A Crop Water Requirement Forecasting Method and System Based on Agricultural Internet of Things

Abstract

This invention relates to the field of water demand prediction technology, and particularly to a method and system for predicting crop water demand based on agricultural IoT. It is applied to precise water management in complex soil structures such as terraced fields. The invention proposes the following scheme: by collecting soil moisture state data at different depths, a water retention time matrix is ​​generated. Combined with dynamic weights of interlayer water migration, the water absorption matrix is ​​updated in real time. Through time-series sliding and cosine similarity calculations, the water retention time matrix is ​​dynamically corrected to ensure accurate soil moisture allocation. A graph theory algorithm is introduced to construct a directed graph model of water flow paths, and a binary mask is used to generate an interference matrix to optimize the matching of water absorption and water retention time, improving irrigation efficiency and water resource utilization. This application can dynamically adjust irrigation strategies according to the moisture state between soil layers, avoiding water waste and improving the accuracy of water supply for crop growth.

Description

Technical Field

[0001] This invention relates to the field of water demand forecasting technology, and in particular to a method and system for crop water demand forecasting based on agricultural Internet of Things. Background Technology

[0002] In agricultural irrigation management, precise water allocation is crucial for healthy crop growth. Traditional irrigation methods typically use fixed irrigation volumes, neglecting the dynamic changes in soil moisture across different layers, leading to uneven water distribution and even water waste or crop water shortages. This is especially true in complex farmland environments, such as terraced fields with multi-layered soil structures, where the migration of water from upper to lower layers is more complex, and differences in water retention characteristics and absorption capacity significantly impact irrigation effectiveness. Existing technologies largely rely on static water models, failing to dynamically adapt to changes in soil moisture and crop growth needs. Therefore, how to accurately adjust water allocation in real-time under dynamic conditions has become a key issue in improving irrigation efficiency, conserving water resources, and ensuring healthy crop growth.

[0003] For example, patent application CN110648020A provides a method and apparatus for predicting the water requirement of greenhouse crops. It inputs collected greenhouse microclimate environmental parameter data and crop physiological parameter data into a greenhouse crop water requirement prediction model, and outputs the corresponding greenhouse crop water requirement based on the greenhouse microclimate environmental parameter data and crop physiological parameter data, thereby predicting the greenhouse crop water requirement. The greenhouse crop water requirement prediction model is trained using greenhouse microclimate environmental parameter data and crop physiological parameter data labeled with greenhouse crop water requirement. This technical solution pre-trains the greenhouse crop water requirement prediction model using greenhouse microclimate environmental parameter data and crop physiological parameter data labeled with greenhouse crop water requirement, and then uses the microclimate environmental parameters and crop physiological parameters as input parameters to output the predicted greenhouse crop water requirement, thus knowing how much water the crop has actually lost, achieving the purpose of precision irrigation.

[0004] All of the above patents suffer from the problem described in the background: they ignore the dynamic changes in soil moisture at different levels, leading to uneven water distribution. To solve this problem, this application designs a crop water requirement prediction method and system based on agricultural Internet of Things. Summary of the Invention

[0005] The technical problem this invention addresses is the shortcomings of existing technologies. It provides a method and system for predicting crop water requirements based on the Internet of Things (IoT) in agriculture. By collecting soil moisture state data at different depths, a water retention time matrix is ​​generated. Combined with dynamic weights for interlayer water migration, the water absorption matrix is ​​updated in real time. Through time-series sliding and cosine similarity calculations, the water retention time matrix is ​​dynamically corrected to ensure accurate soil moisture allocation. A graph theory algorithm is introduced to construct a directed graph model of water flow paths, and a binary mask is used to generate an interference matrix, optimizing the matching of water absorption and water retention time, thereby improving water resource utilization. The system can dynamically adjust water use strategies based on the moisture state between soil layers, avoiding water waste and improving the accuracy of water supply for crop growth.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A crop water requirement prediction method based on agricultural IoT is applied to any layer of farmland in a terraced field with sequentially ordered heights. Each farmland plot is equipped with a sensing terminal that is communicatively connected to a processing terminal. The sensing terminal is equipped with a soil moisture sensor, which collects moisture status data at different depths. The method includes:

[0008] Collect moisture status data for each layer of farmland and generate a moisture retention time matrix;

[0009] Starting from the bottom layer, iterative operations are performed. The current water retention time matrix of the current layer, the current water retention time matrix of the previous layer, and the water absorption matrix of the current layer are combined to update the current water retention time matrix of the previous layer until the current layer is the top layer. The water absorption matrix is ​​calculated by a dynamic siphon model driven by the water potential difference between adjacent layers.

[0010] The irrigation amount for each farmland plot is determined based on the current water retention time matrix of each plot to implement smart irrigation.

[0011] The process of collecting moisture status data for each layer of farmland and generating a moisture retention time matrix includes:

[0012] Moisture state data at each depth are subjected to time-series standardization to form a time-series dataset, which includes the continuous moisture content at the corresponding depth within the monitoring period.

[0013] The rate of change of water content at adjacent depths within the same layer is calculated based on the time series dataset, and a water gradient change matrix is ​​constructed.

[0014] Based on the preset water retention time benchmark model, the deviation between the time series dataset and the water retention time benchmark model is calculated, and a water retention time deviation matrix is ​​generated.

[0015] The moisture gradient change matrix and the moisture retention time deviation matrix are aligned according to the time and depth dimensions to generate a moisture retention time matrix.

[0016] The step of dimensionally aligning the moisture gradient change matrix and the moisture retention time deviation matrix according to the time and depth dimensions includes:

[0017] The gradient feature values ​​and deviation feature values ​​corresponding to the same time point and the same depth layer are fused to generate a water retention time matrix. The row dimension of the water retention time matrix is ​​the monitoring time point, the column dimension is the depth layer, and the channel dimension includes gradient change features and deviation features.

[0018] The water absorption matrix is ​​calculated using a dynamic siphon model driven by the potential difference between adjacent layers, including:

[0019] The composite potential energy difference is calculated based on the difference in average soil moisture content and vertical height between the current layer and the previous layer, wherein the composite potential energy difference includes the vector superposition of matrix potential and gravitational potential.

[0020] Calculate the inter-layer siphon admittance coefficient based on the crop root density distribution and soil viscosity coefficient of the previous layer;

[0021] The composite potential difference and the inter-level siphon admittance coefficient are coupled to calculate the matrix elements of the water absorption matrix and generate the initial water absorption matrix.

