A li ss whole cycle agricultural planning support system and method

Abstract

The application discloses a LISS full-cycle agricultural planning support system, a core model construction module of the planning support system realizes multi-model collaborative operation through a GIS space model, an agricultural economic model, an AI prediction model and an adaptation adjustment module; an application service module provides scheme generation, conflict early warning, benefit evaluation and dynamic deduction; and a closed-loop optimization module realizes whole-process iteration of design-deduction-feedback-optimization. The application can realize intelligent decision support for the whole life cycle of agricultural planning, covers multi-model collaborative operation, economic-social-ecological comprehensive evaluation and dynamic adaptation of regional characteristics. The core model construction module of the application binds GIS layers and agricultural economic decision variables in real time with unified land space-time coding; a GIS output constraint matrix is injected into the economic model, and the optimization result is reversely marked back to the GIS; an AI prediction model predicts production capacity and price accordingly, and deviation is corrected by weight after adaptation adjustment, so that the three aspects are simultaneously converged, and the existing system problems are solved.

Description

Technical Field

[0001] This invention relates to the field of agricultural information technology, and in particular to a LISS full-cycle agricultural planning support system and method. Background Technology

[0002] Agricultural planning involves multiple dimensions, including optimal allocation of land resources, industrial structure layout, ecological environment assessment, and economic benefit forecasting, making it a complex and systematic project. Currently, the application of information technology in agriculture has made significant progress. In terms of planning support, some existing systems are attempting to introduce GIS spatial analysis technology to provide basic spatial layout visualizations of agricultural land; others are using single economic evaluation models or crop growth models to conduct preliminary benefit estimates for planning schemes. These technological explorations have, to some extent, improved the informatization level of agricultural planning.

[0003] Despite the progress made in agricultural informatization by existing technologies, the following significant technical shortcomings still exist in intelligent decision support for the entire lifecycle of agricultural planning:

[0004] First, the functional coverage is incomplete, lacking full life-cycle support capabilities. Existing agricultural systems mostly focus on production execution stages such as "planting monitoring and water and fertilizer management," with their functional design centered on data collection and presentation of the production process, lacking systematic coverage of the entire agricultural planning process. Specifically, existing systems cannot effectively support a complete planning loop from preliminary research, scheme design, and effect prediction to dynamic optimization and iterative review; they can only achieve data display for a single stage, making it difficult to form a systematic decision-making chain from research to optimization. This functional limitation means that planning still relies heavily on human experience, failing to fully utilize information technology for scientific decision-making.

[0005] Secondly, the technical architecture is fragmented, lacking a multi-model collaboration mechanism. Existing systems generally suffer from technical architectural flaws such as "disconnection between data collection and model application" and "separation between spatial planning and economic assessment." Most systems simply overlay GIS and basic data analysis functions, lacking effective data exchange and computational collaboration mechanisms between functional modules, and failing to form a technical architecture for multi-model collaborative computation. This technological fragmentation prevents spatial layout optimization and economic feasibility assessment from being carried out simultaneously, making it difficult for planning schemes to balance spatial rationality and economic benefits. This often results in contradictory situations where the spatial layout is reasonable but economically infeasible, or where economic benefits meet standards but the spatial layout is conflicting.

[0006] Third, the evaluation dimensions are one-sided, and the three-dimensional comprehensive quantitative capability is insufficient. Traditional planning systems focus on single spatial layout optimization, lacking the ability to systematically quantify the economic benefits, social benefits (such as employment-creating effects), and ecological impacts (such as resource consumption intensity and carbon sink contribution) of planning schemes. Existing evaluation models are mostly single-dimensional assessments, unable to simultaneously calculate economic indicators such as input-output ratio and investment payback period, ecological indicators such as water resource consumption, pollution reduction, and carbon sink, and social indicators such as employment creation and farmers' income increase. More importantly, existing systems lack dynamic extrapolation and iterative optimization functions, making it impossible to predict the long-term benefits and potential risks of the scheme during the next 3-5 year implementation period, leading to frequent problems such as unmet benefits, resource waste, and ecological damage after the scheme is implemented.

[0007] Fourth, the models suffer from rigid adaptability and insufficient support for personalized needs. Most existing models are general-purpose designs with fixed parameters, making them unable to dynamically adjust to different regional characteristics (such as plains, mountains, and specialty agricultural areas) and different planning objectives (such as ensuring stable grain supply, expanding cash crops, and developing ecological agriculture). This adaptability deficiency leads to a disconnect between the system's output and actual regional needs, severely undermining support for personalized planning requirements and making it difficult to meet the diverse and differentiated application scenarios of agricultural planning. For example, existing CNN-LSTM models have shortcomings in predicting agricultural yields in mountainous areas: First, CNNs perform indiscriminate convolutions across the entire region, treating ecological protection areas outside the arable land red line, abandoned land on steep slopes exceeding 25°, and planned arable land equally, resulting in a large amount of invalid noise being mixed into spatial feature extraction; Second, the fragmented mountainous terrain results in multiple microclimate units within the same scene image, and existing models lack terrain modulation mechanisms sensitive to slope, causing spatial features of steep slope areas to be confused with those of gentle slope areas, resulting in a systematic bias in yield prediction; Third, existing models simply concatenate the spatial feature vectors extracted by CNNs to the input of LSTM without considering the adjustment of temporal memory strength by the overall spatial reliability, causing LSTM to waste memory capacity on unreliable spatial features, resulting in insufficient stability in temporal prediction.

[0008] In summary, existing agricultural planning support technologies have systemic deficiencies in terms of functional coverage, technology integration, evaluation dimensions, and model adaptation. There is an urgent need for a new type of intelligent decision support system for agricultural planning that can achieve full coverage of the planning process, deep integration of multiple models, three-dimensional comprehensive evaluation, and dynamic regional adaptation. Summary of the Invention

[0009] This invention provides a LISS full-cycle agricultural planning support system and method to solve the technical problems of existing agricultural planning support systems, such as incomplete functional coverage, fragmented technical architecture, one-sided evaluation dimensions, and rigid model adaptation.

