Adjusting cluster division method for hidden abandoned risk

Abstract

The invention relates to an adjustment cluster division method for hidden abandoned risks, and the method comprises the steps: extracting hidden abandoned historical sample points from a territorial survey data set; performing contribution rate and correlation test on the initial driving factors in the initial driving factor data set; performing risk probability prediction on the recessive abandoned land by using a preset maximum entropy machine learning model; carrying out main control factor nonlinear response analysis for resisting the hidden abandoned risk; according to the non-linear response characteristics, constructing a hidden abandoned land risk adjustment effect function considering the adjustment income and adjustment difficulty of the main control factor; and identifying a main control factor adjustment effect cluster meeting a preset medium-high recessive abandoned land risk condition, and dividing the main control factor adjustment effect cluster to generate a cluster management and control strategy of the recessive abandoned land. Therefore, the problems that the recessive abandoned land in a planting-free state is neglected and the prior knowledge of the historical recessive abandoned land is not utilized to carry out risk prediction and regulation management and control research in the prior art are solved.

Description

Technical Field

[0001] The present application relates to the technical field of farmland protection and utilization, and in particular to a method for regulating cluster division of hidden abandonment risks. Background Art

[0002] The stable production capacity of cultivated land is essential for achieving the United Nations Sustainable Development Goal 2 (Zero Hunger). However, driven by multiple factors such as low grain production returns and the outflow of rural labor, the phenomenon of abandoned cultivated land is widespread worldwide. Abandoned cultivated land refers to the state of desolation caused by a lack of effective farming or management practices. This abandonment not only directly leads to a reduction in grain planting area and production potential, but also exacerbates declining phenomena such as rural hollowing out, marginalization, and non-agriculturalization of cultivated land in the context of urbanization, posing a significant threat to the stable production capacity of cultivated land. Therefore, controlling abandoned cultivated land to address the food crisis has become a key issue for scholars and policymakers.

[0003] Cultivated land abandonment can be divided into two stages based on its occurrence. The first stage is latent abandonment, which occurs when cultivated land remains unplanted and not fallow for two consecutive years, but the land type remains unchanged. The second stage is explicit abandonment, characterized by a change in land type. During this stage, cultivated vegetation gradually transitions to grassland, shrubs, or forest. While research and practice on explicit abandonment are extensive, latent abandonment, characterized by the absence of cultivated land, has received less attention, often overlooking its impact on food security. Furthermore, existing research and practice primarily focus on abandoned cultivated land, lacking approaches to predicting and regulating abandonment risk. In recent years, several studies have employed econometric models such as Probit, Logistic, Logit, and Tobit, as well as spatial explanatory models such as GWR and MGWR, to explore the driving mechanisms of abandonment from both global and local perspectives, aiming to provide a basis for spatial management. These studies have provided sound explanatory models for the abandonment phenomenon, but have failed to fully leverage prior knowledge of the driving mechanisms to further diagnose the future probability of abandonment. The booming field of machine learning models offers a solution to this problem. Machine learning involves training models using massive data sets, enabling them to understand the underlying relationships within the data and accurately predict outputs for new input data. Therefore, machine learning models can be used to learn the relationships between existing abandonment cases and driving factors, predicting abandonment risk in a probabilistic manner.

[0004] In summary, existing studies have not only ignored the hidden abandonment of land without planting, but have also not used the prior knowledge of historical hidden abandonment to carry out risk prediction and regulation and control research. Summary of the Invention

[0005] This application provides a method for regulating cluster division of hidden abandonment risks to solve the problem that related technologies not only ignore hidden abandonment in the non-planted state, but also have not yet used the prior knowledge of historical hidden abandonment to carry out risk prediction and regulation and control research.

[0006] An embodiment of the present application provides a method for dividing hidden abandonment risk into regulation clusters, comprising the following steps: collecting a land survey dataset and an initial driving factor dataset of a target study area; extracting historical sample points of hidden abandonment from the land survey dataset; based on the historical sample points, performing contribution rate and correlation tests on the initial driving factors in the initial driving factor dataset to generate the main controlling factors of the hidden abandonment; using a preset maximum entropy machine learning model to perform risk probability prediction on the hidden abandonment to generate a risk probability prediction result of the hidden abandonment; based on the risk probability prediction result, conducting a nonlinear response analysis of the main controlling factors for resisting hidden abandonment risk to determine the nonlinear response characteristics of the main controlling factors to the hidden abandonment risk; constructing a hidden abandonment risk regulation effect function that takes into account the regulation benefits and regulation difficulty of the main controlling factors based on the nonlinear response characteristics; based on the hidden abandonment risk regulation effect function, identifying the main controlling factor regulation effect clusters that meet the preset medium and high hidden abandonment risk conditions, and dividing the main controlling factor regulation effect clusters to generate a cluster management strategy for the hidden abandonment.

[0007] Optionally, in one embodiment of the present application, the extracting of historical sample points of hidden abandonment from the land survey dataset includes: extracting the cultivated land patches of the uncultivated farmland based on the planting status attributes of grain crops, non-grain crops, grain-non-grain rotation, fallow, intercropping of forest and grain, and uncultivated farmland in the land survey dataset; determining the historical hidden abandoned farmland of hidden abandonment based on the cultivated land patches of the uncultivated farmland, and extracting the historical sample points of hidden abandonment from the historical hidden abandoned farmland.

