Agricultural resource dynamic allocation method based on Internet of Things

Abstract

The invention discloses an agricultural resource dynamic allocation method based on the Internet of Things, and particularly relates to the technical field of agricultural resource allocation. The method comprises the following steps: collecting farmland environment parameters through Internet of Things nodes, and constructing a resource response weight vector; generating a resource allocation priority sequence in combination with a crop stage growth model; mapping the priority sequence and a resource available matrix, and constructing a resource dynamic allocation matrix; analyzing the spatial topology characteristic parameters of the allocation matrix, adjusting the operation parameters of the nodes of the Internet of Things based on an energy consumption control model, and executing a corresponding resource allocation task; the method realizes accurate, efficient and low-energy-consumption allocation of agricultural resources, is suitable for a multi-plot and heterogeneous crop planting environment, and has good real-time performance and expandability.

Description

Technical Field

[0001] This invention relates to the field of agricultural resource allocation technology, and more specifically to a method for dynamic allocation of agricultural resources based on the Internet of Things. Background Technology

[0002] Currently, the precise allocation of resources such as water, fertilizer, light, and heat in agricultural planting is a key factor in improving crop yield and quality. Traditional agricultural resource allocation methods rely heavily on manual experience or timed control, lacking real-time response to the actual growth status of crops, which can easily lead to resource waste or insufficient supply. For example, in plateau edge areas, arid and semi-arid zones, or seasonally water-scarce regions, the spatial and temporal distribution of agricultural resources is highly uneven. If allocation is not timely, it will seriously affect crop root development and surface temperature and humidity cycles, and may even cause crop growth disorders and large-scale yield reduction.

[0003] Although existing agricultural monitoring systems based on IoT technology have emerged, most still employ a centralized allocation and control model, resulting in a disconnect between data acquisition and response strategies and an inability to achieve dynamic resource matching at the "microclimate unit" level. Furthermore, IoT terminals often utilize general-purpose chips or modules, which are insufficient in low-power operation, edge intelligent analysis, and heterogeneous data fusion. This makes them ill-suited for the precise resource allocation needs in extreme scenarios, particularly in niche environments such as fragmented agricultural plots on plateaus, remote mountainous areas, or regions with high-frequency climate fluctuations, where traditional systems are virtually ineffective. Summary of the Invention

[0004] The purpose of this invention is to provide a method for dynamic allocation of agricultural resources based on the Internet of Things (IoT) to address the shortcomings of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic allocation of agricultural resources based on the Internet of Things, comprising: By deploying IoT nodes in the target farmland, a set of environmental parameters P is collected, including soil moisture, temperature, light intensity, and plant electrical conductivity. Based on the set of environmental parameters P, and combined with the preset small-scale meteorological factor perturbation function R, a resource response weight vector Q for the target farmland is constructed. Based on the resource response weight vector Q and combined with the agricultural crop stage growth model T, a resource allocation priority sequence L matching the current crop stage is generated. The resource allocation priority sequence L is mapped to the existing agricultural resource availability matrix U to construct the dynamic resource allocation matrix M of the target farmland; Obtain the spatial topology feature parameters C of the resource allocation unit in the allocation matrix M, and adjust the scheduling frequency and energy consumption strategy of IoT nodes in the target area according to the spatial topology feature parameters C to generate an energy consumption control model Z. Based on the energy consumption control model Z, the operating parameters of the IoT nodes are adjusted, and various dynamic resource allocation tasks defined in the resource allocation matrix M are executed.

[0006] Preferably, the construction of the resource response weight vector Q for the target farmland includes: The multidimensional sensing data in the environmental parameter set P are synchronously corrected according to the collection timestamp, and the dimensionality is reduced by principal component analysis to obtain the dimensionality-reduced parameter vector P′. The parameter vector P′ is coupled with a preset small-scale meteorological factor perturbation function R for modeling. The perturbation function R includes perturbation weights for micro-wind speed, instantaneous precipitation, and local radiation intensity. A regularized regression algorithm is used to construct the environmental response matrix E. Based on the environmental response matrix E, the response factor group that affects resource allocation is extracted, and based on the resource demand model of the current growth stage of the crop, the response factor group is weighted and summed to finally generate the resource response weight vector Q of the target farmland, which is used to characterize the priority allocation order of various agricultural resources under the current environmental conditions.

