A zero-carbon water resource comprehensive management system for agricultural irrigation areas

Abstract

The present application relates to the field of water resource management, and particularly relates to a zero-carbon water resource comprehensive management system for agricultural irrigation areas, comprising: an irrigation area basic data calculation module which calculates soil latent heat flux and crop basic evapotranspiration of the irrigation area respectively according to detected air temperature, air pressure, solar radiation and wind speed; an irrigation area data prediction module which generates predicted water storage and utilization amount of the irrigation area through an irrigation area prediction model based on a ConvLSTM architecture; a water demand fusion prediction module which generates comprehensive predicted water demand of the irrigation area through a fusion mapping model based on a gating mechanism and multi-scale convolution; and a zero-carbon water resource management module which calculates predicted net irrigation amount of the irrigation area based on the comprehensive predicted water demand to perform irrigation management on photovoltaic water pumps and photovoltaic valves of the irrigation area. The present application improves the precision of water resource management of the irrigation area, optimizes the scheduling energy consumption of the zero-carbon photovoltaic water pumps and photovoltaic valves, and realizes the optimization of the zero-carbon transformation of the agricultural irrigation area from the head of irrigation energy consumption.

Description

Technical Field

[0001] This invention relates to the field of water resource management, and in particular to a zero-carbon water resource integrated management system for agricultural irrigation areas. Background Technology

[0002] In water resource management in agricultural irrigation areas, with the advancement of "dual-carbon" smart agriculture, the industry has found that the scheduling efficiency of irrigation equipment directly affects water resource utilization efficiency and carbon emission levels. Traditional irrigation areas rely heavily on manual experience or simple timed control for pump and valve scheduling, lacking a coordinated consideration of crop water demand dynamics and weather changes, leading to prominent problems of high energy consumption and high emissions.

[0003] Currently, the irrigation district has introduced clean energy equipment such as photovoltaic water pumps, but there are still significant shortcomings in scheduling and management.

[0004] Photovoltaic water pumps rely on solar power generation for operation. However, solar energy is highly intermittent and fluctuates. If precise prediction and scheduling cannot be combined with the exact water requirements of crops, irrigation interruptions may occur due to insufficient zero-carbon energy supply, affecting crop growth. In other words, precise irrigation based on real-time predicted crop needs can save water, reduce pumping energy consumption, and achieve a precise match between photovoltaic pumping and photovoltaic power generation.

[0005] Therefore, how to improve the water demand forecasting of agricultural irrigation areas in order to achieve precise, energy-saving, zero-carbon photovoltaic water resource management is a technical problem that needs to be solved. Summary of the Invention

[0006] To this end, the present invention provides a zero-carbon water resource integrated management system for agricultural irrigation areas. By integrating soil latent heat flux and basal transpiration using a ConvLSTM model, the system predicts crop water demand, improves the accuracy of water resource management in irrigation areas, reduces waste and imbalance, optimizes the scheduling and energy consumption of zero-carbon emission photovoltaic water pumps and photovoltaic valves, and promotes the zero-carbon transformation of agricultural irrigation areas by optimizing irrigation energy consumption.

[0007] To achieve the above objectives, this invention proposes a zero-carbon water resource integrated management system for agricultural irrigation areas, comprising:

[0008] The irrigation district basic data calculation module is used to calculate the soil latent heat flux and the basic transpiration of crops in the irrigation district based on the detected air temperature, air pressure, solar radiation and wind speed, respectively.

[0009] The irrigation district data prediction module is used to generate predicted water storage and utilization of the irrigation district based on multidimensional meteorological data and crop data and an irrigation district prediction model based on the ConvLSTM architecture.

[0010] The water demand fusion prediction module is used to generate the comprehensive predicted water demand of the irrigation area by using the soil latent heat flux, the basic evapotranspiration and the predicted water storage and utilization through a fusion mapping model based on feature fusion gating mechanism and multi-scale convolution.

[0011] The zero-carbon water resource management module is used to calculate the predicted net irrigation volume of the irrigation area based on the comprehensive predicted water demand, and to manage the photovoltaic water pumps and photovoltaic valves of the irrigation area based on the predicted net irrigation volume.