[0010] In view of the above technical problems, embodiments of the present invention provide a LISS full-cycle agricultural planning support system, comprising:

[0011] The data acquisition module integrates a targeted web crawler module, an IoT device interface module, a third-party data integration module, and a drone data receiving module to collect multi-source data, including land resource data, agricultural economic data, policy and regulatory data, and ecological environment data. The data acquisition module supports HTTP and MQTT protocol access.

[0012] The data processing module includes an intelligent cleaning module, a blockchain traceability module, a data standardization module, and a cross-source data association module, which respectively perform intelligent cleaning, blockchain traceability, data standardization, and cross-source data association processing on multi-source data. Among them, the intelligent cleaning module uses a semantic recognition algorithm based on an agricultural lexicon to remove redundancy and detect outliers in multi-source data. The blockchain traceability module records the collection nodes, processing flow, and update time of multi-source data on the blockchain for full-link evidence storage and generates a unique traceability identifier.

[0013] The core model building module includes a GIS spatial model module, an agricultural economic model module, an AI prediction model module, and an adaptation and adjustment module. These modules enable collaborative computation among multiple models and dynamically adjust model parameters based on regional characteristics and planning objectives. The GIS spatial model module constructs a 3D spatial model based on ArcGIS Engine components, supporting realistic rendering of terrain undulations, buildings, and vegetation elements. The agricultural economic model module uses the analytic hierarchy process (AHP) to determine the weighting of economic, ecological, and social assessments. The AI ​​prediction model module employs a CNN-LSTM fusion model to achieve temporal and spatial dual-dimensional prediction.

[0014] The application service module includes an agricultural industry layout scheme generation module, a spatial conflict early warning module, a three-dimensional benefit assessment module, a dynamic simulation module, and a scheme management module. These modules provide users with services such as planning scheme generation, compliance verification, benefit assessment, dynamic simulation, and scheme management. The agricultural industry layout scheme generation module automatically generates 2-3 differentiated alternative schemes based on the input planning objectives using reinforcement learning algorithms. The spatial conflict early warning module employs ontological modeling methods to construct conflict judgment rules and trigger red, yellow, and blue three-level early warnings. The dynamic simulation module supports 3-5 year full-cycle multi-scenario simulations.

[0015] The closed-loop optimization module is used to realize the full-process iteration of design, deduction, feedback and optimization. When the evaluation results of the three-dimensional benefit evaluation module do not meet the expected goals, it automatically feeds back to the core model building module to adjust the parameters and regenerate the optimization scheme.

[0016] This invention also provides a LISS full-cycle agricultural planning support method, which utilizes the aforementioned LISS full-cycle agricultural planning support system for planning support, including:

[0017] S1. The data acquisition module captures multi-source data of the target planning area, including land resource data, agricultural economic data, policy and regulatory data, and ecological environment data.

[0018] S2. The data processing module performs semantic recognition and cleaning, outlier detection and standardization on multi-source data, establishes cross-source data associations, and performs blockchain-based evidence storage.

[0019] S3. Through the core model building module, the processed multi-source data is used to build a GIS spatial model module, an agricultural economic model module, an AI prediction model module, and a model adaptation and adjustment module, and the model parameters are dynamically adjusted according to regional characteristics and planning objectives.

[0020] S4. Based on the adjusted model parameters, the application service module generates alternative planning schemes, conducts spatial conflict early warning and three-dimensional benefit assessment, and carries out full-cycle dynamic simulation.

[0021] S5. The closed-loop optimization module determines whether the alternative planning scheme has achieved the expected goal. If it has not, it provides feedback to adjust the model parameters and regenerate a new planning scheme until the requirements are met. Then, the final planning scheme is output through the visualization interaction module.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] (1) At the model fusion level, a real-time bidirectional binding between the GIS spatial constraint matrix and the economic model decision variables is established through unified spatiotemporal coding, and the spatial features extracted by CNN are used as the regularization input for LSTM time series prediction, thus realizing the synchronous convergence of spatial layout, economic optimization and production capacity prediction. Through the three-level coupling architecture of spatial constraint tensor → composite mask → input gate modulation, the direct intervention of GIS spatial hard constraints on the internal computation flow of deep learning models is realized. Specifically: The first level of coupling involves using a binary mask R(x,y) of the arable land red line and a sigmoid attention mechanism to shield prohibited development areas during the convolution stage, preventing feature contamination from policy-violating plots. The second level of coupling uses the exponential decay function of the terrain modulation factor G(x,y) to establish a nonlinear mapping between slope deviation and feature response intensity, automatically weakening the feature contribution of steep, fragmented plots. The third level of coupling injects the overall confidence scalar c_avg into the LSTM input gate, transforming spatial constraint information into a temporal memory-gated signal. When spatial confidence is low (e.g., large areas of steep slopes or red line constraint areas), the LSTM automatically reduces the memory weight of spatial features at the current moment, prioritizing prediction based on historical temporal patterns. This three-level coupling architecture solves the technical bottleneck in mountain agriculture planning where 'spatial constraints cannot be deeply integrated into temporal prediction'. This architecture is based on ArcGIS Engine components to build a 3D digital twin model, supporting realistic rendering of terrain undulations, buildings, and vegetation elements. It significantly improves the visualization of spatial-economic data linkage and solves the technical problems of the existing system's spatial planning and economic assessment being disconnected and the prediction model being unable to handle spatial heterogeneity. It reduces the spatial-economic consistency verification cycle of planning schemes from several days to minutes, and reduces the misjudgment rate of 5-year production capacity prediction from 8%-12% of the traditional single model to ≤2%.

