A Maize Irrigation and Fertilization Decision-making Method and System Based on Future Meteorological Data

Abstract

The present application discloses a method and system for maize irrigation and fertilization decision-making based on future meteorological data, which relates to the fields of agricultural informatization and intelligent agriculture. The method includes: constructing a future meteorological prediction model based on the combined deep learning network of Informer + TSLANet + TimesBlock; using this model to extract multi-dimensional features of historical meteorological data for a long time series and capture long-term dependence relationships, and predicting future meteorological data; predicting maize yields through the DSSAT model, and then formulating an irrigation and fertilization plan for maize using the Monte Carlo + NSGA-III double-layer combined multi-objective optimization algorithm, and obtaining the optimal irrigation and fertilization plan after optimization. The present application enhances the representation ability of the future meteorological prediction model, improves the accuracy of future meteorological data prediction, and further improves the accuracy and foresight of maize yield prediction and irrigation and fertilization decision-making prediction.

Description

Technical Field

[0001] This application relates to the fields of agricultural informatization and intelligent agriculture technologies, and particularly to a method and system for maize irrigation and fertilization decision-making based on future meteorological data. Background Art

[0002] Maize is widely planted in agriculture as an important food and feed crop. Due to its high yield, wide adaptability, and economic value, it is commonly cultivated globally. As one of the crops with the largest planting area and highest yield in the world, maize not only provides basic food for billions of people globally but is also an important industrial raw material and energy crop. Irrigation and fertilization have become key measures to ensure high maize yields. Compared with relatively extensive irrigation methods, not only is water resource wasted, but the risk of soil nitrogen loss is also increased. In addition, unreasonable use of nitrogen fertilizers not only has a negative impact on agriculture and soil ecology but may also lead to excessive nitrate content in groundwater, threatening human health.

[0003] Existing meteorological prediction processes have problems such as poor representation ability and inaccurate prediction results. As a result, most existing irrigation and fertilization decisions are analyzed and formulated based on historical data, and the decision-making schemes lack real-time and forward-looking nature, making it difficult to adapt to the impact of climate change on maize planting. Moreover, research on irrigation and fertilization mainly focuses on individual irrigation decisions or individual fertilization decisions, which have certain limitations in practical applications and are difficult to simultaneously consider water resource utilization efficiency and fertilizer use efficiency. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for maize irrigation and fertilization decision-making based on future meteorological data, which enhances the representation ability of the future meteorological prediction model, improves the accuracy of future meteorological data prediction, and thus improves the accuracy and forward-looking nature of maize yield prediction and irrigation and fertilization decision-making prediction.

[0005] To achieve the above purpose, the following solutions are provided in this application.

[0006] In a first aspect, the present application provides a method for making decisions on corn irrigation and fertilization based on future meteorological data, and the method includes: obtaining historical meteorological data; constructing a future meteorological prediction model based on an Informer + TSLANet + TimesBlock combined deep learning network; using the future meteorological prediction model to extract multi-dimensional features of long time series and capture long-term dependence relationships from the historical meteorological data, and predicting future meteorological data; using the DSSAT model to predict the corn yield based on the future meteorological data to obtain the prediction result of the current-season corn yield; formulating an irrigation and fertilization plan for corn by using a Monte Carlo + NSGA-III double-layer combined multi-objective optimization algorithm based on the future meteorological data and the prediction result of the current-season corn yield, and obtaining the optimal irrigation and fertilization plan after optimization.

[0007] In a second aspect, the present application provides a computer system, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the method for making decisions on corn irrigation and fertilization based on future meteorological data described in any one of the above.

[0008] According to the specific embodiments provided by the present application, the following technical effects are disclosed.

