Yak precise grazing scheduling method and system based on multi-source perception

Abstract

The application discloses a yak precise grazing scheduling method and system based on multi-source perception, relates to the technical field of intelligent management and protection, and comprises the following steps: deploying a multi-source perception component in a grazing area, synchronously collecting multi-source heterogeneous data of structure perception, function perception and interaction perception, mining feature elements through a multi-thread parallel data processing component, and determining baseline grazing elements including ecological network coupling degree, disease prevalence probability and the like. The elements are converted into fuzzy membership degree coefficients, the corresponding carrying capacity adjustment gear of comprehensive ecological load is analyzed, a scheduler array is triggered to complete scheduling decision, a grazing scheduling strategy is generated, and the mobile terminal interface is displayed. The application solves the technical problems that the existing grazing scheduling relies on experience to determine the carrying capacity, does not consider the grassland ecological coupling relationship, has no dynamic control means, and has poor scheduling accuracy and adaptability, and achieves the technical effects of precise and dynamic grazing scheduling, and consideration of ecological stability and breeding management efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology, and in particular to a method and system for precise grazing and scheduling of yaks based on multi-source sensing. Background Technology

[0002] Yak grazing is widely used in the high-altitude pastoral areas of the Qinghai-Tibet Plateau. Precise scheduling is crucial for grassland ecological protection, yak breeding efficiency, and scientific management of pastoral areas. Multi-source data processing and intelligent grazing management are the core support for achieving precise scheduling. Current grazing management relies heavily on manual experience and single monitoring methods, lacking multi-source sensing and systematic data processing mechanisms, resulting in relatively extensive grazing management. The complex ecological structure of alpine meadows and the highly dynamic nature of air and soil environments make it impossible for traditional methods to simultaneously collect multi-source heterogeneous data and conduct refined feature mining. Data processing is incomplete, management decisions are lagging, and it is difficult to accurately determine carrying capacity, rotational grazing routes, and rest grazing strategies. The obtained scheduling data is incomplete and unreliable, failing to meet the actual needs of precise yak grazing scheduling and scientific and intelligent management in high-altitude pastoral areas. Summary of the Invention

[0003] This application provides a method and system for precise grazing scheduling of yaks based on multi-source sensing, which solves the technical problems of existing grazing scheduling relying on experience to determine the carrying capacity, not considering the grassland ecological coupling relationship, lacking dynamic control means, and having poor scheduling accuracy and adaptability.

[0004] The first aspect of this application provides a method for precise grazing scheduling of yaks based on multi-source sensing. The method includes: deploying multi-source sensing components in the grazing area to synchronously collect multi-source heterogeneous data, wherein the multi-source heterogeneous data includes structural sensing data, functional sensing data, and interaction sensing data; based on the multi-source heterogeneous data, performing multi-threaded parallel feature element targeted mining according to a data processing component to determine baseline grazing elements, wherein the baseline grazing elements include ecological network coupling degree, disease prevalence probability, home-field advantage breakpoint offset, and functional group balance index; converting the baseline grazing elements into fuzzy membership coefficients, analyzing the carrying capacity adjustment level based on comprehensive ecological load, triggering scheduling decisions of the scheduler array, and determining a grazing scheduling strategy; and displaying the grazing scheduling strategy on a mobile terminal interface.

[0005] The second aspect of this application provides a yak precision grazing scheduling system based on multi-source sensing. The system includes: a multi-source heterogeneous data acquisition module, used to synchronously collect multi-source heterogeneous data by deploying multi-source sensing components in the grazing area, wherein the multi-source heterogeneous data includes structural sensing data, functional sensing data, and interaction sensing data; a baseline grazing element calculation module, used to determine baseline grazing elements by performing multi-threaded parallel feature element targeted mining based on the multi-source heterogeneous data and a data processing component, wherein the baseline grazing elements include ecological network coupling degree, disease prevalence probability, home-field advantage breakpoint offset, and functional group balance index; a grazing scheduling strategy acquisition module, used to convert the baseline grazing elements into fuzzy membership coefficients, analyze the carrying capacity adjustment level based on comprehensive ecological load, trigger the scheduling decision of the scheduler array, and determine the grazing scheduling strategy; and a grazing scheduling strategy display module, used to display the grazing scheduling strategy on a mobile terminal interface.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] This application collects multi-source ecological data by deploying sensing components in grazing areas, and obtains core grazing elements and ecological load data through multi-threaded parallel feature mining and fuzzy logic transformation. It calculates the livestock carrying capacity adjustment level and time-space temperature parameters, and conducts iterative optimization decisions based on the status of each grazing sub-area. This results in the precise generation of grazing scheduling schemes adapted to alpine grasslands, making yak grazing scheduling more scientific and ecologically and production-wise more coordinated. It achieves the technical effect of precise and dynamic grazing scheduling, while taking into account both ecological stability and breeding management efficiency. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the method for precise grazing and scheduling of yaks based on multi-source perception provided in this application embodiment.

[0010] Figure 2 This is a schematic diagram of the structure of the yak precision grazing scheduling system based on multi-source perception provided in the embodiments of this application.

[0011] Figure labeling: Module 1 for multi-source heterogeneous data acquisition, Module 2 for baseline grazing element calculation, Module 3 for grazing scheduling strategy acquisition, and Module 4 for grazing scheduling strategy display. Detailed Implementation

[0012] This application provides a method and system for precise grazing scheduling of yaks based on multi-source sensing, which solves the technical problems of existing grazing scheduling relying on experience to determine the carrying capacity, not considering the grassland ecological coupling relationship, lacking dynamic control means, and having poor scheduling accuracy and adaptability.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0015] Example 1, as Figure 1 As shown, a method for precise yak grazing scheduling based on multi-source sensing is provided, wherein the method includes:

[0016] By deploying multi-source sensing components in grazing areas, multi-source heterogeneous data is collected synchronously, including structural sensing data, functional sensing data, and interaction sensing data.

[0017] Specifically, firstly, a standard monitoring grid of 1m x 1m was divided in the grazing area at 5m intervals. Metal supports were installed at each grid node, and a vegetation cover gauge was installed at a height of 0.5m on each support. Soil temperature and humidity sensors were inserted 10cm underground. Simultaneously, litter collection boxes, disease monitoring scales, and arthropod trapping nets were placed. All sensing components were connected to the same central data acquisition terminal via wireless transmission modules, completing the standardized grid deployment of multi-source sensing components. Next, the vegetation height and cover of each grid node were measured every two hours using the vegetation cover gauge. Monthly, 1m x 1m quadrats were selected for vegetation harvesting. The samples were weighed on-site to calculate vegetation biomass. The plant species within the quadrats were manually counted and recorded, and the number of individuals of each species was counted and their relative proportions calculated. The above measurement and statistical results were summarized to form structural sensing data.

[0018] Then, the soil temperature and humidity sensor automatically collects soil temperature and humidity values ​​every hour. After taking soil samples, they are sieved using a 2 mm mesh screen. The abundance of soil nematode functional groups is manually separated and counted. Samples from litter collection frames are collected monthly, dried, weighed, and the litter decomposition rate is calculated. Soil samples are plate-inoculated using beef extract peptone medium and cultured at 25 degrees Celsius for 72 hours to complete the counting, obtaining soil microbial β-diversity related data. The above detection and calculation results are integrated to form functional perception data.

[0019] Subsequently, the severity of plant diseases was manually determined by comparing the 0 to 4 level disease monitoring scale with standard disease samples. The life cycle types of pathogens were manually distinguished by the morphological characteristics of pathogen spores. Five arthropod sweeping operations were carried out each month using arthropod sweeping nets. The captured samples were manually classified and statistically analyzed to obtain the relative abundance data of the five functional groups of arthropods. The above observation and statistical results were compiled into interactive perception data.

[0020] Finally, the central data acquisition terminal sends synchronous acquisition instructions to each sensing component every six hours. Each component synchronously completes on-site measurement, sample collection, and data statistics. Those skilled in the art input various types of data into the terminal to form data sequences of various types. The central data acquisition terminal first calculates the arithmetic mean of a single data sequence, and then calculates the corresponding standard deviation based on the mean. The effective range of the data is then determined to be from the arithmetic mean minus three standard deviations to the arithmetic mean plus three standard deviations. Values ​​outside this effective range are identified as outliers and removed. Subsequently, all data are uniformly converted into CSV format and the fields are standardized and regularized to achieve synchronous acquisition and preprocessing of multi-source heterogeneous data.

[0021] By deploying standardized on-site measurement methods and data processing rules, we can obtain comprehensive multi-dimensional ecological data of grazing areas, providing stable and reliable data support for subsequent data processing and grazing management.

[0022] Based on the multi-source heterogeneous data, and according to the data processing component, feature element targeted mining is performed under multi-threaded parallelism to determine baseline grazing elements, wherein the baseline grazing elements include ecological network coupling degree, disease prevalence probability, home advantage breakpoint offset, and functional group balance index.

[0023] In this embodiment, the data processing component is an industrial control computer or an embedded data processing module that integrates data computation, analysis, and feature mining programs.

