Intelligent site selection method and system for cultivated land occupation and compensation balance

Abstract

The invention discloses an intelligent site selection method and system for farmland occupation and compensation balance, and belongs to the technical field of intelligent site selection, and the method comprises the steps: S1, collecting and preprocessing multi-source spatio-temporal data, so as to construct a basic database; s2, constructing a spatial adjacency relation graph, inputting data in the basic database into a graph neural network model, and generating a dynamic evaluation weight vector for a to-be-selected plot; s3, calculating a final utility value of each evaluation index, and calculating a multi-scale ecological balance score of the land parcel by adopting a multiplicative aggregation model in combination with the dynamic evaluation weight vector; s4, predicting the future quality evolution trend of the land parcels by adopting a long-short-term memory network model, and the method can accurately capture and respond to the spatial-temporal dynamic characteristics of the cultivated land ecosystem, so that the evaluation result is more fit with the physical reality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent site selection, in particular to an intelligent site selection method and system for arable land occupation and compensation balance. BACKGROUND

[0002] Arable land occupation and compensation balance is a key national policy to ensure national food security and ecological balance. Traditional site selection methods rely on expert experience or static multi-factor weighted evaluation based on analytic hierarchy process. Such methods have significant technical bottlenecks. First, arable land quality presents high spatial heterogeneity and dynamic evolution over time. Traditional static weight models are difficult to capture this spatio-temporal dynamic feature. Second, arable land evaluation needs to consider micro soil properties and macro regional ecological functions. Existing methods often lead to one-sided evaluation results due to single scale. Third, the evaluation process integrates remote sensing images, ground monitoring, social and economic data, and other multi-source heterogeneous data. The inherent errors of each data source will accumulate and amplify in the decision-making chain, resulting in a high degree of uncertainty in the final results. In addition, the health status of arable land ecosystem often shows nonlinear characteristics and threshold effects. For example, soil nutrients below a certain threshold will cause a cliff-like decline in land productivity. Simple linear weighted models cannot accurately represent such abrupt phenomena. Therefore, there is an urgent need for a site selection technology for arable land occupation and compensation balance that can overcome the above challenges and achieve dynamic, multi-scale, high reliability and forward-looking decision-making.

[0003] The present application aims to overcome the shortcomings of the prior art and provide an intelligent site selection method and system for arable land occupation and compensation balance that can comprehensively handle spatio-temporal heterogeneity, multi-scale coupling, data uncertainty and ecological threshold effects, to improve the accuracy, ecological protection ability and long-term sustainability of site selection. SUMMARY

[0004] The present application aims to provide an intelligent site selection method and system for arable land occupation and compensation balance to solve the problems raised in the background.

[0005] The technical solution of the present application is as follows: S1, collect and preprocess multi-source spatio-temporal data to construct a basic database; S2, construct a spatial adjacency graph and input the data in the basic database into a graph neural network model to generate a dynamic evaluation weight vector for the selected land; S3, calculate the final utility value of each evaluation index and combine the dynamic evaluation weight vector to calculate the multi-scale ecological balance score of the land using a multiplicative aggregation model; S4, use a long short-term memory network model to predict the future quality evolution trend of the land; S5, respectively quantify data uncertainty, model uncertainty and spatial uncertainty, and comprehensively calculate the foregoing three uncertainties to generate an overall uncertainty index; S6, calculate a comprehensive resilience index, and linearly weight the multi-scale ecological balance score, future quality evolution trend and comprehensive resilience index according to a strategic parameter to generate a forward-looking decision score; S7, determine the reliability of the site selection suggestion according to the overall uncertainty index, and sort the candidate plots according to the forward-looking decision score to output a final site selection suggestion list.

[0006] Preferably, the multi-source spatio-temporal data includes: historical occupation and compensation project geographic spatial information, plot-scale soil physicochemical properties, landscape pattern index based on remote sensing image interpretation, and regional-scale ecosystem service function data and climate change data.

[0007] Preferably, S3 specifically includes: S31, for key indicators with threshold effect, obtain the normalized measurement value and the corresponding ecological critical threshold value; S32, when the measurement value is lower than the ecological critical threshold value, a first preset function is used to calculate the final utility value with exponentially decaying utility; S33, when the measurement value is not lower than the ecological critical threshold value, a second preset function is used to calculate the final utility value with non-linearly increasing utility.

[0008] Preferably, S5 specifically includes: S51, quantifying data uncertainty according to the reliability of the data source; S52, using Monte Carlo rejection technique, by repeatedly randomly inactivating part of the neurons in the graph neural network model and observing the output variance, to estimate the model uncertainty; S53, calculate the normalized standard deviation of the value of the key indicator of the target plot in all plots in the spatial neighborhood as the spatial uncertainty; S54, square sum and square root operation on data uncertainty, model uncertainty and spatial uncertainty to generate an overall uncertainty index.

[0009] Preferably, the comprehensive resilience index in S6 is obtained by weighted sum of the normalized values of multiple resilience-related indicators, and the weights of each indicator are determined by domain experts.

[0010] Preferably, the determination of the reliability of the site selection suggestion in S7 specifically includes: S71, compare the overall uncertainty index with a preset uncertainty threshold value; S72, when the overall uncertainty index is less than the uncertainty threshold value, the site selection suggestion is determined to be reliable; S73, when the overall uncertainty index is not less than the uncertainty threshold, marking the site selection suggestion as high uncertainty and suggesting manual review.

