Intelligent fertilizing method with self-adaptive fertilizer stirring capability

Abstract

An intelligent fertilizing method with self-adaptive fertilizer stirring capability comprises a self-adaptive fertilizer stirring method and a fertilizer deliquescence prevention method, and the self-adaptive fertilizer stirring method comprises the following steps: (1) carrying out fusion preprocessing on sensor data, and monitoring fertilizer information; (2) performing physical information diffusion control to realize uniform mixing of the fertilizer; (3) carrying out element learning spectrum decoupling; chaos vibration is suppressed, and the service life of equipment is prolonged; (5) carrying out multi-mode control closed loop integration; (6) field deployment and adaptive optimization; the method for preventing the deliquescence of the fertilizer comprises the following steps: (1) establishing a multi-mode deliquescence risk prediction model; (2) dynamic deliquescence threshold value self-adaption; (3) humidity gradient driving stirring; (4) deliquescence inhibition efficiency evaluation; and (5) stirring synergism. The self-adaptive change of the stirring blades in the fertilizer box of the fertilizer applicator can be realized when the fertilizer is stirred, the stirring efficiency is remarkably improved, the fertilizer deliquescence caused by high environment humidity and the like when the fertilizer is stored in the fertilizer box is avoided, and the fertilizer box has a remarkable effect on black land protection.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent fertilization for black soil protection, and particularly to an intelligent fertilization method with adaptive fertilizer stirring ability. Background Art

[0002] In modern agricultural production, as a key device for improving fertilizer utilization rate and promoting crop growth, the performance and intelligent level of fertilizer applicators have an important impact on agricultural production efficiency. However, currently, the fertilizer applicators on the market generally have problems such as the lack of a fertilizer stirring device or insufficient stirring performance, resulting in uneven distribution of fertilizers in the fertilizer tank, affecting the fertilization effect; in addition, fertilizers are prone to deliquescence due to moisture when placed in the fertilizer tank for a long time, reducing fertilizer efficiency and possibly causing corrosion and damage to the fertilizer applicator. These problems are more prominent in humid or rainy seasons, bringing no small trouble to agricultural production.

[0003] Therefore, it is particularly important to develop an intelligent fertilization method with adaptive fertilizer stirring ability. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide, in view of the deficiencies of the prior art, an intelligent fertilization method with adaptive fertilizer stirring ability, which can achieve precise stirring and uniform fertilization of fertilizers through the combination of advanced sensor technology, intelligent control algorithms, and optimized fertilizer stirring device design, and can also perform adaptive adjustment according to conditions such as fertilizer types and humidity, effectively coping with the problem of fertilizer deliquescence, and providing a more efficient and reliable fertilization solution for agricultural production.

[0005] The technical problem to be solved by the present invention is realized through the following technical solutions. The present invention is an intelligent fertilization method with adaptive fertilizer stirring ability, including an adaptive fertilizer stirring method:

[0006] (1) Perform preprocessing on sensor data fusion to monitor fertilizer information

[0007] Use the Kalman filter fusion formula to integrate sensor data of spectrum, weight, and vibration, eliminate noise, and unify spatio-temporal alignment. Specifically:

[0008] Spatio-temporal alignment, that is, Kalman filter fusion:

[0009]

[0010] Posterior state estimation, used as the fused sensor data;

[0011] Prior state estimation at time t, used as the prediction based on the (t - 1) moment;

[0012] K t: Kalman gain, used to balance the weights of prediction and observation;

[0013] y t : Original sensor data vector, including spectrum + weight + vibration;

[0014] H t : Observation matrix, used to dynamically adjust the weights of each sensor;

[0015] (2) Physical information diffusion control to achieve uniform mixing of fertilizers

[0016] Incorporate Fick's diffusion law into PID control using the diffusion-constrained PID formula, and accelerate the calculation of the Laplacian operator for a 50×50 grid through DSP;

[0017] Diffusion-constrained PID formula:

[0018]

[0019] u(t): Control output signal;

[0020] K p : Proportion;

[0021] K i : Integral;

[0022] K d : Differential gain;

[0023] e(t): Error, i.e., the difference between the set value and the actual value;

[0024] λ: Diffusion term weight coefficient;