[0024] (2) At the scheme generation level, a Markov decision process with a high-dimensional discrete-continuous hybrid decision space is constructed. A Pareto front exploration mechanism with a sparse reward function replaces the traditional single-objective optimization. Combined with reinforcement learning algorithms, differentiated alternative schemes are automatically generated, supporting dynamic extrapolation over a 3-5 year full cycle. A single training session can generate 2-3 sets of differentiated schemes covering economic priority, ecological priority, and equilibrium strategies, improving scheme generation efficiency by more than 80%. Moreover, the scheme set is directly located at the non-dominated solution front, avoiding the trial-and-error costs of manually adjusting weight combinations. This mechanism trains the agent through the proximal policy optimization (PPO) algorithm and saves multiple non-dominated strategies with reward functions located at the Pareto front during the exploration phase, significantly improving the foresight and implementation guarantee of the planning.

[0025] (3) At the conflict identification level, through the ontology library in the field of agricultural planning and RDF semantic reasoning, a three-level red, yellow and blue warning mechanism is adopted (red warning corresponds to hard space encroachment, yellow warning corresponds to policy semantic soft conflict, and blue warning corresponds to potential carrying capacity approaching the threshold). Based on the identification of geometric hard conflict, automatic reasoning of cross-policy semantic soft conflict is realized. The conflict identification coverage rate is increased from 65% of the traditional GIS overlay analysis to 92%, and policy rules are hot updated. It can adapt to newly issued regulations without modifying the underlying spatial operation code. Compared with the traditional system, which can only identify geometric hard conflict, the semantic reasoning capability of this invention fills the technical gap of cross-policy compatibility assessment.

[0026] (4) At the assessment and decision-making level, by constructing a three-dimensional coupled assessment matrix of economy, society and ecology and DEA joint efficiency frontier analysis, the marginal substitution elasticity of the three-dimensional benefits is quantified. Combined with the blockchain traceability module, a unique traceability identifier is generated (with a retention period of ≥10 years) to ensure the credibility and traceability of the assessment data. Combined with the regional dynamic weight adaptation based on decision tree, the assessment conclusions of the same system under different regional types (plains / mountains / specialty production areas) are transferable. The three-dimensional benefit assessment module can calculate more than 20 indicators such as input-output ratio, carbon sink, and number of jobs created, avoiding the decision-making blind spots of traditional independent dimension assessment.

[0027] (5) At the iterative optimization level, by evaluating the directional attribution mechanism of the deviation vector, the feedback signal is accurately mapped to the parameter fine-tuning of the corresponding sub-model. Combined with the closed-loop optimization module, the intelligent iterative mechanism of "design-deduction-feedback-optimization" is realized, avoiding blind re-adjustment of all parameters. The number of closed-loop iterations is reduced from 20-50 times of traditional full parameter scanning to ≤10 times. Moreover, each iteration only partially recalculates the affected model, and the time consumption of the whole cycle optimization is shortened by more than 70%. This mechanism significantly improves the scientificity and efficiency of planning by dynamically adjusting parameters such as crop planting area weight and benefit evaluation index ratio. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a schematic diagram of the overall structure of the LISS full-cycle agricultural planning support system in one embodiment of the present invention;

[0030] Figure 2 This is a flowchart of the LISS full-cycle agricultural planning support method in one embodiment of the present invention. Detailed Implementation

[0031] To make the technical problems solved, the technical solutions, and the beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0032] In the description of this invention, it should be understood that the terms "longitudinal," "radial," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0033] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0034] like Figure 1 As shown, one embodiment of the present invention provides a LISS full-cycle agricultural planning support system, including:

[0035] The data acquisition module 100 integrates a targeted web crawler module, an IoT device interface module, a third-party data docking module, and a drone data receiving module to collect multi-source data, including land resource data, agricultural economic data, policy and regulatory data, and ecological environment data. The data acquisition module supports HTTP and MQTT protocol access. The data acquisition latency is ≤5 minutes, and the data coverage is ≥95%.

[0036] Understandably, by effectively integrating multi-source heterogeneous data (including land resources, agricultural economy, policy and regulations, and ecological environment data), and supporting HTTP and MQTT protocol access, the real-time (latency ≤ 5 minutes) and comprehensive (coverage ≥ 95%) data collection is ensured, providing a high-quality data foundation for agricultural planning.

[0037] The data processing module 200 includes an intelligent cleaning module, a blockchain traceability module, a data standardization module, and a cross-source data association module. These modules perform intelligent cleaning, blockchain traceability, data standardization, and cross-source data association processing on multi-source data, respectively. The intelligent cleaning module uses a semantic recognition algorithm based on an agricultural thesaurus to remove redundancy and detect outliers in the multi-source data. The semantic recognition redundancy removal algorithm in the data processing layer uses semantic similarity matching with a threshold ≥0.85. The outlier detection uses the 3σ principle. The data standardization format is compatible with JSON-LD and GeoJSON. The traceability information retention period is ≥10 years, and the data cleaning accuracy is ≥98%. The blockchain traceability module records the entire chain of data collection nodes, processing flows, and update times on the blockchain, generating a unique traceability identifier.

[0038] Understandably, by integrating four functional modules—intelligent data cleaning, blockchain traceability, data standardization, and cross-source data association—a full-chain quality assurance system for agricultural planning data has been constructed, effectively solving the technical problems of fragmented, low-reliability, and weakly correlated agricultural planning data. Specifically, the intelligent data cleaning module employs a semantic recognition algorithm based on an agricultural thesaurus (semantic similarity matching threshold ≥0.85) combined with the 3σ outlier detection principle, achieving a data cleaning accuracy rate ≥98% and ensuring the quality of data input to the model. The blockchain traceability module records and authenticates the entire data collection process, processing flow, and update time on the blockchain, generating a unique traceability identifier. The traceability information is retained for ≥10 years, meeting the data credibility and traceability requirements for planning decisions. The data standardization module is compatible with JSON-LD and GeoJSON formats, breaking down barriers between multi-source heterogeneous data and supporting the collaborative computation of subsequent GIS spatial models and agricultural economic models. The cross-source data association module establishes binding rules for data such as soil fertility and historical output, achieving systematic integration of fragmented data and ultimately providing high-quality, reliable, and standardized data foundation support for the entire lifecycle of agricultural planning.