[0024] Optionally, for the four types of baseline grazing elements—ecological network coupling degree, disease prevalence probability, home-field advantage breakpoint offset, and functional group balance index—the data processing component uses multi-threaded parallel processing to perform calculations and analyses on multi-source heterogeneous data, sequentially mining ecological network coupling degree, disease prevalence probability, home-field advantage breakpoint offset, and functional group balance index, thereby determining the corresponding baseline grazing elements. The steps for obtaining the four types of baseline grazing elements will be explained in detail in the following sections.

[0025] By converting the baseline grazing elements into fuzzy membership coefficients, the adjustment level of livestock carrying capacity based on comprehensive ecological load is analyzed, triggering the scheduling decision of the scheduler array and determining the grazing scheduling strategy.

[0026] In one embodiment of this application, the baseline grazing elements are first transformed by fuzzy logic to obtain the fuzzy membership coefficients of each grazing sub-region. Then, the fuzzy membership coefficients of each sub-region are weighted and integrated to determine the comprehensive ecological load coefficient and mapped to the carrying capacity adjustment level. After spatiotemporal phase integration and iterative decision-making of the grazing sub-regions, the final grazing scheduling strategy is determined. This step will be described in detail in the following content.

[0027] The grazing scheduling strategy is displayed on a mobile terminal interface.

[0028] Specifically, the finalized grazing scheduling strategy will be synchronously transmitted to the handheld mobile terminals of grazing managers via 4G / 5G wireless communication links. The mobile terminals will display the strategy in a clear, modular, and hierarchical manner on a visual interactive interface. The first screen of the interface will be a map module showing the overall scheduling of the entire grazing area. The map will also display the spatial boundaries of each merged scheduling unit, the corresponding effective carrying capacity adjustment level, and the thermal distribution layer of the comprehensive ecological load coefficient. At the same time, the map will be fully drawn based on the rotational grazing migration routes that have been optimized through multiple iterations. Different colors will be used to distinguish the three corresponding areas: increased grazing, maintaining the status quo, and reduced grazing. The avoidance range of the core ecological protection area and the terrain obstacle area will be highlighted.

[0029] The interface synchronously sets up a module for detailed scheduling parameters, displaying specific values ​​for the recommended carrying capacity per unit area, the number of days of residence in each grazing sub-region, the grid range of the targeted foraging priority area, and the duration of the rest and recovery period for each scheduling unit, as well as the verification scores for ecological adaptability, feasibility, and scheduling matching for rotational grazing migration paths. The interface supports touch zoom, detailed query of regional clicks, manual annotation of scheduling parameters, and cloud synchronization. It also includes a weekly data automatic refresh module to synchronously update baseline element monitoring data and scheduling strategy adjustment prompts for the grazing area, ensuring that grazing managers can use mobile terminals to have a real-time and comprehensive understanding of the grazing scheduling execution requirements and grassland ecological monitoring status for the entire region.

[0030] Furthermore, the method provided in this application embodiment includes:

[0031] Using the structural sensing data as the main dimension of the aboveground indicator sequence, the functional sensing data as the main dimension of the underground indicator sequence, and the interaction sensing data as a correction term for the aboveground indicator sequence, dynamic time warping based on a sliding window is employed to calculate the self-regulating distance and cross-regulating distance between the aboveground and underground indicator sequences to determine the ecological network coupling degree. Specifically, the ecological network coupling degree is obtained through exponential mapping by taking the mean ratio of the cross-regulating distance to the self-regulating distance as a normalization condition.

[0032] Specifically, firstly, the structural perception data were arranged sequentially according to the collection time to construct an initial aboveground indicator sequence, and the functional perception data were arranged according to the same collection time sequence to construct a subsurface indicator sequence. Next, the structural perception data and interaction perception data were subjected to min-max normalization, mapping the numerical ranges of both types of data to the 0-1 interval, eliminating dimensional differences between different indicators. Fixed weights were determined through prior grazing area sample plot determination experiments. These experiments involved selecting three 1-hectare experimental plots with soil type, vegetation composition, and grazing intensity consistent with actual yak grazing areas. Structural perception data, interaction perception data, and functional perception data were simultaneously collected from each plot over a one-month period, with data collected weekly.

[0033] The method for determining the fixed weights is as follows: Three different weight gradients are set in the three test plots, for example, 0.1, 0.2, and 0.3. Each gradient weight is substituted into the calculation of the aboveground indicator sequence corrected by the interaction sensing data. The correlation coefficient between the corrected aboveground indicator sequence and the underground indicator sequence is calculated. The weight with the highest correlation coefficient (i.e., the best correction effect) is selected as the fixed weight. For example, based on the above calibration experiment, the empirical value of 0.2 obtained from the calibration is selected as the fixed weight. The normalized interaction sensing data is multiplied by this fixed weight and then added to the normalized aboveground indicator sequence value at the same time position to complete the numerical correction of the aboveground indicator sequence.

[0034] Then, a sliding window containing 10 time-series data points is set. Using one time-series data point as the sliding step, the corrected aboveground and underground indicator sequences are segmented sequentially along the time-series direction, resulting in multiple sets of continuous window data segments. For each set of window data segments, the matching distance between data points within the aboveground indicator sequence is calculated using Euclidean distance and accumulated. The Euclidean distance calculation process is as follows: take any two data points to be matched, calculate the difference between the two values, square the difference, and then take the square root of the squared result. The resulting value is the Euclidean distance between the two data points. The underground self-regulating distance of the underground indicator sequence is calculated using the same method. Then, the Euclidean matching distance between the aboveground and underground indicator sequence data points within the same window is calculated and accumulated to obtain the cross-regulating distance of that window.

[0035] Next, the arithmetic mean of the above-ground and underground self-normalized distances of all windows is summed to obtain the overall self-normalized distance. Similarly, the arithmetic mean of the cross-normalized distances of all windows is summed to obtain the overall cross-normalized distance. The overall cross-normalized distance is then divided by the overall self-normalized distance to obtain the normalized ratio. Finally, the normalized ratio is substituted into the natural exponential function for numerical calculation. The expression for the natural exponential function is: , where x is the ratio result after normalization, and the result obtained after calculating e to the power of x is directly determined as the ecological network coupling degree.

[0036] By employing a clear sequence correction method, fixed window parameters, and Euclidean distance calculation, combined with ratio normalization and natural index mapping, the quantifiable determination of the coupling degree of ecological networks was achieved.

[0037] Furthermore, the method provided in this application embodiment includes:

[0038] Based on the structural perception data, it is determined whether the foraging pressure exceeds the vegetation compensating growth threshold, which serves as the first judgment feature; based on the functional perception data, the degree of inhibition of underground decomposition channels by grazing is quantified, which serves as the second quantitative feature; based on the interaction perception data, the decline in ecosystem resilience caused by overgrazing is identified, which serves as the third identification feature; the first judgment feature, the second quantitative feature, and the third identification feature are input into a fuzzy inference rule with decoupled early warning status as a prerequisite to determine the probability of disease outbreak.

[0039] Optionally, firstly, take the weekly arithmetic mean of all grid monitoring data in the entire monitoring area for the structure-sensing data, function-sensing data, and interaction-sensing data, and perform minimum-maximum normalization processing on each of them. Subtract the minimum value of the data type within the 12-month continuous monitoring period from the weekly statistical value, and then divide by the difference between the maximum and minimum values ​​within the period. Map all data uniformly to the interval between 0 and 1 to eliminate the dimensional differences between different types of data.

[0040] Next, feeding pressure was calculated based on the normalized structure-aware data. The baseline biomass was obtained through monitoring ungrazing quadrats over the previous three months. Each week, three 1m × 1m subplots were selected within the ungrazing quadrats, vegetation was harvested and weighed, and the arithmetic mean of the biomass of all subplots was taken as the baseline biomass. The difference between the current vegetation biomass and the baseline biomass was calculated, and the ratio of this difference to the baseline biomass was taken as the feeding pressure value. A preset threshold of 0.4 was set for vegetation compensating for growth. When the feeding pressure value was greater than this threshold, the first judgment feature was marked as 1, and otherwise it was marked as 0.

[0041] Then, based on the normalized functionally perceived data, the degree of inhibition is quantified. The standard decomposition rate in ungrazing areas is determined by selecting ungrazing plots with the same soil type and vegetation composition as grazing areas, placing litter collection frames in the same manner, collecting samples monthly, and measuring the decomposition rate. The arithmetic mean of the monitoring over two consecutive months is taken as the standard decomposition rate. The difference between the measured decomposition rate and the standard decomposition rate is calculated, and the ratio of this difference to the standard decomposition rate is used to determine the degree of inhibition of underground decomposition channels by grazing. This value is directly used as the second quantitative feature.

