Paddy field wheel mud dynamic response analysis method and system based on multi-sensor fusion

Abstract

The application discloses a paddy field wheel mud soil dynamic response analysis method and system based on multi-sensor fusion, relates to the technical field of agricultural information perception, and collects mud soil response data from different depths and directions; high correlation response feature subsets are obtained by processing mud soil response data based on a multi-layer information fusion network model, a soil physical field excitation sequence under multi-layer coupled boundary conditions is obtained by inversion based on the feature subsets, multi-time domain parallel response analysis is carried out on a target region; a stress-pore pressure coupling peak value region and its spatial distribution in response output are obtained, and a mud soil weak response fingerprint spectrum is constructed; finally, optimization suggestions are proposed for wheel mud paths, operation rhythms and pressure parameters based on the spectrum, dynamic pre-response evaluation and decision support before paddy field wheel mud operation are realized; the method has the advantages of high coupling identification precision, strong risk perception capability and intelligent operation parameter regulation, and is suitable for intelligent agricultural equipment systems and precise paddy field management scenes.

Description

Technical Field

[0001] This invention relates to the field of agricultural information sensing technology, specifically to a method and system for dynamic response analysis of paddy field mud and soil based on multi-sensor fusion. Background Technology

[0002] Rotation of soil in paddy fields is a crucial step in rice production. It primarily uses mechanical disturbance to shape, loosen, and mix the topsoil with cement, significantly impacting crop root growth, water retention, nutrient distribution, and pest and disease control. However, due to the complex soil structure, fluctuating moisture content, and non-uniform mechanical loading in paddy fields, traditional methods struggle to effectively assess the real-time dynamic response of the soil during rotation. Assessments often rely on experience or post-event observations, lacking quantifiable and predictable scientific evidence.

[0003] With the development of agricultural informatization, existing research has attempted to introduce sensor monitoring technology to perceive the state of paddy field soil. However, due to the limited information dimensionality obtained from a single sensor source and the inability to achieve joint analysis of multiple physical quantities such as soil stress, pore water pressure, and deformation under dynamic load conditions, the assessment results are subject to lag and uncertainty. Currently, there is an urgent need for a method that integrates multi-source sensor information and combines physical modeling and dynamic response calculation to perform high-precision analysis and simulation of the actual response process of muddy soil under paddy field mud rotation operations, providing support for agricultural machinery parameter optimization and paddy field management strategies. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for dynamic response analysis of paddy field mud and soil based on multi-sensor fusion, so as to overcome the shortcomings of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic response analysis of paddy field mud and soil based on multi-sensor fusion, comprising:

[0006] A multi-layer information fusion network model M, which integrates surface stress disturbance signals, pore water pressure variation trends, shear wave propagation velocity, and soil-water coupling nonlinear indices, is constructed to characterize the dynamic evolution characteristics of multiple physical properties of soil.

[0007] Simultaneously collect mud and soil response data from different depths and spatial locations in the target paddy field area. The mud and soil response data includes surface acceleration, pore water pressure, shear modulus perturbation spectrum, and time domain signal of reflected waves.

[0008] The soil response data is input into a multi-layer information fusion network model M, and feature decoupling and mutual information optimization are performed to obtain a feature subset T.

[0009] Construct dynamic finite response units for mud and soil, and invert the soil physical field excitation sequence under multi-layer coupled boundary conditions based on the physical features in the feature subset T;

[0010] A wheel mud load function is applied to the mud dynamic finite response unit, and multi-time-domain parallel response analysis is performed on the target area to obtain time-varying stress and pore pressure response data in each response unit.

[0011] Identify the stress-pore pressure coupling peak region and its spatial distribution curve in the response analysis results, extract high-risk response units, and form a soil weak response map P.

[0012] Based on the weak response map P of mud and soil, suggestions are made for optimizing the mud rotation path, adjusting the operation rhythm, and setting pressure parameters, so as to realize dynamic pre-response assessment and operation parameter optimization before mud rotation operation in paddy fields.

[0013] Preferably, training sample data of different physical quantities are collected in the paddy field area, including stress disturbance signals, pore water pressure changes, shear wave propagation velocity, and nonlinear coupling index; the various sample data are aligned according to time synchronization to form a multi-scale input sequence; a multi-layer information fusion model M is constructed through a graph neural network, where nodes represent sensing types and edges represent physical coupling relationships.

[0014] Preferably, multiple types of sensor nodes are deployed in the target paddy field area according to a preset grid; surface acceleration, underground pore pressure, shear wave and reflected wave data from different depths are collected; the original mud response data are denoised and normalized to form a multidimensional dataset.

[0015] Preferably, the multidimensional dataset is input into the multi-layer information fusion network model M, and the maximum correlation and minimum redundancy strategy is used to select highly correlated response variables to form a feature subset T.

[0016] Preferably, the spatial attributes in the feature subset T are mapped to the simulation region grid, the similarity between adjacent simulation region grids is calculated, and if the similarity is greater than a preset threshold, it is marked as a muddy dynamic finite response unit.

[0017] Preferably, dual-channel extreme value tracking is performed on the stress-pore pressure response data output by each unit; spatial nodes where local peak values ​​of stress and pore pressure appear simultaneously are marked; highly coupled peak nodes are mapped to spatial distribution curves to extract high-risk response units; and a mud soil weak response map P is constructed using the high-risk response unit number as an index and the coupling strength as a weight.

[0018] Preferably, a path avoidance map is generated for high-risk response units in the weak response map P of the muddy soil; the operation trajectory is generated using the risk level of the map as the path weight and the minimum risk path algorithm.