[0012] Furthermore, the irrigation district prediction model includes a ConvLSTM layer and a feature mapping layer, and the irrigation district data prediction module includes:

[0013] The convolutional gating mechanism generation unit is used to take the multidimensional meteorological data and the crop data as input data, and generate the input gate, forget gate, output gate and candidate state of the updated gating mechanism through the convolution operation of the ConvLSTM layer according to the input data and the hidden state of the previous time step.

[0014] The gated fusion unit is used to generate state updates through an update gating mechanism of the ConvLSTM layer that weights the candidate states based on the input gate and the forget gate weights the states of the previous time step.

[0015] ConvLSTM output unit, used to generate spatiotemporal characteristics of water demand based on the state update and the output gate;

[0016] The crop growth stage enhancement unit is used to encode the spatiotemporal characteristics of water demand and the crop growth stage through a feature mapping layer to generate the predicted water storage and utilization.

[0017] Furthermore, the crop growth stage enhancement unit includes:

[0018] The water requirement feature mapping subunit is used to map the crop growth stages in the irrigation area to the crop growth stage codes through convolutional coding;

[0019] The factor mapping subunit is used to encode the crop growth stages through linear convolution to generate mapping factors and growth stage translation correction factors.

[0020] The prediction subunit is used to adjust the spatiotemporal characteristics of water demand based on the mapping factor and the growth stage translation correction factor to generate the predicted water storage and utilization.

[0021] Furthermore, the predicted water storage and utilization includes the predicted transpiration, and the irrigation district data prediction module further includes:

[0022] The transpiration loss function training unit is used to construct a transpiration loss function based on the predicted transpiration, the actual transpiration of the sample, the standard crop coefficient, and the parameters of the irrigation district prediction model, and to train and optimize the irrigation district prediction model based on the transpiration loss function.

[0023] In particular, the ConvLSTM layer can effectively capture the dynamic correlation characteristics of water demand in irrigation areas in time and space. The gated fusion unit updates the state by weighting, filters redundant information and retains key features, making the generated spatiotemporal features of water demand more in line with the actual change pattern. This provides higher quality basic data for subsequent prediction of water storage and utilization. It incorporates the characteristics of crop growth stages to achieve dynamic adaptation of water demand prediction. Furthermore, it performs special optimization through the transpiration loss function to ensure that the model can maintain stable prediction accuracy under different crop types and complex environments in different irrigation areas.

[0024] Furthermore, the water demand fusion prediction module includes:

[0025] The gate weight calculation unit is used to generate gate weights by passing the concatenated vector of the soil latent heat flux, the basic evapotranspiration and the predicted water storage and utilization through a feature fusion gate mechanism mapping convolution.

[0026] The gated fusion unit is used to perform a weighted fusion of the spliced ​​vector and the predicted water storage and utilization based on the gate weights using a feature fusion gate mechanism to generate fused water demand features.

[0027] A multi-scale convolutional mapping unit is used to aggregate the fused water demand features through multi-scale convolution to generate the comprehensive predicted water demand.

[0028] Furthermore, the multi-scale convolutional mapping unit includes:

[0029] A dilated convolutional mapping subunit is used to convolve the fused water demand features through multi-drillion-rate convolution to generate water demand at multiple time scales.

[0030] A cross-time-step aggregation subunit is used to aggregate the multi-time-scale water demand across time steps using attention weights to generate the comprehensive predicted water demand.

[0031] Furthermore, the water demand fusion prediction module also includes:

[0032] The comprehensive loss function training unit is used to construct a data fitting term based on the regularization calculation of the comprehensive predicted water demand and the actual water demand of the sample, construct a water vapor conservation term based on the water vapor change rate of the irrigation area, construct a comprehensive loss function based on the weighted sum of the data fitting term and the water vapor conservation term, and train the fusion mapping model through the comprehensive loss function.

[0033] Furthermore, the irrigation district basic data calculation module includes:

[0034] The soil latent heat flux calculation unit is used to determine the saturated water vapor pressure difference based on the air temperature conversion, determine the air pressure specific heat capacity at constant pressure based on the air pressure conversion, and calculate the soil latent heat flux using the improved Penman-Monteith formula based on the water vapor pressure difference, the solar radiation, the air specific heat capacity at constant pressure, and the wind speed.