[0042] Subsequently, based on the normalized interaction sensing data, resilience decline was identified. The baseline abundance under ecological stability was obtained through synchronous monitoring of the aforementioned ungrazing plots. Arthropods were collected three times a week using sweep nets, and the abundance of various functional groups was statistically analyzed. The arithmetic mean of the two months of monitoring data was taken as the baseline abundance. The ratio of the measured abundance values ​​of arthropod functional groups in the interaction sensing data to this baseline abundance was calculated. When the ratio was less than 0.6, it was determined that overfeeding caused ecosystem resilience decline, and the third identification feature was marked as 1; otherwise, it was marked as 0.

[0043] Next, the decoupling and early warning status determination operation is performed. The criteria for extreme weather are defined as daily precipitation ≥ 50 mm or daily maximum temperature ≥ 30 degrees Celsius. The criteria for human disturbance are traces of mechanical work or a gathering of ≥ 3 people within the monitoring area. Abnormal data periods meeting either of these criteria are removed, retaining only valid data states indicating ecological stability as prerequisites for fuzzy inference. Then, the first judgment feature, the second quantitative feature, and the third identification feature are used as inputs. The first judgment feature and the third identification feature are binary features (0 or 1), and the second quantitative feature is a continuous numerical feature. Fuzzy subsets of three levels—low, medium, and high—are constructed. The low level corresponds to a numerical range of 0-0.3, the medium level to 0.3-0.7, and the high level to 0.7-1.0.

[0044] For binary-type first judgment features and third identification features, fuzzification is performed. When the feature value is 0, its low-level membership degree is 1, and the membership degrees for medium and high levels are both 0. When the feature value is 1, its high-level membership degree is 1, and the membership degrees for low and medium levels are both 0. For the second quantization feature of continuous numerical types, a triangular membership function is used to calculate the membership degree of each input item, with the vertices of the triangular membership function corresponding to the midpoints of each level interval. Rule matching is completed through fuzzy logic operations that take the smaller and larger values. Finally, the centroid method is used for defuzzification calculation, that is, calculating the weighted average of the membership degrees of all matched rules, and this average value is determined as the disease epidemic probability. For example, when the first judgment feature is marked as 1 (i.e., feeding pressure exceeds the vegetation compensating growth threshold), the second quantization feature corresponds to a high inhibition level, and the third identification feature is marked as 1 (i.e., ecosystem stress decline), a high-risk inference rule is matched. After defuzzification calculation using the centroid method, the disease epidemic probability is determined to be 0.7.

[0045] By clarifying the calibration method of the benchmark parameters, the judgment criteria for decoupling early warning, and the specific rules of fuzzy inference, and combining data processing and inference operations, the probability of disease outbreaks can be quantitatively determined.

[0046] Furthermore, the method provided in this application embodiment includes:

[0047] The degree of substrate supply for decomposition is determined based on the structural perception data, the microbial decomposition potential is determined based on the functional perception data, and the exogenous variables are determined based on the interaction perception data. The contribution changes of each trend term before and after the breakpoint are taken and weighted summed to obtain the home advantage breakpoint offset. Each trend term is identified with a posterior contribution weight. The breakpoint type includes substrate-limited breakpoints or microbial activity-inhibiting breakpoints. Substrate-limited breakpoints trigger supplemental feeding, while microbial activity-inhibiting breakpoints trigger grazing rest.

[0048] In this embodiment, home advantage refers to the ecological stability of a grazing area ecosystem where the supply of decomposition substrate and the decomposition potential of microorganisms work synergistically to stably support the grazing needs of yaks. The home advantage breakpoint offset is a quantified value of the impact of changes in various trend terms on home advantage when this stable state is disrupted, reflecting changes in the ecosystem's ability to support yak grazing.

[0049] Specifically, firstly, the decomposition matrix supply level is calculated based on the normalized structure-aware data. The structure-aware data includes weekly collected vegetation litter biomass, aboveground vegetation biomass, and vegetation cover. The vegetation litter biomass is obtained by collecting surface litter in 1m×1m standard quadrats and weighing it on-site. The litter biomass is weighted and summed with a weight of 0.7 for vegetation litter biomass and 0.3 for aboveground vegetation biomass to obtain the time series of decomposition matrix supply level. The corresponding weights are calibrated by correlation analysis of 12 months of grazing plot monitoring data.

[0050] Microbial decomposition potential was calculated based on normalized functional perception data, which included weekly measurements of soil microbial biomass carbon, soil urease activity, and soil nematode metabolic footprint. Soil nematode metabolic footprint was calculated by multiplying the abundance of each nematode group by the corresponding group's metabolic constant and then summing the results. The results were weighted and summed with a weight of 0.4 for soil microbial biomass carbon, 0.3 for soil urease activity, and 0.3 for soil nematode metabolic footprint to obtain the time series of microbial decomposition potential. The corresponding weights were calibrated through soil ecological function correlation experiments.

[0051] Exogenous variables were calculated based on normalized interaction-sensing data, which included weekly statistics on plant disease severity, yak grazing frequency, and saprophytic arthropod abundance. Plant disease severity was quantified as the proportion of diseased plants in the standard quadrat to the total number of plants. Yak grazing frequency was calculated as the weekly average of the total number of times yaks grazed daily in the monitoring area. The variables were weighted and summed with a weight of 0.4 for plant disease severity, 0.4 for yak grazing frequency, and 0.2 for saprophytic arthropod abundance. When the saprophytic arthropod abundance was 0, a value of 0.01 was uniformly used in the calculation to obtain the time series of exogenous variables. The corresponding weights were calibrated through perturbation sensitivity analysis.

[0052] Next, a sliding window with a length of 5 weeks and a sliding step size of 1 week was used to perform linear fitting of the three time series along the time series direction: the supply of decomposition substrate, the microbial decomposition potential, and the exogenous variable. The week number within the window was the independent variable x, which was successively selected from 1 to 5. The week number of the corresponding index was the dependent variable y. The fitted line y=kx+b was solved by minimizing the sum of squared residuals, where k is the trend slope corresponding to the window. The trend slope obtained by fitting each window was assigned to the last week of the corresponding window. At the same time, the significance threshold of the fitting was set to a coefficient of determination greater than or equal to 0.8. If the coefficient of determination of the fitting result of a certain window was less than 0.8, the invalid data of that window was removed and the average trend slope of the two adjacent valid windows was used instead.

[0053] The step threshold was determined using monitoring data from a stable grazing area over the previous 12 months. A stable grazing area is defined as a grazing area where grazing intensity is consistently controlled at 2 head per hectare, free from extreme weather and human disturbance. Weekly slope changes of each trend term within this area were monitored for 12 consecutive months, and the upper limit of the 95% confidence interval for this data set was used as the step threshold. When the absolute value of the difference in the trend slope between two consecutive weeks for any of the three trend terms is greater than or equal to the step threshold, that week is considered a trend breakpoint. When multiple trend terms trigger the threshold simultaneously, the trend term with the largest absolute value of change is used as the core criterion.

[0054] Next, after the breakpoint determination is completed, the absolute values ​​of the changes in the trend slopes of the substrate supply and microbial decomposition potential are compared. If the absolute value of the change in the substrate supply is greater, the breakpoint is determined to be a substrate-limited breakpoint; if the absolute value of the change in microbial decomposition potential is greater, the breakpoint is determined to be a microbial activity-inhibiting breakpoint; if the absolute values ​​of the two changes are equal, the breakpoint is preferentially determined to be a microbial activity-inhibiting breakpoint. For each determined breakpoint, a time window of 4 weeks before and 4 weeks after the breakpoint is taken, and the arithmetic mean of each trend term within the corresponding window is calculated. The average value of the window after the breakpoint is subtracted from the average value of the window before the breakpoint to obtain the contribution change of each trend term.

[0055] Furthermore, the posterior contribution weights were determined using linear regression analysis based on historical monitoring data of the grazing area over the past 24 months. The ecosystem stability index was used as the dependent variable, which is the arithmetic mean of the substrate supply and microbial decomposition potential. The changes in the contribution of each trend term were used as independent variables. The weight coefficients of each trend term were optimized through linear regression fitting, ensuring that the coefficient of determination of the fitting results was greater than or equal to 0.85. For example, the final weights were: substrate-limited breakpoints corresponded to a substrate supply weight of 0.6, microbial decomposition potential weight of 0.3, and exogenous variable weight of 0.1; microbial activity-inhibiting breakpoints corresponded to microbial decomposition potential weight of 0.6, substrate supply weight of 0.3, and exogenous variable weight of 0.1. The changes in the contribution of each trend term were multiplied by the posterior contribution weight corresponding to the breakpoint type, and the three products were summed to obtain the home-field advantage breakpoint offset.

[0056] Finally, the corresponding yak grazing scheduling operation is executed according to the breakpoint type. When it is determined to be a substrate-limiting breakpoint, the yak supplementary feeding operation is triggered. The supplementary feeding standard is to supplement each yak with 2 kg of hay per day. The supplementary feeding continues until the substrate decomposition potential recovers to above 0.5 after normalization. When it is determined to be a microbial activity inhibition breakpoint, the corresponding grazing area rest operation is triggered. The rest period is set to 15 days. After the microbial decomposition potential recovers to the normal fluctuation range of 0.4 to 0.6 after normalization, the normal grazing operation in the area is resumed.