[0039] The core model building module 300 includes a GIS spatial model module, an agricultural economic model module, an AI prediction model module, and an adaptation and adjustment module. These modules enable collaborative computation among multiple models and dynamically adjust model parameters based on regional characteristics and planning objectives. The GIS spatial model module constructs a 3D spatial model based on ArcGIS Engine components, supporting realistic rendering of terrain undulations, buildings, and vegetation elements. The agricultural economic model module, tailored to the characteristics of Sichuan's mountainous agriculture and socio-economic development level, develops and constructs an evaluation index system for the modernization level of agriculture and rural areas (as shown in Table 1) and an evaluation index system for the balanced development level of urban and rural areas in the region (as shown in Table 2). It also uses the analytic hierarchy process (AHP) to determine the weighting of economic, ecological, and social assessments. The AI ​​prediction model module employs a CNN-LSTM fusion model to achieve temporal-spatial dual-dimensional prediction.

[0040] Table 1 Evaluation Index System for Agricultural and Rural Modernization in a Certain Region of my country

[0041]

[0042]

[0043] The aforementioned indicator system, constructed based on the characteristics of mountain agriculture and socio-economic development in a certain mountainous area, has the following effects on the development of mountain agriculture: First, it accelerates the transformation of ecological value by converting the ecological barrier advantages and ethnic cultural resources of the mountainous area into quantifiable economic value through indicators such as the realization rate of ecological product value and the proportion of green certification, thus providing a path for the transformation of ecological and cultural value. Second, it promptly adjusts and corrects deviations by replacing the mechanization standards of plains with the popularization rate of small agricultural machinery and the mechanization suitability of sloping farmland, avoiding the forced large-scale transfer and park construction in mountainous areas to meet standards, thereby preventing resource misallocation and formalism. Third, it achieves a governance guidance effect by incorporating the revitalization of ethnic culture and the control of soil erosion into the assessment, both protecting the roots of diverse cultures and consolidating the ecological security bottom line of the upper reaches of the Yangtze River, providing a theoretical basis for the differentiated and characteristic modernization of mountain agriculture and rural areas in the next five years.

[0044] Table 2 Evaluation Index System for Balanced Urban-Rural Development Level in the Region

[0045]

[0046] A rural-urban balanced development indicator system based on the actual conditions of a certain region can achieve the following: First, it can resolve the contradiction between protection and development by transforming the ecological barrier advantages of the upper reaches of the Yangtze River into supporting capital for regional balanced development through indicators such as the realized value of ecological products and the proportion of added value of advantageous and distinctive industries. Second, it can accelerate the balanced development of urban and rural areas by replacing simple quantitative comparisons with indicators such as the integration rate of urban and rural water supply and the county-level medical treatment rate, thus promoting the transition of public services from formal equality to substantive equality and narrowing the gap between the mountainous areas surrounding the basin and the core plain area. Third, it can achieve the effect of ethnic unity by incorporating the rate of living inheritance of intangible cultural heritage and the rate of bilingual education into the assessment to ensure that the roots of ethnic culture are not eroded in the process of promoting regional balanced development, and to provide differentiated solutions for the differentiated urban-rural balanced development in ethnic minority areas.

[0047] Understandably, by deeply integrating GIS spatial models, agricultural economic models, AI prediction models, and adaptation and adjustment modules, a "three-in-one" multi-model collaborative computing architecture has been constructed, effectively solving the technical fragmentation problems of "disconnect between data collection and model application" and "separation between spatial planning and economic evaluation" in existing agricultural planning systems. The GIS spatial model module constructs a 3D digital twin model based on ArcGIS Engine components, enabling realistic rendering and quarterly dynamic updates of terrain undulations, buildings, and vegetation elements. It also links with agricultural economic data in real time to support the coordinated optimization of spatial layout and economic benefits. The agricultural economic model module uses the analytic hierarchy process (AHP) to determine the three-dimensional evaluation weights of economy (e.g., 50%), ecology (e.g., 30%), and society (e.g., 20%). It integrates multiple models such as linear programming, DEA, and equilibrium price to achieve quantitative accounting of indicators such as input-output ratio, carbon sink, and number of jobs created. The AI ​​prediction model module adopts a CNN-LSTM fusion architecture. CNN extracts spatial features such as soil nutrients and topography, while LSTM captures temporal patterns such as meteorological factors and historical prices, enabling 1-5 year cycle crop production capacity and market price prediction (false positive rate ≤2%). The adaptation and adjustment module uses a decision tree algorithm to dynamically adjust the model parameter weights based on regional characteristics such as plains / mountains / specialty production areas and planning objectives such as stable grain production / expanded economic crops / ecological development. Ultimately, it forms an intelligent planning decision support system that takes into account spatial rationality, economic feasibility, and ecological sustainability.

[0048] The application service module 400 includes an agricultural industry layout scheme generation module, an agricultural product processing and distribution project site selection module, a spatial conflict early warning module, a three-dimensional benefit assessment module, a dynamic simulation module, and a scheme management module. These modules provide users with services such as planning scheme generation, compliance verification, benefit assessment, dynamic simulation, and scheme management. The agricultural industry layout scheme generation module automatically generates 2-3 differentiated alternative schemes based on the input planning objectives using a reinforcement learning algorithm. The spatial conflict early warning module uses ontological modeling methods to construct conflict judgment rules and trigger red, yellow, and blue three-level early warnings. The dynamic simulation module supports multi-scenario simulations over a 3-5 year full cycle.

