Intelligent optimization method of farmland irrigation water volume based on deep learning

Abstract

The present invention discloses a deep learning-based intelligent optimization method for farmland irrigation water volume. The method comprises the following steps: S1. obtaining a preprocessed farmland multimodal perception dataset; S2. constructing and training a multi-scale gated SIREN continuous moisture content prediction model using the preprocessed farmland multimodal perception dataset as input; S3. dividing the planned irrigation cycle into several time slices, defining a water volume decision vector space, and constructing a differential evolution multi-objective water volume optimization model; S4. outputting a Pareto optimal water volume decision vector set that satisfies multi-objective constraints, and selecting the optimal farmland irrigation water volume from these vectors based on user weights or preset rules; and S5. transmitting the optimal farmland irrigation water volume to an intelligent valve control system, which drives the solenoid valve to irrigate according to the time slice. This invention implements a fully closed-loop intelligent irrigation prediction-decision-execution-feedback mechanism centered on SIREN, achieving significant water-saving and yield-increasing benefits and project deployability.

Description

Technical Field

[0001] The present invention relates to the field of agricultural technology, and in particular to a method for intelligent optimization of farmland irrigation water volume based on deep learning. Background Art

[0002] With the rapid development of smart agriculture technologies, farmland irrigation is gradually evolving from traditional experience-driven to data-driven and model-driven precision control. As a core component of agricultural water use, the quality of irrigation strategies directly impacts crop yields, water resource utilization efficiency, and energy consumption. However, existing irrigation control systems suffer from technical bottlenecks such as insufficient characterization of soil moisture spatial distribution, limited irrigation decision-making optimization capabilities, and the lack of a multi-objective dynamic control framework. These bottlenecks severely hinder the development of efficient, water-saving irrigation systems.

[0003] At present, mainstream farmland irrigation strategies usually rely on tensiometer single-point sensor equipment or irrigation threshold models based on empirical rules to judge the moisture status of the crop root zone. However, such methods are difficult to fully obtain high-frequency moisture gradient changes in the soil layer at a depth of 0–60 cm. In heterogeneous fields with undulating terrain, differences in permeability or shade belt factors, the phenomenon of "over-irrigation in depressions and water shortage on ridge tops" often coexisting in the same area.

[0004] In recent years, some deep learning-based soil moisture modeling methods have been gradually applied to agricultural scenarios, but most models still rely on regular grid division and static data input, making it difficult to reconstruct the soil moisture state with high resolution in continuous space, and are unable to provide a differentiable continuous prediction function interface for use by optimization algorithms; at the same time, some optimization strategies such as traditional genetic algorithms and particle swarm algorithms are prone to falling into local optimality when dealing with high-dimensional nonlinear irrigation water search space, and lack the ability to deeply express multi-objective coupling relationships.

[0005] Therefore, a new method for complex farmland scenarios is urgently needed to solve the core problems of uneven irrigation space, weak optimization capabilities and lack of closed-loop learning mechanism in existing technologies. Summary of the Invention

[0006] One purpose of the present invention is to propose a method for intelligent optimization of farmland irrigation water based on deep learning. The present invention realizes a full-link closed-loop mechanism of intelligent irrigation prediction-decision-execution-feedback with SIREN as the core, which has significant water-saving and yield-increasing benefits and engineering deployability.

[0007] A method for intelligently optimizing farmland irrigation water volume based on deep learning according to an embodiment of the present invention includes the following steps:

[0008] S1. Periodically collect and preprocess a farmland multimodal perception dataset to obtain a preprocessed farmland multimodal perception dataset;

[0009] S2. Using the preprocessed farmland multimodal perception dataset as input, a multi-scale gated SIREN continuous moisture content prediction model was constructed and trained to obtain SIREN model parameters capable of outputting a three-dimensional spatial-temporal continuous moisture content distribution and spatial gradient.

[0010] S3. Divide the planned irrigation cycle into several time slices, define a water quantity decision vector space, and construct a differential evolution multi-objective water quantity optimization model. Each component of the water quantity decision vector corresponds to the farmland irrigation water quantity for a time slice-valve combination. In this differential evolution multi-objective water quantity optimization model, a multi-objective fitness function is constructed based on the multi-scale gated SIREN continuous moisture content prediction model.

[0011] S4. Perform population initialization, differential mutation, crossover recombination, and fitness selection on the differential evolution multi-objective water optimization model. During each fitness evaluation, the candidate water decision vector is injected into the SIREN continuous moisture content prediction service as a conditional vector to obtain the continuous moisture content distribution and spatial gradient and calculate the multi-objective fitness. It is iteratively updated until the termination condition is met. The Pareto optimal water decision vector set that satisfies the multi-objective constraints is output, and the optimal farmland irrigation water volume is selected from the set based on user weights or preset rules.

[0012] S5. Send the optimal farmland irrigation water volume to the intelligent valve control system, driving the solenoid valve to perform irrigation according to the time slice.

[0013] Optionally, the S1 includes the following steps:

[0014] S11. Deploy sensor equipment in the irrigation area to periodically collect field environmental factors. The sensor equipment includes a tensiometer for collecting soil water potential, a frequency domain reflectance moisture sensor for collecting soil moisture content, a near-infrared imaging device for collecting canopy temperature, and a micro-meteorological station for collecting meteorological factors. Sampling environmental information in the irrigation area at a uniform time interval constructs a data set containing N perception data samples d. i The original farmland multimodal perception dataset D raw , the original dataset of farmland multimodal perception includes sampling timestamp t i , the two-dimensional space coordinates x of the sensing point i ,y i , soil water potential ψ i , soil volume moisture content θ i , canopy temperature T i and the meteorological factor vector ξ consisting of temperature, humidity, wind speed and solar radiation i ;

[0015] S12. Time-synchronize the original farmland multimodal perception dataset according to a preset time window, align all types of asynchronously sampled data to a unified time base, and smoothly complete the data at missing time points using a temporal interpolation method. The sampled data are then mapped to regular grid coordinates according to the spatial interpolation grid resolution, and the perception values ​​of the missing spatial locations are estimated using a spatial interpolation algorithm to generate a multimodal spatially completed farmland dataset.

[0016] S13. Perform anomaly elimination processing on the farmland multimodal spatial completion dataset, calculate the deviation ratio of each data component, and determine that the data is abnormal and eliminated when the deviation ratio is greater than the preset deviation threshold. All data items of the farmland multimodal anomaly elimination dataset are converted into a unified physical dimension, and the unit unification processing is completed to obtain the preprocessed farmland multimodal perception dataset D. final .