[0025] The diffusion term is discretized into a grid gradient operation, D is the diffusion coefficient, is the concentration gradient;

[0026] (3) Meta-learning spectral decoupling

[0027] Decouple the N / P / K spectral features through a small number of samples for rapid adaptation to new fertilizers;

[0028] Spectral decoupling loss function:

[0029]

[0030] Loss function, used to train the spectral decoupling model;

[0031] W i : Weight matrix, which is the weight of different spectral features;

[0032] SA i : Decoupled spectral features;

[0033] A: Spectral basis matrix;

[0034] c i : Actual composition reference value;

[0035] β: Regularization coefficient;

[0036] A: Spectral basis matrix;

[0037] L: Laplacian matrix;

[0038] (4) Suppress chaotic vibration and improve equipment life

[0039] Use real-time calculation of Lyapunov exponent and quantum genetic optimization to generate anti-resonance blade trajectories and reduce mechanical losses;

[0040] Real-time calculation of Lyapunov exponent:

[0041]

[0042] λ LLE : Lyapunov exponent, used to measure the chaos degree of the system;

[0043] x k : State of the system at time k;

[0044] δx k-1 : Small perturbation at time k - 1;

[0045] Chromosome coding of quantum genetic optimization, 10 genes correspond to the blade angular velocity at 10 time points,

[0046] Gene: Chromosome coding, solution represented by quantum superposition state;

[0047] θ i : Angular parameter of the i-th gene;

[0048] (5) Closed-loop integration of multi-modal control

[0049] Integrate each algorithm module for real-time and efficient control;

[0050] (6) Field deployment and adaptive optimization

[0051] Continuously improve the system performance through online learning to adapt to different field environments, and use Bayesian parameter optimization and digital twin calibration technology to continuously adjust and optimize system parameters;

[0052] Bayesian parameter optimization:

[0053] The objective function is:

[0054] f(θ) = ω1·Efficiency + ω2·Energy + ω3·Wear

[0055] Model the parameter space using Gaussian processes and automatically adjust every 24 hours;

[0056] Digital twin calibration:

[0057] Establish a virtual stirring model:

[0058]

[0059] x: System state variable;

[0060] u: Control input;

[0061] ∈: Noise or model error;

[0062] Modify the model parameters by comparing the actual and simulation data.

[0063] The technical problem to be solved by the present invention can also be further realized by the following technical solutions. For the intelligent fertilization method with adaptive manure stirring ability described above, the data delay in step (1) < 15 ms, and the signal-to-noise ratio improvement ≥ 20 dB.

[0064] The technical problem to be solved by the present invention can also be further realized by the following technical solutions. For the intelligent fertilization method with adaptive manure stirring ability described above, the coefficient of variation of mixing uniformity in step (2) < 5%, and the energy consumption reduction ≥ 18%.

[0065] The technical problem to be solved by the present invention can also be further realized by the following technical solutions. For the intelligent fertilization method with adaptive manure stirring ability described above, the number of new fertilizer samples adapted in step (3) ≤ 5, and the decoupling interpretability R 2 > 0.85.

[0066] The technical problem to be solved by the present invention can also be further realized by the following technical solutions. For the intelligent fertilization method with adaptive manure stirring ability described above, the vibration acceleration in step (4) < 0.3 mm / s 2 .

[0067] The technical problem to be solved by the present invention can also be further realized by the following technical solutions. For the intelligent fertilization method with adaptive manure stirring ability described above, the control frequency in step (5) ≥ 100 Hz.

[0068] The technical problem to be solved by the present invention can also be further realized by the following technical solutions. For the intelligent fertilization method with adaptive manure stirring ability described above, the method further includes a method for preventing fertilizer deliquescence:

[0069] (1) Multi-modal deliquescence risk prediction model

[0070] Integrate humidity, temperature, and fertilizer component data through a spatio-temporal convolutional gated network to predict the probability of deliquescence;

[0071]

[0072] σ: Sigmoid function, representing the output probability;

[0073] Conv1d: One-dimensional convolution, used to process the humidity sequence;

[0074] GRU: Gated Recurrent Unit, used to process the temperature sequence;

[0075] W: Weight matrix;

[0076] H: Humidity sequence;

[0077] T: Temperature sequence;