[0049] Understandably, through the coordinated operation of five functional modules—intelligent scheme generation, spatial conflict early warning, three-dimensional benefit assessment, dynamic simulation, and scheme management—a full-process planning service system of "generation-verification-assessment-simulation-management" is constructed. This effectively solves the technical defects of existing agricultural planning systems, which are "heavy on design but light on implementation" and lack dynamic simulation and iterative optimization capabilities. The agricultural industry layout scheme generation module, based on reinforcement learning algorithms, can automatically generate 2-3 differentiated alternative schemes according to the planning objectives input by the user, significantly improving planning efficiency and diversity. The spatial conflict early warning module uses ontological modeling methods to construct conflict judgment rules and realize a three-level early warning mechanism (red, yellow, and blue) to effectively prevent potential risks in planning implementation. The three-dimensional benefit assessment module can comprehensively quantify and evaluate the economic, social, and ecological benefits of the planning scheme, ensuring the scientific nature and feasibility of the plan. The dynamic simulation module supports multi-scenario simulations over a full cycle of 3-5 years, helping users to predict the implementation effect of the scheme in advance and optimize the decision-making process. The scheme management module realizes version control and traceability of planning schemes, improving the convenience and transparency of planning management.

[0050] The closed-loop optimization module 500 is used to realize the full-process iteration of design, deduction, feedback and optimization. When the evaluation results of the three-dimensional benefit evaluation module do not meet the expected goals, it automatically feeds back to the core model construction module to adjust the parameters and regenerate the optimization scheme.

[0051] Understandably, the closed-loop optimization module, by constructing a full-process iterative mechanism of "design-deduction-feedback-optimization," automatically triggers feedback signals to the core model building module when the three-dimensional benefit evaluation results do not meet expectations. This dynamically adjusts model parameters and regenerates the optimized solution, forming a continuously improving closed-loop system. This design not only significantly enhances the scientific rigor and feasibility of agricultural planning schemes but also drastically shortens the planning cycle and reduces the cost of manual intervention through automated iteration. Furthermore, it supports the traceability and comparison of multiple versions of the scheme, providing strong technical support for the dynamic optimization and precise decision-making of agricultural planning.

[0052] This invention enables intelligent decision support for the entire lifecycle of agricultural planning, multi-model collaborative computation, comprehensive three-dimensional assessment of economic, social, and ecological aspects, and dynamic adaptation to regional characteristics. The core model construction module establishes a real-time binding relationship between the GIS spatial layer and agricultural economic decision variables through unified land parcel spatiotemporal coding. The constraint matrix output by the GIS spatial model (such as the arable land red line and ecological buffer distance) is directly injected into the linear programming boundary conditions of the agricultural economic model as parameters. The optimization results of the economic model are synchronously back-annotated to the GIS spatial layer. The AI ​​prediction model uses the optimized spatial layout as input to predict production capacity and price. The prediction deviation is analyzed by the adaptation and adjustment module and then directionally corrected to adjust the GIS spatial weights and the objective function weights of the economic model. This achieves the simultaneous convergence of spatial rationality, economic feasibility, and prediction reliability, solving the problems of incomplete functional coverage, fragmented architecture, one-sided assessment, and rigid model adaptation in existing systems.

[0053] In one embodiment, the multi-source data specifically includes:

[0054] Land resource data includes data on soil types, topography, and arable land boundaries obtained from public information released by the natural resources department through web crawling, as well as topographic data with a resolution of 0.5 meters generated by combining drone aerial photography.

[0055] Agricultural economic data includes data on prices of agricultural products on production and sales platforms, costs at each stage of the industrial chain, agricultural machinery rental prices, planting area, output, output value, per capita disposable income of farmers, and other relevant economic statistics on agriculture and rural areas.

[0056] Policy and regulatory data, including data crawled from agricultural subsidy standards, planning and control red lines, and environmental protection policy requirements of governments at all levels.

[0057] Ecological and environmental data include real-time meteorological data from meteorological departments, water / air quality data from environmental protection departments, elevation data, sunlight data, and soil moisture, nutrient, and pH data collected through ground sensors.

[0058] In one embodiment, the GIS spatial model module in the core model building module includes:

[0059] The spatial-economic data linkage module enables real-time linkage between GIS spatial layers and agricultural economic data, supporting the simultaneous display of historical output, cost-benefit, and resource consumption economic indicators of any plot in the 3D model.

[0060] In one embodiment, the agricultural economic model module in the core model construction module includes:

[0061] The production optimization module uses a linear programming model to optimize crop planting area and livestock breeding scale.

[0062] The efficiency evaluation module uses the DEA model to calculate the technical and scale efficiency of the planning scheme.

[0063] The market adaptation module predicts agricultural product market price fluctuations based on an equilibrium price model.

[0064] The three-dimensional evaluation module is used to calculate economic indicators such as input-output ratio, investment payback period, and net present value; ecological indicators such as water resource consumption, fertilizer pollution reduction, carbon sink, employment creation, and farmers' income increase. This module constructs a coupled economic-social-ecological benefit evaluation matrix, uses Data Envelopment Analysis (DEA) to calculate the three-dimensional joint efficiency frontier of the planning scheme, identifies trade-off areas between dimensions, and calculates marginal substitution elasticity. The regional adaptation and adjustment module constructs a regional feature classification model based on CART decision trees. It automatically adjusts the AHP weight matrix according to different topographical and geomorphological planning regional types (plain agricultural areas, mountainous and hilly areas) and different planning objectives (stabilizing grain supply, expanding cash crop coverage, and developing ecological agriculture), and performs weight sensitivity analysis, outputting the evaluation robustness boundary within a ±10% disturbance range. Therefore, this module not only outputs static three-dimensional evaluation values ​​but also quantifies the precise substitution relationship of economic, ecological, and social benefits under different regional types, avoiding the decision-making blind spots caused by traditional independent-dimensional evaluations and enabling the evaluation conclusions to be adaptable across regions.

[0065] The regional adaptation and adjustment module constructs a regional feature classification model based on a decision tree, and automatically adjusts the weights of each evaluation indicator according to the type of the planned region and the planning objectives.