[0017] Optionally, the S2 includes the following steps:

[0018] S21. Preprocessing farmland multimodal perception dataset D final Add synchronized irrigation depth to each sample Rainfall forecast for the next 24 hours j Compared with the reference evapotranspiration e j , construct the training sample set S train ;

[0019] S22. For each group of plane coordinates x j ,y j and depth coordinate z j The space-time coordinate (x j ,y j ,z j ,t j ) performs Fourier feature mapping to generate a high-frequency vector φ j :

[0020] φ j =[sin(2πB[x j ,y j ,z j ,t j ] T ),cos(2πB[x j ,y j ,z j ,t j ] T )];

[0021] Where B is a fixed random frequency matrix;

[0022] S23. Based on high frequency vector φ j, construct a multi-scale gated SIREN moisture content continuous prediction model:

[0023]

[0024] Among them, f ΘSIREN (φ,p cum ,r,e) represents the multi-scale gated SIREN continuous water content prediction model for predicting the volumetric water content of crop root zone, α l (p cum ,r,e)=σ(a l p cum +b l r+c l e+d l ) is the gating factor, σ(·) is the Sigmoid function, is the parameter set of the model to be trained, represents the subnetwork constructed for the lth soil layer, which is used to model the response characteristics of water content in different soil layers of 0–20 cm, 20–40 cm, and 40–60 cm;

[0025] S24. Soil volume moisture content θ j As the supervision signal, train the multi-scale gated SIREN continuous moisture content prediction model Minimize the model loss function

[0026]

[0027] Among them, λ is the physical constraint regularization weight, K(θ j ) is the hydraulic conductivity function corresponding to the soil volumetric moisture content, M represents the total number of samples used to train the multi-scale gated SIREN continuous moisture content prediction model, represents the time change rate of soil volume moisture content at the jth sample, represents the flow divergence term controlled by soil water potential, and the second term is the physical consistency residual term, which is used to evaluate whether the predicted value conforms to the Richards soil water migration control equation;

[0028] S25. After the multi-scale gated SIREN continuous moisture content prediction model is trained and converged, the mutual information value I(θ,E) between the soil volumetric moisture content prediction results and the unit yield energy consumption E is sorted, and the top K frequency components in the first-layer frequency matrix B are retained to obtain the pruned frequency matrix B. K , and update the model parameters to the compressed model parameter set

[0029] S26. Model parameter set Perform 8-bit weight quantization and insert the MC-Dropout structure to generate the final multi-scale gated SIREN moisture content continuous prediction model G. The calling interface is defined as:

[0030]

[0031] Among them, p opt is a vector element of an optimized water quantity decision, To predict soil volumetric moisture content, is its gradient vector in spatial coordinates, is the prediction standard deviation based on MC-Dropout estimation.

[0032] Optionally, S3 includes the following steps:

[0033] S31. Plan the farmland irrigation cycle T irrig Divide into T non-overlapping time slice sets {τ1,τ2,…,τ T}, the duration of each time slice is Δτ t , satisfying each time slice τ t Corresponding to a solenoid valve control operation, used to perform phased farmland irrigation tasks;

[0034] S32. Set the number of independent electromagnetic valves deployed in the farmland irrigation area to V. Each time slice can control V independent farmland irrigation channels, and construct the farmland irrigation water decision vector. in, represents the time slice τ t The amount of irrigation water for the vth solenoid valve;

[0035] S33. Define upper and lower limits for each electromagnetic valve-time slice combination to satisfy the time slice τ t The amount of irrigation water for the farmland of the vth electromagnetic valve in the time slice τ t Minimum farmland irrigation water volume of the vth solenoid valve Maximum farmland irrigation water volume The upper and lower boundaries are set based on the water supply capacity of the pumping station, the waterlogging tolerance of crops and the soil infiltration threshold;

[0036] S34. Construct a differential evolution multi-objective water optimization model and transform the farmland irrigation water decision vector p opt As the encoding structure of each individual in the differential evolution multi-objective water optimization model, the farmland irrigation water decision vector p optIt contains a combination of T time slices and V valves. Each dimension represents the farmland irrigation water volume of a certain time slice-valve combination. In the differential evolution population, all individuals are initialized within the water volume boundary constraint interval to form the initial population. The number of individuals in the initial population is N. The search space is defined by the upper and lower limits of all T×V groups of valve-controlled water volumes.

[0037] S35. In the differential evolution multi-objective water quantity optimization model, the individuals in each generation of the population are used as the basis, and the mutation vector is generated through the differential mutation operation. The differential mutation operation relies on the water quantity difference between three different individuals. After the mutation, each individual is subjected to the set crossover probability C. r In the crossover operation, each dimension of the candidate water volume decision vector will randomly select its source between the original individual and the mutation vector, and at least one dimension will come from the mutation vector. After the crossover operation is completed, the offspring water volume decision vector is formed and enters the next round of fitness evaluation.

[0038] S36. In the fitness evaluation phase, the multi-scale gated SIREN moisture content continuous prediction model G is called, and the water content decision vector p is used. opt The farmland irrigation water volume corresponding to each time slice-valve combination is used as one of the model inputs, and the predicted volume moisture content under the water volume configuration is obtained by combining the spatial-temporal position, cumulative irrigation depth, reference evapotranspiration and rainfall forecast. Predicting spatial gradients in volumetric water content and the standard deviation of the predicted volumetric moisture content Predict volumetric water content Predicting spatial gradients in volumetric water content and the standard deviation of the predicted volumetric moisture content Coupled with the four optimization objectives, a multi-objective fitness function F(p opt ):

[0039] F(p opt )=ω1f1+ω2f2+ω3f3+ω4f4;

[0040] Among them, the multi-objective preference weight vector ω = [ω1, ω2, ω3, ω4], f1 is the sum of squares of moisture content deviation, f2 is the surface runoff risk control item, f3 is the transpiration satisfaction maximization item, and f4 is the irrigation energy consumption and total water volume control item.

[0041] Optionally, the moisture content deviation sum of squares term is expressed as the sum of squares of deviations between the target volume moisture content and the predicted volume moisture content;

[0042] The surface runoff risk control item is expressed as the degree of deviation between the predicted spatial gradient of volumetric moisture content and the target moisture uniformity;

[0043] The transpiration satisfaction maximization term is expressed as the moisture content prediction uncertainty corresponding to the standard deviation of the predicted volume moisture content, which is used to guide the optimization search towards the high uncertainty area;

[0044] The irrigation energy consumption and total water volume control items are expressed as comprehensive indicators of irrigation water volume per unit area of ​​farmland and energy consumption per unit output, and are used to balance crop yield, water-saving efficiency and energy consumption.

[0045] Optionally, the S4 includes the following steps:

[0046] S41. Based on the differential evolution multi-objective water quantity optimization model and multi-objective fitness function, the water quantity population is initialized and an intergenerational evolution process is performed. In each generation, differential mutation and crossover recombination operations are used to generate candidate water quantity individuals. The multi-scale gated SIREN continuous water content prediction model is used as a prediction service to obtain water content prediction values, spatial gradients, and uncertainty estimates. Based on these, the multi-objective fitness function calculation and fitness optimization are completed.