[0078] C: Fertilizer component vector;

[0079] Tensor concatenation;

[0080] (2) Dynamic deliquescence threshold adaptation

[0081] Automatically adjust the humidity alarm threshold according to different fertilizer types through a composition-dependent threshold formula:

[0082]

[0083] H threshold : Adaptive humidity threshold;

[0084] H0: Basic threshold;

[0085] α: Adjustment coefficient;

[0086] S hygroscopic : Deliquescence score;

[0087] ω i : Deliquescence coefficient of each component;

[0088] C i : Content or concentration of the i-th component in the fertilizer;

[0089] (3) Humidity gradient-driven stirring

[0090] Install a multi-point humidity sensor array and independently speed-controlled partition stirring blades in the fertilizer bin to intelligently adjust the stirring intensity according to the humidity distribution in the bin;

[0091] Control formula:

[0092]

[0093] ω i : Rotational speed of the i-th zone;

[0094] ω base : Base rotational speed;

[0095] K h , K g Gain coefficient of humidity deviation and gradient;

[0096] H i : Humidity value of the i-th zone;

[0097] Average humidity;

[0098] Humidity gradient;

[0099] (4) Evaluation of deliquescence inhibition efficiency

[0100] Online evaluate the stirring effect through the deliquescence inhibition factor and dynamically adjust the strategy;

[0101]

[0102] CDIF: Evaluation of deliquescence inhibition effect, the smaller the value, the better;

[0103] H(t): Real-time humidity;

[0104] H threshold : Deliquescence threshold;

[0105] H max : Maximum humidity;

[0106] T: Evaluation time window;

[0107] (5) Stirring coordination

[0108] The parameters are self-adaptive to achieve the effect of stirring coordination. The higher the humidity, the higher the frequency to break larger droplets.

[0109] Compared with the prior art, the present invention can achieve the self-adaptive change of the stirring blades in the fertilizer tank of the fertilizer applicator when stirring fertilizers, significantly improve the stirring efficiency, and avoid the deliquescence of fertilizers caused by high environmental humidity and other situations when storing fertilizers in the fertilizer tank, which has remarkable effects on the protection of black soil. Specific implementation mode

[0110] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0111] An intelligent fertilization method with adaptive fertilizer stirring ability, including an adaptive fertilizer stirring method and a method for preventing fertilizer deliquescence,

[0112] I. The steps of the adaptive fertilizer stirring method are as follows:

[0113] Step 1: Preprocessing of multi-sensor data fusion to monitor fertilizer information

[0114] Use the Kalman filter fusion formula to integrate sensor data such as spectrum, weight, and vibration, eliminate noise, and unify spatio-temporal alignment, which can reduce data latency and improve the signal-to-noise ratio;

[0115] The verification indicators are data latency, i.e., the timestamp comparison < 15ms, and the improvement of the signal-to-noise ratio, i.e., calculating the variance ratio of the original / fused data ≥ 20dB;

[0116] Spatio-temporal alignment (Kalman filter fusion):

[0117]

[0118] Posterior state estimation (fused sensor data);

[0119] Prior state estimation at time t (prediction based on the t-1 moment);

[0120] K t : Kalman gain (balancing the weights of prediction and observation);

[0121] y t : Original sensor data vector (spectrum + weight + vibration);

[0122] H t : Observation matrix (dynamically adjusting the weights of each sensor);

[0123] Step 2: Physical information diffusion control (PIDM)

[0124] Use the diffusion-constrained PID formula to incorporate Fick's law of diffusion into PID control to achieve efficient and energy-saving stirring, and accelerate the calculation of the Laplacian operator for a 50×50 grid through DSP;

[0125] The verification indicators are that the coefficient of variation CV of multi-point sampling for mixing uniformity is < 5%, and the reduction in energy consumption, with the current integral compared to traditional PID, is ≥ 18%;

[0126] Diffusion-constrained PID formula:

[0127]

[0128] u(t): Control output signal (stirring intensity);

[0129] K p : Proportion;

[0130] K i : Integral;

[0131] K d : Differential gain;

[0132] e(t): Error (difference between set value and actual value);

[0133] λ: Diffusion term weight coefficient;