[0057] By clarifying the calculation methods of core indicators, the rules for the entire process of breakpoint determination, the logic of weight calibration and the standards for scheduling execution, and combining ecological monitoring and data processing methods, the calculation of the breakpoint offset of the home advantage and the precise scheduling of yak grazing were realized.

[0058] Furthermore, the method provided in this application embodiment includes:

[0059] Based on the structure-sensing data, the vegetation heterogeneity index is measured; based on the function-sensing data, the litter carbon-nitrogen ratio index is measured; based on the interaction-sensing data, the resource availability index is measured; based on the combination patterns of different data levels of the vegetation heterogeneity index, litter carbon-nitrogen ratio index, and resource availability index, the functional group balance index is determined, wherein the combination patterns include habitat simplification imbalance, substrate quality imbalance, or resource overload imbalance.

[0060] In this embodiment, the functional group balance index is a quantitative indicator that characterizes the degree of synergistic stability among vegetation community, soil decomposition function, and ecological resource supply in a grazing ecosystem. Habitat simplification imbalance refers to ecological function imbalance caused by excessively low heterogeneity in vegetation species composition and spatial distribution. Substrate quality imbalance refers to imbalance caused by excessively high litter carbon-nitrogen ratio leading to a decline in soil substrate decomposition capacity. Resource overload imbalance refers to imbalance caused by ecological resource supply exceeding the yak grazing demand and the ecosystem's absorption capacity.

[0061] Specifically, firstly, vegetation heterogeneity index is measured based on normalized structure-aware data. Ten 1m×1m monitoring plots are randomly set up in the monitoring area each week. The vegetation species types in each plot are identified and the coverage percentage of each type of vegetation is counted. The Simpson diversity index is used to calculate the vegetation heterogeneity value of a single plot by subtracting the sum of squares of the coverage percentages of each type of vegetation in a single plot from 1. Then, the arithmetic mean of the values ​​of the 10 plots is calculated as the vegetation heterogeneity index for that week. The larger the index value, the higher the species and spatial heterogeneity of the vegetation.

[0062] The litter carbon-nitrogen ratio index is measured based on normalized functional perception data. Surface litter samples are collected weekly within the monitoring area. The total carbon content of the litter is determined using the potassium dichromate volumetric method, and the total nitrogen content is determined using the Kjeldahl method. The original carbon-nitrogen ratio is obtained by dividing the total carbon content by the total nitrogen content. The maximum and minimum values ​​of the original carbon-nitrogen ratio are obtained through continuous monitoring of 12 months of ungrazing baseline plots. The original carbon-nitrogen ratio is then subjected to minimum-maximum normalization to obtain the litter carbon-nitrogen ratio index. A higher index value indicates a worse quality of the litter decomposition matrix.

[0063] The resource availability index is measured based on normalized interaction perception data. The vegetation foraging surplus is calculated by subtracting the measured vegetation biomass after foraging from the baseline biomass. The available phosphorus content in the soil is detected using the molybdenum-antimony colorimetric method, and the available potassium content is detected using the flame photometry method. The abundance of saprophytic arthropods is obtained by statistically analyzing the mean value after sweeping three times a week. The normalized values ​​of the above three categories are multiplied by their corresponding weights and then summed. The weights are calibrated by linear regression analysis of 24 months of historical monitoring data. For example, the weight of vegetation foraging surplus is 0.4, the weight of available phosphorus and potassium content in the soil is 0.3, and the weight of saprophytic arthropod abundance is 0.3. Finally, the resource availability index is obtained. The higher the index value, the higher the degree of availability of regional ecological resources.

[0064] Next, the vegetation heterogeneity index, litter carbon-nitrogen ratio index, and resource availability index are divided into three data levels: values ​​less than or equal to 0.33 are considered low-level, values ​​greater than 0.33 and less than or equal to 0.66 are considered medium-level, and values ​​greater than 0.66 are considered high-level. The type of imbalance is determined based on the combination pattern of the data levels of the three indices. When only the vegetation heterogeneity index is low-level and the other two indices are medium-level or high-level, it is determined to be habitat simplification imbalance. When only the litter carbon-nitrogen ratio index is high-level and the other two indices are low-level or medium-level, it is determined to be substrate quality imbalance. When only the resource availability index is high-level and the other two indices are low-level or medium-level, it is determined to be resource overload imbalance. When multiple indices trigger imbalance conditions at the same time, the priority of substrate quality imbalance, habitat simplification imbalance, and resource overload imbalance is used for determination. When all three indices are medium-level, it is determined to be an ecological equilibrium state.

[0065] Finally, a weighted summation combined with imbalance correction was used to determine the functional group balance index. The basic weights were calibrated through ecological sensitivity analysis, with vegetation heterogeneity index weighted at 0.4, litter carbon-nitrogen ratio index weighted at 0.3, and resource availability index weighted at 0.3. The basic balance value was obtained by multiplying the three indices by their corresponding weights and summing them. The imbalance correction coefficient was calibrated through grazing plot control experiments. Under ecological balance conditions, the basic balance value was directly used as the functional group balance index. For habitat simplification imbalance, the basic balance value was multiplied by a correction coefficient of 0.7; for substrate quality imbalance, it was multiplied by a correction coefficient of 0.6; and for resource overload imbalance, it was multiplied by a correction coefficient of 0.5. The corrected value was the final functional group balance index.

[0066] By refining the index detection and calculation methods, improving the data order determination rules and parameter calibration basis, and combining the ecological index calculation and weighted correction methods, the functional group equilibrium index can be quantitatively determined.

[0067] Furthermore, the method provided in this application embodiment includes:

[0068] By performing fuzzy logic transformation on the baseline grazing elements, fuzzy membership coefficients are determined, wherein each grazing sub-region corresponds to a set of fuzzy membership coefficients; for each grazing sub-region, the fuzzy membership coefficients are weighted and integrated to determine the comprehensive ecological load coefficient, and the numerical range is mapped to the carrying capacity adjustment level; the carrying capacity adjustment level is integrated based on the spatiotemporal phase of the grazing sub-region, and the grazing scheduling strategy is determined through iterative decision-making.

[0069] Specifically, firstly, based on the aforementioned steps, four core indicators of the baseline grazing elements are calculated: ecological network coupling degree, disease prevalence probability, home-field advantage breakpoint offset, and functional group balance index. The calculation results are then normalized and mapped to the 0-1 range. Next, unified data preprocessing is performed on the four core indicators collected weekly from each grazing sub-region. During data preprocessing, for outliers with missing monitoring data or values ​​exceeding the normalization range, the average of valid data from three adjacent monitoring periods in the same region is used for replacement. For extreme outliers in a single monitoring session of a single indicator, the Grubbs criterion is used for removal to ensure the validity of the input data. The min-maximum normalization method is used: the weekly monitoring value of a single indicator is subtracted from the minimum value of that indicator within a 12-month continuous monitoring period, and then divided by the difference between the maximum and minimum values ​​within that period. This maps all indicator values ​​to the 0-1 range, eliminating dimensional differences between different indicators and providing a unified data foundation for subsequent fuzzy logic transformation.

[0070] For each grazing subregion, the four preprocessed core indicators are subjected to a complete fuzzy logic transformation. For each indicator, three levels of fuzzy subsets are constructed: low (0-0.3), medium (0.3-0.7), and high (0.7-1.0). The boundary value of 0.3 belongs to the medium-level fuzzy subset, and the boundary value of 0.7 belongs to the high-level fuzzy subset. A triangular membership function is used to calculate the membership degree values ​​for each indicator corresponding to different fuzzy subsets. The three vertices of the membership function for the low-level are 0, 0.15, and 0.3; for the medium-level, they are 0.3, 0.5, and 0.7; and for the high-level, they are 0.7, 0.85, and 1.0.

[0071] Subsequently, a basic fuzzy rule base covering all fuzzy subset combinations of the four indicators was established. The core inference rules include: outputting a fuzzy result of medium level when all four indicators are at a medium level; outputting a fuzzy result of high level when both ecological network coupling degree and home-field advantage breakpoint offset are at a high level; outputting a fuzzy result of low level when ecological network coupling degree, disease prevalence probability, and home-field advantage breakpoint offset are all at a low level and the functional group balance index is at a medium-high level; and outputting a fuzzy result of medium level when ecological network coupling degree is at a medium level and the other three indicators are at a low-medium level. Rule conflict handling employs priority ranking, with the combination rule of ecological network coupling degree and home-field advantage breakpoint offset having the highest priority, and rules related to the functional group balance index having the lowest priority. Next, the Mamdani minimum-maximum inference method is used to perform fuzzy logic inference. First, the minimum value of the membership degree of each indicator in each rule is taken as the applicable strength of the rule. Then, the maximum value of the strength of all applicable rules is taken as the comprehensive inference result. The centroid method is used to complete the defuzzification. The centroid of the area enclosed by the membership function curve of the inference result and the horizontal axis is calculated. The corresponding horizontal axis value is the single-value fuzzy membership degree coefficient of each indicator. The four indicators of each grazing sub-region form a set of fuzzy membership degree coefficients.