[0066] In one embodiment, the AI ​​prediction model module in the core model building module includes:

[0067] The temporal-spatial dual-dimensional prediction module adopts a CNN-LSTM fusion architecture. The CNN module extracts the spatial features of soil nutrients, topography, and crop distribution, while the LSTM module captures the temporal patterns of meteorological factors, historical prices, and planting area. Based on multi-source data, it predicts crop production capacity and market prices, with a prediction period covering 1-5 years and a misjudgment rate of ≤2%.

[0068] The disaster impact prediction module is used to preset drought, flood, and pest and disease disaster scenarios and simulate the impact of different disaster levels on the effectiveness of the plan.

[0069] The reinforcement learning scheme generation module is used to automatically generate differentiated alternative schemes covering crop layout, infrastructure configuration, and industrial chain extension based on the input planning objectives and constraints of the reinforcement learning algorithm. This module defines agricultural planning decisions as a Markov decision process: the state space is composed of a land parcel attribute vector, a market state vector, and a policy constraint matrix; the action space includes discrete actions such as crop type selection and industrial chain extension level, and continuous actions such as planting area ratio and infrastructure investment intensity; the reward function R = α·NPV + β·ecosystem service value + γ·job creation - δ·spatial conflict penalty, where the weight coefficients α, β, and γ are dynamically configured by the regional adaptation adjustment module according to the planning objective type; the agent is trained using the proximal policy optimization (PPO) algorithm, and during the exploration phase, multiple non-dominated policies whose reward function is at the Pareto front are saved, ultimately outputting 2-3 sets of differentiated alternative schemes covering different objective preferences. Compared to traditional multi-objective programming, which requires manual enumeration of weight combinations and multiple solutions, this module can generate a set of schemes covering economic priority, ecological priority, and equilibrium strategies with a single training, improving scheme generation efficiency by more than 80%.

[0070] This invention also provides a LISS full-cycle agricultural planning support method, such as... Figure 2 As shown, the planning support provided by the aforementioned LISS full-cycle agricultural planning support system includes:

[0071] S1. The data acquisition module captures multi-source data of the target planning area, including land resource data, agricultural economic data, policy and regulatory data, and ecological environment data.

[0072] S2. The data processing module performs semantic recognition and cleaning, outlier detection and standardization on multi-source data, establishes cross-source data associations, and performs blockchain-based evidence storage.

[0073] S3. Through the core model building module, the processed multi-source data is used to build a GIS spatial model module, an agricultural economic model module, an AI prediction model module, and a model adaptation and adjustment module, and the model parameters are dynamically adjusted according to regional characteristics and planning objectives.

[0074] In one embodiment, step S3 further includes the following steps:

[0075] S301. Establish an integrated framework for multiple agricultural economic models, use the analytic hierarchy process (AHP) to determine the economic, ecological and social assessment weights, and construct a regional adaptation and adjustment module based on decision trees.

[0076] S302. In the CNN-LSTM fusion architecture, the CNN encoder performs multi-layer convolution and pooling on spatial raster data of soil nutrients, topography, and crop distribution to extract spatial feature vectors. These feature vectors, after dimensionality reduction by fully connected layers, serve as the bias adjustment signal for the LSTM input gate, performing spatial regularization and noise reduction on the time-series data of meteorological factors, historical prices, and planting area. The LSTM decoder predicts crop yield and market price based on the denoised time-series features. The model is trained using agricultural production data, meteorological data, and market data from the past 10 years, and hyperparameters are optimized through 5-fold cross-validation. In test scenarios with fragmented mountainous terrain and strong spatial heterogeneity, compared to a single LSTM model, this fusion architecture can reduce the misjudgment rate of yield prediction.

[0077] S4. Based on the adjusted model parameters, the application service module generates alternative planning schemes, conducts spatial conflict early warning and three-dimensional benefit assessment, and carries out full-cycle dynamic simulation.

[0078] In one embodiment, step S4 further includes the following steps:

[0079] S401. The agricultural industry layout scheme generation module uses reinforcement learning algorithms to generate alternative planning schemes. After inputting the planning objectives, it completes multi-model calculations and generates 2-3 sets of differentiated alternative schemes covering spatial layout, benefit assessment reports and implementation step suggestions.

[0080] S402. The spatial conflict early warning module constructs an ontology library for the agricultural planning field, defining core concepts such as farmland protection, ecological management, and infrastructure using the OWL ontology language, as well as semantic relationships such as prohibited spatial overlap, minimum buffer distance, and policy compatibility levels. Based on the RDF rule reasoning engine, it not only performs spatial hard conflict scanning at the GIS geometric level but also performs cross-policy semantic soft conflict reasoning. Conflict determination results trigger three-level early warning systems (red, yellow, and blue) according to risk level: red warnings correspond to hard spatial encroachment, yellow warnings correspond to policy semantic soft conflicts, and blue warnings correspond to potential carrying capacity approaching the threshold. Compared to traditional GIS spatial overlay analysis, which can only identify geometric hard conflicts, this invention improves the conflict identification coverage from 65% to 92% through semantic reasoning, and achieves instant adaptation to new policies through hot updates of ontology rules, without modifying the underlying spatial calculation code.

[0081] S403. Conduct multi-scenario simulations over a 3-5 year period, dividing the calculation conditions according to 0.6 to 1.2 times the design parameters, and simulating the implementation effect of the scheme under different meteorological disasters, soil changes, and price fluctuations.

[0082] S5. The closed-loop optimization module determines whether the alternative planning scheme has achieved the expected goal. If it has not, it provides feedback to adjust the model parameters and regenerate a new planning scheme until the requirements are met. Then, the final planning scheme is output through the visualization interaction module.

[0083] In one embodiment, step S5 further includes the following steps:

[0084] S501. Set expected target thresholds, including net present value ≥ 0, capacity compliance rate ≥ 95%, and ecological indicators meeting constraints;

[0085] S502. When the evaluation result does not reach the threshold, a feedback signal is automatically triggered to the core model building module to adjust the crop planting area weight, the proportion of benefit evaluation indicators and model parameters.