[0134] The diffusion term is discretized into a grid gradient operation (D is the diffusion coefficient, is the concentration gradient);

[0135] Step 3: Meta-Spectral decoupling

[0136] Decouple the N / P / K spectral features with a small number of samples to achieve fast adaptation to new fertilizers;

[0137] The verification indicators are that the time to adapt to new fertilizers, i.e., the number of samples required to reach 90% accuracy, is ≤ 5, and the decoupling interpretability, i.e., the correlation coefficient R with the standard spectral library 2 > 0.85;

[0138] Spectral decoupling loss function:

[0139]

[0140] Loss function (used to train the spectral decoupling model);

[0141] W i : Weight matrix (weights of different spectral features);

[0142] SA i : Decoupled spectral features;

[0143] A: Spectral basis matrix (requiring meta-learning);

[0144] c i : Actual composition reference value (such as N / P / K content);

[0145] β: Regularization coefficient;

[0146] A: Spectral basis matrix (meta - learning optimization);

[0147] L: Laplacian matrix (forcing spectral sparsity of different components);

[0148] Step 4: Chaos vibration suppression (Chaos - OPT)

[0149] Use Lyapunov exponents to calculate in real - time and generate anti - resonance blade trajectories with quantum genetic optimization to reduce mechanical losses;

[0150] The verification index is that the vibration acceleration RMS < 0.3mm / s 2 , and the bearing life is extended by ≥ 25%;

[0151] Real - time calculation of Lyapunov exponents:

[0152]

[0153] λ LLE : Lyapunov exponent (measuring the chaos degree of the system);

[0154] x k : The state of the system at time k;

[0155] δx k-1 : The small perturbation at time k - 1;

[0156] Chromosome encoding of quantum genetic optimization (10 genes corresponding to the blade angular velocity at 10 time points);

[0157]

[0158] Gene: Chromosome encoding (representation of solutions in quantum superposition states);

[0159] θ i : The angular parameter of the i - th gene;

[0160] Step 5: Multi - modal control closed - loop integration;

[0161] Integrate each algorithm module to achieve real - time control;

[0162] The verification index is that the control frequency ≥ 100Hz and the system stability is high, that is, it runs continuously for 72 hours without failure;

[0163] Step 6: Field deployment and adaptive optimization

[0164] Continuously improve the system performance through online learning;

[0165] The verification metrics are long-term energy efficiency improvement, i.e., a continuous 2% decrease in the monthly average energy consumption comparison, and the fault self-diagnosis rate, i.e., the automatic recognition accuracy of abnormal events ≥ 90%;

[0166] Bayesian parameter optimization:

[0167] The objective function is:

[0168] f(θ) = ω1·Efficiency + ω2·Energy + ω3·Wear

[0169] θ: Parameters to be optimized (such as control parameters);

[0170] ω1, ω2, ω3: Weight coefficients (trade-off between efficiency, energy consumption, and wear);

[0171] Use Gaussian process to model the parameter space and automatically adjust every 24 hours;

[0172] Digital twin calibration:

[0173] Establish a virtual stirring model:

[0174]

[0175] x: System state variables (such as fertilizer mixing state);

[0176] u: Control input (such as stirring speed);

[0177] ∈: Noise or model error;

[0178] Correct the model parameters by comparing the actual and simulation data.

[0179] II. Methods for Preventing Fertilizer Caking

[0180] Step 1: Multimodal Caking Risk Prediction Model

[0181] Through the spatio-temporal convolutional gated network, comprehensively integrate humidity, temperature, and fertilizer composition data to predict the probability of caking;

[0182]

[0183] σ: Sigmoid function (output probability);

[0184] Conv1d: One-dimensional convolution (process humidity sequence);

[0185] GRU: Gated Recurrent Unit (process temperature sequence);

[0186] W: Weight matrix (combine fertilizer composition C);

[0187] H: Humidity sequence;

[0188] T: Temperature sequence;

[0189] C: Fertilizer component vector;

[0190] Tensor splicing;

[0191] Step 2: Dynamic deliquescence threshold adaptation

[0192] Automatically adjust the humidity alarm threshold according to different fertilizer types through the component-dependent threshold formula;

[0193]

[0194] H threshold : Adaptive humidity threshold;