[0011] A smart site selection system for cultivated land occupation and compensation balance comprises: A data acquisition and preprocessing module is configured to collect and preprocess multi-source spatio-temporal data to construct a basic database. A spatio-temporal weight adaptive learning module is configured to construct a spatial adjacency graph and input data in the basic database into a graph neural network model to generate a dynamic evaluation weight vector for a to-be-selected land plot. A multi-scale ecological balance evaluation module is configured to calculate final utility values of each evaluation index and calculate a multi-scale ecological balance score of the land plot by using a multiplicative aggregation model in combination with the dynamic evaluation weight vector. A cultivated land quality evolution prediction module is configured to predict a future quality evolution trend of the land plot by using a long short-term memory network model. An uncertainty quantification module is configured to quantize data uncertainty, model uncertainty and spatial uncertainty respectively and comprehensively calculate the three uncertainties to generate an overall uncertainty index. A decision support module is configured to calculate a comprehensive resilience index, generate a forward-looking decision score, judge reliability according to the overall uncertainty index and output a final site selection suggestion list according to the forward-looking decision score.

[0012] The present application provides a smart site selection method and system for cultivated land occupation and compensation balance, which has the following improvements and advantages compared with the prior art. 1. The core breakthrough of the present application is to introduce a graph neural network model to generate a dynamic evaluation weight vector. The model abstracts each land plot as a node in the graph and the interaction between land plots as an edge. By learning a large amount of historical data, it can calculate a set of most suitable weights according to the state of each land plot at a specific time point and the influence of its neighborhood. This makes the evaluation process free from the constraints of subjective setting and static framework, and can accurately capture and respond to the spatio-temporal dynamic characteristics of the cultivated land ecosystem, so that the evaluation results are more consistent with the physical reality. 2. The present application synthesizes errors from three orthogonal dimensions of data uncertainty, data source quality, model uncertainty, algorithm stability and spatial uncertainty, and local spatial heterogeneity. This makes each site selection suggestion accompanied by a clear confidence interval. When the overall uncertainty index exceeds the preset threshold, the system suggests manual review, thereby establishing a risk control mechanism for human-machine collaboration. This is a substantial progress from giving answers to giving answers with confidence. 3. The present application balances and optimizes between multiple strategic objectives such as ensuring current production, pursuing future yield potential, and ensuring long-term ecological stability, so that the site selection decision is not only scientific, but also accurately aligned with macro-policy objectives. BRIEF DESCRIPTION OF DRAWINGS

[0013] The present application will be further explained in conjunction with the accompanying drawings and embodiments: Figure 1 is a flowchart of a smart site selection method for cultivated land occupation and compensation balance according to the present application. DETAILED DESCRIPTION

[0014] In order to make the objectives, technical solutions and advantages of the present application more clear and explicit, the present application will be further described in detail below in conjunction with specific embodiments.

[0015] Embodiment 1:

[0016] Please refer to Figure 1 The present application provides a smart site selection method for cultivated land occupation and compensation balance, comprising the following steps: S1, collecting and preprocessing multi-source spatio-temporal data to construct a basic database; S2, constructing a spatial adjacency relationship graph and inputting the data in the basic database into a graph neural network model to generate a dynamic evaluation weight vector for the to-be-selected land block; S3, calculating the final utility value of each evaluation index, and combining the dynamic evaluation weight vector to calculate the multi-scale ecological balance score of the land block using a multiplicative aggregation model; S4, using a long short-term memory network model to predict the future quality evolution trend of the land block; S5, quantifying data uncertainty, model uncertainty and spatial uncertainty respectively, and comprehensively calculating the aforementioned three uncertainties to generate a total uncertainty index; S6, calculating a comprehensive resilience index, and linearly weighting the multi-scale ecological balance score, the future quality evolution trend and the comprehensive resilience index according to a strategic parameter to generate a forward-looking decision score; S7, judging the reliability of the site selection recommendation according to the total uncertainty index, and sorting the candidate land blocks according to the forward-looking decision score to output a final site selection recommendation list; To realize the above-mentioned method, a preferred embodiment is provided; the method aims to overcome the limitations of traditional static evaluation, and realizes dynamic, multi-scale, high-reliability and forward-looking decision-making for cultivated land occupation and compensation balance project site selection; In step S1, multi-source spatio-temporal data is collected and pre-processed to construct a basic database; the purpose of this step is to provide a standardized and high-quality data basis for subsequent analysis and modeling; in this embodiment, data collection and preprocessing are performed by a special data acquisition and preprocessing module; this module is responsible for integrating data of various sources and formats, and performing data cleaning, spatial coordinate registration, time series alignment, and numerical normalization, etc., to ensure that all data are analyzed under a unified spatio-temporal reference; In step S2, a spatial adjacency graph is constructed, and the data in the basic database is input into a graph neural network model to generate a dynamic evaluation weight vector for the selected land plots; this step aims to solve the core problem that the evaluation weight in the traditional method is fixed and cannot reflect spatio-temporal heterogeneity; the spatial adjacency graph is constructed, which is defined as a graph structure:

[0017] wherein the node set represents each selected arable land plot, and the edge set represents the mutual relationship between the plots in geographical space or ecological function; this relationship can be determined based on whether the Euclidean distance between the plots is less than a preset distance threshold or whether there is geographical adjacency; the determination method of the preset distance threshold is objective and repeatable; the threshold is determined by performing spatial autocorrelation analysis on the plots in the study area, such as calculating the Moran index Morans I; by calculating the Moran index under different distance scales, the distance value at which the spatial positive correlation is most significant and starts to decay is found, which is considered as the characteristic distance at which the plots have significant interaction, and thus is set as the preset distance threshold; for example, if it is analyzed that the soil nutrient content of the plots within a range of 500 meters presents the strongest spatial aggregation, then the threshold can be set as 500 meters; the pre-trained graph neural network model is used to process this graph; To enable those skilled in the art to implement it, the graph neural network model can specifically adopt an architecture containing two layers of graph attention networks; the node features input into the GNN are one-dimensional vectors composed of the time series of soil physicochemical properties of each plot in the past 5 years, such as organic matter, total nitrogen content, and vegetation coverage index after flattening; the first layer GAT network uses 8 attention heads to output a 64-dimensional feature vector; the second layer GAT network uses 1 attention head to output the final dynamic evaluation weight vector, the dimension of which is equal to the number of evaluation indicators, which ensures that the model can generate a precise corresponding and adaptive weight value for each evaluation indicator. In the training process, the mean square error is used as the loss function, the Adam optimizer is selected, the initial learning rate is set to 0.005, and 200 iterations are performed for training to ensure model convergence. The selection of the loss function aims to minimize the difference between the model predicted weight and the ideal weight calculated based on the historical project success cases. The network parameters of the GNN model are determined through supervised learning training on a dataset containing a large number of historical land acquisition and compensation projects and their subsequent evolutionary results. Characteristic data of the corresponding plots in the basic database, such as the time-series change vector composed of multi-period soil physical and chemical property monitoring data, are input into the model as node features. The model learns the plot's own properties and its interactions with neighboring plots to output a dynamic evaluation weight vector. This vector is multidimensional, with each dimension corresponding to the weight of an evaluation indicator, and can adaptively adjust with changes in temporal and spatial context. In step S3, the final utility value of each evaluation indicator is calculated, and a multiplicative aggregation model is used in combination with the dynamic evaluation weight vector to calculate the multi-scale ecological balance score of the plot. The purpose of this step is to achieve a comprehensive cross-scale evaluation and reflect the short board effect that limits the overall value when there are serious defects in key dimensions. The internal logic is to calculate the final utility value for each evaluation indicator. This value is a dimensionless value that has been standardized and reflects the contribution of the indicator. For indicators with threshold effects, a nonlinear transformation will be performed. Based on the above utility value, a multiplicative aggregation model is used to aggregate the dynamic evaluation weight vector from step S2 and the final utility value of each indicator. The mathematical expression of the multiplicative aggregation model is the weighted geometric mean model; for Multi-scale ecological balance scores for plots with evaluation indicators The calculation formula is:

[0018] in, It is The final utility value of the evaluation indicators comes from steps S31-S33, where n represents the total number of evaluation indicators owned by the land parcel; i represents the index of the i-th evaluation indicator; represents the measurement value of the i-th evaluation indicator; is generated by the graph neural network model, step S2, and Dynamic evaluation weights corresponding to each indicator; this formula ensures the utility value of any indicator When it approaches 0, the total score It will also approach 0 sharply, thus accurately reflecting the short board effect; The model is designed so that serious deficiencies in any dimension significantly lower the overall score, resulting in a multi-scale ecological balance score that comprehensively reflects the overall ecological health of a site at the micro-plot, meso-landscape, and macro-regional scales. In step S4, the long short-term memory network model is used to predict the future quality evolution trend of the plot; in order to make the decision forward-looking, this step aims to evaluate the long-term sustainability of the plot; the long short-term memory network model is constructed, which is particularly good at processing and predicting time series data; The LSTM model is composed of two stacked LSTM layers, each containing 100 neurons, and a fully connected layer is connected after the last layer to output the prediction result; to prevent overfitting, a dropout layer with a dropout rate of 0.2 is added after each LSTM layer; the input data format of the model is a three-dimensional tensor with dimensions of plot sample number, time step, and feature number, where the time step is set to the past 5 years, i.e. 5 time points, and the feature number is the soil organic matter content, vegetation cover index, soil total nitrogen content, soil available phosphorus content, and soil available potassium content collected at each time point. 5 key indicators; the goal of model training is to predict the single quantitative quality indicator in the future 5th year, and the mean square error is used as the loss function and the Adam optimizer is used to optimize the model parameters; The model takes multi-source monitoring time series data such as soil organic matter content and vegetation cover index as input to predict the quality evolution trend of the plot in the future preset period, such as 5-10 years; the output result is a quantitative indicator representing the quality change rate, which is normalized and used for final decision scoring; In step S5, data uncertainty, model uncertainty, and spatial uncertainty are quantified respectively, and the three uncertainties are integrated to generate the overall uncertainty indicator; in order to ensure the reliability of the decision, this step quantifies the main sources of uncertainty in the site selection evaluation process; data uncertainty is evaluated according to the reliability, accuracy, and timeliness of the data source; model uncertainty is evaluated by technical means such as Monte Carlo dropout method to evaluate the stability of the model itself prediction such as GNN and LSTM; spatial uncertainty is quantified by calculating the variation degree of the key indicators of the target plot in its spatial neighborhood; the three uncertainties are then combined to generate the overall uncertainty indicator, which is used to measure the credibility of the final site selection recommendation; In step S6, the comprehensive resilience index is calculated, and the multi-scale ecological balance score, the future quality evolution trend and the comprehensive resilience index are linearly weighted according to the strategic parameters to generate the forward-looking decision score; this step integrates all evaluation dimensions to form the final basis for sorting; as a specific implementation, the comprehensive resilience index is calculated, which quantifies the ability of the plot to resist and recover from external disturbances such as extreme climate events, usually derived by weighting multiple indicators such as biodiversity, soil aggregate stability, etc.; this index is linearly weighted with the current quality calculated in step S3, i.e. the multi-scale ecological balance score, and the future trend predicted in step S4; the strategic parameters used for weighting are set by the decision maker according to his strategic preferences, for example, prioritizing current yield or valuing long-term sustainability more; in order to assist the decision maker in scientific setting, the system can provide a variety of pre-set standardized decision scenario modes, each mode corresponding to a fixed set of strategic parameters, ensuring wherein, is the weight of the multi-scale ecological balance score reflecting the current comprehensive quality of the plot; is the weight of the future quality evolution trend predicting the future development potential of the plot; is the weight of the comprehensive resilience index evaluating the long-term stability of the plot; Yield priority mode: suitable for emergency food production tasks, focusing on current plot quality. Parameter settings can be: ; Sustainable development mode: suitable for ecologically sensitive areas or long-term planning, focusing on future potential and stability. Parameter settings can be: ; Balanced development mode: taking into account current, future and risk. Parameter settings can be: ; The decision maker can directly choose one mode, or fine-tune based on these modes, so that his preferences can be quantified and realized; the forward-looking decision score is calculated; In step S7, the reliability of the site selection recommendation is judged according to the overall uncertainty index, and the candidate plots are sorted according to the forward-looking decision score to output the final site selection recommendation list; this is the decision output step; the system compares the overall uncertainty index generated in step S5 with the pre-set threshold value; if it is lower than the threshold value, it is considered that the evaluation result of the plot is reliable; otherwise, it is marked as high uncertainty and recommended for manual review; all candidate plots judged to be reliable are sorted in descending order according to the forward-looking decision score generated in step S6 to form a clear and well-founded final site selection recommendation list for the decision maker; The present application constructs a complete and closed-loop intelligent site selection method through the above steps; based on multi-source spatio-temporal data, the spatio-temporal dynamics are captured by using a graph neural network, a multi-scale multiplicative aggregation model is used to realize comprehensive evaluation, a long short-term memory network is used to give foresight to the decision, and uncertainty quantification is introduced to ensure the reliability of the results; compared with the traditional method which relies on static weight and expert experience, the present application can significantly improve the scientificity, accuracy and long-term sustainability of the balance of cultivated land occupation and compensation, and provides advanced technical support for guaranteeing national food security and ecological balance.