[0022] The initial water absorption matrix is ​​corrected by quantifying the water retention time matrix of the current layer and the current water retention time matrix of the previous layer, thus generating a water absorption matrix.

[0023] The step of correcting the initial water absorption matrix to generate a water absorption matrix includes:

[0024] The difference correction factor is calculated based on the cosine similarity between the current layer's water retention time matrix and the previous layer's current water retention time matrix.

[0025] An interlayer water transport model is established using the Bernoulli equation. The difference correction factor is input into the interlayer water transport model, and the absorption coefficient of the current layer to the previous layer is output.

[0026] The initial water absorption matrix is ​​weighted and assigned values ​​based on the absorption coefficient to generate a water absorption matrix.

[0027] The updating of the current water retention time matrix of the previous layer includes:

[0028] The water retention time matrix of the current layer and the previous layer is time-slid through a time window. The dynamic weight matrix of water migration between layers is calculated within the time window. The dynamic weight is determined based on the mutual information entropy of the water gradient change characteristics and retention time deviation characteristics of the upper and lower layers during the same period.

[0029] The dynamic weight matrix and the water absorption matrix are subjected to tensor product operation to generate a composite coefficient matrix for interlayer water transport.

[0030] The current water retention time matrix of the previous layer is updated element by element based on the composite coefficient matrix.

[0031] The current water retention time matrix of the previous layer is updated element-wise based on the composite coefficient matrix, including:

[0032] The product of the upper-layer matrix elements and the corresponding elements of the composite coefficient matrix is ​​added to the compensation amount of the current-layer matrix elements after root water absorption correction, to form the updated upper-layer water retention time matrix.

[0033] The method further includes:

[0034] During irrigation, real-time water status data for each layer of farmland is collected to generate and update the water retention time matrix.

[0035] Based on the updated water retention time matrix, the irrigation strategy for the corresponding farmland plots is adjusted.

[0036] The generation and updating of the water retention time matrix includes:

[0037] When the water content of any depth layer in a farmland plot is detected to be greater than a preset saturation threshold and the hydraulic gradient of the surface runoff path is greater than a preset critical value, it is determined that the layer forms a temporary runoff channel.

[0038] A directed graph model of a temporary runoff channel is constructed using graph theory algorithms. The model uses field ridge fissures and furrow depressions as nodes and real-time flow velocity as edge weights.

[0039] Based on the directed graph model, an interference matrix is ​​generated using a binary mask, where a 1 in the binary mask indicates the absence of temporary runoff channel interference, and a 0 indicates the presence of temporary runoff channel interference.

[0040] The corresponding row in the water absorption matrix is ​​corrected based on the interference matrix to generate an updated water absorption matrix;

[0041] An updated water retention time matrix is ​​generated based on the updated water absorption matrix.

[0042] A crop water requirement prediction system based on agricultural IoT, comprising a sensor acquisition module, a matrix generation module, a hierarchical iteration module, and a water requirement prediction module, wherein:

[0043] The sensor acquisition module is deployed in each layer of farmland and includes a multi-depth soil moisture sensor array, a pore water pressure sensor and a surface runoff monitoring unit, used to collect moisture state data, interlayer hydraulic gradient data and surface runoff characteristics at different depths.

[0044] The matrix generation module is connected to the sensing and acquisition module, and performs the following operations:

[0045] Time-series standardization processing was performed on multi-depth moisture data to construct a moisture gradient change matrix and a residence time deviation matrix.

[0046] The correlation strength between the simultaneous moisture gradient between upper and lower layers and the retention deviation is calculated using the mutual information entropy algorithm, and a dynamic weight matrix is ​​generated.

[0047] Temporary runoff channels are modeled using graph theory algorithms to generate binary mask interference matrices.

[0048] The hierarchical iteration module is used to generate an initial water absorption matrix driven by a dynamic siphon model based on the composite potential energy difference and root admittance coefficient, fuse the interlayer dynamic weight matrix through tensor product operation, and iteratively correct the upper layer water retention time matrix using an element-by-element update rule.

[0049] The water demand prediction module is used to calculate the water demand based on the feature values ​​corresponding to each layer depth in the updated water retention time matrix.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] This invention monitors soil moisture status in real time and dynamically adjusts water demand based on the water absorption capacity and migration characteristics of each soil layer, effectively avoiding uneven water distribution and water waste. By introducing a difference correction factor and a water flow path interference matrix, it can finely adjust the water migration process in complex terrain, ensuring that crops receive sufficient water supply at different growth stages. Attached Figure Description

[0052] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0053] Figure 1 This is a flowchart illustrating the crop water requirement prediction method based on agricultural Internet of Things in Embodiment 1 of the present invention.

[0054] Figure 2This is a flowchart illustrating the calculation of the water retention time matrix in Embodiment 1 of the present invention.

[0055] Figure 3 This is a flowchart of the water absorption matrix calculation in Embodiment 1 of the present invention;

[0056] Figure 4 This is a flowchart of the iterative update process in Embodiment 1 of the present invention. Detailed Implementation

[0057] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0058] Example 1

[0059] Please see Figure 1 This invention provides an embodiment of a crop water requirement prediction method based on agricultural Internet of Things (IoT). This method is applied to any layer of farmland in a terraced field with sequentially ordered heights. All farmland plots have equal areas. Each plot is equipped with a sensing terminal that is communicatively connected to a processing terminal. The sensing terminal is equipped with a soil moisture sensor, which collects moisture status data at different depths. The specific steps of the method are as follows:

[0060] S1: Collect moisture status data for each layer of farmland and generate a moisture retention time matrix;

[0061] In this embodiment, soil moisture sensors are deployed at different depths within the terraced fields to collect real-time moisture status data at different levels. This data covers the moisture content at each depth and is processed through time-series standardization to form a time-series dataset, characterizing the moisture changes in each soil layer during the monitoring period. After generating a water retention time matrix, the water retention characteristics of different soil layers and their changes over time can be clearly reflected.

[0062] S2: Perform iterative operations starting from the bottom layer to calculate the water absorption matrix of the current layer;

[0063] In this embodiment, the iterative operation starts from the bottom layer, generating a water absorption matrix by calculating the water absorption capacity between the current layer and the previous layer. The water absorption matrix for each layer is calculated based on the water state of the current layer, the water state of the previous layer, and the water transfer characteristics. Specifically, a dynamic siphon model driven by the potential difference between adjacent layers can accurately simulate the migration and adsorption process of water in the soil. By considering the dynamic changes in soil characteristics, the limitations of traditional static models are avoided, thereby improving the accuracy of irrigation management.