[0195] H0: Base threshold;

[0196] α: Adjustment coefficient;

[0197] S hygroscopic : Deliquescence score (component weighted sum);

[0198] ω i : Deliquescence coefficient of each component;

[0199] C i : Content or concentration of the i-th component in the fertilizer;

[0200] Step 3: Humidity gradient-driven stirring

[0201] Install a multi-point humidity sensor array and independently controllable speed partition stirring blades in the fertilizer bin to intelligently adjust the stirring intensity according to the humidity distribution in the bin;

[0202] Control formula:

[0203]

[0204] Where:

[0205] ω i : Rotation speed of the i-th zone;

[0206] ω base : Base rotation speed;

[0207] K h ,K g Gain coefficient of humidity deviation and gradient;

[0208] H i : Humidity value of the i-th zone;

[0209] Average humidity;

[0210] Humidity gradient;

[0211] Step 4: Evaluation of deliquescence inhibition efficiency

[0212] Online evaluate the stirring effect through the deliquescence inhibition factor (CDIF) and dynamically adjust the strategy;

[0213]

[0214] CEIF: Evaluation of deliquescence inhibition effect (the smaller the value, the better);

[0215] H(t): Real-time humidity;

[0216] H threshold : Deliquescence threshold;

[0217] H max : Maximum humidity;

[0218] T: Evaluation time window;

[0219] Step 5: Stirring coordination

[0220] The parameters are self-adaptive to achieve the effect of stirring coordination. The higher the humidity, the higher the frequency to break larger droplets.

[0221] Among them, the data source is collected by multi-sensors and processed by a microprocessor for meta-learning spectral decoupling, chaotic vibration suppression, field deployment and adaptive optimization, multi-modal deliquescence risk prediction model, dynamic deliquescence threshold adaptation, deliquescence inhibition efficiency evaluation, etc., which can achieve rapid adaptation to new fertilizers, improved system stability, reduced power consumption, increased fault self-diagnosis rate, improved deliquescence risk prediction ability and better stirring coordination effect;

[0222] Functions of each verification index:

[0223] 1. Preprocessing of sensor data fusion

[0224] Verification index:

[0225] Data delay < 15ms: Ensure real-time synchronization of sensor data and avoid control lag caused by delay;

[0226] Signal-to-noise ratio improvement ≥ 20dB: Eliminate noise through Kalman filtering, improve data accuracy, and provide reliable input for subsequent control;

[0227] Function: Ensure the real-time performance and accuracy of data acquisition, which is the basis of adaptive control.

[0228] 2. Physical information diffusion control (PIDM)

[0229] Verification index:

[0230] Coefficient of variation of mixing uniformity CV < 5%: Measure the uniformity of fertilizer component distribution after mixing to ensure consistent fertilization effect;

[0231] Energy consumption reduction ≥ 18%: Compare with traditional PID control to verify the energy-saving effect and reduce long-term usage costs;

[0232] Function: Optimize mixing efficiency and energy consumption, and improve system economy.

[0233] 3. Meta-learning spectral decoupling

[0234] Verification indicators:

[0235] Number of new fertilizer samples adapted ≤ 5: Reduce the learning cost for new fertilizers and improve system flexibility and adaptability;

[0236] Decoupling interpretability R 2 > 0.85: Verify the accuracy of spectral feature separation to ensure a high match with the standard component library;

[0237] Function: Quickly adapt to different fertilizer types and ensure the reliability of component analysis.

[0238] 4. Chaos vibration suppression (Chaos-OPT)

[0239] Verification indicators:

[0240] Vibration acceleration RMS < 0.3mm / s 2 : Reduce the mechanical vibration amplitude and reduce physical damage to the equipment;

[0241] Bearing life extension ≥ 25%: Prolong the service life of key components and reduce maintenance frequency by suppressing resonance;

[0242] Function: Improve the durability and stability of the mechanical system.

[0243] 5. Multi-modal control closed-loop integration

[0244] Verification indicators:

[0245] Control frequency ≥ 100Hz: Ensure that the control system responds quickly to environmental changes and maintains real-time regulation ability;

[0246] Continuous operation for 72 hours without failure: Verify the stability and reliability of the system under high-intensity operation;

[0247] Function: Ensure long-term stable operation of the system and adapt to complex field environments.