[0047] S42. In each iteration, the non-dominated solution set is updated in real time according to the non-dominated judgment principle to construct the Pareto optimal water quantity decision vector set P. Pareto , and record the changing trend of each objective function for convergence monitoring;

[0048] S43. When the number of iterations reaches the maximum generation G or the average improvement value of the objective function of consecutive δ generations is lower than the convergence threshold ∈, the differential evolution multi-objective water quantity optimization model is judged to have met the convergence conditions, the iteration is terminated and the final Pareto solution set P is output. Pareto ;

[0049] S44. According to the multi-objective preference weight vector ω=[ω1,ω2,ω3,ω4] set by the user, or combined with the agricultural expert rules of the crop growth stage, the Pareto optimal water quantity decision vector set P Pareto Select the final optimal farmland irrigation water vector p opt* .

[0050] Optionally, the S6 includes the following steps:

[0051] S61. The optimal farmland irrigation water decision vector The intelligent valve control system is sent to the intelligent valve control system through the communication gateway. The intelligent valve control system contains V electromagnetic valve nodes. Each valve node is in its own time slice τ. t Internal water volume Implement quantitative irrigation control;

[0052] S62. In each time slice τ t Inside, the intelligent valve control system collects the actual irrigation flow Combined with the time slice length Δτ t Calculate actual farmland irrigation water volume Optimal farmland irrigation water Compare and if there is any deviation The flow feedback correction logic is triggered to update the current valve control parameters to offset the deviation, where ∈ q is the threshold;

[0053] S63. After the irrigation cycle ends, re-collect the farmland multimodal perception dataset D new , using the latest farmland multimodal perception dataset D new The measured soil volume moisture content at each space-time point The predicted moisture content output by calling the multi-scale gated SIREN moisture content continuous prediction model G Compare and calculate the residual sequence ε j , and calculate the mean square error of all residual samples. If the residual mean square error ε MSE >δ model , that is, exceeding the model accuracy threshold δ model , then the high-frequency spectrum incremental retraining process of the multi-scale gated SIREN continuous moisture content prediction model is triggered;

[0054] S64. During the high-frequency spectrum incremental retraining process, only the parameter set of the high-frequency activation channels in the multi-scale gated SIREN continuous water content prediction model is used. Fine-tune and retain the first K frequency components based on the current mutual information sorting structure. The updated frequency matrix is ​​expressed as At the same time, a pruning operation is performed to remove low-contribution channels and 8-bit weight quantization is performed to generate an updated set of model parameters.

[0055] S65. Redeploy the updated multi-scale gated SIREN continuous moisture content prediction model to the field edge computing gateway for use in the water fitness evaluation process in the next round of differential evolution multi-objective water optimization model, thereby realizing a continuous closed-loop learning and updating mechanism for farmland irrigation water regulation.

[0056] Optionally, S5 further includes extracting the optimal farmland irrigation water output classification standard based on the distribution characteristics of the Pareto solution set in historical iterations, the model prediction error distribution, and the crop growth stage, and defining farmland irrigation water recommendation rules under different control modes:

[0057] Drought recovery type: If the average residual of the past three days is negative and the evaporation is ≥ 120% of the daily average historical value, the upper limit of the water component of each time slice will be increased by 10%;

[0058] Water-saving stable type: If the absolute value of the residual is less than the threshold and the prediction uncertainty is Under the premise of maintaining the target transpiration satisfaction constraint, the total farmland irrigation water volume shall be compressed to no more than 15%;

[0059] Energy consumption constraint type: If the energy consumption index per unit output is more than 20% higher than the historical average, some moisture uniformity indicators will be sacrificed according to the target priority, and the optimal solution priority sorting strategy will be switched from ω2>ω3 to ω4>ω1.

[0060] The beneficial effects of the present invention are:

[0061] (1) Based on the traditional SIREN network, the present invention introduces a multi-scale hierarchical structure and an irrigation-meteorological gating mechanism to construct a continuous moisture content prediction model that can perceive the response characteristics of soil at different depths in a hierarchical manner. By integrating the control residual of the Richards equation as a physical consistency term during the model training process, the model can not only perform high-frequency reconstruction in space-time coordinates, but also has physical interpretation capabilities, effectively avoiding prediction drift. Compared with traditional neural networks without prior knowledge, the proposed model has a lower mean square error in moisture content prediction in heterogeneous soil scenarios, and the spatial gradient continuity is significantly improved. It can be used as a stable fitness evaluator for the optimization algorithm.

[0062] (2) The present invention designs a differential evolution optimization structure with prediction results as a feedback closed loop, injects the irrigation water decision vector into the SIREN model, dynamically obtains the soil moisture prediction, spatial gradient and prediction variance corresponding to each individual, and constructs a multi-objective fitness function that integrates the four objectives. It meets the multiple optimization needs under different crops and climatic conditions through adjustable weights, significantly improving the scientificity and flexibility of irrigation decision-making.

[0063] (3) The present invention designs a closed-loop feedback control structure with an intelligent valve control system as the execution body. The water volume execution results of the solenoid valve are collected during the actual irrigation cycle, and the prediction residual of the SIREN model is calculated in real time in combination with the latest field perception data. When the residual mean square exceeds the threshold, the incremental retraining mechanism with the high-frequency channel as the core is automatically triggered to ensure that the model maintains stable performance in the face of extreme climate change or sudden changes in soil conditions. In combination with pruning and quantization compression strategies, the updated model can be redeployed in the MCU-level edge gateway to ensure the continuous linearity, lightweight and high efficiency of the field prediction and control process. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0065] Figure 1This is a flow chart of an intelligent optimization method for farmland irrigation water based on deep learning proposed by the present invention. DETAILED DESCRIPTION

[0066] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0067] refer to Figure 1 , a method for intelligent optimization of farmland irrigation water based on deep learning, comprising the following steps:

[0068] S1. Periodically collect and preprocess a farmland multimodal perception dataset to obtain a preprocessed farmland multimodal perception dataset;

[0069] S2. Using the preprocessed farmland multimodal perception dataset as input, a multi-scale gated SIREN continuous moisture content prediction model was constructed and trained to obtain SIREN model parameters capable of outputting a three-dimensional spatial-temporal continuous moisture content distribution and spatial gradient.

[0070] S3. Divide the planned irrigation cycle into several time slices, define a water quantity decision vector space, and construct a differential evolution multi-objective water quantity optimization model. Each component of the water quantity decision vector corresponds to the farmland irrigation water quantity for a time slice-valve combination. In this differential evolution multi-objective water quantity optimization model, a multi-objective fitness function is constructed based on the multi-scale gated SIREN continuous moisture content prediction model.

[0071] S4. Perform population initialization, differential mutation, crossover recombination, and fitness selection on the differential evolution multi-objective water optimization model. During each fitness evaluation, the candidate water decision vector is injected into the SIREN continuous moisture content prediction service as a conditional vector to obtain the continuous moisture content distribution and spatial gradient and calculate the multi-objective fitness. It is iteratively updated until the termination condition is met. The Pareto optimal water decision vector set that satisfies the multi-objective constraints is output, and the optimal farmland irrigation water volume is selected from the set based on user weights or preset rules.