[0007] Preferably, generating a resource allocation priority sequence L that matches the current crop stage includes: The resource response weight vector Q is mapped item by item to each resource sensitivity coefficient in the crop stage growth model T to obtain the initial resource score set Sinit; wherein, the item by item mapping is performed by multiplying the resource weight of each resource item by the corresponding growth stage sensitivity to obtain the corresponding score value. The initial resource score set Sinit is normalized to obtain the normalized score set Snorm. Apply time smoothing and jitter suppression mechanisms to Snorm to obtain a stable score set Ssmooth, which includes taking a weighted moving average of the scores over the most recent w sampling times; Based on the stable score set Ssmooth, the resource items are sorted in descending order, and a resource allocation priority sequence L is generated according to the sorting result. The priority sequence lists the resources and their expected allocation levels in descending order of score.

[0008] Preferably, the resource dynamic allocation matrix M for constructing the target farmland includes: Using the resource allocation priority sequence L as the main sorting index, each type of resource unit in the resource availability matrix U is compared and filtered to determine the resource set U′ that has the conditions for being called within the current allocation cycle. The resource set U′ only contains resource units whose inventory is greater than the set minimum threshold Vmin. Based on allocation priority and resource availability, an allocation intensity value is calculated for each resource type in the resource set U′. The blending intensity value This is the normalized result of the product between the resource score and the inventory level. The allocation intensity value of each resource A set of plot numbers G is mapped to the target farmland, and a resource allocation sub-matrix Mg is generated by combining the crop type and historical allocation records of each plot, where each element represents the type and relative ratio of resources that the plot should receive in the current period. All sub-matrices of resource allocation for each land parcel are merged according to their parcel numbers to form a dynamic resource allocation matrix M.

[0009] Preferably, obtaining the spatial topological feature parameters C of the resource allocation units in the allocation matrix M includes: Based on the distribution of non-zero elements in the resource allocation matrix M, a resource allocation heat map H is constructed. The resource allocation heat map H is a two-dimensional spatial mapping map, and its coordinate axes represent the spatial location of the land parcel and the resource type, respectively. The heat value is the frequency of resource allocation per unit area. Clustering algorithms are applied to the resource allocation heatmap H to extract high-density resource areas and form spatial clusters Ccluster, which are used to identify the spatial concentration and regional boundaries of resource-intensive allocation units. Calculate parameters such as the location of the resource allocation center of gravity, average allocation density, and boundary complexity for each spatial cluster, and synthesize them to obtain a set of spatial topological characteristic parameters C, including allocation center offset, resource allocation concentration index, heterogeneity index, and boundary connectivity.

[0010] Preferably, the scheduling frequency and energy consumption strategy of IoT nodes within the target area are adjusted according to the spatial topology feature parameter C to generate an energy consumption control model Z, including: Based on the allocation intensity level, the target area is divided into three sensing areas: high-density, medium-density, and low-density, and different scheduling frequency levels are assigned to them. ,in ; For IoT nodes in high-density areas, an edge-aware enhancement mode is enabled, including local cache enhancement, redundant sensing fusion, and short-term prediction activation strategies; for nodes in low-density areas, a low-power standby mode is enabled, and energy consumption is reduced through a dynamic periodic wake-up mechanism. Using spatial topology feature parameter C, regional sensing density level, node historical scheduling frequency and node battery capacity as input variables, an energy consumption prediction function f(Z) is constructed. The energy consumption prediction function f(Z) adopts a weighted energy consumption regression model to calculate the estimated energy consumption per unit time under different strategy combinations; Based on a dual-objective optimization function that minimizes energy consumption and maximizes response capability, the optimal scheduling parameter set for each node in the current allocation cycle is generated. The optimal parameter set includes scheduling frequency, wake-up duration, transmission power level, and edge processing enabled status. The scheduling parameter set is written into the control register of each IoT node as the output execution configuration of the energy consumption control model Z.

[0011] Preferably, the various dynamic resource allocation tasks defined in the execution resource allocation matrix M include: The node operation parameter group Z={Fopt, Popt, Topt, Mode} output by the energy consumption control model Z is parsed into control instructions, where Fopt is the optimal scheduling frequency, Popt is the communication power level, Topt is the minimum wake-up duration of the node, and Mode is the operation mode flag. The control commands are sent to each IoT node in the target area. Under the control of the new parameters, the nodes start the scheduling logic and execute the corresponding resource control tasks according to the allocation level and allocation time window in the resource allocation matrix M, including irrigation, water and fertilizer application, supplemental lighting and pesticide spraying.

[0012] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention constructs a dynamic allocation method for agricultural resources based on IoT chips. By combining environmental parameter sets, crop growth models, resource response weight vectors, and spatial topology analysis, it realizes a closed-loop adaptive management mechanism for agricultural resources from "sensing—analysis—regulation—feedback". In particular, the introduction of a spatial topology parameter-driven energy consumption control model Z significantly improves the energy efficiency control capability of IoT nodes in multi-plot, high-frequency allocation scenarios, avoiding problems such as resource mismatch, response lag, and node overload existing in traditional centralized allocation systems.