[0019] The multi-source spatio-temporal data includes: historical occupation and compensation project geographic spatial information, soil physicochemical properties at plot scale, landscape pattern index based on remote sensing image interpretation, and regional scale ecosystem service function data and climate change data; In the present embodiment, the multi-source spatio-temporal data in step S1 is specifically defined; the selection of data directly determines the depth and breadth of model analysis; in the present embodiment, the construction of the basic database covers four key categories of data: The historical occupation and compensation project geographic spatial information refers to the location, range, implementation time, project type and post-project benefit evaluation data of the past completed cultivated land occupation and compensation projects; the role is to provide samples with real result labels for the training of machine learning models such as GNN and LSTM, so that the model can learn the relationship between site selection advantages and disadvantages and long-term evolution results of the plot; The soil physicochemical properties at plot scale refer to high-precision data collected for each plot, such as soil organic matter, total nitrogen, available phosphorus, available potassium content, pH value, soil texture and bulk density, etc.; the role is to provide the most direct basis for evaluating the current basic soil fertility and production potential of the plot, and it is the core data input at the microscopic scale in multi-scale evaluation; The landscape pattern index based on remote sensing image interpretation refers to the quantitative indicators describing the landscape structure of the region where the plot is located, extracted by analyzing and processing multi-period high-resolution satellite remote sensing images, such as patch density, edge index, aggregation index and Shannon diversity index, etc.; the role is to evaluate the ecological connectivity and habitat quality of the plot from the meso-landscape scale, reflecting its role in the wider ecological network; The regional scale ecosystem service function data and climate change data refer to macro data covering the entire study area; the ecosystem service function data include spatial distribution evaluation results of functions such as water conservation, soil and water conservation, and biodiversity maintenance; the climate change data include historical and future estimated precipitation, temperature and other meteorological factors; the role is to evaluate the macro ecological background of the plot and its long-term climate risks from the macro regional scale, providing support for the comprehensiveness and foresight of the evaluation; By explicitly defining the specific composition of the above multi-source spatio-temporal data, the comprehensiveness and multi-scale characteristics of the input information are ensured; such data combination enables the model to comprehensively consider the suitability of the plot from multiple levels such as soil micro-attribute, surrounding landscape structure, regional macro-ecological function, and climate background, thereby bringing the technical effect of significantly improving the input information dimension and quality of the evaluation model, making the basis for site selection decision-making more sufficient and stereoscopic, and the final evaluation result more accurate and reliable; The spatio-temporal adaptive dynamics of the evaluation weight is realized; the conventional technology generally adopts an analytic hierarchy process or the like to assign static weights to each evaluation index, which do not change once determined; such processing mode cannot reflect the high heterogeneity of the cultivated land quality in space and the dynamic evolution in time; the core breakthrough of the present scheme lies in introducing a model based on a graph neural network for generating a dynamic evaluation weight vector; the model abstracts each plot as a node in a graph and abstracts the interaction between plots as an edge, and through learning a large amount of historical data, can calculate a set of most suitable weights in real time according to the state of each plot at a specific time point and the influence of its neighborhood; this makes the evaluation process free from the shackles of subjective setting and static framework, and can accurately capture and respond to the spatio-temporal dynamic characteristics of the cultivated land ecosystem, so that the evaluation result is more consistent with the physical reality.

[0020] S3 specifically includes: S31, for a key index with a threshold effect, obtaining a normalized measurement value and a corresponding ecological critical threshold value; S32, when the measurement value is lower than the ecological critical threshold value, a first preset function is used to calculate a final utility value with exponentially decaying utility; S33, when the measurement value is not lower than the ecological critical threshold value, a second preset function is used to calculate a final utility value with non-linearly increasing utility; In the present embodiment, the process of calculating the final utility value in step S3 is specified, in particular for a key index with an ecological threshold effect; the threshold effect refers to the fact that the influence of the change of some ecological indexes, such as the soil nutrient content, on the function of the ecological system is not linear, and once the value is lower than or higher than a certain critical point, the state of the system can change dramatically; in order to accurately simulate this non-linear response, the present embodiment introduces a non-linear utility function with ecological threshold perception; In step S31, for the key indicators with threshold effect, the normalized measurement value and the corresponding ecological critical threshold value are obtained; the key indicators are, for example, soil organic matter content or concentration of a certain type of pollutant; the normalized measurement value refers to the dimensionless value obtained by scaling the original observation data to the [0, 1] interval through mathematical transformation such as minimum-maximum normalization; the ecological critical threshold value is a key reference point, which is determined according to relevant ecological research, national or local agricultural production technical regulations or expert knowledge, and is also subjected to the same normalization processing as the measurement value; In steps S32 and S33, the final utility value is calculated by a piecewise function ; To illustrate the calculation logic, when the measurement value is lower than the ecological critical threshold value , a first preset function is used:

[0021] to calculate the utility value; when the measurement value is not lower than the ecological critical threshold value , a second preset function is used:

[0022] to calculate the utility value; wherein, exp: exponential function; i: index of the ith key indicator; is the normalized measurement value of the ith key indicator; is the normalized ecological critical threshold value corresponding to ; and are preset dimensionless coefficients, respectively controlling the rate of utility decay and growth, the values of which are calibrated according to the strictness level of ecological protection policy in specific application scenarios, for example, for basic farmland that needs to be strictly protected, a larger value can be set to implement strong punishment for any behavior below the threshold; To objectively map the policy level and the coefficient, a calibration table can be predefined; for example, three protection levels are defined: I level, within the ecological red line area, most stringent; II level, permanent basic farmland; III level, general farmland; through expert consultation and historical data fitting, specific coefficient ranges are set for each level; for example: I level area: , with a high value to strongly punish any condition below the threshold; II level area: ; III level area: , allowing relatively moderate utility change; ​In specific applications, according to the protection level to which the land belongs, the median value in the range is taken or fine-tuning is made according to more refined policy documents; When the measured value is lower than the ecological critical threshold , a first preset function is adopted; the function is an exponential decay function, characterized in that the utility value will sharply decrease and tend to zero when is further away and lower than ; this simulates the scenario of a cliff-like drop in farmland productivity or ecological health condition when a key indicator is lower than the critical level; When the measured value is not lower than , i.e. greater than or equal to the ecological critical threshold , a second preset function is adopted; the function is a nonlinear growth function, characterized in that the utility value will rapidly increase when exceeds , but the growth rate gradually slows down and tends to 1; this reflects the positive benefits brought by the indicator exceeding the threshold, and also embodies the general law of diminishing marginal returns; By introducing this nonlinear utility function, the nonlinear response and threshold effect in the farmland ecological system can be more realistically simulated and quantified, overcoming the defects of traditional linear weighted models that cannot accurately represent such abrupt phenomena; this makes the calculation results of the multi-scale ecological balance score more sensitive and accurate to changes in key indicators, thereby improving the scientificity and risk avoidance ability of site selection decisions; A nonlinear evaluation model that can reflect the internal laws of the ecological system is constructed; existing technologies mostly use linear weighted summation to calculate the total score, which implies a premise that the contribution of each indicator to the overall quality is independent and linear; this is contrary to the actual behavior of the ecological system, especially unable to represent the threshold effect; this scheme solves this problem in two aspects; when calculating the multi-scale ecological balance score , a multiplicative aggregation model is adopted, the physical connotation of which is that the overall health status of the system depends on the coordinated performance of all key dimensions, and a serious shortcoming in any dimension, i.e. a very low utility value, will cause a sharp drop in the overall score, which precisely corresponds to the short board effect or limiting factor law in the ecological system; for key indicators with threshold effect, the final utility value is calculated by a piecewise nonlinear function; below the ecological critical threshold , the utility is exponentially decaying:

[0023] wherein, : indicator measured value, : critical threshold, : attenuation coefficient; above the threshold, it presents a nonlinear growth:

[0024] wherein, : index measurement value, : critical threshold, : growth coefficient; this design enables the model to sensitively capture the mutant behavior of the ecological state near the critical point, avoiding the misjudgment of potential ecological risks due to linear approximation.

[0025] S5 specifically includes: S51, quantifying data uncertainty according to the reliability of the data source; S52, using Monte Carlo rejection technique, by repeatedly randomly inactivating part of the neurons in the graph neural network model and observing the output variance, to estimate the model uncertainty; S53, calculating the normalized standard deviation of the value of the target plot key indicator in all plots within the spatial neighborhood, as the spatial uncertainty; S54, square sum and square root operation on data uncertainty, model uncertainty and spatial uncertainty to generate overall uncertainty indicators; In this embodiment, the generation process of the overall uncertainty indicator in step S5 is described in detail; this process is performed by the uncertainty quantification module, aiming to comprehensively evaluate the reliability of the decision from the data, model and space three dimensions; In step S51, quantifying data uncertainty according to the reliability of the data source (D) ); The reliability of the data source is a score comprehensively evaluated according to the data source authority, measurement accuracy, spatial resolution and time stamp freshness according to the preset scoring rules; For example, authoritative ground station data from national soil survey is given high reliability and low uncertainty, while data indirectly inverted through low resolution remote sensing images is given lower reliability and higher uncertainty; By assigning uncertainty weights to data from different sources and integrating them, the quantified data uncertainty indicator D is obtained ; The comprehensive calculation formula can be the weighted square sum of the uncertainty of the data sources used by the evaluated plot and then square root, the formula is as follows:

[0026] wherein, is the normalized weight of the th data source to the final evaluation contribution, is the uncertainty score of the data source, a value between 0 and 1, obtained by looking up the metadata such as its source, accuracy and timeliness, for example, national measured data low resolution remote sensing interpretation data ; the formula can reasonably aggregate the uncertainty of different data sources; k: index of the kth data source; In step S52, the Monte Carlo dropout technique is used to estimate the model uncertainty ; the Monte Carlo dropout technique is a method widely used in deep learning for estimating model uncertainty; in this embodiment, for the graph neural network model used in step S2, when making predictions, a portion of the neurons are randomly discarded, i.e. temporarily disabled, with a preset probability, e.g. p = 0.1; this process is repeated multiple times, e.g. N = 50 times, each time resulting in a slightly different dynamically evaluated weight vector; by calculating the variance of the N output results, the stability of the model prediction can be quantified, and the square root of the variance is defined as the model uncertainty ; therefore, the larger the standard deviation of the output result, the less stable the model's prediction for the plot, and the higher the model uncertainty ; In step S53, the normalized standard deviation of the value of the target plot key indicator in all plots within the spatial neighborhood is calculated as the spatial uncertainty ; this step aims to quantify the uncertainty caused by spatial heterogeneity; the definition of the spatial neighborhood is all other plots within a certain geographic radius, e.g. 500 meters, around the target plot; select one or more key indicators sensitive to spatial variation, e.g. soil organic matter, and calculate the standard deviation of the values of this indicator in the target plot and all neighboring plots; the larger the standard deviation, the stronger the spatial heterogeneity of the region, and the lower the representativeness of the measurement value of a single plot, thus the higher the spatial uncertainty ; to eliminate the dimension effect, the standard deviation needs to be normalized; In step S54, the data uncertainty, model uncertainty and spatial uncertainty are squared and square-rooted to generate the overall uncertainty indicator ; the three uncertainties are considered as independent error sources; the square sum and square root commonly used in error propagation theory, i.e. the Euclidean distance method, is used to combine them, and the calculation formula is as follows:

[0027] wherein, is the overall uncertainty, is the data uncertainty, is the model uncertainty, is the spatial uncertainty; By the above specific steps, the present application can unify uncertainties of different sources and different properties into a measurable framework; the technical effect brought by this approach is: the overall, quantitative evaluation of the reliability of site selection decision is realized; not only the quality of input data and the stability of the model itself are considered, but also the consideration of spatial heterogeneity is innovatively introduced, so that the final overall uncertainty index can more accurately reflect the decision risk, providing a clear direction for the subsequent manual review link, significantly enhancing the robustness and credibility of the entire intelligent site selection system; The probability and uncertainty quantification throughout the decision-making process are introduced; traditional site selection methods usually give deterministic ranking results, but do not provide a measure of the reliability of the results, and the decision maker has no way to know the risk hidden behind the evaluation results; the present scheme establishes a complete uncertainty quantification framework, and realizes the overall uncertainty index by calculating the overall uncertainty index

[0028] Among them, : data uncertainty, : model uncertainty, : spatial uncertainty; the theoretical basis of this formula is derived from the error propagation theory, which synthesizes the errors in three orthogonal dimensions, i.e., data uncertainty, model uncertainty, and spatial uncertainty; this makes each site selection suggestion accompanied by a clear confidence interval; when the overall uncertainty index exceeds the preset threshold, the system will suggest manual review, thereby establishing a risk control mechanism for human-machine cooperation; this is a fundamental progress from giving answers to giving answers with confidence.

[0029] The comprehensive resilience index in S6 is obtained by weighted summation of the normalized values of multiple resilience-related indicators, and the weights of each indicator are determined by domain experts; As an embodiment of determining the weights by using the Delphi method, a group of consensus weights reached by 10 soil science and ecology experts for the resilience-related indicators mentioned in step S6 may be as follows: Biodiversity index: Soil aggregate stability index: Landscape connectivity index: Soil water regulation capacity index: Ensure that the sum of all weights is 1, these weight values make the calculation process of the comprehensive resilience index completely determined; In this embodiment, the comprehensive resilience index in step S6​​​​ The calculation method of the comprehensive resilience index is specifically explained; the purpose of the comprehensive resilience index is to quantify the ability of cultivated land ecosystems to maintain their key functions and structures and recover from external disturbances such as drought, flooding, pests and diseases. Land with strong resilience has better long-term stability and sustainability; In this embodiment, The calculation of is a multi-criteria weighted summation process. To further illustrate, a series of resilience-related indicators directly related to ecological resilience are selected. These indicators may include: biodiversity index, soil aggregate stability index, landscape connectivity index, soil moisture regulation capacity index, etc. The original data of these indicators are obtained from the basic database constructed in step S1. The raw measured values ​​of these indicators were normalized and converted into dimensionless values ​​for comparison and weighting. The comprehensive resilience index was calculated using the following weighted summation formula:

[0030] in, is the comprehensive resilience index; j: the index of the jth resilience-related indicator; For the The weight of the indicator, For the Normalized value of each indicator; in, It is The key lies in the determination of the weights, i.e. The weight of each indicator is determined by domain experts; domain experts refer to scholars or technicians with profound knowledge and practical experience in fields such as ecology, soil science, and agronomy. The weight can be determined by using mature expert scoring techniques such as the Delphi method and the analytic hierarchy process. Multiple experts will independently score and demonstrate the relative importance of each indicator for long-term resilience, and ultimately reach a consensus and ensure that the sum of all weights is 1. ; By defining the calculation method for the comprehensive resilience index in this way, the present invention achieves the following technical benefits: It transforms the complex ecological concept of resilience into an actionable and quantifiable evaluation metric. By integrating multiple, expertly validated indicators, it introduces a new evaluation dimension for site selection that focuses on long-term stability and risk resilience. This makes the final forward-looking decision score more comprehensive and insightful, helping to identify high-quality sites that are not only of excellent current quality but also have the potential to remain healthy over the long term. Extend the decision-making perspective from current status assessment to future trend prediction and long-term resilience considerations; existing technologies often only focus on the current snapshot of the plot and lack the assessment of long-term sustainability; this solution uses the long-short-term memory network model to predict the future quality evolution trend of the plot Make predictions; further, construct a comprehensive resilience index;

[0031] in, : jth resilience index, : corresponding weights to quantify the ability of the plot to resist and recover from external disturbances; the final forward-looking decision score The current quality 、Future Trends and resilience Linear weighting of the three;

[0032] in, :Strategic parameters; The practical significance of this model is that it provides a highly flexible decision-making framework, and decision makers can adjust the strategic parameters The value of the site selection is determined by weighing and optimizing multiple strategic goals, such as ensuring current production, pursuing future production potential, and ensuring long-term ecological stability, so that the site selection decision is not only scientific but also precisely aligned with the macro-policy goals.