[0072] S5. Send the optimal farmland irrigation water volume to the intelligent valve control system, driving the solenoid valve to perform irrigation according to the time slice.

[0073] In this embodiment, S1 includes the following steps:

[0074] S11. Deploy sensor equipment in the irrigation area to periodically collect field environmental factors. The sensor equipment includes a tensiometer for collecting soil water potential, a frequency domain reflectance moisture sensor for collecting soil moisture content, a near-infrared imaging device for collecting canopy temperature, and a micro-meteorological station for collecting meteorological factors. Sampling environmental information in the irrigation area at a uniform time interval constructs a data set containing N perception data samples d. i The original farmland multimodal perception dataset D raw , the original farmland multimodal perception dataset includes the sampling timestamp t i , the two-dimensional space coordinates x of the sensing point i ,y i , soil water potential ψ i , soil volume moisture content θ i , canopy temperature T i and the meteorological factor vector ξ consisting of temperature, humidity, wind speed and solar radiation i ;

[0075] S12. Time-synchronize the original farmland multimodal perception dataset according to a preset time window, align all types of asynchronously sampled data to a unified time base, and smoothly complete the data at missing time points using a temporal interpolation method. The sampled data are then mapped to regular grid coordinates according to the spatial interpolation grid resolution, and the perception values ​​of the missing spatial locations are estimated using a spatial interpolation algorithm to generate a multimodal spatially completed farmland dataset.

[0076] S13. Perform anomaly elimination processing on the farmland multimodal spatial completion dataset, calculate the deviation ratio of each data component, and determine that the data is abnormal and eliminated when the deviation ratio is greater than the preset deviation threshold. All data items of the farmland multimodal anomaly elimination dataset are converted into a unified physical dimension, and the unit unification processing is completed to obtain the preprocessed farmland multimodal perception dataset D. final .

[0077] In this embodiment, S2 includes the following steps:

[0078] S21. Preprocessing farmland multimodal perception dataset D final Add synchronized irrigation depth to each sample Rainfall forecast for the next 24 hours j Compared with the reference evapotranspiration e j , construct the training sample set S train ;

[0079] S22. For each group of plane coordinates x j ,y j and depth coordinate z j The space-time coordinate (x j ,y j,z j ,t j ) performs Fourier feature mapping to generate a high-frequency vector φ j :

[0080] φ j =[sin(2πB[x j ,y j ,z j ,t j ] T ),cos(2πB[x j ,y j ,z j ,t j ] T )];

[0081] Where B is a fixed random frequency matrix;

[0082] S23. Based on high frequency vector φ j , construct a multi-scale gated SIREN moisture content continuous prediction model:

[0083]

[0084] in, represents the multi-scale gated SIREN continuous water content prediction model for predicting the volumetric water content of crop root zone, α l (p cum ,r,e)=σ(a l p cum +b l r+c l e+d l ) is the gating factor, σ(·) is the Sigmoid function, is the parameter set of the model to be trained, represents the subnetwork constructed for the lth soil layer, which is used to model the response characteristics of water content in different soil layers of 0–20 cm, 20–40 cm, and 40–60 cm;

[0085] S24. Soil volume moisture content θ j As the supervision signal, train the multi-scale gated SIREN continuous moisture content prediction model Minimize the model loss function

[0086]

[0087] Among them, λ is the physical constraint regularization weight, K(θ j ) is the hydraulic conductivity function corresponding to the soil volumetric moisture content, M represents the total number of samples used to train the multi-scale gated SIREN continuous moisture content prediction model, represents the time change rate of soil volume moisture content at the jth sample, represents the flow divergence term controlled by soil water potential, and the second term is the physical consistency residual term, which is used to evaluate whether the predicted value conforms to the Richards soil water migration control equation;

[0088] S25. After the multi-scale gated SIREN continuous moisture content prediction model is trained and converged, the mutual information value I(θ,E) between the soil volumetric moisture content prediction results and the unit yield energy consumption E is sorted, and the top K frequency components in the first-layer frequency matrix B are retained to obtain the pruned frequency matrix B. K , and update the model parameters to the compressed model parameter set

[0089] S26. Model parameter set Perform 8-bit weight quantization and insert the MC-Dropout structure to generate the final multi-scale gated SIREN moisture content continuous prediction model G. The calling interface is defined as:

[0090]

[0091] Among them, p opt is a vector element of an optimized water quantity decision, To predict soil volumetric moisture content, is its gradient vector in spatial coordinates, is the prediction standard deviation based on MC-Dropout estimation.

[0092] In this embodiment, S3 includes the following steps:

[0093] S31. Plan the farmland irrigation cycle T irrig Divide into T non-overlapping time slice sets {τ1,τ2,…,τ T}, the duration of each time slice is Δτ t , satisfying each time slice τ t Corresponding to a solenoid valve control operation, used to perform phased farmland irrigation tasks;

[0094] S32. Set the number of independent electromagnetic valves deployed in the farmland irrigation area to V. Each time slice can control V independent farmland irrigation channels, and construct the farmland irrigation water decision vector. in, represents the time slice τ t The amount of irrigation water for the vth solenoid valve;

[0095] S33. Define upper and lower limits for each electromagnetic valve-time slice combination to satisfy the time slice τ tThe amount of irrigation water for the farmland of the vth electromagnetic valve in the time slice τ t Minimum farmland irrigation water volume of the vth solenoid valve Maximum farmland irrigation water volume The upper and lower boundaries are set based on the water supply capacity of the pumping station, the waterlogging tolerance of crops and the soil infiltration threshold;

[0096] S34. Construct a differential evolution multi-objective water optimization model and transform the farmland irrigation water decision vector p opt As the encoding structure of each individual in the differential evolution multi-objective water optimization model, the farmland irrigation water decision vector p opt It contains a combination of T time slices and V valves. Each dimension represents the farmland irrigation water volume of a certain time slice-valve combination. In the differential evolution population, all individuals are initialized within the water volume boundary constraint interval to form the initial population. The number of individuals in the initial population is N. The search space is defined by the upper and lower limits of all T×V groups of valve-controlled water volumes.

[0097] S35. In the differential evolution multi-objective water quantity optimization model, the individuals in each generation of the population are used as the basis, and the mutation vector is generated through the differential mutation operation. The differential mutation operation relies on the water quantity difference between three different individuals. After the mutation, each individual is subjected to the set crossover probability C. r In the crossover operation, each dimension of the candidate water volume decision vector will randomly select its source between the original individual and the mutation vector, and at least one dimension will come from the mutation vector. After the crossover operation is completed, the offspring water volume decision vector is formed and enters the next round of fitness evaluation.