[0008] Optionally, in one embodiment of the present application, the contribution rate and correlation test of the initial driving factors in the initial driving factor data set to generate the main controlling factors of the latent abandonment includes: importing the historical sample points of latent abandonment and the initial driving factors into the target operation software to obtain the contribution rate of the initial driving factors to the latent abandonment probability prediction results, and eliminating the driving factors with a contribution rate lower than the target percentage to generate a primary screening driving factor; extracting the driving factor values ​​of the historical sample points of latent abandonment, and extracting the driving factor pairs with the absolute value of the correlation higher than the target value in the driving factor values, and eliminating the driving factor pairs with a contribution rate lower than the target value in the driving factor pairs to generate a secondary screening driving factor; generating the main controlling factors of the latent abandonment based on the primary screening driving factors and the secondary screening driving factors.

[0009] Optionally, in one embodiment of the present application, the risk probability prediction of the latent abandonment is performed using a preset maximum entropy machine learning model to generate a risk probability prediction result of the latent abandonment, including: inputting the historical sample points of the latent abandonment and the main controlling factors into the preset maximum entropy machine learning model, setting preset sample points to train and predict the preset maximum entropy machine learning model to generate a trained model, and using the remaining sample points as a test set to verify the trained model to generate a verified model; outputting the risk probability simulation result of the latent abandonment according to the verified model, and using the area under the receiver operating characteristic curve (ROC) AUC to test the prediction accuracy of the risk probability simulation result to generate the risk probability prediction result.

[0010] Optionally, in one embodiment of the present application, determining the nonlinear response characteristics of the main control factor to the hidden abandonment risk includes: using the preset maximum entropy machine learning model to determine the response curve of the main control factor to the risk probability; displaying the development trend of the hidden abandonment risk probability as the main control factor changes according to the response curve, and determining the nonlinear response characteristics of the main control factor to the hidden abandonment risk based on the development trend.

[0011] Optionally, in one embodiment of the present application, the calculation formula of the implicit abandonment risk adjustment effect function is:

[0012]

[0013] Among them, AdjEff i,j represents the moderating effect of the dominant factor i with value j, and They represent the expected risk changes before and after the adjustment of the main control factor i in the negative and positive adjustment effect functions, respectively. and They represent the adjustable proportions of the main control factor i in the negative and positive regulation effect functions respectively, n represents the number of segments in the regulation range, t is the main control factor value corresponding to the lowest risk in the response curve of the main control factor i, j represents any value of the main control factor i, min is the minimum value of the main control factor i, and max is the maximum value of the main control factor i.

[0014] Optionally, in one embodiment of the present application, the identification of the clusters of regulatory effects of the main controlling factors that meet the preset medium and high hidden abandonment risk conditions includes: dividing the global hidden abandonment risk into three risk segments, and determining the abandonment risk area that meets the preset medium and high hidden abandonment risk conditions based on the last two risk segments of the three risk segments; inputting the regulatory effects of the main controlling factors in the abandonment risk area into a preset model, and based on the preset model, outputting the clusters of regulatory effects of the main controlling factors using a preset SOM+K-means two-stage clustering algorithm.

[0015] On the one hand, this application develops a method for predicting the probability of latent abandonment risk based on a machine learning model, addressing the gaps in risk prediction in previous abandonment research. Building on previous research on the driving mechanisms of farmland abandonment, this method introduces the MaxEnt machine learning model. This method not only learns the nonlinear correlation between historical latent abandonment sample points and the main controlling factors, but also further applies this prior knowledge to predict the future risk probability of latent abandonment. On the other hand, this application proposes a method for clustering latent abandonment risk, providing a scientific basis for formulating refined spatial management strategies. This method innovatively establishes a latent abandonment risk adjustment effect function that takes into account both the adjustment benefits and adjustment difficulty of the main controlling factors. It also uses the SOM+K-means spatial clustering algorithm to divide the dominant factor adjustment effect clusters for medium- and high-risk farmland, and then proposes common and differentiated management strategies to curb farmland abandonment. This addresses the problem that related technologies not only ignore latent abandonment in the uncultivated state but also fail to utilize prior knowledge of historical latent abandonment to carry out risk prediction and adjustment management research.

[0016] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0018] Figure 1 This is a flow chart of a method for adjusting cluster division of hidden abandonment risk provided in accordance with an embodiment of the present application;

[0019] Figure 2 This is a flowchart of the risk probability prediction and adjustment cluster division of hidden abandonment according to one embodiment of the present application;

[0020] Figure 3 A distribution map of historical hidden abandoned farmland and sample points according to one embodiment of the present application;

[0021] Figure 4is a correlation heat map of the initial driving factors according to one embodiment of the present application;

[0022] Figure 5 This is a spatial distribution diagram of the driving factor (main controlling factor) after screening according to one embodiment of the present application;

[0023] Figure 6 This is a probability distribution diagram of hidden abandonment risk according to one embodiment of the present application;