[0013] 2. This invention employs a modular distributed sensing and edge scheduling strategy, supporting personalized execution of resource allocation tasks by region, crop type, and growth stage. This not only improves the efficiency of agricultural resource utilization but also reduces system operation and maintenance costs. The overall system boasts advantages such as high scalability, high stability, and low energy consumption, making it suitable for various application scenarios including precision agriculture, smart greenhouses, and arid zone agriculture. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0015] Figure 1This is a flowchart of the method of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] For examples, please refer to Figure 1 As shown in this embodiment, the method for dynamic allocation of agricultural resources based on the Internet of Things includes: By deploying IoT nodes in the target farmland, a set of environmental parameters P is collected, including soil moisture, temperature, light intensity, and plant electrical conductivity. Based on the set of environmental parameters P, and combined with the preset small-scale meteorological factor perturbation function R, a resource response weight vector Q for the target farmland is constructed to characterize the current microclimate unit's dependence on various agricultural resources. Based on the resource response weight vector Q and combined with the agricultural crop stage growth model T, a resource allocation priority sequence L matching the current crop stage is generated. The resource allocation priority sequence L is mapped to the existing agricultural resource availability matrix U to construct the dynamic resource allocation matrix M of the target farmland; Obtain the spatial topology feature parameters C of the resource allocation unit in the allocation matrix M, and adjust the scheduling frequency and energy consumption strategy of IoT nodes in the target area according to the spatial topology feature parameters C to generate an energy consumption control model Z. Based on the energy consumption control model Z, the operating parameters of the IoT nodes are adjusted, and various dynamic resource allocation tasks defined in the resource allocation matrix M are executed.

[0018] In this invention, in response to the dynamic resource allocation needs of the target farmland, multiple IoT chip nodes are first deployed in the farmland area. Each node integrates a sensing unit, a signal acquisition unit, an edge processing unit, and a wireless communication unit. The IoT chip adopts a low-power heterogeneous architecture, which can adapt to the continuous operation requirements of complex outdoor environments.

[0019] The sensing unit is used to collect multiple environmental parameters inside the farmland, including: Soil moisture: Real-time monitoring of soil moisture content using a capacitive soil moisture sensor to reflect the water supply status of crop roots; Ambient temperature: Monitoring the current climate temperature through digital temperature sensors helps determine farmland evaporation and crop metabolic rhythms; Light intensity: The light intensity level is detected by using a photoresistor or photodiode array to provide a reference for crop photosynthesis assessment and irrigation timing; Plant electrical conductivity: The water conduction within the plant is detected by contact or non-contact electrical conductivity sensors to indirectly characterize the physiological health of the crop.

[0020] After the above parameters are collected, they are first processed by the chip's built-in edge processing module for preliminary filtering, noise reduction, and feature extraction to form a structured dataset. n is the total number of data points, where each item This represents a sample of heterogeneous environmental parameters collected at a certain moment.

[0021] The collected data set P is then transmitted to the gateway node or the upper-level scheduling system via a wireless communication unit. This method enables distributed sensing and preliminary processing of multi-dimensional ecological parameters of farmland without relying on a central server.

[0022] All sensor data in parameter set P are synchronized by timestamp, i.e., a complete data sample matrix is ​​constructed with a uniform sampling period (e.g., once every 5 minutes), with each row representing complete sample data at a single time point. To eliminate redundancy and dimensionality effects between different parameters, principal component analysis (PCA) is used to reduce the dimensionality of the environmental data.

[0023] The specific process of PCA includes the following steps: The original parameter set P is normalized so that all data satisfy a standard normal distribution with zero mean and unit variance; a covariance matrix is ​​constructed to analyze the correlation between different parameters; the eigenvectors and eigenvalues ​​of the covariance matrix are calculated and arranged in descending order of eigenvalue size; the top k principal components with a cumulative contribution rate of over 90% are selected as the dimensionality-reduced environmental parameter vector P′.

[0024] To capture the impact of farmland microclimate on the supply and demand of agricultural resources, this invention predefines a small-scale meteorological factor perturbation function R. This function is used to model the short-term perturbation effects of factors such as rainfall, wind speed, and local radiation, and its structure is as follows: ,in: The instantaneous wind speed disturbance function is defined as the rate of change of the maximum wind speed over the past hour. The short-term rainfall intensity function is defined as the rate of change of rainfall per unit area over the past 30 minutes. This represents the local solar radiation perturbation function, based on the frequency and intensity of light intensity fluctuations.

