,2,..,N}; the historical ET0 benchmark value of the corresponding date is calculated using the following formula: ; in, ∆ is the slope of the saturated water vapor pressure curve; R n is the net radiation on the crop surface (MJ / m 2 ·day); G is the soil heat flux (MJ / m 2 ·day); r is the hygrometer constant (kPa / ℃); T mean is the daily average humidity (℃); u 2 is the wind speed (m / s); es is the saturated water vapor pressure (kPa); e a is the actual water vapor pressure (kPa).

[0021] And organize it into a historical N-day ET0 data set {ET0(t)| t=1,2,..,N}; Exemplarily, preprocessing daily meteorological data may include: (1) Missing value processing: Check the collected meteorological data {MF(t)}. If missing values ​​are found, use linear interpolation to fill in the missing values.

[0022] (2) Outlier detection: The box plot method is used to detect and process outliers in the meteorological data {MF(t)}, and outliers that are beyond the reasonable range are replaced with the median of the variable.

[0023] (3) Data normalization: Normalize the data and normalize all data to the range [0,1].

[0024] For example, by setting the prediction step size k, the daily meteorological data of the previous Nk days {MF(t), t=1,2,..,Nk} and the historical ET0 sequence of the previous Nk days {ET0(t)|t=1,2,..,Nk} can be intercepted.

[0025] S2: Perform empirical mode decomposition (EMD) on the historical reference crop evapotranspiration (ET0) series to determine the multi-scale time-frequency characteristics of historical ET0.

[0026] like Figure 2 As shown, ET0(t) is decomposed by EMD to obtain a set of intrinsic mode functions {IMFi(t), i=1,2,..} components and residual terms R n (t). The specific operations are as follows:

[0027] ①Extreme point extraction: Identify the local maximum and minimum values ​​of the signal and construct the upper and lower envelopes; ② Mean calculation: Fit the envelope mean m1(t) through cubic spline interpolation; ③Component screening: iterative calculation , where x(t) is the original input ET0(t), until the intrinsic mode function analytical condition is met, that is, the difference between the number of extreme points and the number of zero-crossing points is ≤1, and the envelope mean is zero, the IMF1(t) component is obtained. If not, then h 1 (t) is taken as the new input x(t), and repeat ① to ③;

[0028] ④Calculate the total residual R 1(t), R1(t)=x(t)-IMF1(t), let R 1(t) is used as the new x(t), and steps ① to ④ are repeated to obtain the new IMFi(t) component. ⑤ Decomposition termination condition: residual term R n The decomposition is stopped when (t) is a monotonic function or its amplitude is less than the threshold. At the same time, the obtained IMFi(t) components are optimized, and the permutation entropy value is calculated for each IMFi(t) component. The components with permutation entropy values ​​higher than the threshold are selected as valid inputs. The specific calculation method of permutation entropy (PE) is as follows:

[0029] Among them, m is the embedding dimension, p v is the probability of occurrence of the symbol sequence.

[0030] A permutation entropy threshold is set, and components with permutation entropy values ​​higher than the threshold are screened out as valid inputs to remove high-frequency noise interference and reduce computational redundancy.

[0031] S3: Serialize the daily meteorological data, and use the multi-head self-attention mechanism to dynamically assign and fuse the weights of the serialized daily meteorological data to generate dynamic features of meteorological factor data.

[0032] For example, the daily meteorological data {MF(t), t=1, 2, …, Nk} can be used as the input of the multi-head self-attention mechanism MHSA. Figure 3 As shown in Figure 2, each self-attention mechanism of the multiple self-attention mechanisms MHSA independently captures the correlation between different positions in the input data {MF(t), t=1, 2, …, Nk} and extracts global features. The specific calculation process of the self-attention mechanism is as follows:

[0033] ; Among them, the query Q, key K, and value V vectors are the results of multiplying the input data MF(t) with the trainable weight matrices WQ, WK, and WV respectively.

[0034] Calculate the dot product between the query vector Q and the key vector K to get the attention score matrix. To prevent the gradient from disappearing, divide the dot product result by the scaling factor ( d k is the vector dimension). The attention scores are then normalized using the Softmax function to obtain the attention weight matrix.

[0035] Weighted summation: MHSA stitches together the various self-attention mechanisms and performs a weighted sum on their results to generate the final output representation. The calculation method is as follows ; Among them, Concat(·) represents the concatenation operation, L is the number of self-attention, and W is the linear mapping weight of the final output.