[0033] The reliability of the site selection recommendations in S7 includes: S71. comparing the overall uncertainty index with a preset uncertainty threshold; S72. When the overall uncertainty index is less than the uncertainty threshold, the site selection suggestion is determined to be reliable; S73. When the overall uncertainty index is not less than the uncertainty threshold, the site selection suggestion is marked as high uncertainty and manual review is recommended; In this embodiment, the logic of determining the reliability of the site selection suggestion in step S7 is explained; this determination process is a key step in the quality control of the system output; In step S71, the overall uncertainty index is compared with a preset uncertainty threshold; the overall uncertainty index is the uncertainty index calculated in step S5. The value comprehensively reflects the uncertainty of data, model and space; the preset uncertainty threshold is a key decision parameter; the setting of this threshold is not arbitrary, but is determined based on a rigorous method; in this embodiment, It is calibrated by backtesting a large number of historical site selection cases; the goal is to strike a balance between the efficiency of automated screening and the risk of misjudgment or omission; for example, It is set to the value corresponding to the ability to correctly identify 95% of successful site selection cases as reliable in the historical case data set. Statistical quantiles of values, thereby controlling decision risks; In steps S72 and S73, an explicit binary decision logic is performed: When the overall uncertainty index is less than the uncertainty threshold, that is, , the system automatically determines the site selection suggestion for the plot as reliable; this means that based on the currently available data and model analysis, the assessment result of the plot, that is, the forward-looking decision score, has a high confidence level and can directly enter the subsequent sorting and list generation stages; When the overall uncertainty index is not less than, that is, greater than or equal to, the uncertainty threshold, that is, , the system marks the site proposal as high uncertainty and recommends manual review. This does not mean that the site is necessarily unsuitable, but rather that the current automated assessment process cannot provide a sufficiently reliable conclusion. Once marked as high uncertainty, the system will list it separately and may provide the main source of uncertainty, such as data quality issues or excessive model prediction variance, so that domain experts can conduct targeted field surveys, supplementary data collection, or model verification to make the final manual judgment. Through the clear reliability judgment process described above, a risk management mechanism combining automated assessment with manual supervision has been established. This avoids blind trust in model output results and defines the boundaries of machine decision-making through quantitative and well-founded thresholds. This not only ensures the high reliability of the final output list of site recommendations, but also greatly improves the efficiency and pertinence of expert manual review by accurately pointing to cases with high uncertainty, thereby optimizing human-machine collaborative decision-making.

[0034] Example 2:

[0035] A smart site selection system for balancing cultivated land occupation and compensation, including: Data acquisition and preprocessing module, used to collect and preprocess multi-source spatiotemporal data to build a basic database; The spatiotemporal weight adaptive learning module is used to construct a spatial adjacency graph and input data from the basic database into the graph neural network model to generate dynamic evaluation weight vectors for the selected plots; The multi-scale ecological balance assessment module is used to calculate the final utility value of each assessment indicator and use a multiplicative aggregation model in combination with the dynamic assessment weight vector to calculate the multi-scale ecological balance score of the plot; The cultivated land quality evolution prediction module is configured to predict a future quality evolution trend of the land plot by using a long short-term memory network model. The uncertainty quantification module is configured to quantize data uncertainty, model uncertainty and spatial uncertainty respectively, and to comprehensively calculate the three uncertainties to generate an overall uncertainty index. The decision support module is configured to calculate a comprehensive resilience index, generate a forward-looking decision score, judge reliability according to the overall uncertainty index, and output a final site selection suggestion list according to the forward-looking decision score. The present application also provides a smart site selection system for implementing the above method; the system can be deployed on a server or a cloud platform and is composed of a series of software modules that are independent in function and tightly coupled in data, and collectively perform the smart site selection process. The data acquisition and preprocessing module is a unified data entry of the system and is responsible for performing step S1 in the method; the module is built-in with a variety of data interfaces that can access databases of different sources, such as land resources databases, weather databases and file formats such as Shapefile, GeoTIFF and CSV, and contains a series of data processing tools for realizing automatic cleaning, coordinate registration, time series alignment and normalization of data, to generate a standardized basic database that can be called by other modules. The spatio-temporal weight adaptive learning module is a dynamic weight generation unit of the system and is responsible for performing step S2; the core of the module is a weight learning engine based on a graph neural network; it first automatically constructs a spatial adjacency relationship graph according to the geographic information of the land plot, then calls a pre-trained GNN model to process the spatio-temporal data in the basic database, dynamically generates an evaluation weight vector that changes with time and spatial context for each candidate land plot, and passes it to the multi-scale ecological balance evaluation module; The multi-scale ecological balance evaluation module is responsible for performing step S3; the module receives the dynamic weight vector from the spatio-temporal weight module and the multi-scale index data in the basic database; it internally implements a multiplicative aggregation algorithm and embeds a nonlinear utility function for key indicators; the core function of the module is to calculate a multi-scale ecological balance score that comprehensively reflects the current quality of the land plot; The cultivated land quality evolution prediction module is responsible for performing step S4 and provides a forward-looking perspective for decision-making; the core of the module is a pre-trained long short-term memory network model; it extracts historical time series data of the land plot from the basic database, predicts the evolution trend of its future quality, and outputs the quantized trend result to the decision support module. The uncertainty quantification module is responsible for performing step S5 and is a technical component for guaranteeing the reliability of system output; the module integrates three uncertainty evaluation algorithms, corresponding to data uncertainty, model uncertainty based on Monte Carlo dropout technology and spatial uncertainty based on spatial statistics respectively; the overall uncertainty index is calculated by merging formula and provided to the decision support module; The decision support module is an integrated and decision output unit of the system and is responsible for performing steps S6 and S7; the module calculates a comprehensive resilience index according to expert weights; as an integrator, the module linearly weights the current quality score, future trend and resilience index input by other modules to generate a final forward-looking decision score; at the same time, the module receives the output of the uncertainty quantification module and judges the reliability of the evaluation result of each plot according to a preset threshold; the module sorts all reliable site selection suggestions according to the forward-looking decision score, generates and visualizes a final site selection suggestion list; The system provided by the present application divides the complex intelligent site selection task into a series of clear and achievable functional units through the above modular structure design; the modules work cooperatively to completely realize the above technical solution; the system deeply integrates advanced artificial intelligence algorithms with geographic information systems and ecological models to form an automated, intelligent and highly reliable decision support tool; the overall technical effect is: an end-to-end cultivated land occupation and compensation balance intelligent site selection solution is provided, which changes the site selection process from a traditional mode relying on subjective experience to a scientific decision-making mode driven by data and supported by models, significantly improving the efficiency, accuracy, comprehensiveness and forward-looking nature of site selection work.