[0072] For each grazing subregion, a weighted integral operation was performed to determine the comprehensive ecological load coefficient. The weighting of each coefficient was pre-calibrated through a 12-month alpine meadow grazing ecological sensitivity control experiment. Three parallel plots (10m × 10m) were set up for the control experiment, maintaining consistent site conditions and initial vegetation community composition. Three grazing intensity gradients (light, moderate, and heavy) were established, corresponding to livestock carrying capacities of 1 head per hectare, 2 heads per hectare, and 3 heads per hectare, respectively. Vegetation biomass, soil microbial biomass, and vegetation degradation indicators were monitored monthly. Pearson correlation analysis was used to calculate the correlation coefficients between each coefficient and the degree of vegetation degradation, and the weighting of each coefficient was obtained after normalization. Using a continuous 4-week monitoring period as the integration interval, the week number as the independent variable, and the sum of the fuzzy membership coefficients of each week multiplied by their corresponding weights as the dependent variable, the trapezoidal numerical integration method is used to complete the definite integral calculation. The integral result is divided by the length of the 4-week integration interval and normalized to obtain the comprehensive ecological load coefficient in the interval of 0 to 1. The higher the value, the greater the ecological carrying capacity pressure of the corresponding grazing sub-region.

[0073] Subsequently, the first low threshold and second high threshold, determined through a pre-calibrated control experiment of alpine meadow grazing, were established. The control experiment and the weight calibration experiment used the same plot setup and grazing gradient. The core criterion was that the alpine meadow vegetation community did not undergo reverse succession. The quantitative standard for not undergoing reverse succession was that the increase in the proportion of poisonous grass species did not exceed 2% during the experimental period, and the proportion of perennial high-quality forage species did not show a continuous downward trend. The load range for stable ecosystem carrying capacity was determined through gradient grazing experiments, and the first low threshold and second high threshold were calibrated. A mapping rule was established between the comprehensive ecological load coefficient value range and the carrying capacity adjustment level. Ranges with a comprehensive ecological load coefficient value less than the first low threshold were mapped to the increased grazing level; ranges with a comprehensive ecological load coefficient value greater than or equal to the first low threshold and less than or equal to the second high threshold were mapped to the maintenance level; and ranges with a comprehensive ecological load coefficient value greater than the second high threshold were mapped to the reduced grazing and rest level. This completed the mapping correspondence between the comprehensive ecological load coefficient and the carrying capacity adjustment level. When the comprehensive ecological load coefficient reaches extreme values ​​close to 0 or 1, the corresponding increases in grazing and decreases in grazing are fixed respectively, and no further adjustments are made to ensure the stability of the scheduling strategy.

[0074] Furthermore, the method provided in this application embodiment includes:

[0075] The ecological network coupling degree is mapped to the grazing reduction intensity coefficient, the disease epidemic probability is mapped to the disease-driven grazing increase coefficient, the home advantage breakpoint offset is mapped to the urgency coefficient of grazing rest, and the functional group balance index is mapped to the diversity correction coefficient. Among them, the determination of the carrying capacity adjustment level includes: if the value range is less than the first low threshold, increase grazing by one level; if the value range is greater than the first low threshold and the second high threshold, maintain the status quo; if the value range is greater than the second high threshold, reduce grazing by one level and extend the grazing rest period.

[0076] In one embodiment, directional mapping is performed on the four core indicators of baseline grazing elements to obtain corresponding fuzzy membership coefficients. The normalized value of the ecological network coupling degree is mapped to a grazing reduction intensity coefficient using a linear positive correlation mapping function. The grazing reduction intensity coefficient is equal to the normalized value of the ecological network coupling degree, with a value range of 0 to 1. The normalized value of the disease epidemic probability is mapped to a disease-driven grazing increase coefficient using a linear negative correlation mapping function. The disease-driven grazing increase coefficient is equal to 1 minus the normalized value of the disease epidemic probability, with a value range of 0 to 1. The normalized value of the home-field advantage breakpoint offset is mapped to a grazing rest urgency coefficient using a linear positive correlation mapping function. The grazing rest urgency coefficient is equal to the normalized value of the home-field advantage breakpoint offset, with a value range of 0 to 1. The normalized value of the functional group balance index is mapped to a diversity correction coefficient using a linear positive correlation mapping function. The diversity correction coefficient is equal to the normalized value of the functional group balance index, with a value range of 0 to 1. During data processing, for outliers such as missing monitoring data or values ​​exceeding the normalization range, the average of valid data from three adjacent monitoring periods in the same region is used for replacement. For extreme outliers in a single monitoring of a single indicator, the Grubbs criterion is used for removal to ensure the validity of the input data.

[0077] Next, the four mapped coefficients constitute the aforementioned set of fuzzy membership coefficients. For each grazing sub-region, a weighted integral operation is performed on the four coefficients to determine the comprehensive ecological load coefficient. The weighting was pre-calibrated through a 12-month alpine meadow grazing ecological sensitivity control experiment. Three parallel plots of 10m x 10m were set up for the control experiment, maintaining consistent site conditions and initial vegetation community composition. Three grazing intensity gradients (light, moderate, and heavy) were established, corresponding to livestock carrying capacities of 1 head per hectare, 2 heads per hectare, and 3 heads per hectare, respectively. Vegetation biomass, soil microbial biomass, and vegetation degradation indicators were monitored monthly. Pearson correlation analysis was used to calculate the correlation coefficients between each coefficient and the degree of vegetation degradation. After normalization, the weights for the grazing reduction intensity coefficient (0.4), disease-driven grazing increase coefficient (0.2), grazing rest urgency coefficient (0.3), and diversity correction coefficient (0.1) were obtained. Using a continuous 4-week monitoring period as the integration interval, the week as the independent variable, and the sum of the four coefficients multiplied by their corresponding weights each week as the dependent variable, the trapezoidal numerical integration method is used to complete the definite integral calculation. The integral result is divided by the length of the 4-week integration interval and normalized to obtain the comprehensive ecological load coefficient in the interval from 0 to 1.

[0078] Then, a 12-month alpine meadow grazing control experiment was conducted to determine the threshold levels. The control experiment and the weighting calibration experiment used the same plot settings and grazing gradients. The core criterion was that the alpine meadow vegetation community would not undergo reverse succession, thus determining the load range for the stable carrying capacity of the ecosystem. The first low threshold was set at 0.3, and the second high threshold was set at 0.7. Based on the calibrated thresholds, the livestock carrying capacity adjustment levels were determined. When the comprehensive ecological load coefficient was less than the first low threshold, it was determined to be an adjustment level of increasing grazing by one level. When the comprehensive ecological load coefficient was greater than or equal to the first low threshold and less than or equal to the second high threshold, it was determined to be an adjustment level of maintaining the status quo. When the comprehensive ecological load coefficient was greater than the second high threshold, it was determined to be an adjustment level of reducing grazing by one level and extending the rest period.

[0079] For boundary cases where the comprehensive ecological load coefficient is exactly equal to the threshold, a judgment standard consistent with the fuzzy subset assignment rule is adopted: a value of 0.3 corresponds to the maintenance level, and a value of 0.7 corresponds to the reduction and rest level, ensuring uniformity of implementation standards. The adjustment range for the first grazing increase level is a 10% increase in carrying capacity per unit area, while the adjustment range for the first grazing decrease level is a 15% decrease. The extended rest period corresponds to an extension of 10 days from the basic 15-day rest period. The adjustment parameters are jointly calibrated through alpine meadow pasture regeneration rate control experiments and soil microbial community recovery experiments. The experiments use the same plot settings as the weight calibration experiments, continuously monitoring pasture regeneration rate and soil microbial community recovery cycle to determine the reasonable range of the corresponding adjustment parameters. When the comprehensive ecological load coefficient reaches extreme values ​​close to 0 or 1, the corresponding first grazing increase and first grazing decrease levels with extended rest periods are fixed, and no further adjustments are made, ensuring the stability of the scheduling strategy.

[0080] Finally, after mapping the carrying capacity adjustment levels for each grazing sub-region, spatiotemporal phase integration based on the grazing sub-regions is performed. In the spatial dimension, using the centroid coordinates of each grazing sub-region as the benchmark, grazing sub-regions with a horizontal distance of less than 50 meters, adjacent geographical locations, and consistent carrying capacity adjustment levels are merged into the same scheduling unit. In cases where adjacent sub-regions have inconsistent levels, the level of the sub-region with the higher comprehensive ecological load coefficient is used as the unified scheduling level after merging, avoiding fragmented scheduling from affecting grazing efficiency. In the temporal dimension, the carrying capacity adjustment level determination results of each sub-region for four consecutive weeks are integrated, and occasional level fluctuations in a single week are eliminated. Stable levels that remain consistent for two consecutive weeks or more are used as the effective scheduling level for the sub-region or the merged scheduling unit, providing a unified and standardized input basis for subsequent iterative decision-making to determine grazing scheduling strategies.

