Crop planting suggestion generation method adapting to different climate conditions

Abstract

The invention relates to the technical field of agricultural planting, and discloses a crop planting suggestion generation method adapting to different climate conditions. Real-time climate data are collected through a multi-source climate sensor, climate dynamic feature data are generated through a spatial-temporal feature fusion algorithm, and planting strategy parameters are obtained by inputting the climate dynamic feature data into a pre-trained mixed decision model. And constructing a multi-objective optimization model taking the crop adaptability matching degree and the resource utilization efficiency as objectives, and globally optimizing a planting scheme by adopting a hierarchical dynamic programming algorithm integrating fuzzy logic constraint and an adaptive decomposition strategy. A multi-level regulation and control framework comprising a decision-making layer, an adaptation layer and an execution layer is constructed, the decision-making layer generates a global planting sequence, the adaptation layer uses a sliding window optimization algorithm to adjust local parameters, the execution layer regulates and controls soil humidity and illumination intensity based on a robust feedback control algorithm, and finally a planting control instruction is output. And accurate and efficient climate adaptability planting decision is realized.

Description

Technical Field

[0001] The present invention relates to the field of agricultural planting technology, and in particular to a method for generating crop planting suggestions adapted to different climatic conditions. Background Art

[0002] Against the backdrop of global climate change, agricultural production faces unprecedented challenges. Climate conditions vary widely across regions, and traditional crop cultivation methods often lack the ability to precisely adapt to the changing climate. This leads to unstable crop yields and inefficient resource utilization.

[0003] Climatic conditions have a crucial impact on crop growth. Climatic factors such as temperature, humidity, precipitation, sunlight, and wind speed play a key role in the different growth stages of crops. Take temperature, for example: excessively high or low temperatures can affect crop metabolism, leading to slow growth or even death. In some areas, frequent extreme high temperatures can cause heat damage to crops that were previously suitable for cultivation, resulting in a significant drop in yield. Uneven rainfall distribution also poses numerous challenges to agricultural production. Excessive rainfall can easily lead to floods and damage the growing environment for crops, while insufficient rainfall can cause drought, affecting crop water absorption and photosynthesis.

[0004] Traditional methods of generating planting recommendations are mostly based on experience or simple statistical analysis of climate data. Farmers typically choose crop varieties and schedule planting times based on years of planting experience. This approach lacks the dynamic monitoring and analysis of real-time climate data. Even in some regions where simple climate data is used, it only calculates average values for basic data such as temperature and precipitation, failing to capture the dynamic changes in the climate. For example, when determining whether a region is suitable for growing a certain crop, only referring to the region's average temperature and precipitation over many years, while ignoring the fluctuations in temperature and precipitation in recent years, can result in poor growth of crops during actual planting due to their inability to adapt to abnormal climate conditions.

[0005] With the advancement of information technology and agricultural science and technology, precision agriculture has become the future direction of agricultural development. Precision agriculture requires providing precise planting recommendations based on diverse climate conditions, soil conditions, and crop growth requirements to maximize crop yields and optimize resource utilization. However, existing technologies struggle to meet the demands of precision agriculture. While multi-source climate sensors can collect vast amounts of climate data, effectively integrating and analyzing this data to extract valuable information for crop cultivation remains a challenge. Furthermore, existing planting decision models and optimization algorithms suffer from low computational efficiency and poor accuracy when dealing with complex climate conditions and multi-objective optimization problems. For example, when considering multiple objectives such as crop adaptability and resource efficiency, traditional decision models often struggle to balance conflicting objectives, resulting in inadequate planting plans.

[0006] Traditional control methods for soil moisture and light intensity are unable to adapt to complex and changing climates. Soil moisture and light intensity are important factors affecting crop growth, but existing control technologies lack real-time and precision. For example, traditional irrigation systems often operate at fixed times and water volumes, failing to adjust to real-time changes in soil moisture, which can easily lead to water waste or crop water shortages. Regarding light control, existing shading and supplemental lighting equipment cannot dynamically adjust light intensity based on weather changes and crop growth needs, impacting crop photosynthesis and growth. Summary of the Invention

[0007] The object of the present invention is to provide a method for generating crop planting suggestions that are adapted to different climatic conditions to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above-mentioned object, the present invention provides the following technical solution: a method for generating crop planting suggestions adapted to different climatic conditions, the method comprising: Collect real-time climate data of the target area through multi-source climate sensors; Performing multi-dimensional correlation analysis on the real-time climate data based on a spatiotemporal feature fusion algorithm to generate climate dynamic feature data; Inputting the climate dynamic characteristic data into a pre-trained hybrid decision model to generate planting strategy parameters; constructing a multi-objective optimization model based on the planting strategy parameters, wherein the multi-objective optimization model takes crop adaptability matching and resource utilization efficiency as optimization objectives, and globally optimizes the planting plan using a hierarchical dynamic programming algorithm, wherein the hierarchical dynamic programming algorithm integrates fuzzy logic constraints and an adaptive decomposition strategy; and outputting optimal planting plan data based on the multi-objective optimization model; A multi-level control framework is constructed based on the optimal planting plan data. The multi-level control framework includes a decision layer, an adaptation layer and an execution layer. The decision layer generates a global planting sequence based on the planting strategy parameters, the adaptation layer uses a sliding window optimization algorithm to dynamically adjust local planting parameters, and the execution layer realizes the regulation of soil moisture and light intensity based on a robust feedback control algorithm. The planting control instructions are output through the multi-level control framework to complete the generation of climate-adaptive planting suggestions.

[0009] Preferably, the multi-dimensional correlation analysis of the real-time climate data based on the spatiotemporal feature fusion algorithm to generate climate dynamic feature data includes: Perform spatiotemporal interpolation processing on temperature, humidity and precipitation data to generate a continuously distributed climate field; perform time-frequency transformation on wind speed data and light data to construct a frequency domain energy distribution map; Constructing a dual-modal feature extraction network, wherein the first modality uses wavelet transform to extract the periodic characteristics of the continuously distributed climate field, and the second modality uses non-negative matrix decomposition to extract the frequency domain correlation characteristics of the frequency domain energy distribution map; The periodic features and frequency domain correlation features are integrated through the feature cross-fusion layer to generate a joint feature matrix; the state transition modeling of the joint feature matrix is performed based on the hidden Markov model to output climate dynamic feature data including climate fluctuation trends and extreme event probabilities.