[0036] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A smart site selection method for balancing cultivated land occupation and compensation, characterized by: The steps include: S1. Collect and preprocess multi-source spatiotemporal data to build a basic database; S2. Construct a spatial adjacency graph and input the data from the basic database into the graph neural network model to generate a dynamic evaluation weight vector for the selected plots. S3. Calculate the final utility value of each evaluation indicator and use a multiplicative aggregation model in combination with the dynamic evaluation weight vector to calculate the multi-scale ecological balance score of the plot; S4. Use the long short-term memory network model to predict the future quality evolution trend of the land parcel; S5. Quantify data uncertainty, model uncertainty, and spatial uncertainty separately, and perform a comprehensive calculation of the three uncertainties to generate an overall uncertainty index; S6. Calculate the comprehensive resilience index and linearly weight the multi-scale ecological balance score, future quality evolution trend, and comprehensive resilience index based on strategic parameters to generate a forward-looking decision-making score; S7. Determine the reliability of the site selection recommendations based on the overall uncertainty index and rank the candidate sites based on the forward-looking decision scores to output the final list of site selection recommendations.

2. The intelligent site selection method for balancing cultivated land occupation and compensation according to claim 1 is characterized in that: Multi-source spatiotemporal data include: geospatial information of historical land occupation and compensation projects, soil physical and chemical properties at the plot scale, landscape pattern index based on remote sensing image interpretation, and ecosystem service function data and climate change data at the regional scale.

3. The intelligent site selection method for balancing cultivated land occupation and compensation according to claim 1 is characterized in that: S3 specifically includes: S31. For key indicators with threshold effects, obtain normalized measurement values ​​and corresponding ecological critical thresholds; S32. When the measured value is lower than the ecological critical threshold, a first preset function is used to calculate a final utility value with exponential decay of utility; S33. When the measured value is not lower than the ecological critical threshold, a second preset function is used to calculate a final utility value in which the utility increases nonlinearly.

4. The intelligent site selection method for balancing cultivated land occupation and compensation according to claim 1 is characterized in that: S5 specifically includes: S51. Quantify data uncertainty based on the reliability of the data source; S52. Monte Carlo dropout technique is used to estimate model uncertainty by randomly inactivating some neurons in the graph neural network model multiple times and observing the output variance; S53, calculating the normalized standard deviation of the key indicator of the target plot for all plots in the spatial neighborhood as the spatial uncertainty; S54. Perform square root operations on the data uncertainty, model uncertainty, and spatial uncertainty to generate an overall uncertainty index.

5. The intelligent site selection method for balancing cultivated land occupation and compensation according to claim 1 is characterized in that: The comprehensive resilience index in S6 is obtained by weighted summing of the normalized values ​​of multiple resilience-related indicators, with the weights of each indicator determined by domain experts.

6. The intelligent site selection method for balancing cultivated land occupation and compensation according to claim 1 is characterized in that: The reliability of the site selection recommendations in S7 includes: S71. comparing the overall uncertainty index with a preset uncertainty threshold; S72. When the overall uncertainty index is less than the uncertainty threshold, the site selection suggestion is determined to be reliable; S73. When the overall uncertainty index is not less than the uncertainty threshold, the site selection suggestion is marked as high uncertainty and manual review is recommended.

7. A smart site selection system for cultivated land occupation and compensation balance, applied to a smart site selection method for cultivated land occupation and compensation balance according to any one of claims 1 to 6, characterized in that: include: Data acquisition and preprocessing module, used to collect and preprocess multi-source spatiotemporal data to build a basic database; The spatiotemporal weight adaptive learning module is used to construct a spatial adjacency graph and input data from the basic database into the graph neural network model to generate dynamic evaluation weight vectors for the selected plots; The multi-scale ecological balance assessment module is used to calculate the final utility value of each assessment indicator and use a multiplicative aggregation model in combination with the dynamic assessment weight vector to calculate the multi-scale ecological balance score of the plot; The cultivated land quality evolution prediction module is used to predict the future quality evolution trend of the land using the long-short-term memory network model; Uncertainty quantification module, which is used to quantify data uncertainty, model uncertainty and spatial uncertainty respectively, and comprehensively calculate the above three uncertainties to generate an overall uncertainty index; The decision support module is used to calculate the comprehensive resilience index, generate a forward-looking decision score, judge the reliability based on the overall uncertainty index, and output the final list of site selection recommendations based on the forward-looking decision score.

Images