[0024] Figure 7 1 is a response curve diagram of the main control factors to the hidden abandonment risk according to one embodiment of the present application;

[0025] Figure 8 This is a diagram showing the regulatory effect of the leading factors of the hidden abandonment risk according to one embodiment of the present application;

[0026] Figure 9 This is a cluster distribution diagram of the main control factors for medium and high hidden abandonment risk areas according to one embodiment of the present application;

[0027] Figure 10 This is a radar chart of normalized main control factors of each adjustment cluster according to one embodiment of the present application. DETAILED DESCRIPTION

[0028] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0029] The following describes a method for regulating cluster division of hidden abandonment risk in an embodiment of the present application with reference to the accompanying drawings. In view of the problem that the related technologies mentioned in the above background technology not only ignore the hidden abandonment in the non-planting state, but also have not used the prior knowledge of historical hidden abandonment to carry out risk prediction and regulation and control research, the present application provides a method for regulating cluster division of hidden abandonment risk. In this method, a machine learning model is introduced to learn the relationship between historical hidden abandonment sample points and main control factors, and the future hidden abandonment risk is predicted in a probabilistic form, which makes up for the lack of risk prediction in previous abandonment research. At the same time, an innovative method for regulating cluster division of hidden abandonment risk is proposed, aiming to formulate a refined spatial control strategy to curb the abandonment phenomenon and help ensure national and regional food security. As a result, it solves the problem that the related technologies not only ignore the hidden abandonment in the non-planting state, but also have not used the prior knowledge of historical hidden abandonment to carry out risk prediction and regulation and control research.

[0030] Specifically, Figure 1A flow chart of a cluster division method for regulating hidden abandonment risks provided in an embodiment of the present application.

[0031] like Figure 1 As shown, the method for adjusting cluster division of hidden abandonment risk includes the following steps:

[0032] In step S101, a land survey dataset and an initial driving factor dataset of the target study area are collected.

[0033] In the actual implementation process, Figure 2 As shown, the embodiment of the present application can collect multi-source geospatial data of the target study area, including a land survey dataset and an initial driving factor dataset, and perform data preprocessing.

[0034] For example, the embodiment of the present application takes the Wuhan metropolitan area as the study area, collects the land change survey data set of the Wuhan metropolitan area in 2019 and 2020, and extracts the cultivated land patch vector data therefrom. At the same time, according to the agricultural development status of the Wuhan metropolitan area, 23 initial driving factors of hidden abandonment were selected, including 1 terrain factor, 3 climate factors, 5 soil factors, 4 tillage factors, 4 management factors, 3 economic factors and 3 population factors, as shown in Table 1. All spatial data are uniformly interpolated or resampled to a spatial resolution of 30m and uniformly projected as CGCS2000_3_Degree_GK_Zone_38. Table 1 is the initial driving factor table, where, as shown in Table 1:

[0035] Table 1

[0036]

[0037]

[0038]

[0039] Among them, g ii is the number of similar adjacent patches of patch type i, Maxg ii is the maximum number of similar adjacent patches of patch type i.

[0040]

[0041] Among them, e ij is the side length of patch j in patch type i, and A is the total area of ​​the patches.

[0042] In step S102, historical sample points of hidden abandoned land are extracted from the land survey dataset.

[0043] Among them, the embodiment of the present application can extract historical sample points of hidden abandonment from the land survey data set, so as to more accurately evaluate the actual use of cultivated land and provide a scientific basis for policy making.

[0044] Optionally, in one embodiment of the present application, historical sample points of hidden abandonment are extracted from a land survey dataset, including: extracting cultivated land patches of uncultivated farmland based on the planting status attributes of grain crops, non-grain crops, grain-non-grain rotation, fallow, intercropping of forest and grain, and uncultivated farmland in the land survey dataset; determining the historical hidden abandoned farmland of hidden abandonment based on the cultivated land patches of the uncultivated farmland, and extracting historical sample points of hidden abandonment from the historical hidden abandoned farmland.

[0045] For example, in the embodiment of the present application, the measurement process for extracting hidden abandoned land historical sample points based on the land change survey dataset is as follows:

[0046] (1) Identify and extract historical hidden abandoned farmland. The cultivated land planting status attributes in the land survey dataset are divided into "planting grain crops", "planting non-grain crops", "grain and non-grain rotation", "fallow", "forest-grain intercropping" and "uncultivated". The cultivated land patches with the attribute of "uncultivated" were extracted from the cultivated land patch vector data of the Wuhan metropolitan area in 2019 and 2020, and the intersection patches of uncultivated cultivated land for two consecutive years were regarded as historical hidden abandoned farmland. The results are as follows: Figure 3 shown.

[0047] (2) Selecting sample points of historical hidden abandoned farmland. Historical hidden abandoned farmland includes both large-scale contiguous patches and a large number of scattered patches. Therefore, in order to ensure the uniformity and representativeness of the sample points and prevent the sample points from being overly concentrated in contiguous patches, the Wuhan metropolitan area was divided into 57,996 1 km × 1 km grids, and the center point of each grid was extracted. Then, 487 center points that intersected with historical hidden abandoned farmland were extracted as sample points. The results are as follows: Figure 3 shown.