[0035] Furthermore, the irrigation district basic data calculation module also includes:

[0036] The basal transpiration calculation unit is used to multiply the saturated vapor pressure difference by solar radiation and divide by the sum of the saturated vapor pressure difference and the ecliptic constant to generate an energy term. The energy term is then used to calculate the basal transpiration using the improved and extended Penman-Monteith formula.

[0037] Furthermore, the zero-carbon water resource management module includes:

[0038] The net irrigation volume prediction calculation unit is used to calculate the net irrigation volume prediction of the irrigation district based on the detected irrigation district water holding capacity, irrigation efficiency, effective precipitation and the comprehensive predicted water demand.

[0039] A photovoltaic power prediction calculation unit is used to determine the hydraulic power demand based on the predicted net irrigation volume, and to set the sum of the operating power of the photovoltaic water pump and the photovoltaic valve to be less than the hydraulic power demand.

[0040] In particular, by using a gating mechanism for precise weighting, the importance of different data in water demand forecasting is identified, avoiding interference from low-quality information on the fusion results. This makes the generated fused water demand features more reflective of the actual water demand in the irrigation area. Through dilated convolution mapping, the water demand variation patterns at different time scales can be captured, enabling the comprehensive water demand forecast to accurately respond to short-term water demand changes and grasp long-term water demand trends. Through comprehensive loss function optimization, the forecast accuracy and physical rationality are guaranteed.

[0041] Compared with the prior art, the beneficial effects of the present invention are that it uses the ConvLSTM model to integrate soil latent heat flux and basal transpiration to predict crop water demand, thereby improving the accuracy of water resource management in irrigation areas, reducing waste and imbalance, optimizing the scheduling energy consumption of zero-carbon emission photovoltaic water pumps and photovoltaic valves, and realizing the zero-carbon transformation of agricultural irrigation areas by optimizing the energy consumption of irrigation.

[0042] In particular, this invention, through the ConvLSTM layer, can effectively capture the dynamic correlation characteristics of water demand in irrigation areas in time and space. The gated fusion unit updates the state by weighting, filtering redundant information and retaining key features, so that the generated spatiotemporal features of water demand are more in line with the actual change pattern, providing higher quality basic data for subsequent prediction of water storage and utilization. It incorporates the characteristics of crop growth stages to achieve dynamic adaptation of water demand prediction, and performs special optimization through the transpiration loss function to ensure that the model can maintain stable prediction accuracy under different crop types and complex environments in different irrigation areas.

[0043] In particular, this invention uses a gating mechanism to accurately weight data and identify the importance of different data in water demand prediction, avoiding interference from low-quality information on the fusion results. This makes the generated fused water demand features more reflective of the actual water demand in the irrigation area. Through dilated convolution mapping, it can capture the water demand change patterns at different time scales, enabling the comprehensive water demand prediction to accurately respond to short-term water demand changes and grasp long-term water demand trends. Through comprehensive loss function optimization, it ensures prediction accuracy and physical rationality. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the structure of the zero-carbon water resource integrated management system for agricultural irrigation areas according to an embodiment of the present invention;

[0045] Figure 2 This is a flowchart illustrating the zero-carbon water resource integrated management system for agricultural irrigation areas according to an embodiment of the present invention.

[0046] Figure 3 This is a flowchart illustrating the irrigation district prediction model of the zero-carbon water resource integrated management system for agricultural irrigation districts according to an embodiment of the present invention.

[0047] Figure 4 This is a flowchart illustrating the fusion mapping model of the agricultural irrigation district zero-carbon water resource integrated management system according to an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0049] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0050] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0051] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0052] like Figures 1 to 4 As shown, this invention provides a zero-carbon water resource integrated management system for agricultural irrigation areas. By integrating soil latent heat flux and basal transpiration using a ConvLSTM model, it predicts crop water demand, improves the accuracy of water resource management in irrigation areas, reduces waste and imbalance, optimizes the scheduling energy consumption of zero-carbon emission photovoltaic water pumps and photovoltaic valves, and promotes the zero-carbon transformation of agricultural irrigation areas from the perspective of optimizing irrigation energy consumption.