[0064] S3: Combine the current water retention time matrix of the current layer, the current water retention time matrix of the previous layer, and the water absorption matrix of the current layer to update the current water retention time matrix of the previous layer until the current layer is the top layer;

[0065] In this embodiment, the water state of each layer is updated progressively upwards by combining the water retention time matrix of the current layer, the water retention time matrix of the previous layer, and the water absorption matrix of the current layer. This process is carried out under the changing water properties of each soil layer, reflecting the dynamic transfer and retention of water, up to the top layer. This process not only simulates the upward flow characteristics of water in terraced fields but also ensures real-time updates and precise management of water status. This iterative update mechanism can adapt to the complex terrain changes of terraced fields, effectively avoiding the problem of excess water in the upper layers or insufficient water in the lower layers.

[0066] S4: Determine the irrigation amount for each farmland plot based on the current water retention time matrix of each plot to implement intelligent irrigation;

[0067] In this embodiment, based on the water retention time matrix of each farmland plot, the required irrigation amount for each plot is determined by calculating changes in soil moisture state. The water retention time matrix provides precise data support for intelligent irrigation, making irrigation adjustments more aligned with the actual water needs of the soil. This precise irrigation adjustment avoids water waste or poor crop growth caused by uniform irrigation amounts in traditional irrigation methods. In this way, irrigation decisions are more scientific, saving significant amounts of water resources and improving the quality and efficiency of crop growth.

[0068] S5: Collect water status data of each farmland plot in real time during irrigation and generate an updated water retention time matrix;

[0069] In this embodiment, an updated water retention time matrix is ​​generated by collecting real-time moisture status data for each layer of farmland. In actual terraced systems, water flow does not follow an ideal, uniform absorption pattern. When the absorption capacity of a certain terraced layer approaches saturation, the water flow tends to seek the path of least resistance, which may form temporary runoff channels. These channels alter the originally expected water flow direction and flow distribution. The absorption matrix aims to describe the dynamic process of water absorption between terraces; to more accurately reflect this process, the influencing factor of temporary runoff channels needs to be considered. Physically speaking, the emergence of temporary runoff channels signifies a change in the way lower terraces absorb water from upper terraces; it reduces the amount of water entering the lower terraces through normal soil absorption pathways.

[0070] In other words, when irrigation reaches a certain level, as the lower terraces absorb more water and the soil gradually becomes saturated, the formation of temporary runoff channels is a natural way of regulating the water flow, allowing excess water to drain away and preventing excessive water accumulation in the terraces.

[0071] S6: Adjust the irrigation strategy for the corresponding farmland plots based on the updated water retention time matrix;

[0072] Specifically, existing technologies have shortcomings in water management and precise regulation in terraced irrigation, especially given the complex soil stratification and water transfer characteristics of terraced fields. Traditional methods struggle to accurately address water distribution and its interactions. Specifically, current irrigation decisions often rely on static models, neglecting the dynamic changes and transfer of water between different soil layers. This makes it difficult to effectively address the complexity of water retention, particularly in terraced fields with significant elevation differences, where the flow and absorption characteristics of water from upper to lower layers are difficult to predict accurately, leading to over-irrigation or under-irrigation.

[0073] In this embodiment, this challenge is solved by iteratively updating the water retention time matrix from bottom to top and dynamically adjusting the water absorption matrix in conjunction with the water status of each layer. Based on the water retention time matrix and water absorption matrix of each layer, continuous iteration and feedback updates ensure that the water retention behavior of each layer can be accurately simulated, and the water status of each layer can be adjusted at any time. This dynamic update mechanism allows the retention time and absorption characteristics of water in each layer to reflect the actual soil conditions in real time, avoiding errors caused by static models and solving the problem of water interaction between different soil layers. This ensures accurate calculation of irrigation amount, thus avoiding the problems of over- or under-irrigation caused by previous model simplification or inaccurate parameter settings.

[0074] Furthermore, by updating the water uptake matrix and combining it with the water retention time matrix, precise regulation can be implemented for different soil layers and varying water dynamics. For example, when there is excessive moisture in the upper soil layer, the water uptake matrix guides water flow to the lower soil layers, preventing excessive accumulation of water in the upper layers, and optimizes water absorption in the lower layers by dynamically adjusting the water retention time matrix. In this process, the transfer of water from top to bottom considers not only the physical properties of the soil but also the flow path and absorption capacity of water at different depths within the terraces. This adaptive regulation method not only better utilizes water resources but also ensures healthy crop growth, guaranteeing that crops receive the most suitable water supply in different seasons and environments, thereby improving the long-term productivity of farmland.

[0075] It should be noted that in this embodiment, the dimensions of all matrices are fixed as the number of farmland plots and the height of the terraces, and matrix elements that cannot be calculated are set to zero.

[0076] Please see Figure 2 The flowchart for calculating the water retention time matrix in this embodiment of the invention is as follows, and the specific steps of S1 are as follows:

[0077] S1.1: Perform time-series standardization on the moisture status data at each depth to form a time-series dataset, which includes the continuous moisture content at the corresponding depth within the monitoring period;

[0078] Specifically, moisture status data are typically affected by factors such as different soil depths, climate conditions, and soil types, resulting in significant variations in data range and distribution. To eliminate these external influences, it is necessary to standardize the moisture data for each soil layer. The standardization process includes removing dimensional differences and transforming the moisture status data for each depth layer to the same scale range, ensuring consistent processing in subsequent steps.

[0079] S1.2: Calculate the rate of change of water content at adjacent depths within the same layer based on the time series dataset, and construct a water gradient change matrix, where the matrix elements are the difference in water content at adjacent depths at the same time, which is used to characterize the water migration gradient features in the vertical direction of the root region.

[0080] Specifically, the moisture gradient variation matrix is ​​an important tool for describing the changes and migration of soil moisture in the vertical direction (i.e., between different depth layers). As water infiltrates from the upper soil layers to the lower layers, it is affected by various factors, including soil permeability, root water absorption capacity, and topographic relief. Therefore, calculating the rate of change in moisture content can effectively reflect the migration rate and direction of water between different soil depths.