[0248] 6. Field deployment and adaptive optimization

[0249] Verification indicators:

[0250] Monthly average energy consumption decreases by ≥ 2%: Continuously optimize energy efficiency through online learning to reduce long-term operating costs;

[0251] Fault self-diagnosis accuracy rate ≥ 90%: Automatically identify abnormal events and reduce the need for manual intervention;

[0252] Function: Achieve system intelligence and self-optimization, and improve user convenience.

[0253] 7. Prediction and control of deliquescence risk

[0254] Verification indicators:

[0255] Minimize the deliquescence inhibition factor (CDIF): Quantify the proportion of time with excessive humidity to evaluate the moisture-proof effect;

[0256] Synergistic effect of humidity gradient-driven stirring: Optimize local humidity control by adjusting the rotation speed in different zones;

[0257] Function: Accurately prevent fertilizer deliquescence and ensure fertilizer storage quality.

[0258] The present invention has also been verified through actual implementation:

[0259] Equipment configuration: Adaptive fertilizer applicator: Integrated with a spectral sensor (detecting fertilizer components), a weight sensor (monitoring the stock in the fertilizer tank), and a vibration sensor (perceiving the stirring state);

[0260] Partitioned stirring device: Install 4 independently speed-controlled stirring blades in the fertilizer tank, equipped with a multi-point humidity sensor array;

[0261] Control unit: Equipped with a DSP chip to run the diffusion-constrained PID algorithm, the meta-learning spectral decoupling model, and the deliquescence risk prediction module;

[0262] Digital twin platform: Synchronize the virtual model through the cloud to calibrate the control parameters in real time.

[0263] Implementation process:

[0264] 1. Equipment initialization and fertilizer loading

[0265] Steps:

[0266] Load compound fertilizer (N-P-K = 20-10-15) into the fertilizer tank, and the system automatically identifies the fertilizer type;

[0267] The spectral sensor scans the fertilizer spectrum, and the meta-learning model calls the pre-trained basis matrix to decouple the N / P / K characteristics (R 2 = 0.89);

[0268] Set the target fertilization amount (200 kg / acre) and humidity threshold (H o= 60% RH, α = 0.3).

[0269] 2. Data Acquisition and Fusion

[0270] Real-time sensor data:

[0271] Spectral data: Detect changes in fertilizer composition (such as spectral shift caused by caking);

[0272] Weight data: Monitor the remaining amount in the fertilizer bin and adjust the feeding speed;

[0273] Vibration data: Capture the working state of the stirring blades;

[0274] Kalman filter fusion:

[0275] Eliminate sensor noise (signal-to-noise ratio increased by 22 dB), synchronize timestamps (delay < 12 ms), and output the fused fertilizer status information.

[0276] 3. Adaptive Stirring Control

[0277] Diffusion-constrained PID algorithm:

[0278] Calculate the stirring intensity based on the fused data;

[0279] Adjust the blade rotation speed in real time to ensure the mixing uniformity (CV = 4.2%), while reducing the energy consumption by 19% (compared with the traditional PID);

[0280] Chaotic vibration suppression:

[0281] Optimize the blade trajectory using the quantum genetic algorithm, vibration acceleration RMS = 0.28 mm / s 2 , bearing life extended by 28%;

[0282] 4. Dynamic Management of Caking Risk

[0283] Caking prediction model:

[0284] Integrate humidity (H = 65% RH), temperature (T = 25°C), and composition (C i = 20 - 10 - 15) data, risk index = 0.72 (threshold 0.6);

[0285] Dynamic response:

[0286] Increase the humidity threshold to H_threshold = 68% RH, increase the blade rotation speed (ω i = 1200 rpm) in high-humidity areas, and break the droplet aggregation;

[0287] CDIF = 0.15 (target < 0.2), effectively suppressing caking.

[0288] 5. Field Adaptive Optimization

[0289] Bayesian parameter optimization:

[0290] Automatically adjust the PID gains (K p , K i , K d ) and the diffusion weight λ every 24 hours. Objective function:

[0291] f(θ) = 0.5·Efficiency + 0.3·Energy + 0.2·Wear

[0292] The monthly energy consumption decreases by 2.3%, and the self-diagnosis accuracy rate of faults is 92%;

[0293] Digital twin calibration:

[0294] Compare the virtual model with the actual data, correct the parameters of the stirring dynamics equation, and improve the control accuracy.