[0048] In step S103, based on historical sample points, the initial driving factors in the initial driving factor data set are tested for contribution rate and correlation to generate the main controlling factors of latent abandonment.

[0049] During the actual implementation process, the embodiment of the present application can perform contribution rate and correlation tests on the initial driving factors in the initial driving factor data set based on historical sample points, and screen out the main controlling factors of hidden abandonment, thereby providing support for subsequent risk probability prediction of hidden abandonment, and further predicting the future risk probability of hidden abandonment.

[0050] Optionally, in one embodiment of the present application, the initial driving factors in the initial driving factor data set are tested for contribution rate and correlation to generate the main controlling factors of latent abandonment, including: importing the historical sample points of latent abandonment and the initial driving factors into the target operating software to obtain the contribution rate of the initial driving factors to the latent abandonment probability prediction results, and eliminating the driving factors with a contribution rate lower than the target percentage to generate a primary screening driving factor; extracting the driving factor values ​​of the historical sample points of latent abandonment, and extracting the driving factor pairs with the absolute value of the correlation in the driving factor values ​​higher than the target value, and eliminating the driving factor pairs with a contribution rate lower than the target value in the driving factor pairs to generate a secondary screening driving factor; generating the main controlling factors of latent abandonment based on the primary screening driving factors and the secondary screening driving factors.

[0051] It can be understood that the target operating software in the embodiment of the present application can be the operating software of MaxEnt; the historical sample points of implicit abandonment in the embodiment of the present application can be historical implicit abandonment sample points.

[0052] As a possible implementation method, the process of screening the initial driving factors to obtain the main controlling factors of hidden land abandonment in the Wuhan metropolitan area in the embodiment of the present application is as follows:

[0053] (1) Contribution rate test of initial driving factors. The historical hidden abandonment sample points and all initial driving factor data of the Wuhan metropolitan area were imported into the MaxEnt operating software to obtain the contribution rate of all initial driving factors to the hidden abandonment probability prediction results. The driving factors with a contribution rate of less than 1% were eliminated to generate the initial screening driving factors. Table 2 is the contribution rate table of the initial driving factors, where, as shown in Table 2:

[0054] Table 2

[0055] Initial driving factors Contribution rate Initial driving factors Contribution rate Var18 17％ Var3 2.1％ Var13 13％ Var8 1.9％ Var14 12.7％ Var2 1.6％ Var20 10.6％ Var1 1.2％ Var22 8.6％ Var19 1.2％ Var23 7.1％ Var17 1％ Var16 4.9％ Var21 0.9％ Var5 3.1％ Var15 0.8％ Var4 2.9％ Var9 0.8％ Var10 2.8％ Var12 0.7％ Var11 2.5％ Var6 0.5％ Var7 2.2％

[0056] (2) Conduct a correlation test on the initial driving factors. In ArcGIS 10.8 software, the Extract MultiValues ​​to Points tool was used to extract all driving factor values ​​of all historical hidden abandonment sample points, and these data were imported into SPSS software for Spearman correlation test. Driving factor pairs with absolute correlation values ​​higher than 0.8 were extracted, and the factor with the lower contribution rate in each factor pair was eliminated to generate secondary screening driving factors. The main controlling factors of hidden abandonment were generated based on the initial screening driving factors and the secondary screening driving factors, that is, the driving factors after the final screening were the main controlling factors. The correlation test results of the initial driving factors are shown in Figure 2. Figure 4 As shown in Table 3 and Figure 5 As shown in Table 3, the driving factors (main controlling factors) after screening are shown in Table 3:

[0057] Table 3

[0058]

[0059] In step S104, a preset maximum entropy machine learning model is used to predict the risk probability of latent abandonment to generate a risk probability prediction result of latent abandonment.

[0060] It can be understood that the preset maximum entropy machine learning model in the embodiment of the present application can be a maximum entropy (MaxEnt) machine learning model.

[0061] During the actual implementation process, the embodiment of the present application can use the maximum entropy machine learning model to predict the risk probability of latent abandonment to generate a risk probability prediction result of latent abandonment. By introducing the MaxEnt machine learning model, not only the nonlinear correlation between historical latent abandonment sample points and the main controlling factors is learned, but also these prior knowledge are further applied to predict the future risk probability of latent abandonment.

[0062] This application example develops a method for predicting the probability of latent abandonment risk based on a machine learning model, addressing the gaps in risk prediction in previous abandonment research. Building on previous research on the driving mechanisms of farmland abandonment, this method introduces the MaxEnt machine learning model. This model not only learns the nonlinear associations between historical latent abandonment sample points and key controlling factors, but also applies this prior knowledge to predict the future risk probability of latent abandonment.

[0063] Optionally, in one embodiment of the present application, a preset maximum entropy machine learning model is used to predict the risk probability of latent abandonment to generate a risk probability prediction result of latent abandonment, including: inputting historical sample points and main controlling factors of latent abandonment into the preset maximum entropy machine learning model, setting the preset sample points to train and predict the preset maximum entropy machine learning model to generate a trained model, and selecting the remaining sample points as a test set to verify the trained model to generate a verified model; outputting the risk probability simulation result of latent abandonment according to the verified model, and using the area under the receiver operating characteristic curve (ROC) AUC to test the prediction accuracy of the risk probability simulation result to generate a risk probability prediction result.