[0019] Preferably, the multi-layer information fusion network model M adopts the graph convolutional network concept, and the node state update formula is: ; This means that in the l-th layer of the network, the new feature vector of node u is equal to the weighted sum of the feature vectors of its neighboring nodes. This means that in the (l+1)th layer of the network, the new feature vector of node u is equal to the weighted sum of the feature vectors of its neighboring nodes; This indicates that the similarity is calculated from the features of node v and node u. Let represent the trainable weight matrix of the l-th layer, and σ be the activation function.

[0020] This invention also provides a dynamic response analysis system for paddy field mud and soil based on multi-sensor fusion, comprising:

[0021] The data-driven modeling module constructs a multi-layer information fusion network model M that integrates surface stress disturbance signals, pore water pressure variation trends, shear wave propagation velocity, and soil-water coupling nonlinear indices to characterize the dynamic evolution characteristics of multiple physical properties of soil.

[0022] The multi-source sensor acquisition module simultaneously acquires mud and soil response data from different depths and spatial locations in the target paddy field area. The mud and soil response data includes surface acceleration, pore water pressure, shear modulus perturbation spectrum, and time domain signal of reflected wave.

[0023] The response modeling module inputs the soil response data into a multi-layer information fusion network model M, performs feature decoupling and mutual information optimization processing, and obtains a highly correlated response feature subset T.

[0024] The response simulation module constructs a dynamic finite response unit for mud and soil, and inverts the soil physical field excitation sequence under multi-layer coupled boundary conditions based on the physical characteristics in the feature subset T.

[0025] The response analysis module applies a wheel mud load function to the mud dynamic finite response unit and performs multi-time-domain parallel response analysis on the target area to obtain time-varying stress and pore pressure response data in each response unit.

[0026] The response map generation module identifies the stress-pore pressure coupling peak region and its spatial distribution curve in the response analysis results, extracts high-risk response units, and forms a soil weak response map P.

[0027] The feedback module proposes suggestions for optimizing the mud wheel path, adjusting the operation rhythm, and setting pressure parameters based on the mud weak response map P, thereby realizing dynamic pre-response assessment and operation parameter optimization before mud wheel operation in paddy fields.

[0028] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0029] 1. This invention constructs a multi-layer information fusion network model and integrates multi-source heterogeneous sensor data to achieve high-precision perception and dynamic modeling of the multi-physics coupling behavior of paddy soil under mud-wheel disturbance conditions. Compared with traditional methods that rely on single-point sensing and static parameter analysis, this invention can extract the coupling relationship between key physical variables such as soil stress and pore pressure, invert the excitation input conditions under real-world operating conditions, and reconstruct the disturbance propagation process through multi-time-domain parallel simulation, significantly improving the accuracy and timeliness of soil response prediction.

[0030] 2. The soil weak response fingerprint spectrum P construction mechanism and intelligent control method of operation parameters proposed in this invention can realize feedforward optimization of paddy field mud rotation path, operation rhythm and pressure strategy, provide agricultural equipment with adaptive scheduling capability based on soil response feedback, effectively reduce the risk of soil structure damage, and improve the safety and refined management level of paddy field operations. Attached Figure Description

[0031] 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.

[0032] Figure 1 This is a flowchart of the method of the present invention.

[0033] Figure 2 This is a flowchart of the system modules of the present invention. Detailed Implementation

[0034] 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.

[0035] Example 1, please refer to Figure 1 As shown in this embodiment, the method for dynamic response analysis of paddy field mud and soil based on multi-sensor fusion includes:

[0036] A multi-layer information fusion network model M, which integrates surface stress disturbance signals, pore water pressure variation trends, shear wave propagation velocity, and soil-water coupling nonlinear indices, is constructed to characterize the dynamic evolution characteristics of multiple physical properties of mud.

[0037] Simultaneously collect mud and soil response data from different depths and spatial locations in the target paddy field area. The mud and soil response data includes surface acceleration, pore water pressure, shear modulus perturbation spectrum, and time domain signal of reflected wave.

[0038] The soil response data is input into a multi-layer information fusion network model M, and feature decoupling and mutual information optimization are performed to obtain a feature subset T.

[0039] Construct dynamic finite response units for mud and soil, and invert the soil physical field excitation sequence under multi-layer coupled boundary conditions based on the physical features in the feature subset T;

[0040] A wheel mud load function is applied to the mud dynamic finite response unit, and multi-time-domain parallel response analysis is performed on the target area to obtain time-varying stress and pore pressure response data in each response unit.

[0041] Identify the stress-pore pressure coupling peak region and its spatial distribution curve in the response analysis results, extract high-risk response units, and form a soil weak response map P.

[0042] Based on the weak response map P of mud and soil, suggestions are made for optimizing the mud rotation path, adjusting the operation rhythm, and setting pressure parameters, so as to realize dynamic pre-response assessment and operation parameter optimization before mud rotation operation in paddy fields.

[0043] In the multi-sensor fusion-based dynamic response analysis method for paddy field mud and soil proposed in this invention, the information fusion model M is the core computational module for realizing data understanding, response prediction, and subsequent simulation-driven processing. To ensure that the model can realistically represent the evolution of multiple physical properties of mud and soil under mud and soil disturbance, it is necessary to construct a graph structure network with spatiotemporal representation capabilities by fusing observation data from different physical fields.