[0081] In this embodiment, the moisture difference between each depth and its adjacent depths is calculated and compared with the time series to obtain the rate of change. Each element of the matrix represents the moisture difference between different depth layers at the same time point, thereby constructing a moisture gradient change matrix.

[0082] S1.3: Based on the preset water retention time benchmark model, calculate the deviation between the time series dataset and the water retention time benchmark model, and generate a water retention time deviation matrix.

[0083] Specifically, the pre-defined water retention time baseline model is a theoretical description of soil water retention behavior. Each soil type, climate condition, and crop type leads to different water retention characteristics. To simulate these complex water retention behaviors, a baseline model of water retention time is first established based on experimental data. The model describes the water retention behavior of soil under ideal conditions and provides a reference standard for practical applications. By comparing the actually collected water time series data with this baseline model, the deviation value can be calculated, thereby revealing possible anomalies in the water retention process. For example, the water retention time of a certain soil layer may greatly exceed the prediction range of the pre-defined model, indicating that there may be problems such as excess water or poor drainage in that layer. The generated water retention time deviation matrix provides a detailed quantitative description of the deviation between the water retention characteristics at each depth and the expected values.

[0084] S1.4: Align the moisture gradient change matrix and the moisture retention time deviation matrix according to the time and depth dimensions to generate a moisture retention time matrix;

[0085] Specifically, the moisture gradient variation matrix contains the dynamic changes in moisture content differences between various depth layers, i.e., how moisture is transferred vertically from upper to lower layers. The moisture retention time deviation matrix, on the other hand, reflects the deviation between the moisture retention time at each depth layer and the ideal retention time model, revealing anomalies in soil moisture management. Therefore, these two matrices describe the dynamic changes and retention characteristics of soil moisture from different perspectives. They are directly corresponding in the depth dimension, but differ in the time dimension and feature representation. To enable the comprehensive processing of these two matrices within a unified analytical framework, they need to be aligned in both the time and depth dimensions. By ensuring that data from the same time point and the same depth layer correspond and are fused, the features of moisture gradient variation and moisture retention time deviation can be combined into a complete moisture retention time matrix.

[0086] In this embodiment, the time dimensions of the two matrices are first aligned. Since the moisture gradient change matrix and the moisture retention time deviation matrix record the changes in soil moisture state at different time points, it is necessary to ensure that the moisture gradient and retention deviation of each soil layer can match at the same time point. By synchronizing the data of the two matrices, and using interpolation or data interpolation techniques, the moisture state and moisture retention deviation corresponding to each time point in the two matrices can be aligned.

[0087] Furthermore, feature stitching is used to fuse the moisture gradient change characteristics and the moisture retention time deviation characteristics. This fusion not only more accurately describes the moisture state of each soil layer at different times, but also reflects any anomalies that may occur during moisture flow.

[0088] Please see Figure 3 The flowchart for calculating the water absorption matrix in this embodiment of the invention is as follows. The water absorption matrix is ​​calculated using a dynamic siphon model driven by the water potential difference between adjacent layers. The specific steps of S2 are as follows:

[0089] S2.1: Calculate the composite potential energy difference based on the difference in average soil moisture content and vertical height between the current layer and the previous layer, wherein the composite potential energy difference includes the vector superposition of matrix potential and gravitational potential;

[0090] Specifically, the composite potential energy difference is calculated by analyzing the water transfer characteristics between different soil layers. This difference is the driving force of water migration. Water migration is affected not only by the soil matrix but also by gravity; therefore, the composite potential energy difference is a comprehensive reflection of both factors.

[0091] In this embodiment, the difference in average soil moisture content between the current layer and the layer above is first calculated, i.e., the difference in moisture content between each layer. The difference in soil moisture content is a key factor affecting water migration; the greater the difference, the faster water migrates from the upper soil layer to the lower layer.

[0092] Furthermore, the vertical height difference between the two layers is calculated, i.e., the relative positional difference between the two soil layers. Water migration depends not only on the concentration difference but also on the height difference between the soil layers.

[0093] By combining these two factors, the resulting composite potential difference takes into account both soil moisture permeability and water potential differences, thus enabling a more accurate description of water migration trends.

[0094] S2.2: Calculate the inter-layer siphon admittance coefficient based on the crop root density distribution and soil viscosity coefficient of the previous layer;

[0095] Specifically, the siphon admittance coefficient is closely related to the physical properties of the soil, including crop root density distribution and soil viscosity coefficient. Root density affects the soil's water absorption capacity; the denser the root system, the higher the water absorption efficiency, thus more effectively promoting water absorption. The soil viscosity coefficient reflects the resistance to water flow in the soil; the higher the viscosity coefficient, the greater the difficulty of water flow, thereby affecting the efficiency of water migration.

[0096] S2.3: Couple the composite potential energy difference and the inter-level siphon admittance coefficient to calculate the matrix elements of the water absorption matrix and generate the initial water absorption matrix;

[0097] Specifically, the composite potential difference and the siphon admittance coefficient are important factors describing soil moisture migration, and their coupling provides a more comprehensive understanding of moisture migration. By coupling these two factors, a preliminary moisture uptake matrix can be obtained, where each element represents the rate of moisture migration between different layers. The composite potential difference provides the driving force for moisture migration, while the siphon admittance coefficient describes the ease of moisture transport between soil layers. Combining the two allows for a more accurate quantification of the intensity of moisture absorption in each layer. The generation of the initial moisture uptake matrix provides a starting point for subsequent corrections and dynamic adjustments, ensuring accurate simulation of the moisture uptake process.

[0098] In this embodiment, the water uptake rate of each soil layer is obtained by multiplying the composite potential energy difference by the siphon admittance coefficient. The composite potential energy difference provides the driving force for water migration, while the siphon admittance coefficient determines the efficiency of water migration. By multiplying the two, the resulting water uptake rate can more accurately reflect the water uptake capacity of each soil layer.

[0099] Furthermore, by coupling the composite potential difference and the siphon admittance coefficient, the resulting matrix elements will form an initial water absorption matrix. Each element of this matrix represents the rate of water migration between adjacent layers, that is, the absorption capacity of each soil layer for the water in the layer above it.