[0064] It is understandable that in the embodiment of the present application, the preset sample points may be 75% of all sample points randomly selected.

[0065] Among them, the prediction process of the hidden abandonment risk probability of hidden wasteland in the Wuhan metropolitan area in the embodiment of the present application is:

[0066] (1) Set the input content and operating parameters of the MaxEnt model. On the one hand, the hidden abandonment sample points and the selected main control factors are used as the input of the MaxEnt model. On the other hand, 75% of the sample points are randomly selected to train the prediction model, and the remaining 25% of the sample points are used as the test set to verify the model. The maximum number of iterations and convergence threshold of the model training are set to 500 and 10 respectively. -5 The number of repeated runs of the model is set to 10, that is, the mean of 10 simulations is taken as the result.

[0067] (2) Output the risk probability results of hidden abandonment and evaluate the prediction accuracy. Based on the above input and settings, the MaxEnt model outputs the simulation results of the risk probability of hidden abandonment as follows: Figure 6 As shown, the result is a grid layer between 0 and 1, and the closer to 1, the higher the risk probability. At the same time, the model uses the area under the receiver operating characteristic curve (ROC curve) (AUC value) to test the accuracy of the simulation results. The AUC value range is 0.5-1, and the closer to 1, the higher the accuracy of the simulation results. When AUC is less than 0.7, it is generally considered that the simulation failed or the accuracy is extremely low; when AUC is 0.7-0.8, the simulation accuracy of the model is general; when AUC is 0.8-0.9, the simulation accuracy of the model is high; when AUC is greater than 0.9, it shows that the model simulation accuracy is extremely high. In the embodiment of the present application, the average AUC value of the repeated runs of the model is 0.839, and the standard deviation is 0.009, indicating that the risk probability prediction accuracy of the hidden abandonment in the Wuhan metropolitan area is high.

[0068] In step S105, based on the risk probability prediction results, a nonlinear response analysis of the main controlling factors for resisting the hidden abandonment risk is carried out to determine the nonlinear response characteristics of the main controlling factors for the hidden abandonment risk.

[0069] During the actual implementation process, the embodiment of the present application can carry out nonlinear response analysis of the main controlling factors to resist the hidden abandonment risk based on the risk probability prediction results, so as to determine the nonlinear response characteristics of the main controlling factors to the hidden abandonment risk, thereby providing support for the formulation of accurate land management strategies.

[0070] Optionally, in one embodiment of the present application, the nonlinear response characteristics of the main controlling factors to the hidden abandonment risk are determined, including: using a preset maximum entropy machine learning model to determine the response curve of the main controlling factors to the risk probability; displaying the development trend of the hidden abandonment risk probability with the change of the main controlling factors according to the response curve, and determining the nonlinear response characteristics of the main controlling factors to the hidden abandonment risk according to the development trend.

[0071] As a possible implementation method, the principle and function of the nonlinear response analysis of the master control factors to the hidden abandonment risk in the embodiment of the present application are as follows: while outputting the hidden abandonment risk probability, the MaxEnt model also generates a response curve of all the master control factors to the risk probability. The response curve reflects the marginal effect of the change of the master control factor. In other words, the response curve shows the development trend of the hidden abandonment risk probability as the master control factor changes, and determines the nonlinear response characteristics of the master control factor to the hidden abandonment risk based on the development trend, which helps to reveal the influence mechanism of the master control factor on the probability of hidden abandonment and provides a basis for adjusting the master control factor to reduce the risk.

[0072] The response curve of the main controlling factor is as follows Figure 7 As shown in the figure, since average annual precipitation and average annual sunshine hours are natural climate factors that are difficult to adjust, only the other controlling factors are discussed. Among them, topographic factors (Var1), tillage factors (Var11), management factors (Var16), and economic factors (Var19) show a positive correlation with the overall trend of hidden abandonment risk in the Wuhan metropolitan area. Soil factors (Var5, Var7, Var8), tillage factors (Var10, Var13), management factors (Var14, Var17), and economic factors (Var18) show a negative correlation with the overall trend of hidden abandonment risk in the Wuhan metropolitan area. Notably, the relationship between per capita rural income (Var20) and per capita rural labor force (Var22) and hidden abandonment risk is not positive or negative, but rather shows a nearly quadratic function relationship with an upward opening. The Var20 response curve shows that increasing per capita rural income in the Wuhan metropolitan area can increase farmers' willingness to plant and effectively reduce the risk of hidden land abandonment. The Var22 response curve shows that rural labor, as the main source of cultivated land, is a crucial factor in preventing hidden land abandonment. However, when per capita rural income or the average rural labor force per household exceeds the bottom threshold of the curve, the risk of hidden land abandonment increases. This phenomenon is related to the diversification of agricultural management models. With the modernization of agriculture, higher-value-added industries such as agricultural product processing and agricultural services have absorbed an increasing number of rural laborers, potentially leading to a loss of key planting sectors. However, the rise in hidden land abandonment risk cannot be attributed solely to a developed rural economy and an abundant rural labor force, as both are key to improving the well-being of farmers. Therefore, while enriching the agricultural management system, we should also prioritize the fundamental role of cultivated land in national food security, guide the transfer of management rights for scattered abandoned cultivated land, promote large-scale planting, and strictly prevent a resurgence of hidden land abandonment.