[0086] S503. Re-execute steps S3-S4 to generate an optimized solution. Terminate the loop when the number of iterations reaches a preset threshold of ≤10 or the expected target is met. Output the final planning solution and save all versions.

[0087] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A LISS whole-cycle agricultural planning support system characterized by, include: The data acquisition module integrates a targeted web crawler module, an IoT device interface module, a third-party data integration module, and a drone data receiving module to collect multi-source data, including land resource data, agricultural economic data, policy and regulatory data, and ecological environment data. The data acquisition module supports HTTP and MQTT protocol access. The data processing module includes an intelligent cleaning module, a blockchain traceability module, a data standardization module, and a cross-source data association module, which respectively perform intelligent cleaning, blockchain traceability, data standardization, and cross-source data association processing on multi-source data. Among them, the intelligent cleaning module uses a semantic recognition algorithm based on an agricultural lexicon to remove redundancy and detect outliers in multi-source data. The blockchain traceability module records the collection nodes, processing flow, and update time of multi-source data on the blockchain for full-link evidence storage and generates a unique traceability identifier. The core model building module includes a GIS spatial model module, an agricultural economic model module, an AI prediction model module, and an adaptation and adjustment module. These modules enable collaborative computation among multiple models and dynamically adjust model parameters based on regional characteristics and planning objectives. The GIS spatial model module constructs a 3D spatial model based on ArcGIS Engine components, supporting realistic rendering of terrain undulations, buildings, and vegetation elements. The agricultural economic model module, tailored to the characteristics of Sichuan's mountainous agriculture and socio-economic development level, develops and constructs an evaluation index system for the modernization level of agriculture and rural areas and an evaluation index system for the balanced development level of urban and rural areas. It also uses the analytic hierarchy process (AHP) to determine the weighting of economic, ecological, and social assessments. The AI ​​prediction model module employs a CNN-LSTM fusion model to achieve temporal and spatial dual-dimensional prediction. The application service module includes an agricultural industry layout scheme generation module, an agricultural product processing and distribution project site selection module, a spatial conflict early warning module, a three-dimensional benefit assessment module, a dynamic simulation module, and a scheme management module. These modules provide users with services such as planning scheme generation, compliance verification, benefit assessment, dynamic simulation, and scheme management. The agricultural industry layout scheme generation module uses reinforcement learning algorithms to automatically generate 2-3 differentiated alternative schemes based on the input planning objectives. The spatial conflict early warning module uses ontological modeling methods to construct conflict judgment rules and trigger red, yellow, and blue three-level early warnings. The dynamic simulation module supports 3-5 year full-cycle multi-scenario simulations. The closed-loop optimization module is used to realize the full-process iteration of design, deduction, feedback and optimization. When the evaluation results of the three-dimensional benefit evaluation module do not meet the expected goals, it automatically feeds back to the core model building module to adjust the parameters and regenerate the optimization scheme.

2. The LISS whole-cycle agricultural planning support system according to claim 1, characterized in that, The multi-source data specifically includes: Land resource data includes soil type, topography, and arable land red line data publicly available from the natural resources department, obtained through web crawling, as well as 0.5-meter resolution topographic data generated by combining drone aerial photography; Agricultural economic data includes data on prices of agricultural products on production and sales platforms, costs at each stage of the industrial chain, agricultural machinery rental prices, and agricultural and rural economic statistics such as planting area, output, output value, and per capita disposable income of farmers. Policy and regulatory data, including publicly released agricultural subsidy standards, planning and control red lines, project application guidelines, and environmental protection policy requirements by governments at all levels; Ecological and environmental data include real-time meteorological data from meteorological departments, water / air quality data from environmental protection departments, elevation data, sunlight data, and soil moisture, nutrient, and pH data collected through ground sensors.

3. The LISS whole-cycle agricultural planning support system according to claim 1, characterized in that, The GIS spatial model module in the core model construction module includes: The spatial-economic data linkage module enables real-time linkage between GIS spatial layers and agricultural economic data, supporting the simultaneous display of historical output, cost-benefit, and resource consumption economic indicators of any plot in the 3D model.

4. The LISS whole-cycle agricultural planning support system according to claim 1, characterized by, The agricultural economic model module in the core model construction module includes: The production optimization module uses a linear programming model to optimize crop planting area and livestock scale. The efficiency evaluation module uses the DEA model to calculate the technical and scale efficiency of the planning scheme. The market adaptation module predicts agricultural product market price fluctuations based on an equilibrium price model. The three-dimensional evaluation module is used to calculate economic indicators such as input-output ratio, investment payback period, and net present value; ecological indicators such as water resource consumption, fertilizer pollution reduction, carbon sink, employment creation, and farmers' income increase. The regional adaptation and adjustment module constructs a regional feature classification model based on a decision tree, and automatically adjusts the weights of each evaluation indicator according to the type of the planned region and the planning objectives.