[0100] S2.4: By quantifying the current water retention time matrix and the current water retention time matrix of the previous layer, the initial water absorption matrix is ​​corrected to generate a water absorption matrix;

[0101] Specifically, the initial matrix is ​​further adjusted to more accurately reflect the actual water absorption capacity of the soil. The water retention time matrix and the water absorption matrix are closely related; the retention time matrix describes the water accumulation characteristics of each soil layer, while the absorption matrix describes the ability of water to be transferred from upper to lower layers. Through quantitative comparison and analysis of these two matrices, the initial water absorption matrix can be further corrected to ensure it truly reflects the absorption capacity of each soil layer in actual water transfer. This correction method not only considers the dynamic changes in water content but also fully considers the actual water transfer efficiency between soil layers. In this way, the water absorption matrix can be adjusted in real time to better meet the needs of actual agricultural production, ensuring the scientific nature and accuracy of irrigation decisions.

[0102] The specific steps in S2.4 are as follows:

[0103] S2.4.1: Calculate the difference correction factor based on the cosine similarity between the current layer's water retention time matrix and the previous layer's current water retention time matrix;

[0104] Specifically, cosine similarity is often used to measure the similarity between two vectors, specifically the degree of similarity between two layers of water retention time matrices. By calculating the cosine similarity between the current layer's water retention time matrix and the previous layer's water retention time matrix, we can quantify the differences between these two matrices in terms of water distribution and retention behavior. A high similarity indicates that the water state changes in the two soil layers follow similar trends, requiring a smaller correction factor; conversely, a low similarity indicates significant differences in water changes between the two soil layers, necessitating a larger correction factor.

[0105] In this embodiment, the water retention time matrix of each layer can be viewed as a vector describing the soil moisture state. Each matrix element represents the water retention state at a specific time and depth. The entire matrix is ​​a high-dimensional data structure, where each row represents the change in water state at a certain time point, and each column represents the water retention at a certain depth. To quantify the similarity between the two matrices, cosine similarity is used to compare the similarity between the current layer and the previous layer's water retention time matrix. By comparing the water retention data in these two matrices, it can be determined whether the trends of soil moisture state changes in the two layers are consistent, thus determining their similarity.

[0106] Furthermore, when calculating cosine similarity, the water retention time matrix of each layer is treated as a vector. The water retention time matrix of the current layer is designated as the first matrix, and the water retention time matrix of the previous layer as the second matrix. Then, the cosine similarity is calculated through the following steps:

[0107] Flatten the first and second matrices into vectors, and treat each element of the matrix as a component of the vector;

[0108] Calculate the dot product between these two vectors;

[0109] Calculate the magnitudes of these two vectors;

[0110] The cosine similarity can be obtained by formulating the dot product of two vectors divided by the product of their magnitudes.

[0111] Furthermore, if the cosine similarity value is close to 1 (i.e., the two are very similar), the difference correction factor is close to 0, meaning that the water retention process in the current layer and the previous layer is almost identical, and no significant correction is needed. If the cosine similarity value is close to 0 (i.e., the two are significantly different), the difference correction factor is close to 1, meaning that the difference in water retention is significant, requiring strong adjustments to correct the absorption matrix and ensure the rationality of water transfer. Specifically, nonlinear mapping is used to ensure that the correction factor adapts to the actual situation of water transfer under different similarity conditions.

[0112] S2.4.2: Establish an interlayer water transport model using the Bernoulli equation, input the difference correction factor into the interlayer water transport model, and output the absorption coefficient of the current layer to the previous layer;

[0113] Specifically, Bernoulli's equation, a fundamental equation describing energy conversion in fluid mechanics, is applied to water transport. It combines water potential difference, soil type, and the physical conditions of water flow to describe the process of water transfer from one layer of soil to another. By establishing this interlayer water transport model, the efficiency of water transfer from the previous layer to the current layer and its influencing factors can be accurately calculated.

[0114] In this embodiment, the interlayer water transport model describes the process of water transfer from the upper layer to the current layer. Its efficiency is influenced by multiple factors, including water potential difference, soil type, and depth difference. Using these input parameters, the model can calculate the water transport efficiency of each soil layer and provide data support for the water absorption capacity of each layer. This model is used to evaluate the water transfer rate between different soil layers, i.e., the water absorption capacity of each soil layer.

[0115] Furthermore, the previously calculated difference correction factor reflects the difference in water retention behavior between the current layer and the layer above. Inputting the difference correction factor into the interlayer water transport model allows for dynamic adjustment of the water transfer efficiency calculation. When the difference in water retention characteristics between the current and upper layers is significant, the difference correction factor increases the degree of correction for water transfer, thus ensuring the accuracy of water uptake; conversely, when water retention characteristics are similar, the difference correction factor is smaller, and the degree of correction for water transfer is also lower. By combining the difference correction factor with other parameters in the Bernoulli equation (such as water potential difference, soil properties, etc.), the absorption coefficient of the current layer to the upper layer is calculated and output. This absorption coefficient represents the current soil layer's ability to absorb water from the upper layer, i.e., the efficiency of water flow from the upper layer to the current layer. A higher absorption coefficient indicates a greater rate of water inflow from the upper layer to the current layer, and vice versa. The basic conditions for water flow are calculated based on the input water potential difference and soil type (such as permeability and porosity). Then, these calculations are adjusted using a difference correction factor to correct for the effects of differences in water retention characteristics between the two soil layers. Finally, the resulting absorption coefficient is used as the basis for the next step of correcting the water absorption matrix.

[0116] S2.4.3: The initial water absorption matrix is ​​weighted and assigned according to the absorption coefficient to generate a water absorption matrix;

[0117] The specific steps for S3 are as follows:

[0118] S3.1: The water retention time matrix of the current layer and the previous layer is time-series-sliding through a time window. The dynamic weight matrix of interlayer water migration is calculated within the time window. The dynamic weight is determined based on the mutual information entropy of the water gradient change characteristics and retention time deviation characteristics of the upper and lower layers during the same period.

[0119] In this embodiment, by introducing the concept of a time window, a time-series sliding operation is implemented on the water retention time matrix to calculate the dynamic weight matrix of interlayer water migration. The changes in the water retention time matrix at different time points and between different depth layers reflect the dynamic transfer of water, and by time-series sliding, short-term fluctuations and long-term trends in the water migration process can be captured;

[0120] Specifically, the water retention time matrix is ​​a high-dimensional matrix that changes over time, recording the water retention status at different depths. To capture the dynamic changes in water migration, the water retention time matrix is ​​slidable along the time dimension, processing multiple data points within each time window to form a new sub-matrix. These sub-matrices provide information on short-term and long-term changes in water migration. Each sub-matrix contains the water retention status within a specific time period, which can be used to analyze water accumulation, loss, or transfer. The size of the sliding time window is set according to actual needs, typically related to factors such as irrigation cycles, seasonal variations, and soil moisture characteristics. Through this sliding window, the water state changes of each soil layer at different time points can be monitored in real time, ensuring that fluctuations in water transport are captured promptly.