[0009] In the present application, historical meteorological data is obtained; a future meteorological prediction model is constructed based on an Informer + TSLANet + TimesBlock combined deep learning network; the future meteorological prediction model is used to extract multi-dimensional features of long time series and capture long-term dependence relationships from the historical meteorological data, and future meteorological data is predicted; by using the TSLANet module to capture the long-term dependence relationships of long time series and using the TimesBlock module to extract the frequency domain features of meteorological data, the understanding ability of the Informer model for long time series is further improved, thereby enhancing the representation ability of the future meteorological prediction model and improving the accuracy of future meteorological data prediction; by using the DSSAT model to predict the corn yield based on the future meteorological data to obtain the prediction result of the current-season corn yield; an irrigation and fertilization plan for corn is formulated by using a Monte Carlo + NSGA-III double-layer combined multi-objective optimization algorithm based on the future meteorological data and the prediction result of the current-season corn yield, and the optimal irrigation and fertilization plan is obtained after optimization; taking the future meteorological data as the input, predicting the corn yield, and further obtaining the optimal irrigation and fertilization plan, which improves the real-time performance, accuracy, and forward-looking of corn yield prediction and irrigation and fertilization decision prediction, and takes into account the water resource utilization efficiency and fertilizer use efficiency. Description of the Drawings

[0010] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0011] Figure 1 Flow schematic of a corn irrigation and fertilization decision-making method based on future meteorological data provided by an embodiment of the present application Figure 1 。

[0012] Figure 2 Flow schematic of a corn irrigation and fertilization decision-making method based on future meteorological data provided by an embodiment of the present application Figure 2 。

[0013] Figure 3 Flow schematic diagram of the future meteorological prediction model provided by an embodiment of the present application.

[0014] Figure 4 Structure schematic diagram of the Informer+TSLANet+TimesBlock combined deep learning network provided by an embodiment of the present application.

[0015] Figure 5 Structure schematic diagram of the TSLANet module provided by an embodiment of the present application.

[0016] Figure 6 Structure schematic diagram of the TimesBlock module provided by an embodiment of the present application.

[0017] Figure 7 Flow schematic diagram of the DSSAT model training provided by an embodiment of the present application.

[0018] Figure 8 Flow schematic diagram of the Monte Carlo and NSGA-III double-layer combined multi-objective optimization algorithm provided by an embodiment of the present application.

[0019] Figure 9 Original meteorological combined optimization result diagram in 2018 provided by an embodiment of the present application.

[0020] Figure 10 Original meteorological combined optimization result diagram in 2019 provided by an embodiment of the present application.

[0021] Figure 11 Structure schematic diagram of a computer system provided by an embodiment of the present application. Detailed implementation manners

[0022] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0023] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0024] Embodiment 1, as Figures 1 - 2 shown, this embodiment provides a method for making corn irrigation and fertilization decisions based on future meteorological data, and the method includes the following steps.

[0025] S1. Obtain historical meteorological data.

[0026] Further, the meteorological data specifically includes: daily maximum temperature, daily minimum temperature, daily average solar radiation, and daily rainfall.

[0027] S2. Construct a future meteorological prediction model based on the Informer + TSLANet (Time Series Lightweight Adaptive Network) + TimesBlock combined deep learning network.

[0028] Further, Figures 3 - 4 as shown, the Informer + TSLANet + TimesBlock combined deep learning network includes a TSLANet module, a TimesBlock module, and an Informer model connected in series in sequence.

[0029] In the actual application process, the Informer + TSLANet + TimesBlock combined deep learning model is an innovative time series analysis model. By integrating the Informer model, TSLANet, and TimesBlock technology, it significantly improves the ability and efficiency to process time series data, fully utilizes the advantages of each technology, and processes the complexity and dynamics of time series data through a unified architecture. Among them, the Informer model is used to handle long time series prediction problems. It has an efficient self-attention mechanism that can capture long-term dependencies, build and train the Informer model, and make a preliminary prediction of meteorological data. The TSLANet module is added to the Informer model. After embedding encoding, it captures the relationships between multi-dimensional data of meteorological data and adds data features. The TimesBlock module is placed after TSLANet and further extracts the frequency domain features of meteorological data through multi-dimensional convolution to improve the model's understanding ability of time series data.

[0030] First, as Figure 5 shown, the application of the TSLANet layer: The present invention first uses the TSLANet layer for preliminary data processing. The TSLANet layer structure includes an interactive convolution block and an adaptive spectral block, and these blocks are equipped with layer normalization processing, enabling the model to effectively identify and analyze periodic and non-periodic patterns in time series data. The design of this layer not only improves the ability to process local features of data but also optimizes the data flow and interpretability of information.