[0073] In step S106, a hidden abandonment risk adjustment effect function is constructed based on the nonlinear response characteristics, taking into account both the adjustment benefit and adjustment difficulty of the main control factors.

[0074] In actual implementation, embodiments of the present application can construct a hidden abandonment risk adjustment effect function based on nonlinear response characteristics that takes into account both the adjustment benefits and adjustment difficulty of the dominant factor. The adjustment effect function of the dominant factor represents the cumulative effect of adjusting the dominant factor to achieve the lowest hidden abandonment risk probability. This function takes into account both the adjustment benefits and adjustment difficulty of the dominant factor. The adjustment benefit is represented by the change in risk before and after adjustment, while the adjustment difficulty is represented by the adjustable ratio. This function is calculated using a segmented accumulation method, that is, the adjustment process is subdivided into multiple segments based on the adjustment range of the dominant factor, and the adjustment effects of all segments are cumulatively calculated. This is done to avoid inaccurate calculation results caused by uneven changes in the response curve. The adjustment effect function is divided into two categories based on the direction (positive or negative) of the corresponding dominant factor value when a certain dominant factor value in the response curve is adjusted to the lowest risk: a negative function, that is, when the factor value is adjusted to the lowest risk, the corresponding factor value shows a negative trend; and a positive function, that is, when the factor value is adjusted to the lowest risk, the corresponding factor value shows a positive trend. The type of adjustment effect function used depends on the specific value of the main control factor in the response curve. This means that in the same effect-response curve, when the main control factor value is at different positions, its adjustment effect function may be different.

[0075] In one embodiment of the present application, the calculation formula of the implicit abandonment risk adjustment effect function is:

[0076]

[0077] Among them, AdjEff i,j represents the moderating effect of the dominant factor i with value j, and They represent the expected risk changes (i.e., adjusted returns) before and after the adjustment of the main control factor i in the negative and positive adjustment effect functions, respectively. and They represent the adjustable proportion (i.e., adjustment difficulty) of the main control factor i in the negative and positive adjustment effect functions, respectively; n represents the number of segments in the adjustment range; t is the main control factor value corresponding to the lowest risk in the response curve of the main control factor i; j represents any value of the main control factor i; min is the minimum value of the main control factor i; and max is the maximum value of the main control factor i.

[0078] It's important to note that the examples presented in this application show that the relationship between economic and demographic factors and the risk of hidden land abandonment is not a positive or negative correlation, but rather an approximately quadratic function with an upward trend. Therefore, the moderating effect of these factors is calculated as a negative function. For a detailed analysis of the reasons for this, see the example analysis in step S105 above.

[0079] According to the above formula, the regulatory effect of the leading factors of the hidden abandonment risk in the Wuhan metropolitan area is quantified. The results are as follows: Figure 8 shown.

[0080] In step S107, based on the implicit abandonment risk adjustment effect function, the main control factor adjustment effect clusters that meet the preset medium and high implicit abandonment risk conditions are identified, and the main control factor adjustment effect clusters are divided to generate a cluster management strategy for implicit abandonment.

[0081] Specifically, the embodiments of the present application can identify the main control factor regulation effect clusters that meet the preset medium and high hidden abandonment risk conditions based on the hidden abandonment risk regulation effect function, and divide the main control factor regulation effect clusters to generate common and differentiated cluster management strategies to curb farmland abandonment.

[0082] In the embodiment of the present application, the normalized main control factors higher than 0.6 in the identified 6 adjustment clusters are used as the main adjustment measures (the radar chart of the normalized main control factors of each cluster is as follows Figure 10 The common and differentiated management and control suggestions for each cluster include thickening the tillage layer, preventing and controlling geological disasters, and promoting the transformation of agricultural mechanization. The differentiated management and control suggestions are as follows:

[0083] Cluster 1, dominated by soil-management factors, is widespread in the western and eastern plains. On the one hand, these regions need to rationally increase the application of organic fertilizers to improve soil fertility and prevent the marginalization of cultivated land returns. On the other hand, it is necessary to significantly increase cultivated land contiguousness to reduce tillage costs and achieve the scale effect of cultivated land.

[0084] Clusters 2 and 3 are regulatory clusters dominated by tillage and management factors. On the one hand, it is crucial to curb the excessive use of fertilizers and pesticides to prevent negative feedback on farmland productivity caused by ecological issues such as soil compaction and heavy metal pollution. On the other hand, this region should prioritize improving the convenience of agricultural production activities (reducing the distance between farmlands) and the smooth flow of agricultural product trade (reducing the distance between production and marketing) as one of the optimization goals for the layout of rural settlements and farm product trade markets.

[0085] Clusters 4 and 5 are moderating clusters dominated by topography, soil, and management factors. These areas are located in hilly terrain, where farmland is not only fragmented but also prone to soil erosion, resulting in severe loss of soil nutrients. Therefore, priority should be given to terracing sloping farmland and consolidating and fertilizing sloping farmland, followed by a steady implementation of farmland fragmentation remediation projects. Furthermore, based on stable farmland output, an agricultural product industry chain should be established and improved to achieve a balanced production and marketing of agricultural products.