[0010] Preferably, the hybrid decision model adopts a staged decision tree structure to perform multi-objective trade-offs on the planting strategy based on a dynamic weight allocation mechanism; the staged decision tree structure includes: Construct a crop-environment relationship graph, where nodes include target crop nodes, climate factor nodes, soil condition nodes, and resource nodes. Node attributes include growth cycle requirements, environmental tolerance thresholds, and resource consumption coefficients. A two-stage decision-making mechanism is adopted. In the first stage, the priority weight of each node is calculated by the entropy weight method. In the second stage, similarity matching of historical planting cases is performed through grey correlation analysis. The node attributes are iteratively updated based on the multi-objective particle swarm optimization algorithm. The Pareto front is used to screen the optimal solution set for each objective function. The conflicting objectives are balanced through constraint relaxation and normalization, and the planting strategy parameters containing multi-dimensional trade-offs are finally output.

[0011] Preferably, the hierarchical dynamic programming algorithm integrating fuzzy logic constraints and adaptive decomposition strategy includes: The planting plan optimization problem is modeled as a multi-stage decision problem, where the decision variables include discrete crop variety selection variables and continuous irrigation amount variables. Initialize the fuzzy membership function and calculate the initial feasible region, and use the adaptive decomposition strategy to dynamically divide the problem subdomain according to the constraint complexity; During the state transfer process, the feasible solution with the highest membership degree is preferentially selected for expansion; during the pruning phase, a confidence interval model is constructed based on the historical solution set, and interval algebra is used to eliminate redundant decision paths; A recursive solver is used to perform hierarchical optimization of the subproblems, and the optimization results of each layer are used as input constraints for the next layer until the global convergence conditions are met.

[0012] Preferably, the sliding window optimization algorithm dynamically adjusts local planting parameters, including: A time-varying crop growth model is constructed, and the crop physiological equations are discretized into a growth state transfer matrix, which includes photosynthesis rate, transpiration efficiency, and nutrient absorption rate; Define the sliding window optimization objective function, which includes the growth state tracking error term, the environmental interference suppression term, and the resource consumption balance term; The parallel gradient descent method is used to solve the optimization problem in a distributed manner, and the parameter update is accelerated through asynchronous iteration.

[0013] Preferably, the execution layer realizes the regulation of soil moisture and light intensity based on a robust feedback control algorithm, including: defining a dual closed-loop sliding mode controller, coupling the soil moisture error and the light intensity error as a composite control variable, constructing an adaptive backstepping control law, dynamically adjusting the control gain and compensation amount according to the error change rate, and suppressing the influence of external disturbances.

[0014] Preferably, the wavelet transform uses a multi-resolution analysis framework to accelerate feature extraction, including: Decompose temperature, humidity and precipitation data into multi-scale wavelet coefficients, with each scale corresponding to a different climate fluctuation pattern; During the reconstruction process, only significant energy coefficients are retained, and noise interference components are removed through threshold filtering; Redundant wavelet transform is used to compensate for the boundary effect and generate a smooth and continuously distributed climate field.

[0015] Preferably, the entropy weight method uses information entropy differences to quantify node priority weights, including: Calculate the information entropy value of each climate factor node to reflect its information contribution to crop growth; The information entropy difference is mapped into the weight distribution coefficient through the entropy weight formula; The weight coefficients are normalized so that the sum of the priority weights is 1.

[0016] Preferably, the confidence interval model is constructed based on fuzzy interval operations, including: Collect the objective function values and constraint satisfaction of historical decision paths as training samples; Construct interval membership function, including the combination of triangular fuzzy numbers and trapezoidal fuzzy numbers; The feasible solution set is screened through interval intersection operation, and the feasibility interval of the unexplored path is predicted; When the feasibility interval does not overlap with the target solution set, the path pruning operation is triggered.

[0017] Preferably, the interval analysis method is implemented using envelope approximation technology, including: modeling uncertain climate parameters as interval variables, with upper and lower bounds determined by historical extreme values, performing linear combination operations on the interval variables to generate envelope constraint boundaries, and forcing the combination results of all interval variables to be within the envelope boundaries during the optimization process.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The method for generating crop planting recommendations that adapt to different climatic conditions proposed in the present invention involves collaborative innovation in multiple links, from data collection, analysis and modeling to optimization and regulation of planting plans, and has far-reaching significance for improving agricultural production efficiency and responding to the challenges of climate change. At the data processing level, multi-source climate sensors are used to collect real-time climate data, and multi-dimensional correlation analysis is performed through a spatiotemporal feature fusion algorithm. Data collected by traditional methods are often isolated and static, and it is difficult to reflect the real trend of climate change. The present invention performs spatiotemporal interpolation processing on temperature, humidity and precipitation data to generate a continuously distributed climate field, which makes the data more consistent with the spatial changes of the actual climate and is no longer limited to discrete measurement points. Time-frequency transformation is performed on wind speed and light data to construct a frequency domain energy distribution map, which can mine the characteristics of the data in the frequency dimension and capture some periodic climate fluctuation information, such as the periodic impact of day and night on light intensity. Through the dual-modal feature extraction network and hidden Markov model, the climate dynamic feature data finally generated contains climate fluctuation trends and the probability of extreme events, which provides a reliable basis for subsequent accurate decision-making. Taking extreme precipitation events as an example, traditional planting decisions are difficult to provide early warning. However, based on the data generated by this invention, farmers can take protective measures in advance, such as strengthening farmland drainage facilities to reduce the damage to crops caused by floods.

[0019] The hybrid decision-making model utilizes a staged decision tree structure and a dynamic weight allocation mechanism, offering significant advantages over traditional single-objective decision-making models. By constructing a crop-environment relationship map, it comprehensively considers multiple factors, including the target crop, climate factors, soil conditions, and resources. Node attributes include growth cycle requirements, environmental tolerance thresholds, and resource consumption coefficients. This refined description enables more scientific decision-making. The entropy weight method is used to calculate node priority weights, objectively reflecting the importance of each factor to crop growth and avoiding subjective judgment bias. Grey correlation analysis compares historical planting cases for similarity, leveraging past experience. A multi-objective particle swarm optimization algorithm iteratively updates node attributes, employing the Pareto front to select the optimal solution set, balancing conflicting objectives, and outputting planting strategy parameters that achieve multi-dimensional trade-offs. In practical applications, in areas with limited water resources, this model can comprehensively consider crop drought tolerance, yield potential, and water consumption to select the most appropriate crop variety and planting density, ensuring yield while improving water efficiency.

[0020] A hierarchical dynamic programming algorithm integrates fuzzy logic constraints and an adaptive decomposition strategy to effectively solve the challenge of optimizing planting plans. The planting plan optimization problem is modeled as a multi-stage decision-making problem, considering discrete crop variety selection variables and continuous irrigation amount variables, which better reflects actual planting scenarios. The adaptive decomposition strategy dynamically divides the problem into subdomains based on constraint complexity, avoiding computational redundancy when solving complex problems. During the state transition process, feasible solutions with the highest membership are prioritized for expansion. During the pruning phase, a confidence interval model is used to eliminate redundant decision paths, improving the algorithm's search efficiency. A recursive solver employs hierarchical optimization, interconnecting the optimization results at each level until global convergence conditions are met. The resulting planting plan optimizes crop adaptability and resource utilization efficiency under complex constraints. For farmland with varying terrain and soil fertility, this algorithm can develop customized planting and irrigation plans for each region, improving overall planting efficiency.