[0121] Furthermore, in the calculation of dynamic weights, the mutual information entropy of the water gradient change characteristics and the residence time deviation characteristics is used to determine the weight magnitude. Mutual information entropy is an indicator that measures the degree of information sharing between two variables and can describe the degree of interdependence between variables. By calculating the mutual information entropy of the water gradient change characteristics and the residence time deviation characteristics between the current layer and the previous layer, the correlation between these two characteristics during water migration can be determined. If the changes in water gradient and residence time deviation have a high correlation (i.e., a large mutual information entropy), it indicates that the changes in water state in the current layer and the previous layer are closely linked, the water migration efficiency is high, and therefore the dynamic weight will be large. Conversely, if the similarity between the two is low and the mutual information entropy is small, it indicates that water migration is less affected by the water state of the upper and lower layers, and the dynamic weight will be small.

[0122] S3.2: Perform a tensor product operation between the dynamic weight matrix and the water absorption matrix to generate a composite coefficient matrix for interlayer water transport;

[0123] In this embodiment, the dynamic weight matrix describes the intensity of water migration changes within different time windows, while the water absorption matrix provides the water absorption capacity and efficiency of each soil layer. Through tensor product operations, the information from these two matrices is fused together to form a new composite coefficient matrix, which can simultaneously reflect the intensity and efficiency of water transport.

[0124] S3.3: Update the current water retention time matrix of the upper layer element by element according to the composite coefficient matrix, and add the product of the upper layer matrix element and the corresponding element of the composite coefficient matrix, and the compensation amount after root water absorption correction of the current layer matrix element to form the updated upper layer water retention time matrix.

[0125] Specifically, each element of the composite coefficient matrix represents the current soil layer's ability to absorb water from the previous layer, or the rate of water migration in the soil. Multiplying the composite coefficient matrix element-wise with the water retention time matrix of the previous layer is to calculate the transfer effect of water retention state in each layer, migrating water from the previous layer to the current layer. This calculation quantifies the influence of the previous layer's water state on the elements of the previous layer's water retention time matrix at the current moment. Larger values ​​in the composite coefficient matrix indicate a stronger absorption capacity of the previous soil layer for water in the current layer, resulting in higher water transfer efficiency and a greater impact of water retention in the previous layer on the current layer; conversely, smaller composite coefficients indicate that water migration from the previous layer to the current layer is inhibited.

[0126] In this embodiment, root water uptake has a significant impact on the distribution and migration of soil moisture in natural soil. Roots absorb water from the soil to meet the growth needs of crops; therefore, the compensation amount after root water uptake correction is a crucial factor. The calculation of the root water uptake correction considers the following aspects:

[0127] Root distribution varies among different crops, with denser root systems resulting in greater water absorption, especially in the topsoil. Root distribution determines the efficiency of water absorption in the soil, which can be determined through experimental data.

[0128] Soil moisture status is closely related to the water absorption capacity of the root system. When the soil moisture content is high, the root system's water absorption efficiency may be low, while in drought conditions, the root system will compensate for the lack of water by increasing water absorption. Therefore, it is necessary to consider the relationship between soil moisture content and root water absorption.

[0129] Root water absorption models typically describe the water absorption behavior of roots using physiological models. In practical calculations, the root water absorption compensation amount can be obtained by simulating the root's water absorption rate (usually a parameter in the water flow model) and then corrected according to the current soil moisture state.

[0130] Furthermore, the compensation amount obtained through root water uptake correction is superimposed on the product of the composite coefficient matrix and the previous layer's water retention time matrix. This further refines the water transfer process, ensuring that the water retention state of each layer depends not only on the efficiency of water transfer (provided by the composite coefficient matrix) but also on root water uptake behavior. This guarantees that the water retention of each soil layer reflects the actual water requirements of the crop. The updated water retention time matrix is ​​generated by multiplying the elements of the upper-layer matrix with the corresponding elements of the composite coefficient matrix, plus the compensation amount corrected for root water uptake. Each element of this matrix represents the final state of soil water retention in that layer, after water migration and root water uptake correction. This updated matrix provides the foundation for subsequent water management and irrigation strategies, accurately guiding water regulation in each soil layer.

[0131] Please see Figure 4 The schematic diagram of the matrix update principle of this invention illustrates a simplified three-layer terraced field structure, consisting of five farmland blocks: the top layer, the first layer, the second layer, the first bottom layer, and the second bottom layer. The iteration process starts from the bottom layer. The first layer farmland blocks update the first layer water retention matrix once and twice based on the first bottom layer absorption matrix, the first bottom layer water retention time matrix, the second bottom layer absorption matrix, and the second bottom layer water retention time matrix transmitted from the first bottom layer and the second bottom layer farmland blocks. The second layer farmland blocks update the second layer water retention matrix once and twice based on the first bottom layer absorption matrix, the first bottom layer water retention time matrix, the second bottom layer absorption matrix, and the second bottom layer water retention time matrix transmitted from the first bottom layer and the second bottom layer farmland blocks. Then, the top layer farmland blocks update the top layer water retention time matrix once and twice based on the first layer absorption matrix, the updated first layer water retention time matrix, the second layer absorption matrix, and the updated second layer water retention time matrix.

[0132] The specific steps for S5 are as follows:

[0133] S5.1: When the water content of any depth layer in a farmland plot is detected to be greater than the preset saturation threshold and the hydraulic gradient of the surface runoff path is greater than the preset critical value, it is determined that the layer forms a temporary runoff channel.

[0134] In this embodiment, the detection of temporary runoff channels is based on soil moisture content and the hydraulic gradient of the surface runoff path. Excess moisture is the main cause of runoff; therefore, when soil moisture reaches saturation, water can no longer effectively infiltrate the soil and will instead form runoff along the surface.

[0135] Specifically, a preset saturation threshold is set. When the water content at a certain depth in the soil exceeds this threshold, it indicates that the soil layer has reached saturation, and water can no longer infiltrate, forming a potential runoff channel. A critical value for hydraulic gradient is also set; when the slope of surface water flow exceeds this value, the kinetic energy and velocity of the water flow increase, making it easier for water to flow along the slope, leading to water concentration. By combining the detection of these factors, when both conditions are met simultaneously, it is determined that a temporary runoff channel has formed in the soil layer.