[0036] S4: Perform forward and reverse time series feature analysis on the dynamic characteristics of meteorological factor data and the multi-scale time-frequency characteristics of historical ET0 respectively and fuse them to generate corresponding meteorological factor characteristics and historical ET0 characteristics; use a gating mechanism to dynamically weighted fuse the meteorological factor characteristics and historical ET0 characteristics to determine the fused characteristics, perform forward and reverse time series feature analysis on the fused characteristics, and obtain the reference crop evapotranspiration ET0 prediction results.

[0037] Optionally, a gating mechanism is introduced to dynamically weight the meteorological factor features and the historical ET0 features to determine the fused features, specifically including: Through the Softmax function, the normalized weight distribution is generated using the following formula: ; Among them, Z1 is the meteorological factor feature, Z2 is the historical ET0 feature, g=[g1,g2], and g1+g2=1. The fused feature is obtained using the following formula: ; Among them, Z fuse is the fused feature.

[0038] like Figure 4 As shown in Figure 2, for the input feature vector of each time step, BiLSTM transmits information in both forward and reverse directions.

[0039] Input processing: The dynamic features of meteorological factor data {MF(t)} extracted by the multi-head self-attention mechanism are used as the input {x t , t=1,2,…, Nk}, then its output Z1={Z 1,t , t=1,2,…,Nk} is: ; in, and It is the result of BiLSTM forward and backward information learning. Definition and h t ,but

[0040] For example, the calculation steps are the same as those of BiLSTM1. The second BiLSTM module (BiLSTM2 module) is used to learn the historical ET0 multi-scale time-frequency features {IMFi(t)} extracted by EMD, and its output is Z2={Z 2,t ,t=1, 2,…,Nk}.

[0041] like Figure 5 As shown in the figure, it is a schematic diagram of the reference crop evapotranspiration ET0 prediction through the feature learning network. The meteorological factor feature Z1 learned by the BiLSTM1 module and the historical ET0 feature Z2 learned by the BiLSTM2 module are dynamically weighted and fused with the heterogeneous features output by the two modules through the designed gating mechanism.

[0042] Through this function, the gating mechanism generates a normalized weight distribution: g = [g1, g2], and satisfies g1 + g2 = 1. The fused feature is expressed as: .

[0043] For example, the constructed Dropout layer is constructed by merging the features Z fuse Apply random perturbations and randomly set some of the eigenvalues ​​to zero to effectively prevent overfitting.

[0044] Construct the third BiLSTM module (BiLSTM3 module) to learn the feature Z after Dropout processing drop To output ET0 prediction results , expressed as: ; S5: Input the future daily meteorological data and the historical ET0 series into the trained reference crop evapotranspiration prediction model to obtain the future reference crop evapotranspiration ET0 prediction results.

[0045] Exemplarily, {MF(t), t=1,2,..,Nk}, {ET0(t)| t=1,2,..,Nk} can be used as training input data for the model of the present invention.

[0046] Initialize model parameters: Randomly initialize all trainable parameters in the model, including the weight matrices WQ, WK, and WV in the MHSA module and the weight matrices WF, WI, and WO in the BiLSTM module.

[0047] Definition of loss function: The mean square error (MSE) is used as the loss function to measure the difference between the model's predicted value and the true value. The calculation formula of the loss function is:

[0048] ; Where n is the number of samples Nk, y t is ET0(t), ET0 result for training Training process: The training set data is input into the model in batches, the prediction value and loss value are calculated through forward propagation, and then the gradient is calculated using the backpropagation algorithm to update the model parameters.

[0049] Select an optimization algorithm: Use the Adam optimization algorithm to train the model and control the convergence speed and stability of the model.

[0050] For example, when predicting ET0 for the next k days, the daily meteorological data {MF(t), t=N-k+1, N-k+2, .., N} from day N-k+1 to day N, and the data {ET0(t), t=N-k+1, N-k+2, .., N} from day N-k+1 to day N can be input into the reference crop evapotranspiration prediction model of dual-modal parallel feature learning and dynamic fusion proposed in the present invention, and the reference crop evapotranspiration {ET0(t), t=1, 2, .., k} for the next k days can be output.

[0051] Model testing: The performance of the model is measured by calculating indicators such as mean square error (MSE), mean absolute error (MAE), and coefficient of determination (R²). The calculation formula for MAE is:

[0052] The formula for calculating R² is: in, is the average value of {ET0(t)| t=1,2,..,N}.