[0044] This invention uses various types of sensor equipment deployed in the target paddy field area to collect training sample data of the following physical quantities during a typical wheel mud disturbance operation:

[0045] Stress disturbance signal: Recorded by accelerometers deployed on the surface, the disturbance stress change curve σ(t) is obtained after mass conversion and filtering.

[0046] Pore ​​water pressure variation: acquired by an embedded pore pressure sensor and recorded as a pore water pressure time series u(t);

[0047] Shear wave propagation velocity: obtained through a small shear wave excitation device and receiver, the wave velocity is calculated from the propagation time difference between different depths. ;

[0048] Nonlinear coupling index: By deriving the hysteresis response parameters between stress and pore pressure (such as coupling delay τ and response ratio η), it serves as an index to characterize the nonlinear relationship of soil-water interaction.

[0049] The sampling frequency is recommended to be controlled between 50Hz and 500Hz to ensure coverage of the main disturbance frequency components; the synchronous acquisition cycle of each physical quantity is recommended to be no less than one complete mud wheel operation cycle (e.g., 10 to 30 seconds) in order to cover the entire process of disturbance loading and recovery.

[0050] Because different types of sensors differ in sampling frequency, data start and end times, and clock precision, directly using raw data will lead to distortion in the fusion model's learning. Therefore, time synchronization alignment processing needs to be performed before entering the model training, including:

[0051] Unified Time Base: Construct a unified reference time series using the system clock of the master node as the time axis. n represents the total number of reference time points;

[0052] Resampling interpolation: Align all physical quantity data to a unified time point using linear interpolation, cubic spline interpolation, or local mean smoothing.

[0053] Sliding window processing: A sliding window method is used to construct multi-scale input segments. Each input segment contains a time series data window of length L, with a window sliding step of Δt, ensuring temporal continuity and local dynamic feature representation;

[0054] Dimension Construction: Each input fragment is represented in matrix form: X∈ , where L represents the length of the time window, m represents the dimension of the physical quantity (such as stress, pore pressure, wave velocity, etc.), and X is the input sample of the final model.

[0055] Through the above processing, a high-quality input data sequence with uniform time alignment, consistent scale, and fused physical quantities can be obtained, which is suitable for the spatiotemporal modeling needs of graph neural structures.

[0056] After obtaining the fused input of multiple physical quantities, this invention employs a graph neural network structure to construct a multi-layer information fusion model M, which is used to mine the coupling relationships, propagation paths, and dynamic response trends among different physical variables within the soil. Its implementation details are as follows:

[0057] The multi-layer information fusion network model M adopts a graph neural network (GNN) structure, which consists of a node set, an edge set, a node feature matrix, and an edge weight matrix.

[0058] Node set V: Each node v∈V corresponds to a sensing physical quantity (such as stress disturbance signal, pore water pressure change, shear wave propagation velocity, nonlinear coupling index).

[0059] Edge set E: The edge (u, v) ∈ E between any two physical quantity nodes represents their physical coupling relationship, such as the interaction between stress and pore pressure.

[0060] Node feature matrix H: Each row corresponds to the feature vector of a node within the time window, with dimensions of L×m, where L is the number of time steps and m is the number of physical quantity features.

[0061] The edge weight matrix A is calculated from historical data, and the initial weights are determined using mutual information values, correlation coefficients, or physical formulas (such as the pore pressure conduction equation).

[0062] Model M adopts the concept of Graph Convolutional Network (GCN), and the node state update formula is as follows: ; This means that in the l-th layer of the network, the new feature vector of node u is equal to the weighted sum of the feature vectors of its neighboring nodes. This means that in the (l+1)th layer of the network, the new feature vector of node u is equal to the weighted sum of the feature vectors of its neighboring nodes; the weights are the edge attention coefficients. Then through the weight matrix A linear transformation is performed, and then the result is obtained using a nonlinear activation function σ (such as ReLU or LeakyReLU). (Attention coefficient): Calculated from the similarity of features between node v and node u. Let represent the trainable weight matrix of the l-th layer, which can be initialized using the Xavier initialization method. σ is the activation function, used to increase the nonlinear fitting capability.

[0063] Model training and optimization specifically involve:

[0064] Loss function: The mean squared error (MSE) or weighted mean squared error (WMSE) is selected and calculated based on the difference between the predicted output and the true high-risk response label.

[0065] Optimizer: The Adam optimizer is used with an initial learning rate of 0.001.

[0066] Regularization: Add Dropout (0.3~0.5 recommended) to each layer to prevent overfitting, and apply L2 regularization to the weight matrix.

[0067] Training data: Use 70% of the samples as the training set, 15% as the validation set, and 15% as the test set; the recommended batch size is 32.

[0068] The output of model M is the fused node feature matrix, which contains the spatiotemporal coupling feature vector of each physical quantity within the current time window.

[0069] These feature vectors are used for:

[0070] Feature decoupling and mutual information optimization (generating feature subset T);

[0071] Excitation inversion (generating soil physical field excitation sequence);

[0072] Risk area prediction (coupling peak identification).

[0073] In the dynamic response analysis method of paddy field mud based on multi-sensor fusion proposed in this invention, obtaining high-quality, highly synchronous multi-physical response data is a prerequisite for accurate modeling and dynamic response calculation.

[0074] In each sensing node, the following key sensing units are configured according to the physical quantity monitoring requirements:

[0075] Surface accelerometer: used to capture the micro-vibration response of the soil surface when agricultural machinery passes by, and its output signal can reflect the load input intensity, wheel track frequency and disturbance duration;

[0076] Buried pore water pressure sensor: Buried in the middle and lower layers of soil, used to sense the rise, conduction and hysteresis of pore water pressure under soil pressure.