[0136] S5.2: Construct a directed graph model of a temporary runoff channel using graph theory algorithms. The model uses field ridge fissures and furrow depressions as nodes and real-time flow velocity as edge weights.

[0137] Specifically, graph theory algorithms are widely used to describe and analyze network and flow problems. By treating each geographical point in the farmland as a node and the water flow relationship between the nodes as an edge, the flow path of water can be effectively described.

[0138] In this embodiment, field ridge fissures and furrow depressions are used as nodes because these geographical features are natural channels or distribution points for water flow, and these locations play a decisive role in the migration of water. Each edge from one node to another represents the path of water flow, and the weight of the edge is defined by the real-time flow velocity. The greater the flow velocity, the greater the weight of the edge, indicating a greater intensity of water flow.

[0139] Furthermore, field ridge fissures are natural dividing lines in farmland and are often one of the paths for water flow. When there is excess water, it flows along these fissures, thus becoming natural channels for water flow. Ridge depressions are low-lying areas in farmland topography, where water accumulation and runoff typically concentrate. These topographic features determine the concentration and distribution direction of water; therefore, in water flow modeling, ridge depressions, as nodes, can reflect the convergence points of water flow. Each field ridge fissure and ridge depression is considered a node in the graph theory model, and they are connected by water flow paths.

[0140] Furthermore, the edges connecting nodes represent the paths of water flow. In practical water flow analysis, the intensity and velocity of the water flow are key factors affecting the efficiency of water migration. Therefore, the weight of each edge is defined as the real-time flow velocity. Flow velocity is the kinetic energy of the water flow and is the main factor affecting the intensity and velocity of the flow. By monitoring the water flow in real time, real-time flow velocity data on each water flow path can be obtained, thereby assigning values ​​to the edge weights.

[0141] Furthermore, a directed graph in graph theory refers to a graph where there are directional connections between nodes, meaning that water flow is unidirectional. In this embodiment, the direction of water flow is from one node to another, typically following the natural path of water flow, such as flowing from a crack in a field ridge to a depression in a furrow, or from one low-lying point to another.

[0142] S5.3: Based on the directed graph model, generate an interference matrix using a binary mask, where a 1 in the binary mask indicates the absence of temporary runoff channel interference, and a 0 indicates the presence of temporary runoff channel interference.

[0143] Specifically, based on the dynamic characteristics of water flow, the directed graph model is analyzed to identify interference areas in the water flow path. That is, the water flow may be interfered with by other water flow channels at certain specific nodes or paths, thereby affecting the efficiency of water flow.

[0144] In this embodiment, by combining the nodes and water flow paths in the graph theory model with the topography and direction of water flow in the farmland, it is determined which areas are affected by water flow disturbance. In particular, disturbance areas may be formed when the water flow path of a node intersects or overlaps with the water flow paths of other nodes. For example, when water flows from a high-slope area to a low-lying area (such as a furrow depression), but the presence of temporary runoff channels causes the water flow path to deviate or become blocked, thus affecting normal water circulation. In this case, the model will mark the path as a disturbance area.

[0145] Furthermore, after identifying the interference areas, a binary mask is used to represent their presence or absence. Each element in the mask corresponds to a water flow path in the graph; a value of 1 indicates that the flow path is free of interference, while a value of 0 indicates that interference exists. The binary mask is generated based on the following factors:

[0146] Is the water flow path affected by temporary runoff channels?

[0147] Will the location of the temporary runoff channel cause deviations in water flow or changes in its path?

[0148] Do other water flow paths intersect with the current path, creating competition or obstruction of water flow?

[0149] S5.4: Correct the corresponding row in the water absorption matrix according to the interference matrix to generate an updated water absorption matrix;

[0150] Specifically, the interference matrix provides a signal indicating which areas of the water flow path interfere with water absorption. By mapping the areas marked as 0 in the interference matrix to the relevant rows of the water absorption matrix and applying weighted adjustments, the water absorption values ​​of these areas are adjusted. Areas marked as 0 receive a stronger adjustment, reflecting the inhibitory effect of water flow on water absorption; while areas marked as 1 indicate no interference, and their water absorption state remains unchanged.

[0151] S5.5: Generate an updated water retention time matrix based on the updated water absorption matrix.

[0152] Example 2

[0153] This invention provides an embodiment of a crop water requirement prediction system based on agricultural Internet of Things (IoT). The system includes a sensor acquisition module, a matrix generation module, a hierarchical iteration module, and a water requirement prediction module, wherein:

[0154] The sensor acquisition module is deployed in each layer of farmland and includes a multi-depth soil moisture sensor array, a pore water pressure sensor and a surface runoff monitoring unit, used to collect moisture state data, interlayer hydraulic gradient data and surface runoff characteristics at different depths.

[0155] The matrix generation module is connected to the sensing and acquisition module, and performs the following operations:

[0156] Time-series standardization processing was performed on multi-depth moisture data to construct a moisture gradient change matrix and a residence time deviation matrix.

[0157] The correlation strength between the simultaneous moisture gradient between upper and lower layers and the retention deviation is calculated using the mutual information entropy algorithm, and a dynamic weight matrix is ​​generated.

[0158] Temporary runoff channels are modeled using graph theory algorithms to generate binary mask interference matrices.

[0159] The hierarchical iteration module is used to generate an initial water absorption matrix driven by a dynamic siphon model based on the composite potential energy difference and root admittance coefficient, fuse the interlayer dynamic weight matrix through tensor product operation, and iteratively correct the upper layer water retention time matrix using an element-by-element update rule.

[0160] The water demand prediction module is used to calculate the water demand based on the feature values ​​corresponding to each layer depth in the updated water retention time matrix.