[0053] like Figure 1 and 2 As shown in the figure, this embodiment proposes a zero-carbon water resource integrated management system for agricultural irrigation areas, including:

[0054] The irrigation district basic data calculation module is used to calculate the soil latent heat flux and the basic transpiration of crops in the irrigation district based on the detected air temperature, air pressure, solar radiation and wind speed, respectively.

[0055] The irrigation district data prediction module is used to generate predicted water storage and utilization of the irrigation district based on multidimensional meteorological data and crop data and an irrigation district prediction model based on the ConvLSTM architecture.

[0056] The water demand fusion prediction module is used to generate the comprehensive predicted water demand of the irrigation area by using the soil latent heat flux, the basic evapotranspiration and the predicted water storage and utilization through a fusion mapping model based on feature fusion gating mechanism and multi-scale convolution.

[0057] The zero-carbon water resource management module is used to calculate the predicted net irrigation volume of the irrigation area based on the comprehensive predicted water demand, and to manage the photovoltaic water pumps and photovoltaic valves of the irrigation area based on the predicted net irrigation volume.

[0058] Specifically, the zero-carbon water resource integrated management system for agricultural irrigation districts described in this embodiment can be applied to the Internet of Things (IoT) agricultural irrigation management system. That is, it can achieve the integrated irrigation management of multi-dimensional meteorological data, stored crop data, and multi-modal data of detected temperature, air pressure, solar radiation, and wind speed in the irrigation district through the Internet of Things.

[0059] It should be noted that transpiration refers to crop evapotranspiration (ET), representing the total amount of soil evaporation and plant transpiration, to ensure that the predicted water demand matches the real-time crop water requirements under various climatic conditions in the irrigation area. Latent heat flux (LE) represents the heat absorbed or released during the water vapor phase change process in the irrigation area. Positive values ​​indicate heat transfer from the Earth's surface to the atmosphere (evaporation / transpiration), while negative values ​​indicate heat transfer from the atmosphere to the Earth's surface (condensation). It reflects whether the farmland soil in the irrigation area is absorbing or evaporating water, as well as the degree of absorption and evaporation, to ensure that the predicted water demand matches the climatic water consumption / transportation situation in the irrigation area.

[0060] It should be noted that the predicted water storage and utilization includes predicted soil moisture and predicted transpiration. Predicted soil moisture refers to the predicted soil water content, and its rate of change is related to crop growth status. Predicted transpiration is used to predict and quantify the water requirements of crops under the influence of various environmental factors. For example, during the vigorous growth stage of the heading stage, crops have a large LAI (Large Isotope Area) and active stomata, and transpiration is increased due to environmental factors such as photosynthetically active radiation, leading to an increase in the amount of irrigation water required for optimal crop growth.

[0061] like Figure 3 As shown, the irrigation district prediction model further includes a ConvLSTM layer and a feature mapping layer, and the irrigation district data prediction module includes:

[0062] The convolutional gating mechanism generation unit is used to take the multidimensional meteorological data and the crop data as input data, and generate the input gate, forget gate, output gate and candidate state of the updated gating mechanism through the convolution operation of the ConvLSTM layer according to the input data and the hidden state of the previous time step.

[0063] The gated fusion unit is used to generate state updates through an update gating mechanism of the ConvLSTM layer that weights the candidate states based on the input gate and the forget gate weights the states of the previous time step.

[0064] ConvLSTM output unit, used to generate spatiotemporal characteristics of water demand based on the state update and the output gate;

[0065] The crop growth stage enhancement unit is used to encode the spatiotemporal characteristics of water demand and the crop growth stage through a feature mapping layer to generate the predicted water storage and utilization.