[0021] The sliding window optimization algorithm and robust feedback control algorithm play key roles in the adaptation layer and execution layer, respectively. The sliding window optimization algorithm constructs a time-varying crop growth model and defines an objective function that includes a growth state tracking error term, an environmental disturbance suppression term, and a resource consumption balance term. Using a parallel gradient descent method for distributed solution, it dynamically adjusts local planting parameters based on crop growth conditions and environmental changes. The robust feedback control algorithm defines a dual closed-loop sliding mode controller, coupling soil moisture error and light intensity error as composite control variables. It constructs an adaptive backstepping control law, dynamically adjusting control gains and compensation based on the error change rate to effectively suppress the effects of external disturbances. In actual planting, when encountering sudden changes in light intensity or abnormal soil moisture, the execution layer can respond quickly, precisely regulating soil moisture and light intensity to create a stable environment for crop growth, ensure normal crop growth, and thereby improve crop yield and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a working principle diagram of the method for generating crop planting suggestions adapted to different climatic conditions according to the present invention; Figure 2 Flowchart for generating climate dynamic feature data for spatiotemporal feature fusion algorithm; Figure 3 It is the workflow diagram of the hybrid decision model; Figure 4 This is the workflow diagram of the hierarchical dynamic programming algorithm. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] See also Figure 1-Figure 4 The present invention provides a technical solution: The present invention relates to a method for generating crop planting suggestions adapted to different climatic conditions. The specific implementation scheme is as follows: Multi-source climate sensors deployed in the target area continuously collect real-time climate data, including temperature and humidity, precipitation, wind speed, and light intensity. The sensor network covers diverse geographic locations within the target area, creating a high-density data collection point network to ensure that the data fully reflects the region's climate characteristics. For example, temperature and humidity sensors are placed at various locations in farmland, wind speed sensors are installed in open areas, and light intensity sensors are installed at light monitoring points. These sensors transmit real-time data to the data processing center via wireless communication modules.

[0025] Real-time climate data is processed using a spatiotemporal feature fusion algorithm. First, temperature, humidity, and precipitation data are spatiotemporally interpolated using kriging or inverse distance weighting (IDW) to generate a continuously distributed climate field, visually reflecting the continuous spatial variation of climate elements. Second, wind speed and light data are subjected to time-frequency transformations, such as fast Fourier transforms, to construct a frequency-domain energy distribution map, revealing the energy distribution characteristics of the data at different frequency components. Next, a bimodal feature extraction network is constructed. The first modality uses wavelet transforms to extract the periodic characteristics of the continuously distributed climate field. Using a multiresolution analysis framework, the data is decomposed into wavelet coefficients at different scales, each corresponding to a different climate fluctuation pattern, such as diurnal, weekly, or seasonal variations. The second modality uses a non-negative matrix factorization algorithm to extract frequency-domain correlation features from the frequency-domain energy distribution map, exploring the correlations between different frequency components. Finally, a feature cross-fusion layer integrates the periodic features with the frequency-domain correlation features to generate a joint feature matrix. This matrix captures the comprehensive characteristics of the climate data in both the spatiotemporal and frequency domains. Finally, the hidden Markov model is used to model the state transition of the joint feature matrix. Through model parameter training, climate dynamic characteristic data including climate fluctuation trends and extreme event probabilities are output, providing a basis for the subsequent generation of planting strategies.

[0026] Dynamic climate data is fed into a pre-trained hybrid decision-making model. This model employs a staged decision tree structure. First, a crop-environment relationship graph is constructed. Nodes in the graph include target crop nodes, climate factor nodes, soil condition nodes, and resource nodes. Each node's attributes detail information such as the crop's growth cycle requirements, environmental tolerance thresholds, and resource consumption coefficients. For example, the target crop node records a crop's specific temperature and humidity requirements at different growth stages, while the climate factor node records the historical data distribution ranges for each climate factor. A two-stage decision-making mechanism is employed. In the first stage, the priority weight of each node is calculated using the entropy weight method. The information entropy value of each climate factor node reflects its information contribution to crop growth. The entropy weight formula then maps the entropy differences into weight distribution coefficients, which are then normalized so that the sum of the weights equals 1. In the second stage, gray correlation analysis is used to perform similarity matching on historical planting cases to identify the historical cases most similar to the current climate conditions. Node attributes are iteratively updated based on the multi-objective particle swarm optimization algorithm. The Pareto front is used to screen the optimal solution set for each objective function. Conflicting objectives are balanced through constraint relaxation and normalization. The final output is planting strategy parameters containing multi-dimensional trade-offs, such as crop variety selection priority, planting time window, resource input ratio, etc.

[0027] A multi-objective optimization model was constructed based on the planting strategy parameters, with crop adaptability and resource utilization efficiency as optimization objectives. A hierarchical dynamic programming algorithm was used to globally optimize the planting plan. The planting plan optimization problem was first modeled as a multi-stage decision problem, with decision variables including discrete crop variety selection variables and continuous irrigation amount variables. Fuzzy membership functions were initialized and the initial feasible domain was calculated. An adaptive decomposition strategy was then used to dynamically partition the problem into subdomains based on constraint complexity. During the state transition process, feasible solutions with the highest membership were preferentially selected for expansion. During the pruning phase, a confidence interval model was constructed based on the historical solution set. This model, based on fuzzy interval operations, collects objective function values and constraint satisfaction degrees from historical decision paths as training samples and constructs an interval membership function composed of triangular and trapezoidal fuzzy numbers. Interval intersection operations were used to screen feasible solution sets and predict the feasibility intervals of unexplored paths. Path pruning was triggered when the feasibility intervals did not overlap with the target solution set. A recursive solver is used to perform hierarchical optimization of the sub-problems. The optimization results of each layer serve as the input constraints of the next layer until the global convergence conditions are met. Finally, the optimal planting plan data is output, including specific crop planting combinations, planting time arrangements, resource allocation plans, etc.