[0031] Specifically, the TSLANet module consists of two main parts: an interactive convolution block (InteractiveConvolution Block, ICB) and an adaptive spectral block (Adaptive Spectral Block, ASB). These two parts are connected and processed through layer normalization (Layer Norm). Through the collaborative work of ICB and ASB, TSLANet not only enhances the local feature extraction ability of time series data but also optimizes the processing ability of global frequency domain features to enhance the feature extraction and processing ability of time series data.

[0032] 1) Composition and function of the Interactive Convolution Block (ICB): The ICB contains two one-dimensional convolutional layers, which are used to extract local features in time series data. One-dimensional convolutional layers are particularly suitable for processing time series data because they can effectively capture the temporal dependencies in the sequence. In addition, each convolutional layer is followed by a GELU (Gaussian Error Linear Unit) activation function, which helps introduce non-linearity and enables the model to learn more complex data patterns. Finally, the residual connection in the ICB (represented by an addition operation through cross-connection) helps preserve the original information of the input signal, reduce the information loss that may occur during training, and thus increase the training depth of the model without causing performance degradation.

[0033] 2) Composition and function of the Adaptive Spectrum Block (ASB): The ASB uses the Fast Fourier Transform (FFT) and the Inverse Fast Fourier Transform (IFFT), which enable the model to analyze and process signals in the frequency domain. Frequency domain analysis is particularly effective for identifying and processing periodic and aperiodic patterns in time series data. At the same time, through the adaptive masking technique, the ASB can dynamically adjust the importance distribution in the frequency domain, emphasizing important frequency components and suppressing unimportant frequency components. This mechanism improves the sensitivity of the model to important features in the signal, enhancing the prediction accuracy and generalization ability of the model.

[0034] Second, as Figure 6 shown, the integration of the TimesBlock technology: The present invention further integrates the TimesBlock module in the Informer structure to optimize the process of data flowing through the encoder and decoder. The TimesBlock module focuses on processing the temporal dimension changes in time series data. Through multi-scale analysis of the temporal dimension, it improves the accuracy and robustness of the model in predicting future trends.

[0035] As a multi-functional time processing unit, TimesBlock not only enhances the temporal features of the signal through multi-scale analysis but also optimizes the processing of periodicity and frequency components through FFT and Softmax. It improves the model's understanding depth and prediction ability for time series data.

[0036] 1) Fast Fourier Transform (FFT) component: This module first uses the FFT to transform the input time series data from the time domain to the frequency domain. This step is crucial for revealing the periodicity and main frequency components of the data, laying a foundation for subsequent data analysis and feature extraction.

[0037] 2) Parameter-Efficient Inception Block: After the FFT, the time series data is fed into a parameter-efficient Inception block. This block is specifically designed for feature extraction at different time scales, capturing a wide range of time series dynamics through multi-scale convolutional kernels. The high parameter efficiency of this design ensures that the model optimizes the use of computing resources while maintaining high performance.

[0038] 3) Adaptive Masking and Softmax Operation: To further emphasize the important frequency components in the data and suppress unimportant information, an adaptive masking technique is used to dynamically select key frequencies. The Softmax operation is then used to normalize the importance of these frequency components, ensuring the reliability and accuracy of the model output.

[0039] 4) Inverse Fast Fourier Transform (IFFT): After the important frequency components are selected and weighted, the IFFT is used to transform the processed frequency domain data back into the time domain. This reconstruction step is crucial to ensure that the model can accurately recover the signal in the time domain.

[0040] 5) Output Layer: The data that has undergone the above series of fine processing is finally output for direct use by Informer. This ensures that the data output from the TimesBlock maintains a high degree of information integrity and usability.

[0041] Third, the integration of the core technologies of Informer: The data processed by the TSLANet layer is input into the encoder and decoder of the Informer model. In this part, the present invention adopts a multi-head probabilistic sparse self-attention mechanism, which significantly enhances the model's ability to capture long-term dependencies while maintaining high computational efficiency. This method ensures the efficiency and powerful performance of the model in dealing with various complex time series tasks.