[0025] Each perturbation function All perturbation functions are standardized, with values ​​ranging from 0 to 1. This perturbation function set R is trained and refined using historical data from a local agricultural meteorological database. To adapt to different regions, seasons, and crop growth cycles, the weight parameters in the perturbation functions are dynamically updated using a Long Short-Term Memory (LSTM) network model.

[0026] The steps for building an LSTM network are as follows: The input sample is the time series of environmental disturbance factors over the past m days (e.g., the past 30 days); the network structure is set as follows: input layer (dimension 1). to The network consists of two hidden state gating units and an output layer. The training objective is to predict the trend of perturbation factor changes in the next hour and update the weighting coefficients of the current perturbation function in real time. The model training uses the mean squared error (MSE) loss function and an adaptive learning rate optimization algorithm.

[0027] The dimensionality-reduced parameter vector P′ is fused with the perturbation function set R to construct the environmental response matrix E. Specifically, an adaptive regularized regression algorithm (Elastic Net Regression) is used, with the parameter vector P′ as the independent variable and the output values ​​of each function in the perturbation function R as the dependent variable, to establish a multi-factor response relationship.

[0028] This algorithm integrates L1 norm (Lasso) and L2 norm (Ridge) regularization constraints. The specific steps are as follows: Construct a regression objective function, adding L1 and L2 regularization terms to control variable sparsity and stability; fit the parameter weight coefficients using training samples. Output response matrix E, where each item represents the degree of response of a certain environmental parameter to a certain disturbance factor.

[0029] Next, based on the resource types required by the crop at its current growth stage (such as seedling stage, heading stage, grain-filling stage, etc.), a preset crop stage-specific resource demand model T is invoked. Model T includes the following typical structure: water sensitivity factor. Nutrient requirement weight Photosynthesis sensitivity Hormone regulation response threshold The response matrix E and the model T are weighted and fused to finally output the resource response weight vector. ,in: This indicates the priority weight of water resource allocation under the current environment; Indicates the priority of nutrient application; Indicates the weight of illumination compensation demand; This indicates the priority of growth-promoting factors (such as auxin and gibberellin) application. The weight vector Q has a value range of 0 to 1 and satisfies the normalization condition that the sum is 1, which can serve as a direct control reference for agricultural resource scheduling systems.

[0030] To adapt to the differences in land parcels across different micro-regions, this invention also introduces a land parcel label parameter G, which is constructed based on parameters such as soil texture, crop type, and historical climate adaptability. The label parameter G is then fused with the response weight vector Q to achieve differentiated allocation of agricultural resources.

[0031] For example, if the land parcel label G indicates that the target site is located in a dry sandy soil area, it will be enhanced in the weight vector. Prioritize (water resources) while appropriately reducing [their use]. Distribution of (fertilizer).

[0032] Resource Response Weight Vector Each element represents the priority weight of water, nutrients, light, and hormones in the current environment and crop condition. All weight values ​​are normalized to the range of 0 to 1, and the sum is 1.

[0033] The crop stage growth model T is used to characterize the relative sensitivity of crops to various resources at different growth stages. The structure of model T is defined as follows: ;in, These are the stage-specific sensitivity coefficients for water, nutrients, light, and hormones, ranging from 0 to 1, indicating the importance of that resource to crop growth at the current crop stage. For example, during the heading stage, the water sensitivity coefficient... High (e.g., 0.9), while light sensitivity Relatively low (e.g., 0.4).

[0034] The model T can be built based on the experience base of an agricultural expert system, or it can be obtained through regression learning using historical crop yield data. Each crop type in the model is equipped with multiple stage labels (such as seedling stage, tillering stage, jointing stage, heading stage, grain-filling stage, etc.) and bound to the corresponding α coefficient set.

[0035] The resource response weight vector Q is mapped item by item to the sensitivity coefficients in model T, that is, the actual demand scores of various resources at the current crop stage are calculated to form the resource score set Sinit: ;in, For example, when =0.7, When =0.9, =0.63, indicating that water resources are relatively important in the current context. This product operation numerically couples the environmental driver Q with the biological demand model T.

[0036] To avoid error accumulation caused by absolute scores, Sinit needs to be normalized: Snorm = Normalize(Sinit); the normalization method uses max-min linear normalization, that is, for each element in the set... Use the following formula to convert: Where smin and smax are the minimum and maximum values ​​in Sinit, respectively, and the range of the score set Snorm is unified to 0 to 1 after transformation.

[0037] Subsequently, to mitigate the impact of instantaneous fluctuations and outliers on the ranking results, a time-sliding window mechanism is introduced to smooth the Snorm. A weighted moving average method is used to process data across w consecutive sampling periods, with a recommended window width w of 3 to 5. The weight allocation follows an exponential decay pattern; for example, the score calculation expression for the current time t is: Where λ is the attenuation coefficient, with a value between 0.6 and 0.8. The processing result forms a stationary score set Ssmooth.