[0077] Shear modulus disturbance measurement unit: By applying low-frequency shear wave excitation to the sampling area and measuring the wave velocity change, the trend of shear modulus change is indirectly obtained, reflecting the soil consolidation and softening process;

[0078] Seismic reflection wave receiver: The receiver sensor collects the reflected wave signal in the soil caused by mud disturbance, and uses time-domain signal decoding to invert the wave velocity and coupling of the soil layer interface.

[0079] Through the aforementioned layered deployment, synchronous sampling, and dynamic acquisition strategy, this invention achieves high timeliness, high spatial resolution, and multi-physical dimension capability in acquiring raw data on the mud disturbance response of target paddy fields. The acquired data can not only be directly used for subsequent fusion model training and physical simulation modeling, but also has the potential for further applications such as constructing data-driven risk assessment models and paddy field mud response databases.

[0080] In the multi-sensor fusion-based dynamic response analysis method for paddy field mud and soil proposed in this invention, the high-dimensionality, multi-source, and strong coupling characteristics of mud and soil response data determine the complexity and nonlinearity of subsequent analysis. To achieve effective dynamic response modeling, these multi-source response data must be processed in depth to extract core feature variables with physical meaning and computational value, and an analysis unit structure with clear structure and well-defined response behavior must be constructed based on these variables. To this end, this invention proposes a response data modeling mechanism that integrates feature decoupling, mutual information optimization, and structural partitioning. The core of this mechanism consists of three parts: feature decoupling processing, mutual information optimization screening, construction of a highly correlated response feature subset T, and generation of finite mud and soil dynamic finite response units for simulation.

[0081] First, the raw response data originates from the aforementioned multi-layer, multi-type sensor array. Each sensor point records multiple physical dimensions of data, including surface soil acceleration, underground pore water pressure, shear wave velocity disturbance, and reflected wave waveform information, along with spatial location information and sampling timestamps. Faced with such a complex dataset, directly using it for simulation modeling suffers from high redundancy, feature mixing, and ambiguous physical interpretation. Therefore, this invention, supported by model M, first performs feature decoupling and physical semantic classification on the raw data.

[0082] Specifically, the multi-layer graph structure integrated within model M establishes spatial topological relationships and data distribution correlations between sensing nodes. Based on this, a graph convolutional neural network structure is used to propagate and enhance the node data. Each node in the graph represents the data state of a sensor at a given moment, and edge weights represent spatial distance, signal correlation, or coupling degree. Through several layers of graph convolution calculations, the system can extract high-order nonlinear features reflecting coupling phenomena between nodes and perform physical semantic classification accordingly, dividing all response variables into the following four categories:

[0083] Stress-dominated types: such as peak soil acceleration and stress impact ratio, are mainly directly related to mechanical disturbance.

[0084] Pore ​​pressure-dominated categories: such as pore water pressure increment and pore pressure response delay time, which reflect the water migration process;

[0085] Shear wave sensitive parameters: such as shear modulus perturbation rate and wave velocity variation, which reflect the elastic evolution of soil structure;

[0086] Hybrid coupling types, such as the synchronization index of acceleration and pore pressure, and the asymmetry of reflected waves, exhibit complex relationships between multiple physical quantities.

[0087] After completing the above classification, these sub-feature sets are uniformly named feature set F1. Next, a subset of high-value features that are significantly correlated with wheel mud disturbance behavior and can be used for response prediction needs to be selected from feature set F1. This invention uses mutual information analysis to measure the correlation between each feature variable and the wheel mud disturbance input variable. Mutual information is a measure of the nonlinear correlation between two variables, capable of revealing deep connections that linear correlation methods cannot identify. In the calculation process, wheel mud disturbance is first modeled as a disturbance input function G(t), where t represents the operating time and G(t) represents the wheel mud load function per unit time; then, the mutual information value between each feature and G(t) is calculated, and a mutual information matrix of all features and the input function is established.

[0088] Based on the spatial distribution patterns of each feature in the feature subset T, clustering analysis methods (such as density peak clustering or spectral clustering) are used to identify local regions with similar response patterns. For example, distinct physical behavior boundaries are formed in high-feature regions such as areas with frequent pore pressure peaks and stress response abrupt change zones, thus defining unit boundaries.

[0089] In each cell, the dominant feature category (stress, pore pressure, shear wave, or hybrid coupling) is recorded and labeled.

[0090] The entire finite mud dynamic finite response unit ultimately forms a multi-field response space network oriented towards the dynamic action of wheel mud, with the following characteristics:

[0091] The unit exhibits strong consistency in its internal response behavior, making it suitable for local dynamic simulation.

[0092] There are real physical coupling boundaries between units, and the multi-region response transmission can be simulated through coupling equations;

[0093] Based on feature subset construction, it has data-driven modeling capabilities and eliminates the need for manual mesh generation.

[0094] In this invention, constructing a physical field excitation sequence for driving a finite response element of finite mud is the key to realizing dynamic simulation of soil response. Due to the obvious stratification of paddy field mud and the non-uniform, nonlinear, and multi-field coupling characteristics of mud wheel disturbance, traditional models often struggle to accurately define boundary input conditions, relying solely on empirically set fixed stress loading or pore pressure boundaries, leading to significant differences between simulation results and measured behavior.