[0098] S36. In the fitness evaluation phase, the multi-scale gated SIREN moisture content continuous prediction model G is called, and the water content decision vector p is used. opt The farmland irrigation water volume corresponding to each time slice-valve combination is used as one of the model inputs, and the predicted volume moisture content under the water volume configuration is obtained by combining the spatial-temporal position, cumulative irrigation depth, reference evapotranspiration and rainfall forecast. Predicting spatial gradients in volumetric water content and the standard deviation of the predicted volumetric moisture content Predict volumetric water content Predicting spatial gradients in volumetric water content and the standard deviation of the predicted volumetric moisture content Coupled with the four optimization objectives, a multi-objective fitness function F(p opt ):

[0099] F(p opt )=ω1f1+ω2f2+ω3f3+ω4f4;

[0100] Among them, the multi-objective preference weight vector ω = [ω1, ω2, ω3, ω4], f1 is the sum of squares of moisture content deviation, f2 is the surface runoff risk control item, f3 is the transpiration satisfaction maximization item, and f4 is the irrigation energy consumption and total water volume control item.

[0101] In this embodiment, the moisture content deviation sum of squares term is expressed as the sum of squares of the deviations between the target volumetric moisture content and the predicted volumetric moisture content;

[0102] The surface runoff risk control item is expressed as the degree of deviation between the predicted spatial gradient of volumetric moisture content and the target moisture uniformity;

[0103] The transpiration satisfaction maximization term is expressed as the uncertainty of moisture content prediction corresponding to the standard deviation of the predicted volume moisture content, which is used to guide the optimization search towards the area with high uncertainty;

[0104] Irrigation energy consumption and total water volume control items are expressed as comprehensive indicators of irrigation water volume per unit area of ​​farmland and energy consumption per unit output, which are used to balance crop yield, water-saving efficiency and energy consumption.

[0105] In this embodiment, the square sum of moisture content deviation is:

[0106]

[0107] Among them, θ target is the target volumetric moisture content, representing the optimal moisture level required by the crop at the current growth stage; f1 represents the overall cumulative error of the moisture content of each valve control area deviating from the target value, which measures the irrigation accuracy;

[0108] In this implementation, the surface runoff risk control items are:

[0109]

[0110] in, is the vertical component of the spatial gradient, indicating the degree of water increase within a unit depth; γ runoff is the surface runoff threshold, exceeding which will trigger the runoff risk, and f2 measures the potential waterlogging or water overload risk;

[0111] In this implementation, the transpiration satisfaction maximization item is:

[0112]

[0113] Among them, e t represents the reference evapotranspiration, representing the theoretical water demand intensity of crops; ∈ is a very small positive number used to prevent division by zero; f3 represents the water supply level under unit reference evapotranspiration, and minimizing its negative value is equivalent to the fullest water supply;

[0114] In this implementation, the irrigation energy consumption and total water volume control items are:

[0115]

[0116] Among them, α v is the energy consumption coefficient required for irrigation per unit water volume of the vth valve channel, which is obtained from field energy consumption measurement; f4 measures the energy consumption intensity during the irrigation process of the total pumping station.

[0117] In this embodiment, S4 includes the following steps:

[0118] S41. Based on the differential evolution multi-objective water quantity optimization model and multi-objective fitness function, the water quantity population is initialized and an intergenerational evolution process is performed. In each generation, differential mutation and crossover recombination operations are used to generate candidate water quantity individuals. The multi-scale gated SIREN continuous water content prediction model is used as a prediction service to obtain water content prediction values, spatial gradients, and uncertainty estimates. Based on these, the multi-objective fitness function calculation and fitness optimization are completed.

[0119] S42. In each iteration, the non-dominated solution set is updated in real time according to the non-dominated judgment principle to construct the Pareto optimal water quantity decision vector set P. Pareto , and record the changing trend of each objective function for convergence monitoring;

[0120] S43. When the number of iterations reaches the maximum generation G or the average improvement value of the objective function of consecutive δ generations is lower than the convergence threshold ∈, the differential evolution multi-objective water quantity optimization model is judged to have met the convergence conditions, the iteration is terminated and the final Pareto solution set P is output. Pareto ;

[0121] S44. According to the multi-objective preference weight vector ω=[ω1,ω2,ω3,ω4] set by the user, or combined with the agricultural expert rules of the crop growth stage, the Pareto optimal water quantity decision vector set P Pareto Select the final optimal farmland irrigation water vector p opt* .

[0122] In this embodiment, S6 includes the following steps:

[0123] S61. The optimal farmland irrigation water decision vector The intelligent valve control system is sent to the intelligent valve control system through the communication gateway. The intelligent valve control system contains V electromagnetic valve nodes. Each valve node is in its own time slice τ. t Internal water volume Implement quantitative irrigation control;

[0124] S62. In each time slice τ t Inside, the intelligent valve control system collects the actual irrigation flow Combined with the time slice length Δτ t Calculate actual farmland irrigation water volume Optimal farmland irrigation water Compare and if there is any deviation The flow feedback correction logic is triggered to update the current valve control parameters to offset the deviation, where ∈ q is the threshold;

[0125] S63. After the irrigation cycle ends, re-collect the farmland multimodal perception dataset D new , using the latest farmland multimodal perception dataset D new The measured soil volume moisture content at each space-time point The predicted moisture content output by calling the multi-scale gated SIREN moisture content continuous prediction model G Compare and calculate the residual sequence ε j , and calculate the mean square error of all residual samples. If the residual mean square error ε MSE >δ model , that is, exceeding the model accuracy threshold δ model , then the high-frequency spectrum incremental retraining process of the multi-scale gated SIREN continuous moisture content prediction model is triggered;

[0126] S64. During the high-frequency spectrum incremental retraining process, only the parameter set of the high-frequency activation channels in the multi-scale gated SIREN continuous water content prediction model is used. Fine-tune and retain the first K frequency components based on the current mutual information sorting structure. The updated frequency matrix is ​​expressed as At the same time, a pruning operation is performed to remove low-contribution channels and 8-bit weight quantization is performed to generate an updated set of model parameters.

[0127] S65. Redeploy the updated multi-scale gated SIREN continuous moisture content prediction model to the field edge computing gateway for use in the water fitness evaluation process in the next round of differential evolution multi-objective water optimization model, thereby realizing a continuous closed-loop learning and updating mechanism for farmland irrigation water regulation.

[0128] In this embodiment, S5 also includes extracting the optimal farmland irrigation water output classification standard based on the distribution characteristics of the Pareto solution set in historical iterations, the distribution of model prediction errors, and the crop growth stage, and defining the farmland irrigation water recommendation rules under different control modes:

[0129] Drought recovery type: If the average residual of the past three days is negative and the evaporation is ≥ 120% of the daily average historical value, the upper limit of the water component of each time slice will be increased by 10%;

[0130] Water-saving stable type: If the absolute value of the residual is less than the threshold and the prediction uncertainty is Under the premise of maintaining the target transpiration satisfaction constraint, the total farmland irrigation water volume shall be compressed to no more than 15%;

[0131] Energy consumption constraint type: If the energy consumption index per unit output is more than 20% higher than the historical average, some moisture uniformity indicators will be sacrificed according to the target priority, and the optimal solution priority sorting strategy will be switched from ω2>ω3 to ω4>ω1.