[0086] Cluster 6 has significant moderating effects of topography, soil, tillage, management, economy, and population. In other words, this cluster is the most difficult and urgent area to regulate the risk of hidden abandonment. Therefore, this application requires the implementation of all the aforementioned engineering measures in these areas. In addition, due to the siphon effect of Wuhan's high urbanization, these areas have experienced a serious outflow of rural labor and rural hollowing out, falling into a vicious cycle of "labor outflow-reduced income." On the one hand, this application aims to increase farmers' willingness to plant through policies such as raising grain purchase prices and establishing a reclamation subsidy mechanism. On the other hand, it is necessary to guide the transfer of scattered abandoned farmland management rights to efficient agricultural producers, promote large-scale cultivation of farmland, and strictly prevent the resurgence of hidden abandonment trends.

[0087] This application example proposes a method for clustering the regulation of latent abandonment risk, providing a scientific basis for developing refined spatial management strategies. This method innovatively establishes a latent abandonment risk regulation effect function that takes into account both the regulatory benefits and difficulty of the dominant control factors. It then uses the SOM+K-means spatial clustering algorithm to identify clusters of the dominant factor regulation effects for medium- and high-risk cultivated land, thereby proposing both common and differentiated management strategies to curb abandoned cultivated land.

[0088] It should be noted that the preset medium and high hidden abandonment risk conditions can be set by technical personnel in this field according to actual conditions and are not specifically limited here.

[0089] Optionally, in one embodiment of the present application, clusters of regulatory effects of main controlling factors that meet preset medium and high hidden abandonment risk conditions are identified, including: dividing the global hidden abandonment risk into three risk segments, and determining the abandonment risk area that meets the preset medium and high hidden abandonment risk conditions based on the last two risk segments of the three risk segments; inputting the regulatory effects of the main controlling factors in the abandonment risk area into a preset model, and based on the preset model, using the preset SOM+K-means two-stage clustering algorithm to output clusters of regulatory effects of main controlling factors.

[0090] In the actual implementation process, the principle and process of identifying the clusters of the main controlling factors regulating the medium and high hidden abandonment risks based on the SOM+K-means two-stage clustering algorithm in the embodiment of the present application are as follows:

[0091] (1) Identify areas with medium and high hidden abandonment risks. The equal spacing method is used to divide the hidden abandonment risk of the entire region into three levels: low, medium, and high. Areas above the risk threshold of 0.32 between the first two levels are medium and high risk areas, covering an area of ​​approximately 5292 km. 2 , the results are as follows Figure 9 As shown in a.

[0092] (2) Identify the clusters of regulatory effects of the main controlling factors for medium and high hidden abandonment risks. A single or multiple main controlling factors may have a significant regulatory effect on the hidden abandonment risk of a certain patch. Patches with similar combinations of main controlling factors form clusters of regulatory effects of main controlling factors. Therefore, the regulatory effects of all main controlling factors in medium and high risk areas are used as input, and the SOM+K-means two-stage clustering algorithm is used to output regulatory effect clusters. The SOM+K-means algorithm is a two-stage clustering algorithm that combines the advantages of the self-organizing feature map network (SOM) and the K-means algorithm. In the first stage, the self-organizing classification ability of the SOM algorithm is used to train the original massive sample data to achieve coarse clustering of the original data. The dimensionality reduction results of the coarse clustering provide a set of optimized initial cluster centers for the K-means algorithm. In the second stage, the K-means algorithm is used to cluster the coarse clustering results of the SOM again, and the clustering accuracy is improved by iteratively optimizing the positions of the cluster centers. The algorithm requires pre-setting multiple cluster numbers (i.e., the number of clusters) and uses the Davies-Bouldin index to determine the optimal number of clusters to obtain the final clustering results. According to Table 4, when the number of clusters is 6, the Davies-Bouldin index is the lowest and the clustering effect is the best. Therefore, the number of clusters is set to 6, and the SOM+K-means two-stage clustering algorithm is run to obtain the final clustering results. The spatial distribution of each adjustment cluster is shown in the figure below. Figure 9 Table 4 shows the Davies-Bouldin index table for different numbers of clusters, as shown in Table 4:

[0093] Table 4

[0094]

[0095] The calculation formula is as follows:

[0096]

[0097] Where DB represents the Davies-Bouldin index; K is the number of clusters; R ij represents the similarity between clusters i and j; S i and S j are the average distances from all sample points in the i-th and j-th clusters to the cluster center; d ij is the distance between the centers of clusters i and j.

[0098] According to the method for regulating cluster division of latent abandonment risk proposed in the embodiment of the present application, a machine learning model is introduced to learn the relationship between historical latent abandonment sample points and main controlling factors, and the future latent abandonment risk is predicted in a probabilistic form, which makes up for the lack of risk prediction in previous abandonment research. At the same time, an innovative method for regulating cluster division of latent abandonment risk is proposed, aiming to formulate a refined spatial control strategy to curb the abandonment phenomenon and help ensure national and regional food security. Thus, it solves the problem that the relevant technology not only ignores the latent abandonment in the non-planting state, but also has not yet used the prior knowledge of historical latent abandonment to carry out risk prediction and regulation and control research.