[0038] Sort the smooth score set Ssmooth in descending order to generate a resource allocation priority sequence L. The priority sequence L is an ordered arrangement of resource items, with the following structure: ;in This indicates the resource type with the highest score, followed by resources with progressively lower priority.

[0039] Based on the allocation priority sequence L, the various types of resources in the agricultural resource availability matrix U are screened in turn, and resources whose current inventory is less than the minimum deployable threshold Vmin are excluded.

[0040] The principles for setting the minimum threshold Vmin are as follows: For resource types Its Vmin = average daily consumption × safety factor, with a preferred safety factor of 1.5. The average daily consumption can be inferred from the site's sensor data and historical records. For example, if the site... The daily water consumption over the past three days was 200, 220, and 210 liters, respectively, so the average is 210 liters. Therefore, Vmin is set to 210 × 1.5 = 315 liters. Resources satisfying the Vmin condition constitute the effective resource set U′, which then proceeds to the next stage of allocation intensity calculation.

[0041] For each valid resource Based on its sorting position in L and its current inventory level, calculate its allocation intensity value. This is used to measure the priority and execution intensity of the resource in actual allocation. The specific formula is: ;in: For resources The priority score in L is preferably set as a linear value mapped in reverse rank. For example, if L is 4, the first rank gets 1.0, the second rank gets 0.75, the third rank gets 0.5, and the fourth rank gets 0.25. This represents the current available inventory value, with units set according to resource type (e.g., liters, kilograms, watt-hours, etc.). All After summarizing the values, perform a normalization operation: ; Furthermore, to adapt to emergency resource allocation scenarios, this invention introduces a resource scarcity correction factor η. When the current inventory of a resource is lower than 80% of its short-term forecasted demand (in the next period), Denorm is multiplied by η (ranging from 0.5 to 0.9) to reduce its allocation priority and prevent the system from becoming overly reliant on scarce resources.

[0042] Based on the set of plot numbers G and the crop type label C, spatial allocation mapping is performed on each type of resource in Dnorm to generate a resource allocation submatrix Mg for each plot. The structure of each submatrix is ​​as follows: the row index is the resource type; the column index is the allocation parameter, such as the target application amount, application start time, duration, etc.; the element value is the allocation ratio or intensity, indicating the application priority of the resource in that plot.

[0043] The submatrix generation process includes the following logic: referencing the allocation intensity value Dnormᵢ, an initial allocation ratio is generated for each plot; if the crop type of the plot is not sensitive to a certain resource (e.g., wheat has a weak response to light compensation), the corresponding ratio is multiplied by a weakening factor γ (γ∈[0.3,0.7]); if the plot is located in a resource-sensitive area (e.g., sandy soil is sensitive to water), the allocation ratio of that resource is increased; if a plot has historically been over-allocated (calculated from sensor historical records), an upper limit is set for the current allocation ratio and it is marked as "low priority".

[0044] All sub-matrices Mg of the land parcels are merged according to the land parcel number G to form the complete resource allocation matrix M. Matrix M is a two-dimensional array, where: rows represent land parcel numbers ( to ); the column represents the resource type ( to ); each element This represents the execution level or expected delivery volume (unit depends on resource type) of the j-th type of resource allocated to the i-th plot in the current period. To facilitate reading by the execution control system, all values ​​in M ​​will be rounded to the smallest control unit ΔR supported by the resource execution layer. For example, if the minimum delivery unit for a water pump is 10 liters, then the value in the water resource column of M will be an integer multiple of 10.

[0045] In addition, to prevent allocation overload, the matrix must meet the following boundary constraints: the number of non-zero elements in each row does not exceed Tmax (maximum number of resource interventions), preferably 3; the total amount of resource allocation in each column does not exceed Umax (maximum available allocation limit), and Umax is set according to the system load capacity; a minimum interval Δt is allowed between land allocation tasks to prevent the risk of conflict caused by continuous resource application to the same land.

[0046] The completed resource allocation matrix M will be transmitted to the IoT agricultural control system to guide actuators such as irrigation controllers, fertilizer applicators, supplemental lighting systems, and automatic spraying equipment in carrying out corresponding resource distribution tasks. After the allocation process is executed, sensor nodes will monitor and provide feedback on the results in real time, recording environmental change data before and after allocation for model iteration and optimization before the next allocation.

[0047] This implementation method is based on a resource allocation matrix M, which is defined as a two-dimensional structure, where rows represent target plot numbers G, columns represent resource types R, and matrix elements... Representing resources On the plot The level or intensity of the allocation.