[0132] Example 1:

[0133] At 06:00 on the morning of August 6, 2024, a 9.6-hectare corn field in Site A entered its third irrigation window. The irrigation target is summer corn from the end of jointing to the beginning of tasseling. The area is divided into 16 solenoid valve control units, each covering approximately 0.6 hectares. To perform precise irrigation tasks, the farm manager activated the system of the present invention deployed on the field LoRa edge gateway.

[0134] At 6:05 AM, the system automatically collected the latest data from 64 FDR moisture sensors, 32 tensiometers, 16 infrared canopy temperature monitoring nodes, and one meteorological microstation. In one block, sensor "node-fdr-03" returned the following record: timestamp: 2024-08-06 06:00; spatial coordinates: (X=112.1371, Y=36.3758); soil volumetric moisture content: 0.177 cm 3 / cm 3 Soil water potential: -38 kPa; canopy temperature: 31.2°C; surface wind speed: 2.3 m / s; air humidity: 61%; solar radiation intensity: 673 W / m 2 .

[0135] After the system completes data collection and anomaly elimination (excluding the abnormal drift data from "node-tension-11"), unified time alignment and performs spatial interpolation, it automatically constructs the latest round of perception dataset D at 06:12. final .

[0136] Subsequently, the SIREN continuous moisture content prediction model G began to work, performing high-frequency predictions for the target prediction window (06:00 on August 6 to 06:00 on August 7, a total of 24 hours). The spatial coordinates (112.1371, 36.3758, 0.35) and the cumulative irrigation depth of 31.4 mm, the predicted rainfall of 2.1 mm, and the reference evapotranspiration of 6.3 mm were used as input. The model's predicted value at this point was:

[0137] Moisture content spatial gradient Prediction uncertainty

[0138] Based on the spatial average results of 179 prediction points in the entire field, the system assessed that the proportion of areas with moisture content below the lower limit of corn water requirement reached 38.2%. At the same time, spatial gradient mutations occurred in multiple areas, indicating that there was an obvious heterogeneous distribution phenomenon of "ridge top drought-depression waterlogging" within the field.

[0139] The differential evolution multi-objective water flow optimization module is started. The system sets the optimization dimension to 192 (12 time slices × 16 valves), and the objective function weight vector is set to:

[0140] Moisture content deviation weight ω1 = 0.35; moisture content spatial gradient weight ω2 = 0.25; uncertainty induced search weight ω3 = 0.15; unit output energy consumption weight ω4 = 0.25;

[0141] The system began DE optimization with an initial population size of 50, a crossover probability of 0.9, and a maximum number of iterations of 80. Each generation of candidate solutions was fed into Model G for a full 24-hour forecast, and the resulting fitness was used to evolve the population. After convergence at the 67th generation, the system output 16 Pareto-optimal solutions and automatically selected the optimal solution using the agronomist's predefined "water conservation priority at the end of jointing" strategy.

[0142] Finally, the system outputs the optimal water volume vector p opt* , and issue the following for execution (excerpt from the first four valves):

[0143] Table 1 Valve execution time slice according to optimal water flow vector

[0144] Valve number Time Slice 1 Time Slice 2 Time Slice 3 ... Time Slice 12 V-01 22.0mm 20.0mm 19.5mm ... 0.0mm V-02 14.0mm 18.0mm 18.5mm ... 0.0mm V-03 0.0mm 0.0mm 6.5mm ... 0.0mm V-04 16.5mm 20.0mm 21.0mm ... 1.0mm

[0145] At 08:00, irrigation officially began. The system monitored irrigation progress in real time based on valve status and flow meter readings. At 08:27, the system detected that the flow rate deviation of "V-03" in time slice 2 exceeded ±12% (the target was 0, but the actual measurement was 2.4mm). It automatically implemented feedback control, forcing the water volume in subsequent time slices to zero.

[0146] The irrigation cycle was completed at 20:00. At 21:00, the post-irrigation data was collected and combined with the model prediction value to calculate the residual mean square error of the whole field, which was 0.026cm 3 / cm 3 , exceeding the threshold of 0.020. The system automatically triggered SIREN's high-frequency channel incremental retraining, updated the model parameters, and compressed it into an edge deployment version, which took 11 minutes.

[0147] Table 2 is the comparison results between the method of the present invention and the traditional tensiometer threshold irrigation method

[0148] index Method of the present invention Traditional tensiometer threshold irrigation method Total water volume for a single irrigation <![CDATA[14160m 3 ]]> <![CDATA[17720m 3 ]]> Average corn yield 661kg / mu 612kg / mu Spatial standard deviation of moisture content 0.0098 0.0175 Average error of moisture content prediction 0.021 No modeling Valve control flow execution average deviation ±4.1％ Unmonitorable Model update cycle Automatically after each irrigation No update mechanism Optimizing calculation time 18 minutes No optimization module

[0149] This Example 1 proves that in a situation where the spatial heterogeneity of fields is significant and the water distribution is complex, the present invention can effectively avoid the problem of "partial over-irrigation and partial under-irrigation", achieve stable crop yields, save water and reduce energy consumption per unit output, and has significant practicality and engineering promotion value.

[0150] Based on the traditional SIREN network, the present invention introduces a multi-scale hierarchical structure and an irrigation-meteorological gating mechanism to construct a continuous moisture content prediction model that can hierarchically perceive the response characteristics of soil at different depths. By integrating the control residual of the Richards equation as a physical consistency term during the model training process, the model can not only perform high-frequency reconstruction in space-time coordinates, but also has physical interpretation capabilities, effectively avoiding prediction drift. Compared with traditional neural networks without prior knowledge, the proposed model has a lower mean square error in moisture content prediction in heterogeneous soil scenarios, and significantly improves the continuity of spatial gradients. It can be used as a stable fitness evaluator for the optimization algorithm.

[0151] The present invention designs a differential evolution optimization structure with prediction results as a feedback closed loop. The irrigation water decision vector is injected into the SIREN model, and the soil moisture prediction, spatial gradient and prediction variance corresponding to each individual are dynamically obtained. A multi-objective fitness function integrating the four objectives is constructed. Through adjustable weights, it meets the multiple optimization needs under different crops and climatic conditions, significantly improving the scientificity and flexibility of irrigation decision-making.

[0152] The present invention designs a closed-loop feedback control structure with an intelligent valve control system as the executor. It collects the water volume execution results of the solenoid valve during the actual irrigation cycle, and calculates the SIREN model prediction residual in real time based on the latest field perception data. When the residual mean square exceeds the threshold, the incremental retraining mechanism with the high-frequency channel as the core is automatically triggered to ensure that the model maintains stable performance in the face of extreme climate changes or sudden changes in soil conditions. Combined with pruning and quantization compression strategies, the updated model can be redeployed in the MCU-level edge gateway, ensuring the continuous linearity, lightweight and high efficiency of the field prediction and control process.