[0081] Furthermore, the method provided in this application embodiment includes:

[0082] A scheduler array is deployed, wherein the scheduler array consists of a set of schedulers corresponding to different scheduling dimensions, and is built using an adversarial network training method; the carrying capacity adjustment level is input into the scheduler array, and the rotational grazing migration path, recommended carrying capacity, dwell time in each grazing sub-region, targeted foraging priority area and rest and recovery period duration are output in parallel; the recommended carrying capacity, dwell time in each grazing sub-region, targeted foraging priority area and rest and recovery period duration are encoded based on spatiotemporal gradient, and spatial phase-based distribution marking is performed on the rotational grazing migration path to determine the grazing scheduling strategy.

[0083] In this embodiment, the rotational grazing migration path is the orderly movement route of yaks between various grazing sub-areas. The recommended carrying capacity is the number of yaks that a unit area of ​​grassland can comfortably support. The dwell time in each grazing sub-area is the duration for which yaks continuously graze within a single sub-area. The targeted foraging priority area is the area within the grazing sub-area with excellent forage quality suitable for priority grazing. The rest and recovery period is the duration during which grazing is prohibited in the sub-area after grazing ends, awaiting ecological recovery.

[0084] In one embodiment, after completing the spatiotemporal phase integration of the carrying capacity adjustment levels, several independent scheduling units are obtained after merging. Each scheduling unit corresponds to a unique effective carrying capacity adjustment level. The effective level is one of three levels: increasing grazing by one level for two consecutive weeks or more, maintaining the status quo, or reducing grazing by one level and extending the rest period. The number of grazing sub-areas, spatial coordinates, and area data corresponding to each scheduling unit are simultaneously determined. One-hot encoding is performed on the effective level of each scheduling unit. The three levels correspond to three independent input feature nodes: increasing grazing by one level is coded as 1, 0, 0; maintaining the status quo is coded as 0, 1, 0; and reducing grazing by one level and extending the rest period is coded as 0, 0, 1. When multiple scheduling units have different levels, the level with the highest weight is selected as the main input level according to the spatial area ratio, and the other levels are input as auxiliary features simultaneously. For the number and spatial distribution characteristics of grazing sub-regions, the number of sub-regions is directly taken as the total number of scheduling units after integration. The spatial distribution characteristics are quantified by the spatial discretization of the centroid of the scheduling unit. The standard deviation of the centroid coordinates of all scheduling units is calculated. The two values ​​are then subjected to min-max normalization and mapped to the interval between 0 and 1, corresponding to two input feature nodes. Finally, a standardized input feature vector with a dimension of 5 is formed, completing the complete transformation from the spatiotemporal phase integration result to the model input.

[0085] Next, a scheduler array is deployed. This array consists of four parallel schedulers corresponding to different scheduling dimensions: recommended carrying capacity, sub-region dwell time, targeted grazing priority areas, and rest and recovery period duration. All schedulers are trained and built using a generative adversarial network (GAN), a deep neural network model. The GAN is constructed end-to-end with a fully connected architecture, consisting of two core sub-networks: a generator and a discriminator. The generator is a four-layer fully connected neural network. The input layer has a fixed number of 5 nodes, perfectly matching the dimensions of the standardized input feature vector, and uses the ReLU activation function. The first hidden layer has 128 neurons, also using the ReLU activation function, with a Dropout layer (0.2 inactivation rate) added to prevent overfitting. The second hidden layer has 64 neurons, also using the ReLU activation function. The output layer has four branch node groups corresponding to the outputs of the four scheduling dimensions, all using the Sigmoid activation function to normalize the results to the 0-1 range. All layers are sequentially connected using a fully connected approach, without any skip connections. The discriminator is a three-layer fully connected neural network that forms a closed-loop adversarial structure with the generator. The number of nodes in the input layer is fixed at 32, matching the total feature dimension of the generator's output, and it uses a LeakyReLU activation function with a negative slope of 0.2. The hidden layer has 64 neurons and uses the same type of LeakyReLU activation function. The output layer has 1 neuron and uses a Sigmoid activation function to output a discrimination result of 0 to 1. The closer the value is to 1, the more likely the input scheme is the true optimal scheme. The layers are connected sequentially using a fully connected method.

[0086] Then, a standardized training sample set was constructed, which consisted of historical monitoring data from the target grazing area over the past 36 months and corresponding optimal grazing scheduling schemes that had been verified in practice. The total number of samples was no less than 1000, divided into training, validation, and test sets in a 7:2:1 ratio. The input features of the sample set were unique hot codes for the carrying capacity adjustment levels for each historical period, the number of grazing sub-regions, and their spatial distribution characteristics. The quantification coding method was completely consistent with that in the inference stage. The sample set labels were the optimal grazing scheduling schemes for the corresponding periods, containing core parameters for four scheduling dimensions. The quantification criteria for the optimal scheme were: after the scheme was implemented, the monthly decrease in vegetation cover in the entire area did not exceed 5%, the fluctuation range of the proportion of dominant forage grasses did not exceed 10%, and the soil organic matter content did not decrease significantly, while simultaneously maximizing yak carrying capacity. The criteria for not experiencing reverse succession of the vegetation community were: the increase in the proportion of toxic grasses did not exceed 2%, and the proportion of perennial high-quality forage grasses did not show a continuous downward trend. This was used to complete the screening and calibration of all samples.

[0087] Subsequently, an alternating iterative training process was executed to complete the scheduler array setup. Each round of training was divided into two independent stages: First, the discriminator was trained with the generator weights fixed. The real optimal solution in the training set was marked as a positive sample (label 1), and the solution generated by the generator was marked as a negative sample (label 0). The loss was calculated using the binary cross-entropy loss function, and the weights were updated using the Adam optimizer. The learning rate was set to 0.0002, and the batch size was 32. Then, the generator was trained with the discriminator weights fixed. The generator output solution was input into the discriminator, and the goal was to make the discrimination result approach 1. The weights were updated using the same loss function, optimizer, and hyperparameters. The total number of iterations was set to 10,000 rounds. The performance was verified using a validation set every 100 rounds. When the discriminator's recognition accuracy on the validation set stabilized in the range of 45%-55% and the fluctuation range did not exceed 2% for 200 consecutive rounds, the model was considered to have converged, training was stopped, and the generator weight parameters were saved. For anomalies such as pattern collapse and failure to converge that occurred during training, the learning rate was adjusted to 50% of its original value, the sample size was expanded, and the Dropout inactivation rate was adjusted to 0.3 to correct the problem.

[0088] Next, the standardized input feature vector is input into the trained scheduler array. Through forward propagation calculation by the generator, normalized scheduling parameters within the range of 0 to 1 are output in parallel. These parameters include the rotational grazing migration path, recommended carrying capacity, dwell time in each grazing sub-region, targeted foraging priority areas, and rest and recovery period duration. The output process of the rotational grazing migration path will be explained in detail later. Then, all normalized output results are denormalized and converted into executable physical quantities: the recommended carrying capacity is capped at 3 head per hectare (based on the grassland carrying capacity of the target area), and the normalized value is multiplied by the cap to obtain the actual carrying capacity per unit area; the dwell time and rest and recovery period duration are capped at 15 days and 30 days respectively, and the normalized value is multiplied by the cap and rounded to obtain the actual number of days; the targeted foraging priority areas are grid nodes with an output priority greater than 0.7 included in the corresponding set. For abnormal output results, a threshold truncation method is used to restrict the values ​​to the upper and lower limits of the corresponding parameters to ensure compliance with grazing operation logic.

[0089] Finally, the four scheduling parameters that have undergone denormalization are encoded using a spatiotemporal gradient. The time dimension uses a 30-day grazing cycle as the time axis, dividing it into 30 time nodes with unique integer codes. The dwell time and rest period are matched as continuous intervals on the time axis, corresponding to the encoded values ​​of the start and end nodes. The spatial dimension uses a pre-divided 1m × 1m standard monitoring grid. Each grid node corresponds to a unique row and column number integer code. The recommended carrying capacity and targeted grazing priority areas are bound to the corresponding grid codes, completing the spatiotemporal gradient encoding. The encoded scheduling parameters are then distributed and marked based on spatial phase along the rotational grazing migration path. The path start point is the spatial phase zero point. Continuous spatial phase values ​​from 0 to 1 are calculated along the direction of travel using the ratio of cumulative travel distance to total path length. Each sub-region entrance corresponds to a unique phase value. The encoded parameters of the corresponding sub-region are bound to this phase value, integrating all marked content with the rotational grazing migration path to form a complete and executable grazing scheduling strategy.

[0090] By clarifying the connection logic between spatiotemporal phase integration and model input, the complete architecture and training process of generative adversarial networks, the standardized rules for input and output, and the execution method of encoding tags, a reproducible and highly ecologically adaptable yak grazing scheduling strategy based on an adversarial network scheduler array was generated.

[0091] Furthermore, the method provided in this application embodiment includes:

[0092] The first scheduler array receives the livestock carrying capacity adjustment level, determines the first rotational grazing migration path according to the first generation component, performs path verification based on the second judgment component, and determines the first feedback feature; inputs the first feedback feature into the first generation component, iteratively determines the second rotational grazing migration path, and through multiple rounds of iteration, until the determined feedback feature meets the preset standard, and determines the rotational grazing migration path.