[0048] To obtain the spatial distribution characteristics of this allocation behavior, matrix M needs to be mapped to a spatial data model. The specific steps are as follows: The non-zero elements in matrix M are mapped to the actual spatial coordinates of farmland to form a resource allocation heatmap H. The horizontal axis represents the two-dimensional coordinates of the plots (such as the grid location information corresponding to GPS or local identification numbers); the vertical axis represents the resource type number; the value of each heatmap cell is the allocation frequency per unit area of ​​a plot in the current period, or the allocation intensity weighted value (such as the number of allocations per hour × intensity level). The heatmap uses color depth to represent allocation density, with areas of high density values ​​appearing as "hot spots". The density-based spatial clustering algorithm DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is applied to the heatmap H to discover concentrated areas of resource allocation. This algorithm does not require specifying the number of clusters and is suitable for handling irregularly distributed data. Its core parameters are defined as follows: ε: Neighborhood radius, used to determine whether two points are "close neighbors", with a preferred value range of 1 to 5 (unit: plot spacing or meters). MinPts: Minimum number of points, used to determine whether a point is a core point. It is recommended to set it to 3 to 10, and adjust it according to the scene density.

[0049] After DBSCAN clustering, multiple spatial resource clusters are formed, and each cluster represents a resource allocation intensive area.

[0050] For each spatial cluster, the following topological parameters are calculated to form a set of spatial topological characteristic parameters C: adjustment of the centroid offset. : Indicates the offset distance of the current cycle's resource allocation focus relative to the previous cycle, in meters; resource concentration. Defined as the ratio of high-density area to total farmland area; regional heterogeneity index. The difference in allocation intensity across regions is measured using the normalized standard deviation; boundary connectivity. : Measures the degree of boundary contact between spatial clusters, used to identify transitional regions in resource allocation. Set of topological feature parameters. This serves as the basis for adjusting energy consumption strategies.

[0051] After obtaining the parameter set C, the sensing scheduling frequency and operating mode of IoT nodes in the farmland are dynamically adjusted based on their spatial distribution characteristics. The specific adjustment strategy is as follows: Based on allocation density and spatial topology, farmland areas are divided into three sensing zones: High-Density Zone (HDZ): high allocation frequency, high concentration, and strong sensing demand; Medium-Density Zone (MDZ): relatively uniform allocation and moderate resource consumption; Low-Density Zone (LDZ): sparse allocation and low sensing priority. The classification method is as follows: if the allocation density of a certain area is more than 1.5 times the average of the entire area, it is defined as HDZ; if it is less than 0.5 times, it is defined as LDZ; and the rest are MDZ.

[0052] To reduce communication load and energy consumption, different node scheduling frequencies are set in different areas: HDZ area scheduling frequency. Scheduling occurs every 5 minutes; MDZ area scheduling frequency. Scheduling occurs every 15 minutes; LDZ area scheduling frequency. Scheduling occurs every 30 minutes. Scheduling activities include sensor data acquisition, edge processing tasks, wireless communication, and wake-up logic.

[0053] HDZ nodes enable "edge enhancement mode": local data caching, short-term prediction algorithms (such as sliding window averaging or lightweight LSTM), and redundant sensing fusion strategies (such as multi-sensor comparison and trust weighting); LDZ nodes enable "low-power standby mode": default to sleep mode, with the wake-up cycle dynamically adjusted according to resource allocation changes, initially set to wake up once every 30 to 60 minutes, and supporting "change-triggered wake-up" (such as waking up when temperature and humidity exceed a set threshold); MDZ nodes maintain normal operating frequency and do not require additional mode switching.

[0054] To achieve global energy efficiency optimization in strategy execution, this invention designs a space parameter-driven energy consumption control model Z, which is used to generate optimal operating parameter configurations for each IoT node.

[0055] The input variables of the energy consumption control model Z include: the set of spatial topology characteristic parameters C; the region classification of the node (HDZ, MDZ, LDZ); the historical scheduling frequency of the node Fh; the current remaining power of the node E; and the resource allocation density level D.

[0056] The energy consumption control function f(Z) is used to predict the energy consumption of a node per unit time under different operating strategies: Where: F is the current scheduling frequency; P is the communication power consumption; E is the node's current remaining power. , representing the sensitivity to scheduling, communication, and power consumption. The optimal weight values ​​are: , , .