[0028] A multi-level control framework is constructed based on the optimal planting plan data, which includes a decision-making layer, an adaptation layer, and an execution layer. The decision-making layer generates a global planting sequence based on planting strategy parameters, clarifying the planting tasks and objectives for each stage. The adaptation layer dynamically adjusts local planting parameters using a sliding window optimization algorithm. First, a time-varying crop growth model is constructed, and the crop physiological equations are discretized into a growth state transfer matrix, which contains parameters such as photosynthesis rate, transpiration efficiency, and nutrient absorption rate. An optimization objective function is then defined, consisting of a growth state tracking error term, an environmental interference suppression term, and a resource consumption balance term. A parallel gradient descent method is used to perform a distributed solution to the optimization problem. Parameter updates are accelerated through asynchronous iteration, enabling dynamic adjustment of local parameters such as irrigation and fertilization rates. The execution layer regulates soil moisture and light intensity based on a robust feedback control algorithm. A dual-closed-loop sliding mode controller is defined, coupling the soil moisture error and light intensity error into a composite control variable. An adaptive backstepping control law is constructed, dynamically adjusting the control gain and compensation amount based on the error change rate to suppress the influence of external disturbances. For example, soil moisture can be controlled by adjusting the valve opening of the irrigation system, and light intensity can be controlled by adjusting the operating status of the sunshade net or fill light. Planting control instructions are output through a multi-level control framework to generate climate-adaptive planting recommendations.

[0029] The present invention will be further described below in conjunction with Examples 1 to 5:

[0030] Example 1: In the process of multidimensional correlation analysis, when performing spatiotemporal interpolation processing on temperature, humidity and precipitation data, the specific implementation method of using redundant wavelet transform to compensate for boundary effects is as follows:

[0031] To process the spatiotemporal interpolation of temperature, humidity, and precipitation data, it is necessary to first acquire discrete, real-time data points within the target area using multiple climate sensors. These data points may be irregularly distributed spatially. For example, sensors at different locations in a field may be unevenly spaced due to factors such as terrain and crop occlusion. Temporally, data are collected continuously at a fixed sampling frequency (e.g., every minute or every hour), forming a discrete spatiotemporal data series. To generate a continuously distributed climate field, spatiotemporal interpolation algorithms are used to fill in the space and make the discrete data continuous over time. Common interpolation methods include kriging and inverse distance weighted (IDW). These methods estimate the climate parameter values for any location within the region based on the spatial location and attribute values of discrete points, thereby constructing a continuous two-dimensional or three-dimensional climate field model that intuitively reflects the spatial distribution trends and temporal evolution of temperature, humidity, and precipitation.

[0032] After completing spatiotemporal interpolation to generate a continuously distributed climate field, feature extraction is required to identify periodic features. Wavelet transform technology is employed for this purpose. Its core principle is to decompose climate field data into wavelet coefficients at different scales through a multi-resolution analysis framework. Specifically, multi-resolution analysis decomposes a signal into low-frequency approximate components and high-frequency detail components. By repeatedly decomposing the low-frequency components, coefficients at different scales are obtained. For a climate field composed of temperature, humidity, and precipitation data, wavelet coefficients at different scales correspond to different climate fluctuation patterns: large-scale (low-resolution) coefficients reflect long-term trends or macroclimate characteristics, such as seasonal temperature variations and regional precipitation distribution trends; small-scale (high-resolution) coefficients correspond to short-term fluctuations or local details, such as intraday temperature fluctuations and localized heavy precipitation events. This multi-scale decomposition effectively separates the different frequency components in the climate data, facilitating the subsequent extraction of periodic features.

[0033] During the wavelet decomposition process, the lack of contextual information at data boundaries can cause distortion in the reconstructed data, such as oscillations or false extremes, in the boundary regions. This is known as the boundary effect. To address this issue, the redundant wavelet transform (also known as the shift-invariant wavelet transform) is introduced to compensate for this boundary effect. By removing the downsampling operation in the traditional wavelet transform, the redundant wavelet transform makes the transform shift-invariant, thus avoiding boundary distortion caused by data shift. In its implementation, the original climate field data is first expanded by adding mirrored or periodically extended data at both ends of the data sequence to simulate signal characteristics outside the boundary. The expanded data is then decomposed using redundant wavelet basis functions to obtain wavelet coefficients at various scales. During the reconstruction process, the wavelet coefficients in the boundary regions are specially processed. By introducing compensation factors or adjusting reconstruction weights, the influence of boundary effects is offset, ensuring that the reconstructed continuously distributed climate field remains smooth in the boundary regions and avoiding unrealistic climate abrupt changes.

[0034] Furthermore, during the wavelet transform reconstruction process, noise filtering is required to improve data quality. The specific steps are as follows: first, an energy threshold is set. This threshold is determined based on the noise level and signal characteristics of the data and is usually automatically calculated using statistical methods (such as minimax estimation and the SureShrink algorithm). Then, the wavelet coefficients at each scale are traversed, and coefficients with absolute values less than the threshold are considered noise components and set to zero or soft-thresholded, retaining only significant energy coefficients with absolute values greater than the threshold. Finally, the processed wavelet coefficients are used to reconstruct the climate field data, thereby removing noise interference and highlighting the true climate characteristics.

[0035] Through the above steps, a continuous climate field is first generated using spatiotemporal interpolation. Multi-resolution wavelet decomposition is then used to extract periodic features at different scales. Redundant wavelet transforms are used to compensate for boundary effects, and threshold filtering is combined to remove noise. Ultimately, an accurate, smooth, and continuously distributed climate field containing rich periodic information is generated, providing a high-quality data foundation for the extraction of the first modal features in the subsequent dual-modal feature extraction network. This process ensures the characteristic integrity of climate data in the spatiotemporal domain, laying a solid foundation for the subsequent hidden Markov model modeling of climate fluctuation trends and the probability of extreme events. This allows the generated climate dynamic characteristic data to more realistically reflect the climate evolution patterns of the target region, thereby providing a reliable basis for the scientific formulation of planting strategies. The entire processing flow strictly adheres to the basic principles of signal processing. Through mathematical transformations and algorithm optimization, multi-level feature mining and noise suppression are achieved for the original climate data, ensuring the accuracy and reliability of the analysis results.

[0036] Example 2: In the staged decision tree structure of the hybrid decision model, the specific implementation of the entropy weight method is as follows:

[0037] It is necessary to collect historical data of each climate factor node to construct a data set for calculating information entropy. Climate factor nodes include environmental parameter nodes closely related to crop growth, such as temperature, humidity, precipitation, wind speed, and light intensity. For each climate factor node, it is necessary to obtain its historical observation data in the target area. The data time span must cover the entire growth cycle of the crop or multiple growth cycles to ensure that the data can reflect the long-term change patterns and distribution characteristics of the climate factors. For example, the historical data of the temperature node can include temperature values at different times of the day, and the humidity node can include relative humidity values at the corresponding time. The data collection frequency is consistent with the real-time collection frequency of the multi-source climate sensor to ensure the consistency of the data format.

[0038] The historical data of each climate factor node is preprocessed, including data cleaning, missing value filling and normalization. The data cleaning step requires the removal of obviously abnormal outliers. The Z-score method or IQR method based on statistical methods can be used to identify and process abnormal data; missing value filling can be based on data characteristics using linear interpolation, polynomial interpolation, or mean filling based on adjacent time / locations; normalization converts climate factor data of different dimensions into dimensionless standardized data, usually using the minimum-maximum normalization method, with the formula:

[0039] in, is the original data value, is the minimum value of the historical data of the climate factor node, is the maximum value, is the normalized data value. Preprocessing ensures that the data of each climate factor node are comparable and reliable.