[0095] To address the aforementioned problems, this invention proposes a physical field excitation sequence inversion method based on a feature subset T. By introducing a data-driven boundary reconstruction mechanism and a multi-physical variable collaborative driving strategy, a more realistic and coupled expressive boundary excitation function is systematically generated, thereby improving the accuracy and adaptability of soil dynamic response calculation. This method mainly includes the following three stages:

[0096] Construct the perturbation response field and perform spatial extrapolation:

[0097] First, after completing the screening of the feature subset T, a set of core physical variables that are highly correlated with the wheel mud disturbance behavior have been obtained, such as stress impact amplitude, shear modulus disturbance rate, and peak value of pore pressure change.

[0098] The structure of paddy field soil typically consists of a topsoil layer, a sedimentary layer, and a bearing layer, with significant differences in physical properties among these layers. Under the influence of mud wheel disturbance, these soil layers exhibit different response modes, thus requiring coupled boundary conditions in the simulation. Traditional methods often set boundaries based on empirical soil depth, but neglect the dynamic response during actual disturbance, resulting in low accuracy of boundary representation.

[0099] This invention employs an automatic boundary identification algorithm based on feature gradients and coupling indices to invert the boundary layering process from response data. Specifically, it uses the shear modulus perturbation gradient function in the perturbation response field to scan the longitudinal (depth) direction. When the rate of change of the shear modulus along the vertical direction exceeds a set threshold, it is identified as a potential coupling boundary.

[0100] After setting the boundary conditions, a physical field excitation sequence is constructed. The excitation sequence is essentially a dynamic input condition defined on the boundary or region, and its quality directly affects the accuracy of the response simulation. The multi-component coupled excitation function F(t, x, y, z) is obtained, specifically comprising the following components:

[0101] Time-varying pressure wave function P(t): obtained based on stress-dominant characteristics inversion, used to simulate the time-series pressure input formed during the downward propagation of wheel mud disturbance load on the ground surface;

[0102] Shear force loading function S(t): Based on shear modulus perturbation rate data, a time-varying shear excitation input is constructed, which is particularly suitable for analyzing soil structural failure or shear slip.

[0103] Pore ​​pressure boundary response function U(t): Combining the lag characteristics of pore pressure change, a time-varying function of boundary pore pressure change is established to support the simulation of delayed behavior of soil moisture stress response.

[0104] The inversion method uses a feature subset T as the input variable set, constructs a function library trained from the data, and applies boundary response behavior as optimization constraints. It then uses the minimum error fitting principle to deduce the excitation sequence that best matches the actual perturbation process. The final output function F(t, x, y, z) can achieve a refined description of the physical input in the spatiotemporal domain under multi-layered coupled boundary conditions, providing highly reliable simulation input for driving finite mud dynamic finite response units.

[0105] The "wheel mud load function G(t)" referred to in this invention is a function that expresses the change of force exerted on the soil by agricultural machinery (such as wheeled tractors, rotary tillers, etc.) during the mud-wheeling process in paddy fields over time. This function not only considers the periodicity of the force applied by the wheel axle, but also comprehensively considers the tire contact characteristics, water-soil coupling characteristics, and disturbance attenuation behavior.

[0106] The basic structure of the load function G(t) is as follows:

[0107] Wheel and axle periodic function: represents the periodic pressure input when the tire contacts the ground, usually a sine or pulse function;

[0108] Contact disturbance function: describes the nonlinear disturbance process generated when the tire comes into contact with the soil, and has peak effects and asymmetry;

[0109] Disturbance attenuation factor: Used to represent the energy attenuation characteristics of a disturbance as it propagates outward from the point of application of the load, and depends on the soil type and water content.

[0110] To obtain a time-varying function with high expressive power, the Fourier reconstruction method is used to decompose the field-acquired data in the frequency domain and reconstruct a synthesis function G(t) with similar spectral characteristics and phase structure, namely:

[0111] Mud load function Where Ai represents the amplitude of the i-th perturbation component, ωi represents the frequency, φi represents the phase angle, and Σ represents the summation of multiple sine wave terms to construct a complex periodic signal.

[0112] This function has periodicity, adjustable disturbance intensity, and frequency domain controllability, making it convenient for subsequent use as dynamic simulation input in different scenarios.

[0113] Because the input of mud load has strong temporal non-uniformity—for example, the soil in one area may be disturbed first, while another area responds with a lag—traditional methods with a uniform time step will result in low efficiency and loss of accuracy. Therefore, this invention designs a multi-temporal parallel response analysis mechanism, which includes three core parts: time-domain windowing, disturbance segment decomposition, and asynchronous parallel scheduling.

[0114] First, the load function G(t) is divided into minor disturbance segments according to the principal period of the disturbance, with each segment corresponding to the disturbance input characteristics of different time windows. By analyzing the spatial location, historical response delay, and coupling boundary conditions of each element in the array W, a disturbance-element mapping table is established, and different disturbance segments are applied to different response elements in space.

[0115] Secondly, different units enter the solution state as needed, avoiding the waste of computational resources caused by forced synchronization of all units in a unified time-step calculation. Some adjacent units share overlapping perturbation time windows to maintain the coupling relationship without distortion.

[0116] For example, under the influence of a disturbance at the front of the tire, the forward response unit will enter the solution state first, and subsequent units will start with a delay of several time steps. The time offset is controlled by the corresponding disturbance propagation speed in G(t). This asynchronous processing method not only preserves the temporal characteristics of disturbance propagation, but also significantly reduces the overall computational load.