[0057] A dual objective function is constructed: minimize energy consumption per unit time; maximize the dispatch task response coverage (i.e., the coverage duration and perception accuracy of nodes in important dispatch areas). Heuristic optimization algorithms (such as genetic algorithms and particle swarm optimization) are used to search for the optimal configuration combination in the policy space, outputting the parameter set: Z={Fopt,Popt,Topt,Mode}; Fopt: optimal scheduling frequency; Popt: optimal communication power; Topt: optimal wake-up duration; Mode: operating mode (enhanced / normal / standby).

[0058] The parameter set Z is downloaded to the control register of the IoT node to drive the hardware system to run according to the new parameters; the model is refreshed once every adjustment cycle or recalculated under major environmental events (such as sudden rainfall or drought warnings).

[0059] To enable remote control of multiple distributed IoT nodes, this implementation adopts a configuration bus mechanism centered on the IoT edge gateway, which sends control commands to nodes in the target area in a hierarchical manner.

[0060] After receiving the instruction, the node automatically verifies its integrity, parses the control field, writes the parameter value to the control register, and updates the running status in real time.

[0061] After the node completes the parameter update, it immediately activates the local scheduling logic module and executes various resource scheduling tasks according to the following process: The node starts an internal scheduling timer and wakes up at the scheduled frequency according to the Fopt setting. For example, if it is set to wake up once every 10 minutes, it will wake up once every 600 seconds to execute periodic tasks.

[0062] The node compares the current system time with the "resource-time window mapping table" defined in the resource allocation matrix M, and executes the corresponding resource task only when it is currently in the active period of the allocation task.

[0063] The node will call the corresponding execution module according to the allocation level and resource type defined in M ​​to complete the following tasks: Irrigation task: activate the micro solenoid valve or drip irrigation controller, and set the irrigation duration and flow rate according to the allocation level; Water and fertilizer application task: control the fertilizer pump to start, adjust the proportion valve, and accurately mix and apply fertilizer according to crop needs; Supplemental lighting task: turn on the supplemental lighting LED module, adjust the light intensity to the preset level, and the continuous irradiation time is controlled by M; Spraying task: drive the micro sprayer, set the spraying cycle and area range, and clear the pipeline residue after the task is completed.

[0064] The node remains active during task execution, with the minimum stay time controlled by Topt. If the task is not completed but Topt reaches the threshold, it enters a protection state, retaining only the task thread while the rest of the subsystems switch to low-power mode to prevent excessive power consumption.

[0065] After completing each round of tasks, the node will generate the following feedback data packets: task completion status (success / failure); actual resource deployment amount (unit resource); node current power E (%); communication latency / signal quality indicators (RSSI, SNR), etc.

[0066] Feedback data is transmitted back to the edge control gateway via a low-power link and synchronized to the cloud resource management platform to update the resource allocation matrix M or update the parameter training samples of the energy consumption control model Z.