[0042] S3. Use the future meteorological prediction model to extract multi-dimensional features of long time series of historical meteorological data and capture long-term dependencies, and predict future meteorological data.

[0043] Furthermore, step S3 specifically includes the following steps.

[0044] S31. Substitute the historical meteorological data into the future meteorological prediction model.

[0045] S32. Use the TSLANet module to analyze the periodic and non-periodic patterns in the long time series of historical meteorological data. By processing the local features of the long time series, optimize the historical meteorological data from the aspects of data flow and information representation, and obtain the optimized historical meteorological data.

[0046] S33. Use the TimesBlock module to perform multi-scale analysis on the optimized historical meteorological data from the time dimension to obtain the historical meteorological data after multi-scale analysis.

[0047] S34. Substitute the historical meteorological data after multi-scale analysis into the Informer model, and use the multi-head probabilistic sparse self-attention mechanism to capture the long-term dependencies in the long time series of the historical meteorological data after multi-scale analysis, and predict the future meteorological data.

[0048] Optionally, the training process of the future meteorological prediction model is as follows.

[0049] Select meteorological data at different times as the training set and the validation set.

[0050] Use the meteorological data of the previous moment as the input and the meteorological data of the next moment as the output to train the future meteorological prediction model.

[0051] Use the validation set to evaluate the prediction performance of the future meteorological prediction model, adjust the model parameters to improve the prediction accuracy, and obtain the trained future meteorological prediction model.

[0052] S4. As Figure 7 shown, use the DSSAT model to predict the corn yield based on the future meteorological data, and obtain the prediction result of the current season's corn yield.

[0053] Furthermore, the training process of the DSSAT model is as follows.

[0054] Screen the training data and validation data for corn yield prediction based on the corn planting environment data; the corn planting environment data specifically includes: historical meteorological data, historical soil data, corn variety data, farmer planting profile data, and crop growth information.

[0055] Optimize the genetic parameters of the DSSAT model in a fitting and calibration manner based on the training data (simulate the corn growth process in the current season to generate simulation data at different growth stages), and use the validation data to evaluate the accuracy and stability of the DSSAT model, and finally determine the trained DSSAT model.

[0056] Furthermore, the soil data specifically includes: soil type, soil texture, and soil nutrient content; the corn variety data specifically includes: growth cycles of different corn varieties and yields of different corn varieties; the farmer planting profile data specifically includes: planting area, fertilization and irrigation methods, and planting density; the crop growth information specifically includes: growth stage, growth parameters, and yield records.

[0057] In the actual application process, the construction and verification process of the DSSAT model is as follows.

[0058] 1) Data import: Import the collected corn planting environment data (historical soil data, historical meteorological data, corn variety data, farmer planting profile data, and crop growth information) into the DSSAT model.

[0059] 2) Model calibration: Calibrate the DSSAT model using the corn planting environment data, and adjust the model parameters to fit the historical yields of corn crops. By adjusting the model parameters, make the simulated yields output by the model close to the historical actual yields to verify the accuracy of the model.

[0060] 3) Model verification: Use the corn planting environment data not involved in calibration to verify the model to ensure that the model can accurately predict the corn yields under different planting conditions. By comparing the model prediction values with the actual observed values, evaluate the accuracy and stability of the model.

[0061] S5. As Figure 8 shown, based on the future meteorological data and the predicted results of the current season's corn yields, use the Monte Carlo + NSGA-III double-layer combined multi-objective optimization algorithm to formulate the irrigation and fertilization plan for corn, and obtain the optimal irrigation and fertilization plan after optimization.

[0062] Further, step S5 specifically includes the following steps.

[0063] S51. Determine the optimization objectives; the optimization objectives are: maximizing yield, minimizing irrigation water volume, and minimizing fertilization amount.

[0064] Optionally, determining the optimization objectives is used to ensure that while increasing the corn yield and meeting the market demand, reducing the irrigation water volume and improving the water resource utilization efficiency; reducing the fertilizer usage amount and reducing environmental pollution.