[0161] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A crop water requirement prediction method based on agricultural Internet of Things, applied to any layer of farmland in a terraced field with sequentially ordered heights, wherein the farmland is equipped with a sensing terminal communicating with a processing terminal, the sensing terminal being equipped with a soil moisture sensor, the soil moisture sensor being used to collect moisture status data at different depths, characterized in that... The method includes: Collect moisture status data for each layer of farmland and generate a moisture retention time matrix; Iterative operations are performed starting from the bottom layer. The current water retention time matrix of the current layer, the current water retention time matrix of the previous layer, and the water absorption matrix of the current layer are combined to update the current water retention time matrix of the previous layer, until the current layer becomes the top layer. The water absorption matrix is ​​calculated using a dynamic siphon model driven by the potential difference between adjacent layers, including: The composite potential energy difference is calculated based on the difference in average soil moisture content and vertical height between the current layer and the previous layer, wherein the composite potential energy difference includes the vector superposition of matrix potential and gravitational potential. Calculate the inter-layer siphon admittance coefficient based on the crop root density distribution and soil viscosity coefficient of the previous layer; The composite potential difference and the inter-level siphon admittance coefficient are coupled to calculate the matrix elements of the water absorption matrix and generate the initial water absorption matrix. The initial water absorption matrix is ​​corrected by quantifying the water retention time matrix of the current layer and the current water retention time matrix of the previous layer, and a water absorption matrix is ​​generated. Predict the water demand of each farmland plot based on the current water retention time matrix of each plot. The updating of the current water retention time matrix of the previous layer includes: The water retention time matrix of the current layer and the previous layer is time-slid through a time window. The dynamic weight matrix of water migration between layers is calculated within the time window. The dynamic weight is determined based on the mutual information entropy of the water gradient change characteristics and retention time deviation characteristics of the upper and lower layers during the same period. The dynamic weight matrix and the water absorption matrix are subjected to tensor product operation to generate a composite coefficient matrix for interlayer water transport. The current water retention time matrix of the previous layer is updated element by element based on the composite coefficient matrix.

2. The crop water requirement prediction method based on agricultural Internet of Things according to claim 1, characterized in that, The process of collecting moisture status data for each layer of farmland and generating a moisture retention time matrix includes: Moisture state data at each depth are subjected to time-series standardization to form a time-series dataset, which includes the continuous moisture content at the corresponding depth within the monitoring period. The rate of change of water content at adjacent depths within the same layer is calculated based on the time series dataset, and a water gradient change matrix is ​​constructed. Based on the preset water retention time benchmark model, the deviation between the time series dataset and the water retention time benchmark model is calculated, and a water retention time deviation matrix is ​​generated. The moisture gradient change matrix and the moisture retention time deviation matrix are aligned according to the time and depth dimensions to generate a moisture retention time matrix.

3. The crop water requirement prediction method based on agricultural Internet of Things according to claim 2, characterized in that, The step of dimensionally aligning the moisture gradient change matrix and the moisture retention time deviation matrix according to the time and depth dimensions includes: The gradient feature values ​​and deviation feature values ​​corresponding to the same time point and the same depth layer are fused to generate a water retention time matrix. The row dimension of the water retention time matrix is ​​the monitoring time point, the column dimension is the depth layer, and the channel dimension includes gradient change features and deviation features.

4. The crop water requirement prediction method based on agricultural Internet of Things according to claim 1, characterized in that, The step of correcting the initial water absorption matrix to generate a water absorption matrix includes: The difference correction factor is calculated based on the cosine similarity between the current layer's water retention time matrix and the previous layer's current water retention time matrix. An interlayer water transport model is established using the Bernoulli equation. The difference correction factor is input into the interlayer water transport model, and the absorption coefficient of the current layer to the previous layer is output. The initial water absorption matrix is ​​weighted and assigned values ​​based on the absorption coefficient to generate a water absorption matrix.

5. The crop water requirement prediction method based on agricultural Internet of Things according to claim 1, characterized in that, The current water retention time matrix of the previous layer is updated element-wise based on the composite coefficient matrix, including: The product of the upper-layer matrix elements and the corresponding elements of the composite coefficient matrix is ​​added to the compensation amount of the current-layer matrix elements after root water absorption correction, forming the updated upper-layer water retention time matrix.

6. The crop water requirement prediction method based on agricultural Internet of Things according to claim 1, characterized in that, The method further includes: During irrigation, real-time water status data for each layer of farmland is collected to generate and update the water retention time matrix. Based on the updated water retention time matrix, the irrigation strategy for the corresponding farmland plots is adjusted.

7. The crop water requirement prediction method based on agricultural Internet of Things according to claim 6, characterized in that, The generation and updating of the water retention time matrix includes: When the water content of any depth layer in a farmland plot is detected to be greater than a preset saturation threshold and the hydraulic gradient of the surface runoff path is greater than a preset critical value, it is determined that a temporary runoff channel has been formed at that depth layer. A directed graph model of a temporary runoff channel is constructed using graph theory algorithms. The model uses field ridge fissures and furrow depressions as nodes and real-time flow velocity as edge weights. Based on the directed graph model, an interference matrix is ​​generated using a binary mask, where a 1 in the binary mask indicates the absence of temporary runoff channel interference, and a 0 indicates the presence of temporary runoff channel interference. The corresponding row in the water absorption matrix is ​​corrected based on the interference matrix to generate an updated water absorption matrix; An updated water retention time matrix is ​​generated based on the updated water absorption matrix.

8. A crop water requirement prediction system based on agricultural Internet of Things (IoT), used to implement the crop water requirement prediction method based on agricultural IoT as described in any one of claims 1-7, characterized in that, The system includes a sensor acquisition module, a matrix generation module, a hierarchical iteration module, and a water demand prediction module, wherein: The sensor acquisition module is deployed in each layer of farmland and includes a multi-depth soil moisture sensor array, a pore water pressure sensor and a surface runoff monitoring unit, used to collect moisture state data, interlayer hydraulic gradient data and surface runoff characteristics at different depths. The matrix generation module is connected to the sensing and acquisition module, and performs the following operations: Time-series standardization processing was performed on multi-depth moisture data to construct a moisture gradient change matrix and a residence time deviation matrix. The correlation strength between the simultaneous moisture gradient between upper and lower layers and the retention deviation is calculated using the mutual information entropy algorithm, and a dynamic weight matrix is ​​generated. Temporary runoff channels are modeled using graph theory algorithms to generate binary mask interference matrices. The hierarchical iteration module is used to generate an initial water absorption matrix driven by a dynamic siphon model based on the composite potential energy difference and root admittance coefficient, fuse the interlayer dynamic weight matrix through tensor product operation, and iteratively correct the upper layer water retention time matrix using an element-by-element update rule. The water demand prediction module is used to calculate the water demand based on the feature values ​​corresponding to each layer depth in the updated water retention time matrix.

Images