[0093] Optionally, the first scheduler array belongs to the scheduler array in the aforementioned steps. It is an independent sub-scheduler array specifically responsible for generating rotational grazing migration paths. It consists of a first generation component and a second judgment component. Both the first generation component and the second judgment component are built using a fully connected neural network architecture, consistent with the generative adversarial network training logic in the aforementioned steps. The first scheduler array receives the carrying capacity adjustment level after spatiotemporal phase integration, and simultaneously receives the spatial coordinates, area, comprehensive ecological load coefficient, determined residence time, and rest period duration data of each grazing sub-region. It performs one-hot encoding and min-max normalization on all input data to form a standardized input feature vector with fixed dimensions, providing unified basic input data for path generation.

[0094] Then, the standardized input feature vector is input into the trained first generation component to generate the first round of grazing migration paths. The first generation component is a three-layer fully connected neural network. The number of nodes in the input layer perfectly matches the dimension of the standardized input feature vector. The hidden layer has 64 neurons and uses the ReLU activation function. The output layer corresponds to the path node sequence of the grazing area's gridded coordinates and uses the Sigmoid activation function to normalize the output results of the path nodes to the 0-1 range. Through the forward propagation calculation of the first generation component, the normalized result of the output path node sequence is output. The output result is then subjected to inverse normalization processing to match the actual gridded coordinates of the grazing area. All valid path nodes are connected according to the path progression order to form a continuous first round of grazing migration path. The generated path covers the entrance and exit of all grazing sub-areas corresponding to the livestock carrying capacity adjustment level, while avoiding the preset ecological protection core area and terrain obstacle areas.

[0095] Next, the generated first-round grazing migration path is input into the second judgment component to perform path verification and determine the first feedback feature. The second judgment component is a two-layer fully connected neural network. The number of nodes in the input layer matches the node sequence dimension of the first-round grazing migration path, and the output layer corresponds to the output of three path verification indicators, using the Sigmoid activation function. The second judgment component performs three core path verifications: the first is ecological adaptability verification, which calculates the average comprehensive ecological load coefficient of the path coverage area to verify the path's avoidance effect on high ecological load areas; the second is path feasibility verification, which calculates the total length and slope variation of the path to verify whether the path meets the yak grazing mobility requirements; and the third is scheduling matching verification, which verifies whether the order of grazing sub-regions covered by the path matches the corresponding stay time and rest period duration. Through these three verifications, the ecological adaptability score, feasibility score, and scheduling matching score of the path are calculated. All three scores are normalized values ​​in the range of 0 to 1. The three scores are combined to form the first feedback feature, with the closer the value is to 1, the better the overall performance of the path.

[0096] Next, the obtained first feedback feature is concatenated with the original standardized input feature vector to form a new input feature vector, which is then input into the first generation component to iteratively determine the second round of herding migration path. The first generation component optimizes the three scores of the first feedback feature to be close to 1, updates the network weight parameters through backpropagation, re-executes the forward propagation calculation, and outputs the optimized path node sequence. After inverse normalization, the second round of herding migration path is formed. The second round of herding migration path is then input into the second judgment component, and the three-path verification process is repeated to obtain new feedback features, completing a single round of iterative optimization.

[0097] Finally, a preset standard for path verification is established: all three verification scores must be greater than or equal to 0.9, and the score fluctuation between two consecutive iterations must not exceed 0.02. The iterative optimization process is repeated multiple times. In each iteration, the feedback features from the previous round are input into the first generation component to generate an optimized cycle-migration path and perform path verification. Iteration stops immediately when the feedback features from the latest round meet the preset standard, and the cycle-migration path generated in the current round is determined as the final output. If the preset standard is not met even after 100 iterations (the maximum preset number of iterations), the cycle-migration path with the highest average score across all iterations is selected as the final output. The final determined cycle-migration path is synchronously output to the scheduler array to complete subsequent spatiotemporal gradient encoding and spatial phase distribution labeling.

[0098] In summary, the yak precision grazing scheduling method based on multi-source sensing provided in this application has the following technical effects:

[0099] This application collects baseline elements of yak grazing areas through multi-source sensing, determines the comprehensive ecological load coefficient and carrying capacity adjustment level through fuzzy logic transformation and weighted integration, and then inputs them into an adversarial scheduler array for iterative decision-making after spatiotemporal phase integration. Combined with path verification, the rotational grazing migration path is optimized, and spatiotemporal gradient coding and spatial phase marking are completed to accurately determine the grazing scheduling strategy. This improves the ecological adaptability and execution accuracy of yak grazing scheduling, achieving the technical effect of precise and dynamic grazing scheduling while taking into account ecological stability and breeding management efficiency.

[0100] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides a yak precision grazing scheduling system based on multi-source sensing, the system comprising:

[0101] Multi-source heterogeneous data acquisition module 1 is used to synchronously collect multi-source heterogeneous data by deploying multi-source sensing components in the grazing area, wherein the multi-source heterogeneous data includes structural sensing data, functional sensing data and interaction sensing data.

[0102] The baseline grazing element calculation module 2 is used to determine the baseline grazing elements by performing feature element directional mining under multi-threaded parallel processing based on the multi-source heterogeneous data and the data processing component. The baseline grazing elements include ecological network coupling degree, disease prevalence probability, home advantage breakpoint offset and functional group balance index.

[0103] The grazing scheduling strategy acquisition module 3 is used to convert the baseline grazing elements into fuzzy membership coefficients, analyze the livestock carrying capacity adjustment level based on the comprehensive ecological load, trigger the scheduling decision of the scheduler array, and determine the grazing scheduling strategy.

[0104] The grazing scheduling strategy display module 4 is used to display the grazing scheduling strategy on a mobile terminal interface.

[0105] Furthermore, the baseline grazing element calculation module 2 is used to perform the following steps:

[0106] Using the structural sensing data as the main dimension of the aboveground indicator sequence, the functional sensing data as the main dimension of the underground indicator sequence, and the interaction sensing data as a correction term for the aboveground indicator sequence, dynamic time warping based on a sliding window is employed to calculate the self-regulating distance and cross-regulating distance between the aboveground and underground indicator sequences to determine the ecological network coupling degree. Specifically, the ecological network coupling degree is obtained through exponential mapping by taking the mean ratio of the cross-regulating distance to the self-regulating distance as a normalization condition.

[0107] Furthermore, the baseline grazing element calculation module 2 is used to perform the following steps:

[0108] Based on the structural perception data, it is determined whether the foraging pressure exceeds the vegetation compensating growth threshold, which serves as the first judgment feature; based on the functional perception data, the degree of inhibition of underground decomposition channels by grazing is quantified, which serves as the second quantitative feature; based on the interaction perception data, the decline in ecosystem resilience caused by overgrazing is identified, which serves as the third identification feature; the first judgment feature, the second quantitative feature, and the third identification feature are input into a fuzzy inference rule with decoupled early warning status as a prerequisite to determine the probability of disease outbreak.

[0109] Furthermore, the baseline grazing element calculation module 2 is used to perform the following steps:

[0110] The degree of substrate supply for decomposition is determined based on the structural perception data, the microbial decomposition potential is determined based on the functional perception data, and the exogenous variables are determined based on the interaction perception data. The contribution changes of each trend term before and after the breakpoint are taken and weighted summed to obtain the home advantage breakpoint offset. Each trend term is identified with a posterior contribution weight. The breakpoint type includes substrate-limited breakpoints or microbial activity-inhibiting breakpoints. Substrate-limited breakpoints trigger supplemental feeding, while microbial activity-inhibiting breakpoints trigger grazing rest.

[0111] Furthermore, the baseline grazing element calculation module 2 is used to perform the following steps:

[0112] Based on the structure-sensing data, the vegetation heterogeneity index is measured; based on the function-sensing data, the litter carbon-nitrogen ratio index is measured; based on the interaction-sensing data, the resource availability index is measured; based on the combination patterns of different data levels of the vegetation heterogeneity index, litter carbon-nitrogen ratio index, and resource availability index, the functional group balance index is determined, wherein the combination patterns include habitat simplification imbalance, substrate quality imbalance, or resource overload imbalance.

[0113] Furthermore, the grazing scheduling strategy acquisition module 3 is used to perform the following steps:

[0114] By performing fuzzy logic transformation on the baseline grazing elements, fuzzy membership coefficients are determined, wherein each grazing sub-region corresponds to a set of fuzzy membership coefficients; for each grazing sub-region, the fuzzy membership coefficients are weighted and integrated to determine the comprehensive ecological load coefficient, and the numerical range is mapped to the carrying capacity adjustment level; the carrying capacity adjustment level is integrated based on the spatiotemporal phase of the grazing sub-region, and the grazing scheduling strategy is determined through iterative decision-making.