5. The LISS whole-cycle agricultural planning support system according to claim 1, characterized by, The AI ​​prediction model module in the core model building module includes: The temporal-spatial dual-dimensional prediction module adopts a CNN-LSTM fusion architecture. The CNN module extracts the spatial features of soil nutrients, topography, and crop distribution, while the LSTM module captures the temporal patterns of meteorological factors, historical prices, and planting area. Based on multi-source data, it predicts crop yield and market prices, with a prediction period covering 1-5 years and a false positive rate of ≤2%. The spatial constraint coding submodule is used to receive the slope raster S(x,y), soil type raster T(x,y), and farmland boundary line raster R(x,y) output by the GIS spatial model module, and construct a three-dimensional spatial constraint tensor C(x,y)=[S(x,y),T(x,y), R(x,y)], where R(x,y) is a binary mask (1 represents a planarable plot, and 0 represents a plot that is prohibited from development). The spatial attention mask generation submodule is used to convolve the spatial constraint tensor C(x,y) and activate it with Sigmoid to generate a spatial attention mask M(x,y)=σ(Conv_{1×1}(C(x,y))), whose value range is [0,1], representing the confidence weight of each spatial location feature; The terrain modulation factor calculation submodule is used to calculate the terrain modulation factor G(x,y)=exp(−γ·|S(x,y)−S_0|) based on the slope grid S(x,y), where S_0 is the crop suitable slope benchmark value and γ is the slope sensitivity coefficient. The terrain modulation factor is used to weaken the spatial characteristic response of steep slope areas. The composite mask generation submodule is used to multiply the spatial attention mask with the terrain modulation factor element by element to generate a composite mask M′(x,y)=M(x,y)⊙G(x,y); The CNN feature weighting submodule is used to multiply the original spatial feature map F(x,y) extracted by the CNN encoder with the composite mask element by element to obtain the constrained filtered spatial feature map F′(x,y)=F(x,y)⊙M′(x,y); The LSTM input gate modulation submodule is used to calculate the spatial average value of the composite mask c_avg=ΣM′(x,y) / (H·W), and inject this overall confidence scalar into the LSTM input gate control equation: i_t=σ(W_i·x_t+U_i·h_{t−1}+V_i·c_avg+b_i), where V_i is a learnable spatial-temporal coupling weight, which adjusts the intensity of the spatial features entering the memory unit at the current time through the overall confidence. The CNN encoder extracts F′(x,y) through deep convolution, and the LSTM decoder predicts the time series of meteorological factors and historical prices based on the modulated input gate, and outputs the predicted values ​​of crop production capacity and market price. The disaster impact prediction module is used to preset drought, flood and pest disaster scenarios and simulate the impact of different disaster levels on the effectiveness of the plan. The reinforcement learning scheme generation module is used to automatically generate differentiated alternative schemes covering crop layout, infrastructure configuration, and industrial chain extension based on the input planning objectives and constraints of reinforcement learning algorithms.

6. A LISS whole-cycle agricultural planning support method characterized by, Planning support using the LISS full-cycle agricultural planning support system as described in any one of claims 1-5 includes: S1. The data acquisition module captures multi-source data of the target planning area, including land resource data, agricultural economic data, policy and regulatory data, and ecological environment data. S2. The data processing module performs semantic recognition and cleaning, outlier detection and standardization on multi-source data, establishes cross-source data associations, and performs blockchain-based evidence storage. S3. Through the core model building module, the processed multi-source data is used to build a GIS spatial model module, an agricultural economic model module, an AI prediction model module, and a model adaptation and adjustment module, and the model parameters are dynamically adjusted according to regional characteristics and planning objectives. S4. Based on the adjusted model parameters, the application service module generates alternative planning schemes, conducts spatial conflict early warning and three-dimensional benefit assessment, and carries out full-cycle dynamic simulation. S5. The closed-loop optimization module determines whether the alternative planning scheme has achieved the expected goal. If it has not, it provides feedback to adjust the model parameters and regenerate a new planning scheme until the requirements are met. Then, the final planning scheme is output through the visualization interaction module.

7. The LISS whole-cycle agricultural planning support method according to claim 6, characterized in that, Step S3 further includes the following steps: S301. Establish an integrated framework for multiple agricultural economic models, use the analytic hierarchy process (AHP) to determine the economic, ecological and social assessment weights, and construct a regional adaptation and adjustment module based on decision trees. S302. In the CNN-LSTM fusion architecture, the CNN encoder performs multi-layer convolution and pooling on the spatial raster data of soil nutrients, topography, and crop distribution to extract spatial feature vectors. After dimensionality reduction by a fully connected layer, these feature vectors are used as the bias adjustment signal for the input gate of the LSTM to perform spatial regularization and noise reduction on the time series of meteorological factors, historical prices, and planting area. The LSTM decoder predicts crop yield and market price based on the noise-reduced time series features.

8. The LISS whole-cycle agricultural planning support method according to claim 7, characterized in that, Step S4 further includes the following steps: S401. The agricultural industry layout scheme generation module uses reinforcement learning algorithms to generate alternative planning schemes. After inputting the planning objectives, it completes multi-model calculations and generates 2-3 sets of differentiated alternative schemes covering spatial layout, benefit assessment reports and implementation step suggestions. S402. The spatial conflict early warning module constructs an ontology library for the agricultural planning field, defining core concepts such as farmland protection, ecological management, and infrastructure using the OWL ontology language, as well as semantic relationships such as prohibited spatial overlap, minimum buffer distance, and policy compatibility level. Based on the RDF rule reasoning engine, it not only performs spatial hard conflict scanning at the GIS geometric level, but also performs cross-policy semantic soft conflict reasoning. The conflict judgment result triggers a three-level warning system (red, yellow, and blue) according to the risk level: red warning corresponds to hard spatial encroachment, yellow warning corresponds to policy semantic soft conflict, and blue warning corresponds to potential carrying capacity approaching the threshold. S403. Conduct multi-scenario simulations over a 3-5 year period, dividing the calculation conditions according to 0.6 to 1.2 times the design parameters, and simulating the implementation effect of the scheme under different meteorological disasters, soil changes, and price fluctuations.

9. The LISS full-cycle agricultural planning support method according to claim 8, characterized in that, Step S5 further includes the following steps: S501. Set expected target thresholds, including net present value ≥ 0, capacity compliance rate ≥ 95%, and ecological indicators meeting constraints; S502. When the evaluation result does not reach the threshold, a feedback signal is automatically triggered to the core model building module to adjust the crop planting area weight, the proportion of benefit evaluation indicators and model parameters. S503. Re-execute steps S3-S4 to generate an optimized solution. Terminate the loop when the number of iterations reaches a preset threshold of ≤10 or the expected target is met. Output the final planning solution and save all versions.

Images