[0295] Through multi-modal intelligent control and adaptive optimization technology, the comprehensive performance of the fertilization system is significantly improved. Specifically:

[0296] Based on the multi-source sensor data fusion method (spectrum, weight, vibration) of Kalman filter, the signal-to-noise ratio is increased by ≥20 dB and the data delay is <15 ms, ensuring the real-time performance and accuracy of fertilizer state monitoring; combined with the physical information diffusion control (PIDM) algorithm of Fick's diffusion law, the coefficient of variation of fertilizer mixing uniformity is optimized to <5%, and at the same time, the system energy consumption is reduced by ≥18% compared with the traditional PID control, with both high-efficiency mixing and energy-saving characteristics.

[0297] In terms of spectral analysis, the meta-learning-driven spectral decoupling model can complete the separation of N / P / K characteristics (the decoupling interpretability R 2 > 0.85) with ≤5 samples, significantly shortening the adaptation period of new fertilizers by more than 50%.

[0298] For mechanical vibration problems, based on the anti-resonance blade trajectory planning technology optimized by quantum genetics and combined with the real-time calculation of Lyapunov exponents, the vibration acceleration is effectively suppressed to <0.3 mm / s 2 , and the service life of the bearing is extended by ≥25%, ensuring the durability of the equipment under complex working conditions.

[0299] In the field of moisture-proof control, the dynamic component-dependent threshold adjustment strategy (such as H0 = 60% RH → 68% RH) and the humidity gradient-driven partition stirring cooperation mechanism (the rotational speed difference reaches 1200 rpm) make the deliquescence inhibition factor (CDIF) <0.2, and the accuracy rate of deliquescence risk prediction exceeds 90%.

[0300] The present invention autonomously updates control parameters every 24 hours through a Bayesian optimization framework, achieving long-term energy efficiency improvement (monthly average energy consumption reduction ≥ 2%) and fault self-diagnosis accuracy ≥ 90%; combined with digital twin calibration technology, the virtual model is dynamically aligned with actual data, and the control accuracy is improved by 10% - 15%. Practical applications show that this technology increases the fertilizer utilization rate by more than 15% and reduces the waste volume by 20% - 30%. While reducing the risk of soil pollution, it provides a solution with both theoretical innovation and engineering practicality for black soil protection and agricultural sustainable development.

Claims

1. An intelligent fertilization method with self-adaptive manure stirring ability, characterized in that: Including an adaptive fertilizer mixing method: (1) Preprocess the sensor data fusion to monitor fertilizer information Use the Kalman filter fusion formula to integrate the sensor data of spectrum, weight, and vibration, eliminate noise and unify the spatio-temporal alignment. Specifically: Spatio-temporal alignment, that is, Kalman filter fusion: A posteriori state estimation for use as fused sensor data; A priori state estimate at time t, used as a prediction based on the time t-1. K t : Kalman gain, which is used to balance the weights of prediction and observation; y t : The original sensor data vector, including spectrum + weight + vibration; H t : Observation matrix, used to dynamically adjust the weights of each sensor; (2) Physical information diffusion control to achieve uniform mixing of fertilizers Integrate Fick's diffusion law into PID control using the diffusion-constrained PID formula, and accelerate the calculation of the Laplacian operator for a 50×50 grid through DSP; Diffusion-constrained PID formula: u(t): Control output signal; K p : Ratio; K i : Integration; K d : Differential gain; e(t): Error, that is, the difference between the set value and the actual value; λ: Diffusion term weight coefficient; The diffusion term is discretized into a grid gradient operation, where D is the diffusion coefficient, and is the concentration gradient; (3) Meta-learning spectral decoupling Decouple the N / P / K spectral features through a small number of samples for rapid adaptation to new fertilizers; Spectral decoupling loss function: A loss function for training a spectral decoupling model; W i : Weight matrix, which is the weight of different spectral features; SA i : Decoupled spectral features; A: Spectral basis matrix; c i : Reference value of actual components; β: Regularization coefficient; A: Spectral basis matrix; L: Graph Laplacian matrix; (4) Chaotic vibration suppression to improve equipment life Use real-time calculation of Lyapunov exponents and quantum genetic optimization to generate anti-resonance blade trajectories to reduce mechanical losses; Real-time calculation of Lyapunov exponents: λ LLE : Lyapunov exponent, which is used to measure the chaos degree of the system; x k : The state of the system at time k; δx k-1 : Small perturbation at time k-1; Chromosome coding optimized by quantum genetics, where 10 genes correspond to the angular velocity of the blade at 10 time points Gene: Chromosome encoding, and the solution is represented by a quantum superposition state; θ i : The angular parameter of the i-th gene; (5) Multi-modal control closed-loop integration Integrate each algorithm module for real-time and efficient control; (6) Field deployment and adaptive optimization Continuously improve the system performance through online learning to adapt to different field environments, and use Bayesian parameter optimization and digital twin calibration technology to continuously adjust and optimize system parameters; Bayesian parameter optimization: The objective function is: f(θ) = ω1·Efficiency + ω2·Energy + ω3·Wear Use Gaussian process to model the parameter space and automatically adjust it every 24 hours; Digital twin calibration: Establish a virtual mixing model: x: System state variable; u: Control input; ∈: Noise or model error; Correct the model parameters by comparing the actual and simulation data.