[0153] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for intelligent optimization of farmland irrigation water volume based on deep learning, characterized in that: The steps include: S1. Periodically collect and preprocess a farmland multimodal perception dataset to obtain a preprocessed farmland multimodal perception dataset; S2. Using the preprocessed farmland multimodal perception dataset as input, a multi-scale gated SIREN continuous moisture content prediction model was constructed and trained to obtain SIREN model parameters capable of outputting a three-dimensional spatial-temporal continuous moisture content distribution and spatial gradient. The S2 comprises the following steps: S21. Preprocessing farmland multimodal perception dataset D final Add synchronized irrigation depth to each sample Rainfall forecast for the next 24 hours j Compared with the reference evapotranspiration e j , construct the training sample set S train ; S22. For each group of plane coordinates x j ,y j and depth coordinate z j The space-time coordinate (x j ,y j ,z j ,t j ) performs Fourier feature mapping to generate a high-frequency vector φ j ; S23. Based on high frequency vector φ j , construct a multi-scale gated SIREN moisture content continuous prediction model: in, represents the multi-scale gated SIREN continuous water content prediction model for predicting the volumetric water content of crop root zone, α l (p cum ,r,e)=σ(a l p cum +b l r+c l e+d l ) is the gating factor, σ(·) is the Sigmoid function, is the parameter set of the model to be trained, represents the subnetwork constructed for the lth soil layer, which is used to model the response characteristics of water content in different soil layers of 0–20 cm, 20–40 cm, and 40–60 cm; S24. Soil volume moisture content θ j As the supervision signal, train the multi-scale gated SIREN continuous moisture content prediction model Minimize the model loss function S25. After the multi-scale gated SIREN continuous moisture content prediction model is trained and converged, the mutual information value I(θ,E) between the soil volumetric moisture content prediction results and the unit yield energy consumption E is sorted, and the top K frequency components in the first-layer frequency matrix B are retained to obtain the pruned frequency matrix B. K , and update the model parameters to the compressed model parameter set S26. Model parameter set Perform 8-bit weight quantization processing and insert the MC-Dropout structure to generate the final multi-scale gated SIREN moisture content continuous prediction model G; S3. Divide the planned irrigation cycle into several time slices, define a water quantity decision vector space, and construct a differential evolution multi-objective water quantity optimization model. Each component of the water quantity decision vector corresponds to the farmland irrigation water quantity for a time slice-valve combination. In this differential evolution multi-objective water quantity optimization model, a multi-objective fitness function is constructed based on the multi-scale gated SIREN continuous moisture content prediction model. S4. Perform population initialization, differential mutation, crossover recombination, and fitness selection on the differential evolution multi-objective water optimization model. During each fitness evaluation, the candidate water decision vector is injected into the SIREN continuous moisture content prediction service as a conditional vector to obtain the continuous moisture content distribution and spatial gradient and calculate the multi-objective fitness. It is iteratively updated until the termination condition is met. The Pareto optimal water decision vector set that satisfies the multi-objective constraints is output, and the optimal farmland irrigation water volume is selected from the set based on user weights or preset rules. S5. Send the optimal farmland irrigation water volume to the intelligent valve control system, driving the solenoid valve to perform irrigation according to the time slice.

2. The method for intelligent optimization of farmland irrigation water volume based on deep learning according to claim 1, characterized in that: Said S1 comprises the following steps: S11. Deploy sensor equipment in the irrigation area to periodically collect field environmental factors. The sensor equipment includes a tensiometer for collecting soil water potential, a frequency domain reflectance moisture sensor for collecting soil moisture content, a near-infrared imaging device for collecting canopy temperature, and a micro-meteorological station for collecting meteorological factors. Sampling environmental information in the irrigation area at a uniform time interval constructs a data set containing N perception data samples d. i The original farmland multimodal perception dataset D raw , the original dataset of farmland multimodal perception includes sampling timestamp t i , the two-dimensional space coordinates x of the sensing point i ,y i , soil water potential ψ i , soil volume moisture content θ i , canopy temperature T i and the meteorological factor vector ξ consisting of temperature, humidity, wind speed and solar radiation i ; S12. Time-synchronize the original farmland multimodal perception dataset according to a preset time window, align all types of asynchronously sampled data to a unified time base, and smoothly complete the data at missing time points using a temporal interpolation method. The sampled data are then mapped to regular grid coordinates according to the spatial interpolation grid resolution, and the perception values ​​of the missing spatial locations are estimated using a spatial interpolation algorithm to generate a multimodal spatially completed farmland dataset. S13. Perform anomaly elimination processing on the farmland multimodal spatial completion dataset, calculate the deviation ratio of each data component, and determine that the data is abnormal and eliminated when the deviation ratio is greater than the preset deviation threshold. All data items of the farmland multimodal anomaly elimination dataset are converted into a unified physical dimension, and the unit unification processing is completed to obtain the preprocessed farmland multimodal perception dataset D. final .

3. The method for intelligent optimization of farmland irrigation water volume based on deep learning according to claim 1, characterized in that: The minimization model loss function for: Among them, λ is the physical constraint regularization weight, K(θ j ) is the hydraulic conductivity function corresponding to the soil volumetric moisture content, M represents the total number of samples used to train the multi-scale gated SIREN continuous moisture content prediction model, represents the time change rate of soil volume moisture content at the jth sample, The first term represents the flow divergence term controlled by soil water potential, and the second term is the physical consistency residual term, which is used to evaluate whether the predicted value conforms to the Richards soil water migration control equation.