[0065] S52. Based on the future meteorological data and the predicted results of the current season's corn yields, use the Monte Carlo model to adopt the method of repeated random sampling, and randomly generate all possible combinations of irrigation and fertilization dates with potential management strategies as the irrigation and fertilization plan for corn.

[0066] Optionally, the irrigation and fertilization plan for corn is to generate multiple irrigation and fertilization plans according to the random combinations of various irrigation and fertilization dates, covering all possible management strategies.

[0067] S53. Take the irrigation and fertilization plan for corn as the initial population of the NSGA-III multi-objective optimization algorithm, and obtain the optimal trade-off solution (Pareto optimal solution) between multiple optimization objectives after multiple iterative optimizations as the optimal irrigation and fertilization plan.

[0068] Furthermore, after multiple iterations of optimization, the optimal trade-off solution among multiple optimization objectives is obtained as the optimal irrigation and fertilization plan, which specifically includes: using the NSGA-III multi-objective optimization algorithm to evaluate the individuals in the initial population by non-dominated sorting and crowding distance calculation, generating a new generation of solutions through crossover and mutation operations, and using uniformly distributed reference points to maintain the uniformity of the population. Through repeated execution of the algorithm until the preset number of iterations is reached or the solution quality is stable, an optimal trade-off solution among multiple optimization objectives is provided for the decision maker.

[0069] In the actual application process, the actual operation process of step S5 is as follows.

[0070] 1) Determine the optimization objectives.

[0071] 2) Determine the number of irrigation times X1.

[0072] 3) Determine the number of nitrogen application times X2.

[0073] 4) Repeat.

[0074] 5) Generate X1 random dates for irrigation application.

[0075] 6) Generate X2 random dates for nitrogen fertilizer application.

[0076] 7) Run U-NSGA-III with the application rates on the given irrigation and nitrogen application dates as variables.

[0077] 8) Combine the solutions in step 6 with the solutions found in previous iterations.

[0078] 9) Delete non-optimal solutions from the combined solution set.

[0079] More detailed explanations are made for the following several parts.

[0080] First, determine the optimization objectives: Determine three optimization objectives of maximizing yield, minimizing irrigation water volume, and minimizing fertilization amount, ensuring to increase the corn yield, meet the market demand, while reducing the irrigation water consumption and improving the water resource utilization efficiency; reducing the fertilizer usage amount and reducing environmental pollution.

[0081] Second, Monte Carlo random irrigation generation: The Monte Carlo random algorithm is a computational method based on probability and statistical theory, aiming to solve physical and mathematical problems through repeated random sampling. By randomly generating a series of possible combinations of irrigation and fertilization dates, these combinations reflect potential management strategies and provide a basis for the next-step optimization.

[0082] Third, NSGA-III multi-objective optimization processing: NSGA-III is a multi-objective optimization algorithm suitable for dealing with complex optimization problems with three or more objectives. The algorithm starts from a randomly generated population and evaluates individuals through non-dominated sorting and crowding distance calculation to ensure the diversity and quality of solutions. NSGA-III uses crossover and mutation operations to generate a new generation of solutions, and at the same time introduces uniformly distributed reference points to maintain the uniformity of the population. The algorithm is repeatedly executed until a preset number of iterations is reached or the solution quality is stable, providing decision-makers with the optimal trade-off solutions between multiple optimization objectives. In this embodiment, NSGA-III multi-objective optimization is used at the lower layer to find the Pareto optimal solution set for each Monte Carlo randomly generated irrigation and fertilization date.

[0083] Further, before step S2, it also includes: preprocessing historical meteorological data; the preprocessing method is any one or more of smoothing to fill missing values and filtering to reduce noise.

[0084] Optionally, smoothing to fill missing values (interpolation) ensures data integrity by interpolating missing data; filtering to reduce noise specifically removes noise in the data through the application of filtering techniques to enhance data quality.

[0085] In the actual application process, as Figures 9 - 10 shown, the global Pareto optimal solution set after screening the original meteorological data in 2018 and 2019 reflects the relationship between irrigation amount, fertilization amount and crop yield, and the following changes can be observed from the figure.