[0040] After completing data preprocessing, the information entropy value of each climate factor node is calculated to reflect its information contribution to crop growth. The calculation of information entropy is based on the Shannon entropy principle in information theory. The specific steps are as follows: Climate factor nodes ( , is the total number of climate factor nodes), its historical data can be regarded as a discrete probability distribution after normalization. The data of this node is divided into bins, and count the frequency of data in each bin. ( ), then the information entropy of the node The calculation formula is:

[0041] in, The value range is ,when The more evenly the distribution is, The larger the value is, the higher the information uncertainty of the climate factor node is, and the lower the information contribution to crop growth is. On the contrary, if Concentrated in a few intervals, The smaller the value, the higher the information contribution. For example, if the temperature in a certain area is stable within the range suitable for crop growth all year round, the data distribution of the temperature node is concentrated and the information entropy value is low, indicating that temperature is a key factor affecting crop growth and the priority weight should be higher.

[0042] After calculating the information entropy value of each climate factor node, the information entropy difference is mapped into a weight distribution coefficient through the entropy weight formula. The entropy weight formula is:

[0043] in, For the The weight distribution coefficient of each climate factor node, For the The information entropy value of the climate factor node. The physical meaning of this formula is: It represents the information utility value of the climate factor node. The larger the information utility value, the more important the factor is to crop growth, and the higher the weight should be given. By normalizing the denominator, it ensures that the sum of all weight distribution coefficients is 1, satisfying the basic properties of weight.

[0044] After the weight distribution coefficient is calculated, it needs to be normalized to obtain the final priority weight. The normalization process is similar to the normalization in data preprocessing, but here it is mainly to convert the weight distribution coefficient into a priority weight that can be directly used for decision making. ,make sure The normalized priority weights directly reflect the order of importance of each climate factor node in the planting strategy decision. For example, if the priority weight of the temperature node is 0.3, the humidity node is 0.25, the precipitation node is 0.2, the wind speed node is 0.15, and the light intensity node is 0.1, it means that in the decision-making process, the temperature factor is the most important, followed by humidity, precipitation and other factors.

[0045] In the first stage of the phased decision tree structure, the entropy weight method quantifies the priority weights of each node through the above steps, providing a key quantitative basis for the subsequent two-stage decision-making mechanism. These weights will be used to guide the second stage of gray correlation analysis. That is, when performing similarity matching on historical planting cases, the case characteristics are weighted according to the priority weights of each climate factor node, so that the matching results are more focused on climate factors that have a greater impact on crop growth, thereby improving the accuracy and rationality of the generation of planting strategy parameters. The entire entropy weight method implementation process is strictly based on the principles of information theory. Through mathematical formula derivation and data statistical analysis, it achieves objective quantification of the importance of climate factor nodes, avoids the bias of human subjective judgment, and ensures the scientificity and reliability of the hybrid decision-making model in the multi-objective trade-off process.

[0046] Example 3: In the process of integrating fuzzy logic constraints and adaptive decomposition strategies into the hierarchical dynamic programming algorithm, the construction and application of the confidence interval model are specifically implemented as follows: The objective function values and constraint satisfaction of historical decision paths are collected as training samples. These historical decision paths are derived from previous optimization processes for planting plans generated under different climatic conditions. Each path corresponds to a set of decision variable combinations (such as crop variety selection and irrigation amount) and their optimization results. The objective function values include output indicators of the multi-objective optimization model, such as crop adaptability and resource utilization efficiency. Constraint satisfaction reflects whether the decision path meets pre-set constraints (such as soil moisture threshold, light intensity range, and resource consumption limit). For example, a historical decision path selecting a corn and soybean rotation with an irrigation rate of 200 cubic meters per hectare has an objective function value of 0.85 for crop adaptability and 0.78 for resource utilization efficiency. Constraint satisfaction is reflected in soil moisture remaining within the optimal range of 60%-80% and light intensity not falling below the crop light compensation point. Training samples should cover a variety of climatic conditions, crop types, and resource allocation scenarios to ensure sample diversity and representativeness.

[0047] The interval membership function is constructed by combining triangular fuzzy numbers and trapezoidal fuzzy numbers to adapt to different types of uncertainty data distribution. Triangular fuzzy numbers are suitable for describing uncertainty data with a single peak, and its form is ,in is the lower limit value, is the peak point, is the upper limit value, and the membership function is in the interval Increasing linearly, The inner linear decreases; the trapezoidal fuzzy number is suitable for describing uncertainty data with a flat transition interval, in the form of ,in is a left flat interval with a membership of 0. is a linearly increasing interval, is a right flat interval with a membership of 1. is a linear decreasing interval (if it exists). For the objective function value, such as crop adaptability matching, its value range is usually [0,1], which can be described by triangular fuzzy numbers, with the peak point is the mode or mean of the indicator in the historical sample, and are the fluctuation range boundaries determined by the sample standard deviation; for continuous constraints in constraint satisfaction (such as soil moisture), trapezoidal fuzzy numbers can be used to describe them, where and is the ideal interval boundary of the constraint, and It is the boundary between the allowed buffer zones to reflect the flexible nature of the constraint.

[0048] After constructing the interval membership function, interval intersection operations are used to screen feasible solutions and predict the feasibility intervals of unexplored paths. For each newly generated decision path, estimates of its objective function value and constraint satisfaction are first calculated. These estimates are then mapped into the constructed interval membership function to calculate its membership within each fuzzy interval. The core idea of the interval intersection operation is that a decision path is considered feasible only if its objective function value has non-zero membership within the feasible intervals of all objectives and its constraint satisfaction has non-zero membership within the permissible intervals of all constraints. In specific implementation, for each objective function (e.g., crop adaptability), the intersection of its estimated value and the fuzzy interval of the objective is calculated. If the intersection is non-empty, the path is retained as a candidate solution. Similarly, for constraints, the intersection of the constraint satisfaction estimate and the fuzzy interval of the constraint is calculated. If the intersection is empty for any constraint, the path is eliminated. This method allows for rapid screening of feasible solutions that meet the multi-objective optimization requirements and the constraints.

[0049] For unexplored decision paths, the confidence interval model uses the distribution characteristics of historical samples to predict their feasibility intervals. The specific method is to fit the probability density functions of each objective and constraint based on the objective function values and constraint satisfaction of historical samples using kernel density estimation or parameterized distribution assumptions (such as normal distribution or uniform distribution). This then constructs a feasibility interval with a confidence level (such as 95%). For example, for crop adaptability, if the historical samples are used to fit a normal distribution with a mean of 0.75 and a standard deviation of 0.1, the feasibility interval at a 95% confidence level is [0.55, 0.95]. When generating a new decision path, if its estimated objective function value falls outside this interval, the path is considered infeasible at the current confidence level, triggering path pruning.