[0067] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An agricultural resource dynamic allocation method based on the Internet of Things, characterized in that: The method comprises the following steps: Collecting a set of environmental parameters P including soil humidity, temperature, light intensity, and plant conductivity through Internet of Things nodes deployed in the target farmland; Based on the set of environmental parameters P, a resource response weight vector Q of the target farmland is constructed in combination with a preset small-scale meteorological factor disturbance function R; According to the resource response weight vector Q, a resource allocation priority sequence L matching the current crop stage is generated in combination with an agricultural crop stage growth model T; The resource allocation priority sequence L is mapped with an existing agricultural resource available matrix U to construct a resource dynamic allocation matrix M of the target farmland; The spatial topological feature parameters C of the resource allocation unit in the allocation matrix M are obtained, and the scheduling frequency and energy consumption strategy of the Internet of Things nodes in the target area are adjusted according to the spatial topological feature parameters C to generate an energy consumption control model Z; Based on the energy consumption control model Z, the operating parameters of the Internet of Things nodes are adjusted, and various types of resource dynamic allocation tasks defined in the resource allocation matrix M are executed. 2.The method according to claim 1, wherein: The construction of the resource response weight vector Q of the target farmland comprises: Synchronize the multi-dimensional perception data in the set of environmental parameters P according to the collection time stamp, and perform dimension reduction processing on the multi-dimensional perception data by using the principal component analysis method to obtain the parameter vector P'; The parameter vector P' is coupled with the preset small-scale meteorological factor disturbance function R to build a model, the disturbance function R includes disturbance weights for wind speed, instantaneous precipitation, and local radiation intensity, and a regularized regression algorithm is used to build an environmental response matrix E; According to the environmental response matrix E, a response factor group affecting resource allocation is extracted, and based on a resource demand model of the current growth stage of crops, the response factor group is weighted and summed to finally generate a resource response weight vector Q of the target farmland, which is used to represent the priority allocation order of various agricultural resources under the current environmental conditions. 3.The IoT-based dynamic allocation method of agricultural resources according to claim 1, characterized in that: The generation of the resource allocation priority sequence L matching the current crop stage comprises: Mapping the resource response weight vector Q and each resource sensitivity coefficient in the crop stage growth model T item by item to obtain an initial resource score set Sinit; wherein the item-by-item mapping adopts the method of multiplying the resource weight by the corresponding growth stage sensitivity to obtain the corresponding score value for each resource item; The initial resource score set Sinit is normalized to obtain a normalized score set Snorm; Apply time smoothing and jitter suppression mechanism to Snorm to obtain a smooth score set Ssmooth, including taking a weighted moving average of the scores of the last w sampling time points; Based on the smooth score set Ssmooth, the resource items are arranged in descending order, and a resource allocation priority sequence L is generated according to the sorting result, wherein the priority sequence lists the resources and their expected allocation levels in the order of high to low scores. 4.The method of claim 1, wherein: The construction of the resource dynamic allocation matrix M of the target farmland comprises: The resource allocation priority sequence L is taken as a main sorting index, each type of resource unit in the resource available matrix U is compared and screened item by item, and a resource set U' with a callable condition in the current allocation period is determined, the resource set U' only contains resource units with an inventory greater than a set minimum threshold Vmin; Based on the deployment priority and resource availability, a deployment strength value is calculated for each resource type in the resource set U' , which is a normalized product of the resource score and inventory level . The allocation intensity value of each resource A set of plot numbers G is mapped to the target farmland, and a resource allocation sub-matrix Mg is generated by combining the crop type and historical allocation records of each plot, where each element represents the type and relative ratio of resources that the plot should receive in the current period. All sub-matrices Mg of the divided land resources are merged according to the land block numbers to form a resource dynamic allocation matrix M. 5.The IoT-based dynamic allocation of agricultural resources method according to claim 1, characterized in that: The spatial topological feature parameters C of the resource allocation units in the allocation matrix M are obtained, including: Based on the distribution of non-zero elements in the resource allocation matrix M, a resource allocation heat map H is constructed, which is a two-dimensional space mapping graph, the coordinate axes represent the land space position and the resource type respectively, and the heat value is the resource allocation frequency per unit area; A clustering algorithm is applied to the resource allocation heat map H to extract a high-density resource area to form a spatial cluster Ccluster, which is used to identify the spatial concentration degree and area boundary of the resource-intensive allocation unit; The spatial topological feature parameter set C is obtained by calculating the resource allocation center position, average allocation density and boundary complexity of each spatial cluster, including the allocation center offset degree, resource allocation concentration degree index, heterogeneity index and boundary connectivity. 6.The IoT-based dynamic allocation method of agricultural resources according to claim 5, characterized in that: According to the spatial topological feature parameters C, the scheduling frequency and energy consumption strategy of the Internet of Things nodes in the target area are adjusted to generate an energy consumption control model Z, including: According to the deployment heat level, the target area is divided into three types of sensing areas, i.e. high-density area, medium-density area and low-density area, and different scheduling frequency levels are allocated wherein ; The Internet of Things nodes in the high-density area enable the edge perception enhancement mode, including local cache enhancement, redundant perception fusion and short-time prediction enabling strategy; the nodes in the low-density area enable the low-power standby mode to reduce energy consumption through a dynamic period wake-up mechanism; The spatial topological feature parameters C, the regional perception density level, the historical scheduling frequency of the nodes and the node battery remaining amount are taken as input variables to construct an energy consumption prediction function f(Z); The energy consumption prediction function f(Z) adopts a weighted energy consumption regression model to calculate the unit time energy consumption evaluation under different strategy combinations; Based on the double-objective optimization function of minimum energy consumption and maximum response capability, the optimal scheduling parameter set of each node in the current allocation period is generated, including the scheduling frequency, wake-up duration, transmission power level and edge processing start state; The scheduling parameter set is written into the control register of each Internet of Things node as the output execution configuration of the energy consumption control model Z. 7.The IoT-based dynamic allocation method of agricultural resources according to claim 6, characterized in that: The execution of each type of resource dynamic allocation task defined in the resource allocation matrix M includes: The node running parameter set Z output by the energy consumption control model Z is parsed into control instructions, where Fopt is the optimal scheduling frequency, Popt is the communication power level, Topt is the minimum wake-up duration of the node, and Mode is the running mode identification bit; The control instructions are sent to each Internet of Things node in the target area, and the node starts the scheduling logic under the control of the new parameters, and executes the corresponding resource control tasks according to the allocation level and allocation time window in the resource allocation matrix M, including irrigation, water and fertilizer application, light supplement and pesticide spraying tasks.

Images