[0086] (1) Yield and irrigation amount: As the irrigation amount increases, the overall yield also increases, but after reaching a certain irrigation level, the increase in yield tends to slow down or even stagnate. This indicates that there is a phenomenon of diminishing marginal effect in the utilization of irrigation resources, and appropriately controlling the irrigation amount can improve the water resource utilization efficiency.

[0087] (2) Yield and fertilization amount: Similarly, the increase in fertilization amount can promote the increase in yield, but excessive fertilization has limited contribution to yield and may lead to resource waste and environmental problems. Optimizing the fertilization amount is crucial for achieving efficient agriculture.

[0088] (3) Trade-off between irrigation amount and fertilization amount: In the high-yield solution set, it can be found that the irrigation amount and fertilization amount do not increase monotonically, but show a certain degree of complementarity and balance relationship. This indicates that when formulating agricultural management strategies, it is necessary to comprehensively consider the synergistic effect of irrigation and fertilization and find the best input combination.

[0089] The technical effects of this application are as follows.

[0090] This application obtains historical meteorological data; constructs a future meteorological prediction model based on the combined deep learning network of Informer + TSLANet + TimesBlock; uses the future meteorological prediction model to extract multi-dimensional features of long time series and capture long-term dependence relationships from the historical meteorological data, and predicts future meteorological data; by using the TSLANet module to capture the long-term dependence relationships of long time series and using the TimesBlock module to extract the frequency domain features of meteorological data, the understanding ability of the Informer model for long time series is further improved, thereby enhancing the representation ability of the future meteorological prediction model and improving the accuracy of future meteorological data prediction; by using the DSSAT model to predict the corn yield based on the future meteorological data, the prediction result of the current season's corn yield is obtained; based on the future meteorological data and the prediction result of the current season's corn yield, a MonteCarlo + NSGA-III double-layer combined multi-objective optimization algorithm is used to formulate the irrigation and fertilization plan for corn, and the optimal irrigation and fertilization plan is obtained after optimization; using the future meteorological data as the input, the corn yield is predicted, and the optimal irrigation and fertilization plan is further obtained, which improves the real-time performance, accuracy and forward-looking of corn yield prediction and irrigation and fertilization decision prediction, and takes into account the water resource utilization efficiency and fertilizer use efficiency. This application can not only predict in advance the impact of meteorological changes on corn growth, but also help farmers and agricultural managers formulate scientific and reasonable planting plans and response measures before the planting season, effectively reduce the uncertainty brought by climate change, and improve the stability and efficiency of agricultural production. Farmers can maximize the conservation of water resources and fertilizers while ensuring high yields, reduce production costs and environmental pollution, promote the sustainable development of agriculture, and enhance the economic and environmental benefits of agricultural production.

[0091] Example 2. This application also provides a computer system, which can be a server or a terminal, and its internal structure diagram can be as Figure 11As shown. The computer system includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer system is used to provide computing and control capabilities. The memory of the computer system includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer system is used to store video tag processing data. The input / output interface of the computer system is used to exchange information between the processor and external devices. The communication interface of the computer system is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for making decisions on corn irrigation and fertilization based on future meteorological data.

[0092] Those skilled in the art can understand that Figure 11 the structure shown in is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer system to which the solution of this application is applied. The specific computer system may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0093] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0094] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, 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, it should be considered as the scope described in this specification.