[0050] Path pruning is triggered when the feasibility interval does not overlap with the target solution set—that is, when the estimated objective function value of the new path lies completely outside the feasible interval of all target solution sets, or when the estimated constraint satisfaction violates the permissible interval of at least one constraint—then the path is deemed redundant and removed from the search tree. For example, if the target solution set requires a crop adaptability match of at least 0.7, and the estimated value of the new path is 0.65, and the upper limit of its feasibility interval is 0.68, then the path fails to meet the target requirement and should be pruned. Pruning effectively reduces the search space of the hierarchical dynamic programming algorithm, lowers computational complexity, and improves optimization efficiency.

[0051] The entire confidence interval model construction and application process is strictly based on the principles of fuzzy mathematics and statistics. By analyzing historical samples to fit the distribution characteristics of uncertainty data, fuzzy interval operations and probabilistic statistical methods are used to evaluate the feasibility of decision paths and eliminate redundant paths. This model does not rely on specific experimental effect data, but objectively describes the uncertainty in the decision-making process in a data-driven manner, ensuring the scientificity and efficiency of the hierarchical dynamic programming algorithm when dealing with multi-stage decision-making problems. By integrating the confidence interval model, the hierarchical dynamic programming algorithm can dynamically screen feasible solutions during the optimization process, avoid falling into invalid search paths, and thus converge to the global optimal solution or approximate optimal solution more quickly, providing strong support for generating high-quality planting plans.

[0052] Example 4: The implementation method of dynamically adjusting local planting parameters through the sliding window optimization algorithm in the adaptation layer is as follows: Taking a corn-growing region as an example, a time-varying crop growth model was first constructed. This model, based on the physiological characteristics of corn at different growth stages, divides the entire growth cycle (e.g., sowing, seedling, jointing, tasseling, grain filling, and maturity) into multiple discrete time nodes (e.g., daily). Each node corresponds to a growth state transition matrix consisting of a set of physiological parameters. For example, during the seedling stage (days 10-30), the state transition matrix includes parameters such as photosynthesis rate (μmolCO₂ / m² / s), transpiration efficiency (mmolH₂O / g dry matter), and nutrient absorption rate (mg / kg soil / day). These parameters vary dynamically with daily environmental conditions such as temperature, humidity, and light intensity. During model construction, reference was made to corn physiology research literature, combined with historical climate data and field observations for the target region, to determine baseline values and environmental response functions for each physiological parameter at each stage. For example, the photosynthesis rate increases nonlinearly with increasing light intensity and tends to stabilize when the light intensity exceeds the light saturation point. This relationship is described by the Michaelis equation; transpiration efficiency is negatively correlated with air humidity, and the specific relationship is obtained by fitting historical data with a linear regression model.

[0053] When defining the sliding window optimization objective function, a sliding window of fixed length (e.g., 7 days) is first set. The window gradually slides forward over time, and the planting parameters within the window are optimized each time. The objective function contains three core terms: a growth state tracking error term, an environmental disturbance suppression term, and a resource consumption balance term. The growth state tracking error term is used to measure the deviation between the actual crop growth state and the model prediction state. For example, if the model predicts that the number of corn leaves on the 15th day should be 5, but the actual observed value is 4, an error signal is generated. The environmental disturbance suppression term targets climate fluctuations within the window (such as sudden high temperatures and rainfall). By penalizing changes in environmental parameters that deviate from historical average climate conditions, it reduces the excessive adjustment of planting parameters caused by short-term disturbances. The resource consumption balance term constrains the fluctuation range of resource inputs such as irrigation and fertilization within the window to avoid resource waste or environmental pollution caused by excessive irrigation or fertilization on a single day. For example, if the irrigation volume in the first two days of a week is 50 cubic meters per hectare and in the last five days it is 10 cubic meters per hectare, the resource consumption balance item will adjust the parameters to a more even distribution pattern by calculating the standard deviation of the irrigation volume and applying a penalty.

[0054] When using parallel gradient descent for distributed optimization problem solving, the optimization task within a sliding window is first decomposed into multiple subtasks, each corresponding to a planting parameter adjustment at a specific time point (e.g., irrigation amount on day 1, fertilization time on day 2, etc.). These subtasks are assigned to different compute nodes (such as virtual servers on a cloud computing platform) for parallel processing. Each node independently calculates the gradient of the objective function with respect to the corresponding parameter and updates the parameter value based on the gradient direction. For example, for the irrigation amount parameter on day 3, the compute node calculates the gradient based on the current error signal. If the gradient is positive, it indicates that increasing irrigation amount will reduce the objective function value, so the parameter is increased accordingly; if the gradient is negative, the parameter is decreased. Each node updates the parameters asynchronously, without waiting for all nodes to complete their calculations. Instead, it updates and shares the parameter state based on its own computational progress, accelerating convergence. For example, after completing the parameter update for day 1, node 1 immediately synchronizes the results to shared storage. Node 2, processing the parameters for day 2, can access the updated parameters in real time, avoiding the time loss associated with waiting for global synchronization in traditional synchronous iterations.

[0055] During the dynamic adjustment process, the sliding window optimization algorithm requires real-time access to the latest data from multiple climate sensors. For example, if a sudden heavy rainfall on the fourth day causes soil moisture to exceed a threshold, the algorithm automatically triggers an emergency adjustment mechanism: within the current window (assuming the window covers days 3-9), it first updates the weight of the environmental interference suppression term, increasing the penalty for abnormal soil moisture. It then recalculates the irrigation parameters for each day through parallel computing nodes, significantly reducing or even suspending irrigation plans for subsequent days to prevent soil waterlogging from impacting corn root growth. Simultaneously, based on the corn growth model, the algorithm predicts the impact of heavy rainfall on subsequent growth. For example, if it could cause root hypoxia and thus affect nutrient absorption, the algorithm appropriately increases the adjustment weight of nitrogen fertilizer application in the resource consumption balance term, compensating for potential growth defects by optimizing fertilization timing and dosage.

[0056] When corn reaches the jointing stage (assuming days 40-60), the sliding window optimization algorithm automatically adjusts the objective function's weight distribution based on the growth characteristics of that stage. For example, the demand for water and nutrients increases significantly during the jointing stage, so the weights of indicators such as plant height and stem diameter in the growth state tracking error term increase, while the weight of the resource consumption balance term decreases accordingly, prioritizing crop growth needs. At the same time, the algorithm pre-sets a high temperature threshold (e.g., 35°C) in the environmental interference suppression term to address the common high temperature and drought risks during this stage. When the maximum temperature on a particular day within the monitoring window approaches this threshold, the algorithm adjusts irrigation parameters in advance, applying small, frequent irrigations in the early morning or evening to reduce water loss due to leaf transpiration and avoid root damage caused by irrigation during the midday heat.