[0115] Furthermore, the grazing scheduling strategy acquisition module 3 is used to perform the following steps:

[0116] A scheduler array is deployed, wherein the scheduler array consists of a set of schedulers corresponding to different scheduling dimensions, and is built using an adversarial network training method; the carrying capacity adjustment level is input into the scheduler array, and the rotational grazing migration path, recommended carrying capacity, dwell time in each grazing sub-region, targeted foraging priority area and rest and recovery period duration are output in parallel; the recommended carrying capacity, dwell time in each grazing sub-region, targeted foraging priority area and rest and recovery period duration are encoded based on spatiotemporal gradient, and spatial phase-based distribution marking is performed on the rotational grazing migration path to determine the grazing scheduling strategy.

[0117] Furthermore, the grazing scheduling strategy acquisition module 3 is used to perform the following steps:

[0118] The first scheduler array receives the livestock carrying capacity adjustment level, determines the first rotational grazing migration path according to the first generation component, performs path verification based on the second judgment component, and determines the first feedback feature; inputs the first feedback feature into the first generation component, iteratively determines the second rotational grazing migration path, and through multiple rounds of iteration, until the determined feedback feature meets the preset standard, and determines the rotational grazing migration path.

[0119] Furthermore, the grazing scheduling strategy acquisition module 3 is used to perform the following steps:

[0120] The ecological network coupling degree is mapped to the grazing reduction intensity coefficient, the disease epidemic probability is mapped to the disease-driven grazing increase coefficient, the home advantage breakpoint offset is mapped to the urgency coefficient of grazing rest, and the functional group balance index is mapped to the diversity correction coefficient. Among them, the determination of the carrying capacity adjustment level includes: if the value range is less than the first low threshold, increase grazing by one level; if the value range is greater than the first low threshold and the second high threshold, maintain the status quo; if the value range is greater than the second high threshold, reduce grazing by one level and extend the grazing rest period.

[0121] The yak precision grazing scheduling system based on multi-source perception provided in this invention can execute the yak precision grazing scheduling method based on multi-source perception provided in any embodiment of this invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0122] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0123] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for precise yak grazing scheduling based on multi-source sensing, characterized in that, The method includes: By deploying multi-source sensing components in grazing areas, multi-source heterogeneous data is collected synchronously, wherein the multi-source heterogeneous data includes structural sensing data, functional sensing data, and interaction sensing data. Based on the multi-source heterogeneous data and data processing components, feature element targeted mining is performed under multi-threaded parallelism to determine baseline grazing elements, wherein the baseline grazing elements include ecological network coupling degree, disease prevalence probability, home advantage breakpoint offset and functional group balance index. By converting the baseline grazing elements into fuzzy membership coefficients, the carrying capacity adjustment level based on the comprehensive ecological load is analyzed, triggering the scheduling decision of the scheduler array, and determining the grazing scheduling strategy. The grazing scheduling strategy is displayed on a mobile terminal interface.

2. The method for precise yak grazing and scheduling based on multi-source sensing as described in claim 1, characterized in that, The targeted mining of the coupling degree of the ecological network includes: The structural sensing data is used as the main dimension of the above-ground indicator sequence, the functional sensing data is used as the main dimension of the underground indicator sequence, and the interaction sensing data is used as the correction term of the above-ground indicator sequence. Dynamic time warping based on sliding window is used to calculate the self-warping distance and cross-warping distance between the above-ground indicator sequence and the underground indicator sequence to determine the coupling degree of the ecological network. The ecological network coupling degree is obtained by taking the mean ratio of cross-normalization distance to self-normalization distance as the normalization condition and using exponential mapping.

3. The method for precise yak grazing and scheduling based on multi-source sensing as described in claim 1, characterized in that, The targeted mining of the disease prevalence probability includes: Based on the structural sensing data, it is determined whether the foraging pressure exceeds the vegetation compensatory growth threshold, which is used as the first judgment feature. Based on the aforementioned functional perception data, the degree of inhibition of underground decomposition channels by grazing is quantified as a second quantitative feature; Based on the interaction perception data, the decline in ecosystem resilience caused by overfeeding is identified as a third identification feature; The first judgment feature, the second quantification feature, and the third identification feature are input into a fuzzy reasoning rule with the decoupled early warning state as a prerequisite to determine the probability of disease prevalence.

4. The method for precise yak grazing and scheduling based on multi-source sensing as described in claim 1, characterized in that, The targeted mining of the home advantage breakpoint offset includes: The degree of substrate supply for decomposition is determined based on the structural perception data, the microbial decomposition potential is determined based on the functional perception data, and the exogenous variables are determined based on the interaction perception data. The contribution changes of each trend term before and after the breakpoint are taken and weighted summed to obtain the home advantage breakpoint offset. Each trend term is marked with a posterior contribution weight. Among them, the breakpoint types include substrate-limited breakpoints or microbial activity-inhibiting breakpoints. Substrate-limited breakpoints trigger supplemental feeding, while microbial activity-inhibiting breakpoints trigger grazing rest.

5. The method for precise yak grazing and scheduling based on multi-source sensing as described in claim 1, characterized in that, The mining of the functional group equilibrium index includes: Based on the structure-sensing data, measure the vegetation heterogeneity index; Based on the aforementioned functional perception data, the carbon-to-nitrogen ratio of litter is measured; Based on the interaction-aware data, a resource availability index is measured. Based on the combination patterns of different data levels of the vegetation heterogeneity index, litter carbon-nitrogen ratio index, and resource availability index, the functional group balance index is determined. The combination patterns include habitat simplification imbalance, substrate quality imbalance, or resource overload imbalance.

6. The method for precise yak grazing and scheduling based on multi-source sensing as described in claim 1, characterized in that, Determining the grazing scheduling strategy includes: By performing fuzzy logic transformation on the baseline grazing elements, fuzzy membership coefficients are determined, wherein each grazing sub-region corresponds to a set of fuzzy membership coefficients; For each grazing sub-area, the fuzzy membership coefficients are weighted and integrated to determine the comprehensive ecological load coefficient, and the numerical range is mapped to the livestock carrying capacity adjustment level. The carrying capacity adjustment level is integrated in spatiotemporal phase based on the grazing sub-region, and the grazing scheduling strategy is determined through iterative decision-making.

7. The method for precise yak grazing and scheduling based on multi-source sensing as described in claim 6, characterized in that, The grazing scheduling strategy is determined through iterative decision-making, including: Deploy a scheduler array, wherein the scheduler array consists of a set of schedulers corresponding to different scheduling dimensions, and is built using adversarial network training; The carrying capacity adjustment level is input into the scheduler array, and the rotational grazing migration path, recommended carrying capacity, dwell time in each grazing sub-area, targeted foraging priority area and rest and recovery period duration are output in parallel. The recommended carrying capacity, dwell time in each grazing sub-region, targeted foraging priority areas and rest and recovery period duration are encoded based on spatiotemporal gradients, and spatial phase-based distribution markings are performed on the rotational grazing migration path to determine the grazing scheduling strategy.

8. The method for precise yak grazing and scheduling based on multi-source sensing as described in claim 7, characterized in that, Output the rotational herd migration path, including: The first scheduler array receives the livestock carrying capacity adjustment level, determines the first rotational grazing migration path according to the first generation component, performs path verification based on the second judgment component, and determines the first feedback feature. The first feedback feature is input into the first generation component, and the second round of herding migration path is determined iteratively. Through multiple rounds of iteration, the determined feedback feature meets the preset standard, and the herding migration path is determined.

9. The method for precise yak grazing and scheduling based on multi-source perception as described in claim 8, characterized in that, The ecological network coupling degree is mapped to the grazing reduction intensity coefficient, the disease prevalence probability is mapped to the disease-driven grazing increase coefficient, the home advantage breakpoint offset is mapped to the urgency coefficient of grazing rest, and the functional group balance index is mapped to the diversity correction coefficient. The determination of the carrying capacity adjustment level includes: if the value range is less than the first low threshold, increase grazing by one level; if the value range is greater than the first low threshold and the second high threshold, maintain the status quo; if the value range is greater than the second high threshold, reduce grazing by one level and extend the rest period.

10. A yak precision grazing scheduling system based on multi-source sensing, characterized in that, The system is used to implement the yak precision grazing scheduling method based on multi-source sensing as described in any one of claims 1-9, the system comprising: A multi-source heterogeneous data acquisition module is used to synchronously collect multi-source heterogeneous data by deploying multi-source sensing components in grazing areas, wherein the multi-source heterogeneous data includes structural sensing data, functional sensing data, and interaction sensing data. The baseline grazing element calculation module is used to determine the baseline grazing elements by performing multi-threaded parallel feature element targeted mining based on the multi-source heterogeneous data and the data processing component. The baseline grazing elements include ecological network coupling degree, disease prevalence probability, home advantage breakpoint offset and functional group balance index. The grazing scheduling strategy acquisition module is used to convert the baseline grazing elements into fuzzy membership coefficients, analyze the livestock carrying capacity adjustment level based on the comprehensive ecological load, trigger the scheduling decision of the scheduler array, and determine the grazing scheduling strategy. The grazing scheduling strategy display module is used to display the grazing scheduling strategy on a mobile terminal interface.

Images