[0053] like Figure 6 (a) to Figure 6As shown in (c), the test error distribution ranges (based on the first to third quartiles) of the reference crop evapotranspiration prediction model proposed in this invention on three meteorological station datasets are -0.8 to 0.8, -0.9 to 0.9, and -0.9 to 1.0, respectively. In comparison, the test error distribution ranges of the prior art MHSA-BiLSTM model on the same datasets are -0.8 to 0.9, -1.1 to 1.1, and -1.1 to 1.1, respectively. The test results show that the prediction error distribution range of the proposed model is smaller than that of the comparison model MHSA-BiLSTM at all stations, demonstrating its superior prediction accuracy and stability.

[0054] In the ET0 prediction task, the proposed model significantly outperforms the MHSA-BiLSTM model, the BiLSTM model, and the MHSA model. This performance advantage stems from the novel architectural design of the proposed model, centered on the implementation of a dual-modal parallel feature learning mechanism. This effectively integrates key information from both modalities. Comparative experiments further validate the effectiveness of the model components:

[0055] 10. The test error distribution range of the BiLSTM model is larger than that of the model of the present invention and the MHSA-BiLSTM model, highlighting the importance of introducing MHSA for feature learning.

[0056] 11. The performance of the reference crop evapotranspiration prediction model, MHSA-BiLSTM model, and BiLSTM model proposed in this invention are significantly better than that of the single MHSA model, demonstrating the core role of BiLSTM in time series prediction.

[0057] In summary, the proposed reference crop evapotranspiration prediction method, leveraging its unique bimodal parallel feature learning architecture, synergistically leverages the feature learning capabilities of the multi-head self-attention mechanism and the temporal modeling advantages of BiLSTM, ultimately demonstrating excellent performance in the ET0 prediction task. The test error distribution results for all sites consistently show that the model's prediction accuracy ranks as follows: the proposed model > the MHSA-BiLSTM model > the BiLSTM model > the MHSA model, fully demonstrating the significant advantages of the proposed model in accurately predicting ET0.

[0058] The above is a reference crop evapotranspiration prediction method provided in one or more embodiments of this specification. Based on the same idea, this specification also provides a corresponding reference crop evapotranspiration prediction device, including: An acquisition module is used to obtain the historical reference crop evapotranspiration ET0 series and historical daily meteorological factor data, and determine the daily meteorological data based on the daily meteorological factor data; The decomposition module is used to perform empirical mode EMD decomposition on the historical reference crop evapotranspiration ET0 series to determine the multi-scale time-frequency characteristics of historical ET0; The fusion module is used to serialize the daily meteorological data and use the multi-head self-attention mechanism to dynamically assign weights to the serialized daily meteorological data and fuse them to generate dynamic features of meteorological factor data; The prediction module is used to perform forward and reverse time series feature analysis and fusion on the dynamic features of meteorological factor data and the multi-scale time-frequency features of historical ET0, respectively, to generate corresponding meteorological factor features and historical ET0 features; a gating mechanism is used to perform dynamic weighted fusion of meteorological factor features and historical ET0 features, determine the fused features, perform forward and reverse time series feature analysis on the fused features, and obtain the reference crop evapotranspiration ET0 prediction results.

[0059] The specific definitions of the reference crop evapotranspiration prediction device can be found in the definitions of the reference crop evapotranspiration prediction method described above and will not be further elaborated here. Each module within the aforementioned reference crop evapotranspiration prediction device may be implemented in whole or in part via software, hardware, or a combination thereof. Each of these modules may be embedded in or independent of a processor within a computer device in hardware form, or may be stored in a computer device memory in software form, allowing the processor to invoke and execute the corresponding operations of each module.

[0060] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program can be used to execute the reference crop evapotranspiration prediction method provided above.

[0061] The present invention also provides Figure 7 The structural diagram of the computer equipment shown in FIG. Figure 7 As shown, at the hardware level, the computer device includes a processor, an internal bus, a network interface, memory, and non-volatile storage, and may also include other hardware required for operations. The processor reads the corresponding computer program from the non-volatile storage into the internal memory and then runs it to implement the reference crop evapotranspiration prediction method provided in the above embodiment.