4. The method for intelligent optimization of farmland irrigation water volume based on deep learning according to claim 1, characterized in that: The S3 includes the following steps: S31. Plan the farmland irrigation cycle T irrig Divide into T non-overlapping time slice sets {τ1,τ2,…,τ T }, the duration of each time slice is Δτ t , satisfying each time slice τ t Corresponding to a solenoid valve control operation, used to perform phased farmland irrigation tasks; S32. Set the number of independent electromagnetic valves deployed in the farmland irrigation area to V. Each time slice can control V independent farmland irrigation channels, and construct the farmland irrigation water decision vector. in, represents the time slice τ t The amount of irrigation water for the vth solenoid valve; S33. Define upper and lower limits for each electromagnetic valve-time slice combination to satisfy the time slice τ t The amount of irrigation water for the farmland of the vth electromagnetic valve in the time slice τ t Minimum farmland irrigation water volume of the vth solenoid valve Maximum farmland irrigation water volume The upper and lower boundaries are set based on the water supply capacity of the pumping station, the waterlogging tolerance of crops and the soil infiltration threshold; S34. Construct a differential evolution multi-objective water optimization model and transform the farmland irrigation water decision vector p opt As the encoding structure of each individual in the differential evolution multi-objective water optimization model, the farmland irrigation water decision vector p opt It contains a combination of T time slices and V valves. Each dimension represents the farmland irrigation water volume of a certain time slice-valve combination. In the differential evolution population, all individuals are initialized within the water volume boundary constraint interval to form the initial population. The number of individuals in the initial population is N. The search space is defined by the upper and lower limits of all T×V groups of valve-controlled water volumes. S35. In the differential evolution multi-objective water quantity optimization model, the individuals in each generation of the population are used as the basis, and the mutation vector is generated through the differential mutation operation. The differential mutation operation relies on the water quantity difference between three different individuals. After the mutation, each individual is subjected to the set crossover probability C. r In the crossover operation, each dimension of the candidate water volume decision vector will randomly select its source between the original individual and the mutation vector, and at least one dimension will come from the mutation vector. After the crossover operation is completed, the offspring water volume decision vector is formed and enters the next round of fitness evaluation. S36. In the fitness evaluation phase, the multi-scale gated SIREN moisture content continuous prediction model G is called, and the water content decision vector p is used. opt The farmland irrigation water volume corresponding to each time slice-valve combination is used as one of the model inputs, and the predicted volume moisture content under the water volume configuration is obtained by combining the spatial-temporal position, cumulative irrigation depth, reference evapotranspiration and rainfall forecast. Predicting spatial gradients in volumetric water content and the standard deviation of the predicted volumetric moisture content Predict volumetric water content Predicting spatial gradients in volumetric water content and the standard deviation of the predicted volumetric moisture content Coupled with the four optimization objectives, a multi-objective fitness function F(p opt ): F(p opt )=ω1f1+ω2f2+ω3f3+ω4f4; Among them, the multi-objective preference weight vector ω = [ω1, ω2, ω3, ω4], f1 is the sum of squares of moisture content deviation, f2 is the surface runoff risk control item, f3 is the transpiration satisfaction maximization item, and f4 is the irrigation energy consumption and total water volume control item.

5. The method for intelligent optimization of farmland irrigation water volume based on deep learning according to claim 4, characterized in that: The moisture content deviation sum of squares term is expressed as the sum of squares of the deviations between the target volume moisture content and the predicted volume moisture content; The surface runoff risk control item is expressed as the degree of deviation between the predicted spatial gradient of volumetric moisture content and the target moisture uniformity; The transpiration satisfaction maximization term is expressed as the moisture content prediction uncertainty corresponding to the standard deviation of the predicted volume moisture content, which is used to guide the optimization search towards the high uncertainty area; The irrigation energy consumption and total water volume control items are expressed as comprehensive indicators of irrigation water volume per unit area of ​​farmland and energy consumption per unit output, and are used to balance crop yield, water-saving efficiency and energy consumption.

6. The method for intelligent optimization of farmland irrigation water volume based on deep learning according to claim 4, characterized in that: The S4 comprises the following steps: S41. Based on the differential evolution multi-objective water quantity optimization model and multi-objective fitness function, the water quantity population is initialized and an intergenerational evolution process is performed. In each generation, differential mutation and crossover recombination operations are used to generate candidate water quantity individuals. The multi-scale gated SIREN continuous water content prediction model is used as a prediction service to obtain water content prediction values, spatial gradients, and uncertainty estimates. Based on these, the multi-objective fitness function calculation and fitness optimization are completed. S42. In each iteration, the non-dominated solution set is updated in real time according to the non-dominated judgment principle to construct the Pareto optimal water quantity decision vector set P. Pareto , and record the changing trend of each objective function for convergence monitoring; S43. When the number of iterations reaches the maximum generation G or the average improvement value of the objective function of consecutive δ generations is lower than the convergence threshold ∈, the differential evolution multi-objective water quantity optimization model is judged to have met the convergence conditions, the iteration is terminated and the final Pareto solution set P is output. Pareto ; S44. According to the multi-objective preference weight vector ω=[ω1,ω2,ω3,ω4] set by the user, or combined with the agricultural expert rules of the crop growth stage, the Pareto optimal water quantity decision vector set P Pareto Select the final optimal farmland irrigation water vector p opt* .

7. The method for intelligent optimization of farmland irrigation water volume based on deep learning according to claim 6, characterized in that: The S5 comprises the following steps: S51. The optimal farmland irrigation water decision vector The intelligent valve control system is sent to the intelligent valve control system through the communication gateway. The intelligent valve control system contains V electromagnetic valve nodes. Each valve node is in its own time slice τ. t Internal water volume Implement quantitative irrigation control; S52. In each time slice τ t Inside, the intelligent valve control system collects the actual irrigation flow Combined with the time slice length Δτ t Calculate actual farmland irrigation water volume Optimal farmland irrigation water Compare and if there is any deviation The flow feedback correction logic is triggered to update the current valve control parameters to offset the deviation, where ∈ q is the threshold; S53. After the irrigation cycle ends, re-collect the farmland multimodal perception dataset D new , using the latest farmland multimodal perception dataset D new The measured soil volume moisture content at each space-time point The predicted moisture content output by calling the multi-scale gated SIREN moisture content continuous prediction model G Compare and calculate the residual sequence ε j , and calculate the mean square error of all residual samples. If the residual mean square error ε MSE >δ model , that is, exceeding the model accuracy threshold δ model , then the high-frequency spectrum incremental retraining process of the multi-scale gated SIREN continuous moisture content prediction model is triggered; S54. During the high-frequency spectrum incremental retraining process, only the parameter set of the high-frequency activation channels in the multi-scale gated SIREN continuous water content prediction model is used. Fine-tune and retain the first K frequency components based on the current mutual information sorting structure. The updated frequency matrix is ​​expressed as At the same time, a pruning operation is performed to remove low-contribution channels and 8-bit weight quantization is performed to generate an updated set of model parameters. S55. Redeploy the updated multi-scale gated SIREN continuous moisture content prediction model to the field edge computing gateway for use in the water fitness evaluation process in the next round of differential evolution multi-objective water optimization model, thereby realizing a continuous closed-loop learning and updating mechanism for farmland irrigation water regulation.

8. The method for intelligent optimization of farmland irrigation water volume based on deep learning according to claim 7, characterized in that: The S5 also includes extracting the optimal farmland irrigation water output classification standard based on the distribution characteristics of the Pareto solution set in historical iterations, the distribution of model prediction errors, and the crop growth stage, and defining the farmland irrigation water recommendation rules under different control modes: Drought recovery type: If the average residual of the past three days is negative and the evaporation is ≥ 120% of the daily average historical value, the upper limit of the water component of each time slice will be increased by 10%; Water-saving stable type: If the absolute value of the residual is less than the threshold and the prediction uncertainty is Under the premise of maintaining the target transpiration satisfaction constraint, the total farmland irrigation water volume shall be compressed to no more than 15%; Energy consumption constraint type: If the energy consumption index per unit output is more than 20% higher than the historical average, some moisture uniformity indicators will be sacrificed according to the target priority, and the optimal solution priority sorting strategy will be switched from ω2>ω3 to ω4>ω1.

Images