[0095] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the methods and core ideas of the present application; at the same time, for those of ordinary skill in the art, according to the ideas of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A corn irrigation and fertilization decision-making method based on future meteorological data, characterized in that: The corn irrigation and fertilization decision-making method based on future meteorological data includes: Get historical weather data; A future weather forecast model is constructed based on an Informer+TSLANet+TimesBlock combined deep learning network; the Informer+TSLANet+TimesBlock combined deep learning network comprises a TSLANet module, a TimesBlock module and an Informer model connected in series in sequence; The future meteorological data are predicted by using the future meteorological prediction model to extract long-term multidimensional features of historical meteorological data and capture long-term dependencies, which includes: substituting historical meteorological data into the future meteorological prediction model; using the TSLANet module to analyze the periodic and non-periodic patterns in the long-term series of historical meteorological data, and optimizing the historical meteorological data from the data flow and information representation levels by processing the local features of the long-term series to obtain the optimized historical meteorological data; using the TimesBlock module to perform multi-scale analysis on the optimized historical meteorological data from the time dimension to obtain the historical data after multi-scale analysis. Historical meteorological data; Substitute the historical meteorological data after multi-scale analysis into the Informer model, use the multi-head probabilistic sparse self-attention mechanism to capture the long-term dependency of the long time series in the historical meteorological data after multi-scale analysis, and predict the future meteorological data; The training process of the future meteorological prediction model is as follows: select meteorological data at different times as training sets and validation sets; Use the meteorological data at the previous moment as input and the meteorological data at the next moment as output to train the future meteorological prediction model; Use the validation set to evaluate the prediction performance of the future meteorological prediction model, adjust the model parameters to improve the prediction accuracy, and obtain a trained future meteorological prediction model; Based on future meteorological data, the DSSAT model is used to predict corn yield and obtain the corn yield forecast results for the current season; Based on future meteorological data and the forecast results of corn yield in the current season, the irrigation and fertilization plan of corn is formulated by using the Monte Carlo + NSGA-III double-layer combination multi-objective optimization algorithm, and the optimal irrigation and fertilization plan is obtained after optimization, which specifically includes: determining the optimization target; the optimization target is: maximizing yield, minimizing irrigation amount and minimizing fertilizer amount; based on future meteorological data and the forecast results of corn yield in the current season, using the Monte Carlo model to take the repeated random sampling method, and randomly generating all possible irrigation and fertilization date combinations with potential management strategies as the irrigation and fertilization plan of corn; using the irrigation and fertilization plan of corn as the initial population of the NSGA-III multi-objective optimization algorithm, and obtaining the optimal trade-off solution between multiple optimization targets after multiple iterative optimization, as the optimal irrigation and fertilization plan.

2. The method for corn irrigation and fertilization decision-making based on future meteorological data according to claim 1, characterized in that: The meteorological data specifically include: daily maximum temperature, daily minimum temperature, daily average solar radiation and daily rainfall.

3. The method for corn irrigation and fertilization decision-making based on future meteorological data according to claim 1, characterized in that: The training process of the DSSAT model is as follows: Training data and verification data for corn yield prediction are obtained based on corn planting environment data; the corn planting environment data specifically includes: historical meteorological data, historical soil data, corn variety data, farmer planting profile data and crop growth information; The genetic parameters of the DSSAT model were optimized by fitting and calibrating the training data, and the accuracy and stability of the DSSAT model were evaluated using the validation data, and finally the trained DSSAT model was determined.

4. The method for corn irrigation and fertilization decision-making based on future meteorological data according to claim 3 is characterized in that: The soil data specifically includes: soil type, soil texture and soil nutrient content; the corn variety data specifically includes: growth cycle of different corn varieties and yield of different corn varieties; the farmer planting profile data specifically includes: planting area, fertilization and irrigation methods and planting density; the crop growth information specifically includes: growth stage, growth parameters and yield records.

5. The method for corn irrigation and fertilization decision-making based on future meteorological data according to claim 1, characterized in that: After multiple iterations of optimization, the optimal trade-off solution among multiple optimization objectives is obtained as the optimal irrigation and fertilization plan, which specifically includes: The NSGA-III multi-objective optimization algorithm is used to evaluate the individuals in the initial population using non-dominated sorting and crowding distance calculation. A new generation of solutions is generated through crossover and mutation operations, and uniformly distributed reference points are used to maintain the uniformity of the population. The algorithm is repeatedly executed until the preset number of iterations is reached or the solution quality is stable, providing decision makers with the optimal trade-off solution between multiple optimization objectives.

6. The method for corn irrigation and fertilization decision-making based on future meteorological data according to claim 1, characterized in that: Before building a future weather forecast model based on the Informer+TSLANet+TimesBlock combined deep learning network, it also includes: The historical meteorological data is preprocessed; the preprocessing method is: any one or more of smoothing to fill missing values ​​and filtering to reduce noise.

7. A computer system comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the corn irrigation and fertilization decision-making method based on future meteorological data as described in any one of claims 1 to 6.

Images