[0062] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A reference crop evapotranspiration prediction method, characterized in that: include: Obtain the historical reference crop evapotranspiration ET0 series and historical daily meteorological factor data, and determine the daily meteorological data based on the daily meteorological factor data; The empirical mode decomposition (EMD) of the historical reference crop evapotranspiration (ET0) series was performed to determine the multi-scale time-frequency characteristics of the historical ET0. The daily meteorological data is serialized and the multi-head self-attention mechanism is used to dynamically assign weights to the serialized daily meteorological data and fuse them to generate dynamic features of meteorological factor data. The dynamic characteristics of meteorological factor data and the multi-scale time-frequency characteristics of historical ET0 were analyzed and fused in the forward and reverse time series to generate the corresponding meteorological factor characteristics and historical ET0 characteristics. The gating mechanism was used to dynamically weight the meteorological factor characteristics and historical ET0 characteristics to determine the fused characteristics. The fused characteristics were then analyzed in the forward and reverse time series to obtain the reference crop evapotranspiration ET0 prediction results.

2. The reference crop evapotranspiration prediction method according to claim 1, wherein: The empirical mode EMD decomposition of the historical reference crop evapotranspiration ET0 series is performed to determine the multi-scale time-frequency characteristics of the historical ET0, specifically including: Obtain all local maximum and minimum points in the historical reference crop evapotranspiration ET0 series, interpolate the maximum and minimum points respectively using the interpolation method to generate the upper and lower envelopes, and determine the mean of the upper and lower envelopes; Determine the mean envelope according to the mean of the upper envelope and the lower envelope; Subtract the mean envelope from the original ET0 sequence to determine the intermediate signal, and extract the IMF component based on the intermediate signal; The residual signal is determined by subtracting the IMF component from the original ET0 sequence; Based on the IMF components and residual signals, the multi-scale time-frequency characteristics of historical ET0 are determined.

3. The reference crop evapotranspiration prediction method according to claim 1, wherein: The reference crop evapotranspiration ET0 prediction result is obtained by using a feature learning network, which specifically includes: The first BiLSTM module is used to perform forward and reverse time series feature analysis and fusion on the dynamic features of meteorological factor data to generate corresponding meteorological factor features; The second BiLSTM module is used to perform forward and reverse time series feature analysis and fusion on the multi-scale time-frequency features of historical ET0 to generate corresponding historical ET0 features; The gated network is used to dynamically weight the fusion of meteorological factor features and historical ET0 features to determine the fused features; The third BiLSTM module is used to perform forward and reverse time series feature analysis on the fused features to obtain the reference crop evapotranspiration ET0 prediction results.

4. The reference crop evapotranspiration prediction method according to claim 1, wherein: The gate control mechanism is used to dynamically weight the meteorological factor features and the historical ET0 features to determine the fused features, specifically including: Based on the Softmax function, the following formula is used to generate normalized weight distribution: ； in, Z 1 is the meteorological factor characteristic, Z 2 is the historical ET0 characteristic, g =[ g 1, g 2], and g 1+ g 2=1; The fused features are obtained using the following formula: ； in, Z fuse is the fused feature.

5. The reference crop evapotranspiration prediction method according to claim 3, wherein: Before performing forward and reverse time series feature analysis on the fused features, random perturbations are added to the fused features through the Dropout layer to perform a zeroing operation on the feature values.

6. The reference crop evapotranspiration prediction method according to claim 1, wherein: Before serializing the daily meteorological data, preprocessing the daily meteorological data is performed, specifically including: Missing value handling, outlier detection and normalization.

7. The reference crop evapotranspiration prediction method according to claim 1, wherein: The daily meteorological factor data specifically includes: Maximum temperature, minimum temperature, wind speed, relative humidity and sunshine hours.

8. A reference crop evapotranspiration prediction device, characterized in that: include: An acquisition module is used to obtain the historical reference crop evapotranspiration ET0 series and historical daily meteorological factor data, and determine the daily meteorological data based on the daily meteorological factor data; The decomposition module is used to perform empirical mode EMD decomposition on the historical reference crop evapotranspiration ET0 series to determine the multi-scale time-frequency characteristics of historical ET0; The fusion module is used to serialize the daily meteorological data and use the multi-head self-attention mechanism to dynamically assign weights to the serialized daily meteorological data and fuse them to generate dynamic features of meteorological factor data; The prediction module is used to perform forward and reverse time series feature analysis and fusion on the dynamic features of meteorological factor data and the multi-scale time-frequency features of historical ET0, respectively, to generate corresponding meteorological factor features and historical ET0 features; a gating mechanism is used to perform dynamic weighted fusion of meteorological factor features and historical ET0 features, determine the fused features, perform forward and reverse time series feature analysis on the fused features, and obtain the reference crop evapotranspiration ET0 prediction results.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the reference crop evapotranspiration prediction method according to any one of claims 1 to 7 is implemented.

10. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for predicting the reference crop evapotranspiration according to any one of claims 1 to 7 is implemented.

Images