[0117] In this invention, the inversion of the soil physical field excitation sequence is based on the highly correlated response feature subset T, which calculates and backtracks to obtain the soil physical state change curve under specific coupled boundary conditions, and is used to drive subsequent simulation analysis.

[0118] To improve simulation accuracy and multiphysics representation capabilities, this invention introduces a boundary-coupled response update mechanism in the simulation solution. Specifically, after each response element receives the load function input, the system synchronously calculates the following physical responses:

[0119] Pressure field P(x,y,z,t): Calculate the stress tensor distribution at the nodes based on the applied load and the static parameters of the soil;

[0120] Shear stress field S(x,y,z,t): determined by the load direction and soil shear modulus, used to determine the possibility of local sliding and failure;

[0121] Pore ​​water pressure field U(x,y,z,t): Simulates the pore pressure change caused by disturbance compaction, and dynamically updates it by combining the previous pore pressure hysteresis function.

[0122] After each time step is solved, the system transmits the response state of the current element to the adjacent elements according to the boundary condition transfer mechanism between elements, forming a local disturbance response chain. This process is repeated iteratively to finally output the complete dynamic response map of the entire target region within the wheel mud disturbance cycle.

[0123] In this invention, after completing multi-time-domain simulation, time-varying data of the stress field and pore water pressure field of each response unit in the target paddy field area under load were obtained. In order to further identify potential structurally weak areas, stress concentration areas, or abnormal pore pressure response areas in the mud, this invention proposes a risk unit identification method based on coupled peak analysis, and on this basis, constructs a mud weak response fingerprint spectrum P for subsequent visualization, risk assessment, and intelligent avoidance of agricultural operation paths.

[0124] Traditional response analysis typically processes stress field or pore pressure field data separately, lacking dynamic identification methods for the interaction mechanism between the two. However, in paddy field mud disturbance, soil structural instability often occurs in spatiotemporal regions where stress concentration and pore pressure abrupt changes occur simultaneously, such as at the boundary between the topsoil and silt layers or near the groundwater head. Therefore, this invention proposes a dual-channel extreme value tracking algorithm specifically designed to identify stress-pore pressure coupling peak regions.

[0125] Specifically, firstly, a time-series sliding window scan is performed on the stress field σ(x,y,z,t) and pore water pressure field u(x,y,z,t) of each response unit in array W, and the local maxima within each time window (e.g., 5 seconds) are calculated. Let a certain time window be... The local peak value of intermediate stress is The local peak value of pore pressure is If a spatial node is in time Both of the following conditions must be met:

[0126] Greater than the historical average stress value of the unit (such as the previous 10 mud wheel cycles);

[0127] The percentage exceeding the long-term mean of the pore pressure response at that point (e.g., 150%).

[0128] This node is then marked as a coupling peak point, i.e., a "cooperative high-response unit".

[0129] The algorithm runs within multiple time windows, forming a set of coupled peak points C, which is then used as the core input data source for spatial analysis and risk identification.

[0130] After obtaining the set C of coupling peak points, the next step is to identify its spatial distribution characteristics in the simulation area and determine whether it constitutes a high-risk response region. To achieve this goal, this invention introduces a density clustering algorithm (such as DBSCAN or OPTICS) to map the set C of coupling peak points to a three-dimensional coordinate system, and clusters them according to the distance between points and a density threshold to extract continuous high-response sub-regions.

[0131] For each cluster sub-region, the following three types of risk factors are further extracted:

[0132] Local peak amplitude factor : Reflects the intensity at the peak coupling point, and can be defined as ;

[0133] Response duration factor : Calculate the total duration of the coupling state within this region; the longer the duration, the higher the risk.

[0134] Spatial Concentration Factor The density of response cells within a region is estimated based on point density to assess whether it is a localized hotspot of damage.

[0135] Set multidimensional risk thresholds (e.g.) Critical stress pore pressure product Minimum duration of damage If all three conditions are met (minimum cluster density), then the set of response units corresponding to that region is marked as the high-risk response unit set R.

[0136] This mechanism effectively avoids false alarms caused by single-point disturbances, while preserving the complete spatial characteristics of areas with real risk of destruction.

[0137] To achieve structured management, visualization, and intelligent path optimization of response risks, this invention proposes constructing a set of high-risk response units R into an indexable and quantifiable mud soil weak response fingerprint map P. The map unit index number: each response unit is represented by a number ID, which serves as the basic index key of the map.

[0138] Attribute tag set: Each response unit comes with a set of attribute fields, including ( , , , , Peak occurrence time )wait;

[0139] Spatial coordinate labeling: Each response unit is accompanied by its three-dimensional position coordinates (x, y, z) within the simulation region;

[0140] Visualization style suggestion fields include parameters such as color coding level, transparency, and priority display level in the map, which can be called by GIS systems or 3D visualization platforms.

[0141] The final map P can be exported as a standardized data structure (such as JSON, GeoTIFF, NetCDF, etc.), which can be interfaced with agricultural operation path planning systems to provide risk-avoidance-based path recommendations, or used in agricultural machinery adaptive control systems for disturbance self-feedback optimization strategies.

[0142] In conventional agricultural machinery operation path planning, path generation often takes the shortest distance, the fewest number of coverages, or the maximization of operation efficiency as the objective function, without considering the limitations of soil structure on path selection. This can easily cause further disturbance to high-risk response areas, exacerbating soil structure damage and hydraulic channel disorder.