[0057] The entire sliding window optimization algorithm is implemented closely around the dynamic needs of crop growth and real-time environmental changes. Through discrete growth state modeling, multi-objective function design, and distributed parallel computing, dynamic fine-tuning of local planting parameters such as irrigation amount and fertilization time is achieved. This method does not rely on pre-set fixed thresholds or empirical rules. Instead, it automatically learns the interaction patterns between crops and the environment in a data-driven manner, achieving efficient resource utilization and effective suppression of environmental interference while ensuring crop growth goals. Through continuous sliding windows and iterative optimization, the algorithm can adapt to the physiological needs of different growth stages and the changes in climate patterns brought about by seasonal changes, providing crops with refined planting parameter adjustment suggestions, and ensuring the dynamic adaptability and global optimization of the planting plan in the time dimension.

[0058] Example 5: The implementation method of soil moisture and light intensity control based on the robust feedback control algorithm at the execution layer is as follows: Taking a greenhouse tomato plantation as an example, the input variables of a dual closed-loop sliding mode controller are first defined as soil moisture error and light intensity error. The soil moisture error is the difference between the real-time monitored value and the setpoint. For example, if the setpoint is 65% (volume water content), and the sensor measured value is 58%, the error is -7%. The light intensity error is the difference between the measured value and the light saturation point of tomato photosynthesis (e.g., 100,000 lux). If the measured value is 80,000 lux, the error is -20,000 lux. The controller couples these two error signals into a composite control variable. By designing a sliding surface function, the system state converges to the sliding surface within a finite time, thereby achieving coordinated control of soil moisture and light intensity.

[0059] When building an adaptive backstepping control law, the control gain and compensation amount must be dynamically adjusted based on the error rate of change. For example, if the soil moisture error increases rapidly (e.g., a decrease of 1% per minute), indicating a high risk of water loss, the control law automatically increases the control gain of the irrigation system, rapidly increasing the irrigation valve opening from 30% to 50% to accelerate water replenishment. At the same time, compensation is introduced to offset the impact of interfering factors such as soil seepage or crop transpiration. For example, an additional 10% irrigation amount can be added to compensate for the current transpiration rate estimated based on historical data. For light intensity errors, if the measured value drops sharply within a short period of time (e.g., cloud cover causes a drop of 5,000 lux per minute), the controller immediately triggers the adaptive adjustment mechanism of the fill light, increasing the light source drive current to above 90,000 lux within 5 minutes. The current gain is also adjusted based on the error rate of change to avoid excessive adjustment that could damage the light source life.

[0060] In terms of hardware implementation, the execution layer includes a soil moisture sensor, a light intensity sensor, an irrigation solenoid valve, a sunshade net drive motor, and a supplemental lighting system. The soil moisture sensor, using frequency domain reflectometry (FDR), is buried in the primary tomato root zone (10-20 cm below the surface) and collects data every 10 minutes. The light intensity sensor, installed at the top of the greenhouse at the highest light transmittance, monitors changes in external light levels in real time. The irrigation solenoid valve is installed in the main water supply pipe and controls valve opening via a pulse width modulation (PWM) signal, with an adjustable range of 0-100%. The sunshade net is an aluminum foil curtain, which is deployed and retracted via a motor-driven reel, with an adjustable shading range of 20%-80%. The supplemental lighting system uses an LED light source with a spectrum that matches the peak photosynthesis wavelength of tomatoes (red light 660nm, blue light 450nm), and an adjustable light intensity range of 0-150,000 lux.

[0061] When the system detects that the daytime ambient light intensity exceeds 120,000 lux (above the light saturation point for tomatoes), and the light intensity error is positive (+20,000 lux), the controller triggers the shade net adjustment mechanism. First, the shade net deployment range is calculated using a backstepping control law. If the error rate of change is positive (light intensity is still increasing), the shade net is deployed at a faster rate (e.g., 5% per minute) until the light intensity error falls within the ±5,000 lux range. If the error rate of change approaches zero (light intensity is stable), the system switches to fine-tuning mode, making small adjustments (1% at a time) to achieve precise control. At the same time, soil moisture may gradually increase due to reduced evaporation caused by shading. The controller couples these two error signals in real time. If the soil moisture error drops below -5%, irrigation is automatically reduced to prevent excessive moisture from causing root rot.

[0062] At night or during rainy weather, when the ambient light intensity falls below the tomato light compensation point (e.g., 1000 lux) and the light intensity error is negative (-9000 lux), the controller activates the supplemental lighting system. The supplemental lighting control process consists of two stages: first, the light intensity is rapidly increased to 5000 lux at maximum gain to suppress dry matter loss caused by crop respiration; then, the gain is gradually reduced based on the error rate, stabilizing the light intensity at 8000 lux (near the midpoint between the light compensation point and the light saturation point) within 30 minutes to balance energy consumption and photosynthetic efficiency. During this process, if the controller detects an increase in soil moisture due to evaporation caused by heating from the supplemental lighting (e.g., a 2% decrease per hour), the controller will adjust the irrigation frequency, adding a micro-irrigation dose (5 cubic meters per hectare) every four hours to maintain stable moisture levels.

[0063] To address sudden environmental disturbances (such as irrigation pipe leaks and fill light ballast failures), the robust feedback control algorithm mitigates the effects of these disturbances through the robustness of the sliding surface design. For example, when a crack in an irrigation pipe causes the actual water supply to fall below the theoretical value corresponding to the valve opening, the soil moisture error does not decrease as expected but instead continues to increase. Upon detecting the abnormal error rate of change, the controller immediately triggers a fault diagnosis process: first, the solenoid valve is closed, and the pressure sensor is used to detect the pressure drop in the pipe to determine the leak location. Simultaneously, the system switches to a backup pipe or activates an emergency water pump, restoring normal irrigation within 15 minutes. For fill light failures, when a group of light sources ceases operation, resulting in uneven light intensity distribution, the controller identifies the abnormal area using multi-zone sensor data and automatically adjusts the light intensity of adjacent groups to compensate for the luminous flux loss in the faulty area until maintenance personnel can replace the faulty component.

[0064] The entire regulation process achieves dynamic coupled control of soil moisture and light intensity through the synergistic effect of a dual closed-loop sliding mode controller and an adaptive backstepping control law. This method does not rely on a precise mathematical model of the system, but instead automatically adapts to the nonlinearity, time-varying nature, and uncertainty of the greenhouse environment through real-time feedback of errors and their rates of change. For example, seasonal variations in the root distribution depth of tomatoes, the attenuation of light transmittance due to the age of shade nets, and the light decay of LED light sources are all effectively compensated for through the controller's adaptive mechanism. Through continuous monitoring and dynamic adjustment, the system can stabilize soil moisture and light intensity within the optimal range for crop growth despite frequent changes in external climatic conditions, providing a stable growth environment for tomatoes and ensuring that the implementation of the planting plan meets the expected requirements of the multi-objective optimization model.