[0099] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0100] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing a custom logical function or process step, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed in a different order than shown or discussed, including performing functions in a substantially simultaneous manner or in a reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application pertain.

Claims

1. A method for regulating hidden abandonment risk clustering, characterized in that: The following steps are involved: Collect land survey datasets and initial driving factor datasets for the target study area; Extracting historical sample points of hidden abandoned land from the land survey dataset; Based on the historical sample points, performing contribution rate and correlation tests on the initial driving factors in the initial driving factor data set to generate the main controlling factors of the hidden abandonment; Using a preset maximum entropy machine learning model to perform risk probability prediction on the hidden abandonment of land, so as to generate a risk probability prediction result of the hidden abandonment of land; Based on the risk probability prediction results, a nonlinear response analysis of the main controlling factors for resisting the hidden abandonment risk is carried out to determine the nonlinear response characteristics of the main controlling factors to the hidden abandonment risk; Constructing a hidden abandonment risk adjustment effect function that takes into account both the adjustment benefits and adjustment difficulty of the main control factors according to the nonlinear response characteristics; Based on the hidden abandonment risk adjustment effect function, the main control factor adjustment effect clusters that meet the preset medium and high hidden abandonment risk conditions are identified, and the main control factor adjustment effect clusters are divided to generate the cluster management strategy of the hidden abandonment.

2. The method according to claim 1, characterized in that The extracting of historical sample points of hidden abandoned land from the land survey dataset includes: Extracting the cultivated land patches of the uncultivated arable land based on the planting state attributes of grain crops, non-grain crops, grain-non-grain rotation, fallow, forest-grain intercropping, and uncultivated arable land in the land survey dataset; The historical hidden abandoned farmland of the hidden abandonment is determined according to the cultivated land patches of the uncultivated farmland, and the historical sample points of the hidden abandonment are extracted from the historical hidden abandoned farmland.

3. The method according to claim 2, characterized in that The performing of contribution rate and correlation tests on the initial driving factors in the initial driving factor data set to generate the main controlling factors of the hidden abandonment includes: Importing the historical sample points of hidden abandonment and the initial driving factors into the target operating software to obtain the contribution rate of the initial driving factors to the hidden abandonment probability prediction results, and eliminating driving factors with a contribution rate lower than the target percentage to generate preliminary screening driving factors; Extracting the driving factor values ​​of the historical sample points of hidden abandonment, and extracting the driving factor pairs whose absolute correlation values ​​are higher than the target value from the driving factor values, and eliminating the driving factor pairs whose contribution rates are lower than the target value from the driving factor pairs, so as to generate secondary screening driving factors; The main controlling factor of the latent abandonment is generated based on the primary screening driving factor and the secondary screening driving factor.

4. The method according to claim 3, characterized in that The method of using a preset maximum entropy machine learning model to predict the risk probability of the latent abandonment to generate a risk probability prediction result of the latent abandonment includes: Inputting the historical sample points of hidden abandonment and the main controlling factors into the preset maximum entropy machine learning model, setting the preset sample points to train and predict the preset maximum entropy machine learning model to generate a trained model, and using the remaining sample points as a test set to verify the trained model to generate a verified model; The risk probability simulation result of the hidden abandonment is output according to the verified model, and the prediction accuracy of the risk probability simulation result is tested using the area under the receiver operating characteristic curve (ROC) AUC to generate the risk probability prediction result.

5. The method according to claim 1, wherein Determining the nonlinear response characteristics of the main controlling factor to the hidden abandonment risk includes: Determining a response curve of the master control factor to the risk probability using the preset maximum entropy machine learning model; The response curve shows the development trend of the hidden abandonment risk probability as the main control factor changes, and the nonlinear response characteristics of the main control factor to the hidden abandonment risk are determined based on the development trend.

6. The method according to claim 1, characterized in that The calculation formula of the implicit abandonment risk adjustment effect function is: Among them, AdjEff i,j represents the moderating effect of the dominant factor i with value j, and They represent the expected risk changes before and after the adjustment of the main control factor i in the negative and positive adjustment effect functions, respectively. and They represent the adjustable proportions of the main control factor i in the negative and positive regulation effect functions respectively, n represents the number of segments in the regulation range, t is the main control factor value corresponding to the lowest risk in the response curve of the main control factor i, j represents any value of the main control factor i, min is the minimum value of the main control factor i, and max is the maximum value of the main control factor i.

7. The method according to claim 1, characterized in that The identification of the main controlling factor regulation effect clusters that meet the preset medium-to-high hidden abandonment risk conditions includes: Divide the hidden abandonment risk of the entire region into three risk segments, and determine the abandonment risk areas that meet the preset medium and high hidden abandonment risk conditions based on the last two of the three risk segments; The regulatory effects of the main controlling factors in the abandonment risk area are input into a preset model, and based on the preset model, a preset SOM+K-means two-stage clustering algorithm is used to output the cluster of regulatory effects of the main controlling factors.

Images