[0143] This invention uses the set of high-risk response units R marked in the risk map P as the core avoidance basis, and proposes a path optimization strategy based on the risk map. The specific process is as follows:

[0144] Reconstruct the spatial location set of high-risk units in map P into a two-dimensional or three-dimensional grid region;

[0145] The parameters of agricultural machinery operation equipment (such as working width, wheel track, turning radius, etc.) are used to construct the operation trajectory model;

[0146] Based on the intersection of the trajectory model and the coordinates of the weak response region, an impassable region mask is generated.

[0147] The minimum weak response risk path algorithm is executed throughout the entire operating area, using the response risk function as the path cost function, and its value can be expressed as:

[0148] The cost of a path is equal to the sum of the integrals of the risk level function R(x,y) along the path.

[0149] Where R(x,y) represents the response risk level corresponding to location (x,y) of map P. The final selected path combination is the mud-wheel operation path with the ability to avoid high-risk areas.

[0150] Agricultural work rhythm is typically a combination of parameters such as travel speed, wheel track interval, and load input frequency, which has a significant impact on soil disturbance intensity. This invention constructs a disturbance response hysteresis function for work rhythm regulation by using the duration of coupled peak values ​​and rhythm sensitivity labels recorded in graph P.

[0151] The disturbance response hysteresis function is defined as follows: ;

[0152] in Indicates the duration of peak stress. Indicates the duration of peak pore pressure. and These represent the peak increments of stress and pore pressure under local disturbance. Based on the function output, the time window required for soil recovery after the current disturbance is determined.

[0153] If the current operating rhythm causes the time for the next wheel track to enter a certain area to be less than the recovery window output by the hysteresis function, the system will automatically reduce the speed, increase the wheel spacing, or adjust the misalignment interval between the front and rear wheels to avoid the disruptive cycle of "reloading before the mud has recovered".

[0154] This rhythm control mechanism introduces the soil response time window into agricultural rhythm parameter scheduling for the first time, which improves operational stability while reducing repeated disturbances and structural softening caused by rhythm resonance during operations.

[0155] Besides the path and rhythm, the pressure parameters applied to the soil during agricultural machinery operations (such as tire ground pressure, tillage depth, and application intensity) are direct factors affecting soil structural stability and pore water pressure accumulation. In traditional operations, pressure parameters are set by the equipment manufacturer or manually, without considering dynamic feedback from the soil.

[0156] This invention combines the pore pressure surge risk factor and the distribution of shear modulus weakening regions in spectrum P to design a pressure parameter adjustment suggestion mechanism based on spectrum features:

[0157] If a work path will pass through a weak response area, and the pore pressure response factor in that area is higher than a set threshold (e.g., pore pressure fluctuation coefficient > 1.5), it is recommended to reduce the tire inflation pressure by a certain percentage.

[0158] If the modulus attenuation rate in the weak response region is greater than the threshold (e.g., shear modulus decreases by more than 20%), shallow tillage is recommended and the peak value of uniaxial load should be limited.

[0159] If a region with strong hysteresis is detected, it is recommended to adopt an intermittent perturbation strategy to reduce pore pressure accumulation by intermittent pressurization.

[0160] Through the above adjustment strategies, agricultural machinery can adjust its force application behavior according to the actual response status of the plot, realize intelligent pressure control, and greatly reduce structural damage, subsidence or slippage caused by mismatch in pressure input.

[0161] Example 2, please refer to Figure 2 As shown in this embodiment, the paddy field soil dynamic response analysis system based on multi-sensor fusion includes:

[0162] The data-driven modeling module constructs a multi-layer information fusion network model M that integrates surface stress disturbance signals, pore water pressure variation trends, shear wave propagation velocity, and soil-water coupling nonlinear indices to characterize the dynamic evolution characteristics of multiple physical properties of soil.

[0163] The multi-source sensor acquisition module simultaneously acquires mud and soil response data from different depths and spatial locations in the target paddy field area. The mud and soil response data includes surface acceleration, pore water pressure, shear modulus perturbation spectrum, and time domain signal of reflected wave.

[0164] The response modeling module inputs the soil response data into a multi-layer information fusion network model M, performs feature decoupling and mutual information optimization processing, and obtains a highly correlated response feature subset T.

[0165] The response simulation module constructs a dynamic finite response unit for mud and soil, and inverts the soil physical field excitation sequence under multi-layer coupled boundary conditions based on the physical characteristics in the feature subset T.

[0166] The response analysis module applies a wheel mud load function to the mud dynamic finite response unit and performs multi-time-domain parallel response analysis on the target area to obtain time-varying stress and pore pressure response data in each response unit.

[0167] The response map generation module identifies the stress-pore pressure coupling peak region and its spatial distribution curve in the response analysis results, extracts high-risk response units, and forms a soil weak response map P.

[0168] The feedback module proposes suggestions for optimizing the mud wheel path, adjusting the operation rhythm, and setting pressure parameters based on the mud weak response map P, thereby realizing dynamic pre-response assessment and operation parameter optimization before mud wheel operation in paddy fields.