[0065] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0066] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating crop planting suggestions adapted to different climatic conditions, characterized in that: include: Collect real-time climate data of the target area through multi-source climate sensors; Performing multi-dimensional correlation analysis on the real-time climate data based on a spatiotemporal feature fusion algorithm to generate climate dynamic feature data; Inputting the climate dynamic characteristic data into a pre-trained hybrid decision model to generate planting strategy parameters; A multi-objective optimization model is constructed based on the planting strategy parameters. The multi-objective optimization model takes crop adaptability matching and resource utilization efficiency as optimization objectives and uses a hierarchical dynamic programming algorithm to globally optimize the planting plan, wherein the hierarchical dynamic programming algorithm integrates fuzzy logic constraints and an adaptive decomposition strategy. Outputting optimal planting plan data based on the multi-objective optimization model; A multi-level control framework is constructed based on the optimal planting plan data. The multi-level control framework includes a decision layer, an adaptation layer and an execution layer. The decision layer generates a global planting sequence based on the planting strategy parameters, the adaptation layer uses a sliding window optimization algorithm to dynamically adjust local planting parameters, and the execution layer realizes the regulation of soil moisture and light intensity based on a robust feedback control algorithm. The planting control instructions are output through the multi-level control framework to complete the generation of climate-adaptive planting suggestions.

2. The method for generating crop planting suggestions according to claim 1, wherein: The multi-dimensional correlation analysis of the real-time climate data based on the spatiotemporal feature fusion algorithm to generate climate dynamic feature data includes: Perform spatiotemporal interpolation processing on temperature, humidity and precipitation data to generate a continuously distributed climate field; perform time-frequency transformation on wind speed data and light data to construct a frequency domain energy distribution map; Constructing a dual-modal feature extraction network, wherein the first modality uses wavelet transform to extract the periodic characteristics of the continuously distributed climate field, and the second modality uses non-negative matrix decomposition to extract the frequency domain correlation characteristics of the frequency domain energy distribution map; The periodic features and frequency domain correlation features are integrated through the feature cross-fusion layer to generate a joint feature matrix; the state transition modeling of the joint feature matrix is performed based on the hidden Markov model to output climate dynamic feature data including climate fluctuation trends and extreme event probabilities.

3. The method for generating crop planting suggestions according to claim 1, wherein: The hybrid decision model adopts a staged decision tree structure to perform multi-objective trade-offs on planting strategies based on a dynamic weight allocation mechanism; the staged decision tree structure includes: Construct a crop-environment relationship graph, where nodes include target crop nodes, climate factor nodes, soil condition nodes, and resource nodes. Node attributes include growth cycle requirements, environmental tolerance thresholds, and resource consumption coefficients. A two-stage decision-making mechanism is adopted. In the first stage, the priority weight of each node is calculated by the entropy weight method. In the second stage, similarity matching of historical planting cases is performed through grey correlation analysis. The node attributes are iteratively updated based on the multi-objective particle swarm optimization algorithm. The Pareto front is used to screen the optimal solution set for each objective function. The conflicting objectives are balanced through constraint relaxation and normalization, and the planting strategy parameters containing multi-dimensional trade-offs are finally output.

4. The method for generating crop planting suggestions according to claim 1, wherein: The hierarchical dynamic programming algorithm integrates fuzzy logic constraints and adaptive decomposition strategies including: The planting plan optimization problem is modeled as a multi-stage decision problem, where the decision variables include discrete crop variety selection variables and continuous irrigation amount variables. Initialize the fuzzy membership function and calculate the initial feasible region, and use the adaptive decomposition strategy to dynamically divide the problem subdomain according to the constraint complexity; During the state transfer process, the feasible solution with the highest membership degree is preferentially selected for expansion; during the pruning phase, a confidence interval model is constructed based on the historical solution set, and interval algebra is used to eliminate redundant decision paths; A recursive solver is used to perform hierarchical optimization of the subproblems, and the optimization results of each layer are used as input constraints for the next layer until the global convergence conditions are met.

5. The method for generating crop planting suggestions according to claim 1, wherein: The sliding window optimization algorithm dynamically adjusts local planting parameters, including: A time-varying crop growth model is constructed, and the crop physiological equations are discretized into a growth state transfer matrix, which includes photosynthesis rate, transpiration efficiency, and nutrient absorption rate; Define the sliding window optimization objective function, which includes the growth state tracking error term, the environmental interference suppression term, and the resource consumption balance term; The parallel gradient descent method is used to solve the optimization problem in a distributed manner, and the parameter update is accelerated through asynchronous iteration.

6. The method for generating crop planting suggestions according to claim 1, wherein: The execution layer implements soil moisture and light intensity regulation based on a robust feedback control algorithm, including: defining a dual closed-loop sliding mode controller, coupling soil moisture error and light intensity error into a composite control variable, constructing an adaptive backstepping control law, and dynamically adjusting the control gain and compensation amount according to the error change rate to suppress the influence of external disturbances.

7. The method for generating crop planting suggestions according to claim 2, wherein: The wavelet transform uses a multi-resolution analysis framework to accelerate feature extraction, including: Decompose temperature, humidity and precipitation data into multi-scale wavelet coefficients, with each scale corresponding to a different climate fluctuation pattern; During the reconstruction process, only significant energy coefficients are retained, and noise interference components are removed through threshold filtering; Redundant wavelet transform is used to compensate for the boundary effect and generate a smooth and continuously distributed climate field.

8. The method for generating crop planting suggestions according to claim 3, wherein: The entropy weight method uses information entropy differences to quantify node priority weights, including: Calculate the information entropy value of each climate factor node to reflect its information contribution to crop growth; The information entropy difference is mapped into the weight distribution coefficient through the entropy weight formula; The weight coefficients are normalized so that the sum of the priority weights is 1.

9. The method for generating crop planting suggestions according to claim 4, wherein: The confidence interval model is constructed based on fuzzy interval operations, including: Collect the objective function values and constraint satisfaction of historical decision paths as training samples; Construct interval membership function, including the combination of triangular fuzzy numbers and trapezoidal fuzzy numbers; The feasible solution set is screened through interval intersection operation, and the feasibility interval of the unexplored path is predicted; When the feasibility interval does not overlap with the target solution set, the path pruning operation is triggered.

10. The method for generating crop planting suggestions according to claim 5, wherein: The interval analysis method is implemented using envelope approximation technology, including: modeling uncertain climate parameters as interval variables, with upper and lower bounds determined by historical extreme values, performing linear combination operations on the interval variables to generate envelope constraint boundaries, and forcing the combination results of all interval variables to be within the envelope boundaries during the optimization process.

Images