2. The intelligent fertilization method with adaptive manure stirring ability according to claim 1, wherein: The data delay in step (1) < 15ms, and the signal-to-noise ratio improvement ≥ 20dB.

3. The intelligent fertilization method with adaptive manure stirring ability according to claim 1, characterized in that: The coefficient of variation of the mixing uniformity in step (2) < 5%, and the energy consumption reduction ≥ 18%.

4. The intelligent fertilization method with adaptive manure stirring ability according to claim 1, characterized in that: The number of new fertilizer samples adapted in step (3) ≤ 5, and the decoupled interpretability R 2 > 0.

85.

5. The intelligent fertilization method with adaptive manure stirring ability according to claim 1, characterized in that: The vibration acceleration in step (4) < 0.3mm / s 2 .

6. The intelligent fertilization method with adaptive manure stirring ability according to claim 1, characterized in that: The control frequency in step (5) ≥ 100Hz.

7. The intelligent fertilization method with adaptive manure stirring ability according to claim 1, characterized in that: This method also includes a method for preventing fertilizer deliquescence: (1) Multi-modal deliquescence risk prediction model Comprehensively use humidity, temperature, and fertilizer composition data through a spatio-temporal convolutional gated network to predict the probability of deliquescence; σ: Sigmoid function, representing the output probability; Conv1d: One-dimensional convolution for processing humidity sequences; GRU: Gated recurrent unit for processing temperature sequences; W: Weight matrix; H: Humidity sequence; T: Temperature sequence; C: Fertilizer composition vector; Tensor concatenation; (2) Dynamic deliquescence threshold adaptation Automatically adjust the humidity alarm threshold according to different fertilizer types through the composition-dependent threshold formula: H threshold : Adaptive humidity threshold; H0: Basic threshold; α: Adjustment coefficient; S hygroscopic : Efflorescence rating; ω i : deliquescence coefficient of each component; C i : The content or concentration of the i-th component in the fertilizer; (3) Humidity gradient-driven mixing Install a multi-point humidity sensor array and independently speed-controlled partition mixing blades in the fertilizer bin to intelligently adjust the mixing intensity according to the humidity distribution in the bin; Control formula: ω i : Rotational speed of the i-th zone; ω base : Base rotational speed; K h ,K g Gain coefficient for humidity deviation and gradient H i : humidity value of the i-th area; Average humidity; Humidity gradient; (4) Deliquescence inhibition efficiency evaluation Online evaluate the mixing effect through the deliquescence inhibition factor and dynamically adjust the strategy; CDIF: Deliquescence inhibition effect evaluation, the smaller the value, the better; H(t): Real-time humidity; H threshold : Deliquescence threshold; H max : Maximum humidity; T: Evaluation time window; (5) Stirring coordination Parameter adaptation to achieve the effect of stirring coordination, the higher the humidity, the higher the frequency to break larger droplets.

Images