[0169] 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. A method for analyzing the dynamic response of paddy field mud and soil based on multi-sensor fusion, characterized in that: include: A multi-layer information fusion network model M, which integrates surface stress disturbance signals, pore water pressure variation trends, shear wave propagation velocity, and soil-water coupling nonlinear indices, is constructed to characterize the dynamic evolution characteristics of multiple physical properties of mud. Simultaneously collect mud and soil response data from different depths and spatial locations in the target paddy field area. The mud and soil response data includes surface acceleration, pore water pressure, shear modulus perturbation spectrum, and time domain signal of reflected wave. The soil response data is input into a multi-layer information fusion network model M, and feature decoupling and mutual information optimization are performed to obtain a feature subset T. The spatial attributes in the feature subset T are mapped to the simulation region grid, and the similarity between adjacent simulation region grids is calculated. If the similarity is greater than a preset threshold, it is marked as a muddy dynamic finite response unit. A dynamic finite response unit for mud and soil is constructed, and a soil physical field excitation sequence under multi-layer coupled boundary conditions is generated based on the physical features in the feature subset T to drive the dynamic finite response unit for mud and soil. A wheel mud load function is applied to the mud dynamic finite response unit, and multi-time-domain parallel response analysis is performed on the target area to obtain time-varying stress and pore pressure response data in each response unit. Identify the stress-pore pressure coupling peak region and its spatial distribution curve in the response analysis results, extract high-risk response units, and form a soil weak response map P. Based on the weak response map P of mud and soil, suggestions are made for optimizing the mud rotation path, adjusting the operation rhythm, and setting pressure parameters, so as to realize dynamic pre-response assessment and operation parameter optimization before mud rotation operation in paddy fields.

2. The method for dynamic response analysis of paddy field mud and soil based on multi-sensor fusion according to claim 1, characterized in that: Training sample data of different physical quantities are collected in the paddy field area. The physical quantities include stress disturbance signal, pore water pressure change, shear wave propagation velocity and nonlinear coupling index. The various sample data are aligned according to time synchronization to form a multi-scale input sequence. A multi-layer information fusion model M is constructed through graph neural network, where nodes represent sensing types and edges represent physical coupling relationships.

3. The method for analyzing the dynamic response of paddy field mud and soil based on multi-sensor fusion according to claim 2, characterized in that: Multiple types of sensor nodes are deployed in the target paddy field area according to a preset grid; surface acceleration, underground pore pressure, shear wave and reflected wave data from different depths are collected; the original mud and soil response data are denoised and normalized to form a multidimensional dataset.

4. The method for dynamic response analysis of paddy field mud and soil based on multi-sensor fusion according to claim 3, characterized in that: Input the multidimensional dataset into the multilayer information fusion network model M and extract the coupled feature tensor; The maximum correlation and minimum redundancy strategy is adopted to screen highly correlated response variables and form a feature subset T.

5. The method for dynamic response analysis of paddy field mud and soil based on multi-sensor fusion according to claim 1, characterized in that: Dual-channel extreme value tracking is performed on the stress-pore pressure response data output by each unit; spatial nodes where local peak values ​​of stress and pore pressure occur simultaneously are marked; highly coupled peak nodes are mapped to spatial distribution curves to extract high-risk response units; A weak response map P of mud soil was constructed using the high-risk response unit number as the index and the coupling strength as the weight.

6. The method for dynamic response analysis of paddy field mud and soil based on multi-sensor fusion according to claim 5, characterized in that: A path avoidance map is generated for high-risk response units in the weak response map P of the muddy soil; the operation trajectory is generated using the minimum risk path algorithm with the risk level of the map as the path weight.

7. The method for dynamic response analysis of paddy field mud and soil based on multi-sensor fusion according to claim 1, characterized in that: The multi-layer information fusion network model M adopts the concept of graph convolutional networks, and the node state update formula is: ; This means that in the l-th layer of the network, the new feature vector of node u is equal to the weighted sum of the feature vectors of its neighboring nodes. This means that in the (l+1)th layer of the network, the new feature vector of node u is equal to the weighted sum of the feature vectors of its neighboring nodes; This indicates that the similarity is calculated from the features of node v and node u. Let represent the trainable weight matrix of the l-th layer, and σ be the activation function.

8. A paddy field soil dynamic response analysis system based on multi-sensor fusion, used to implement the paddy field soil dynamic response analysis method based on multi-sensor fusion as described in any one of claims 1-7, characterized in that: include: The data-driven modeling module constructs a multi-layer information fusion network model M that integrates surface stress disturbance signals, pore water pressure variation trends, shear wave propagation velocity, and soil-water coupling nonlinear indices to characterize the dynamic evolution characteristics of multiple physical properties of soil. The multi-source sensor acquisition module simultaneously acquires mud and soil response data from different depths and spatial locations in the target paddy field area. The mud and soil response data includes surface acceleration, pore water pressure, shear modulus perturbation spectrum, and reflected wave time domain signal. The response modeling module inputs the soil response data into a multi-layer information fusion network model M, performs feature decoupling and mutual information optimization processing, and obtains a highly correlated response feature subset T. The spatial attributes in the feature subset T are mapped to the simulation region grid, and the similarity between adjacent simulation region grids is calculated. If the similarity is greater than a preset threshold, it is marked as a muddy dynamic finite response unit. The response simulation module constructs a dynamic finite response unit for mud and generates a soil physical field excitation sequence under multi-layer coupled boundary conditions based on the physical features in the feature subset T, which is used to drive the dynamic finite response unit for mud. The response analysis module applies a wheel mud load function to the mud dynamic finite response unit and performs multi-time-domain parallel response analysis on the target area to obtain time-varying stress and pore pressure response data in each response unit. The response map generation module identifies the stress-pore pressure coupling peak region and its spatial distribution curve in the response analysis results, extracts high-risk response units, and forms a soil weak response map P. The feedback module proposes suggestions for optimizing the mud wheel path, adjusting the operation rhythm, and setting pressure parameters based on the mud weak response map P, thereby realizing dynamic pre-response assessment and operation parameter optimization before mud wheel operation in paddy fields.

Images