[0066] Specifically, the process of generating the spatiotemporal characteristics of water demand in the ConvLSTM layer can be represented as follows:

[0067] i t =σ(W xi *X t +W hi *H t-1 +b i )

[0068] f t =σ(W xf *X t +W hf *H t-1 +b f )

[0069] o t =σ(W xo *X t +W ho *H t-1 +b o )

[0070]

[0071] H t =o t ⊙tanh(C t )

[0072] In the formula, i t f t o t , These represent the input gate, forget gate, output gate, and candidate state, respectively. σ represents the sigmoid activation function, used to control the retention or forgetting of information. X t W represents the input data. xi W hi W xf W hf W xo W ho W xc W hc These represent the kernel weights for different gating parameters (input gate, forget gate, output gate, candidate state), * indicates a 3D convolution operation (kernel = 3×3×3 / 5×5×5), b i b f b o b cThese represent the bias terms with different gating parameters, tanh represents the hyperbolic tangent activation function used for nonlinear transformation, and C... t Represents a state update, ⊙ represents the Hadamard product, and C t-1 H represents the state at the previous time step. t This indicates the spatiotemporal characteristics of water demand.

[0073] Specifically, the multidimensional meteorological data includes temperature, humidity, wind speed, sunshine duration, and precipitation (historical and forecast data). This multidimensional meteorological data is standardized and used as input data. The crop data includes crop type and crop wilting point; this crop data is uniquely encoded and used as input data. Therefore, the irrigation district prediction model can serve as an intelligent model capable of sensing "crop growth status," "soil water retention capacity," and "current growth stage."

[0074] In particular, the ConvLSTM layer (Convolutional LSTM) performs convolutional operations on multidimensional meteorological and crop data, combining the hidden state from the previous time step to generate key parameters such as the input gate and forget gate. This effectively captures the dynamic correlation characteristics of irrigation area water demand in time and space. The gated fusion unit further filters redundant information and retains key features by updating the state with weights, making the generated spatiotemporal characteristics of water demand more consistent with actual changes. This avoids the limitation of traditional time series models in capturing insufficient spatial information of irrigation areas, providing higher-quality basic data for subsequent predictions of soil moisture and transpiration, and improving the reliability of predicted water demand.

[0075] like Figure 3 As shown, the crop growth stage enhancement unit further includes:

[0076] The water requirement feature mapping subunit is used to map the crop growth stages in the irrigation area to the crop growth stage codes through convolutional coding;

[0077] The factor mapping subunit is used to encode the crop growth stages through linear convolution to generate mapping factors and growth stage translation correction factors.

[0078] The prediction subunit is used to adjust the spatiotemporal characteristics of water demand based on the mapping factor and the growth stage translation correction factor to generate the predicted water storage and utilization.

[0079] Specifically, the process of generating predicted water storage and utilization values ​​in the prediction subunit can be represented as follows:

[0080] e pheno =E·one_hot(s)

[0081] γ=W γ e pheno +b γ

[0082] δ=W δ e pheno +b δ

[0083]

[0084] In the formula, e pheno This represents the encoding of crop growth stages, E represents the phenological embedding matrix, one_hot(s) represents the one-hot encoding of crop growth stage s, γ and δ represent the mapping factor and growth stage translation correction factor, respectively, and W γ W δ b represents the convolution kernel weights of the mapping factor and the growth stage translation correction factor, respectively. γ b δ These represent the bias terms of the mapping factor and the growth stage translation correction factor, respectively. The model output representing time step t includes predictions of water storage and use, which in turn includes predictions of soil moisture and transpiration. ⊙ represents the Hadamard product, and H... t This indicates the spatiotemporal characteristics of water demand.

[0085] Specifically, the crop growth stages include at least the seedling stage, heading stage, grain-filling stage, and maturity stage, so that the predicted transpiration for predicting water storage and utilization is more in line with the growth stages in which the crop's water demand varies greatly.

[0086] In particular, by using mapping factors, the spatiotemporal characteristics of water demand are amplified in the convolutional features during the critical water demand periods (i.e., heading and grain-filling stages) to reflect soil characteristics related to deep soil moisture and root distribution area water. This allows the model to focus more on the actual available water for the crop. By using growth stage translation correction factors, the deep moisture characteristics are weakened and the weights of channels related to surface soil evaporation and precipitation infiltration are strengthened during the seedling or maturity stages. This matches the dominant influencing factors of soil moisture at this time. Especially before harvest at maturity, the growth stage translation correction factors are used to shift the feature distribution and reduce the predicted baseline of overall transpiration, avoiding misjudgment of the low transpiration characteristics of residual dead leaves as abnormal.

[0087] Specifically, the parameters of the irrigation area prediction model based on the ConvLSTM architecture include: an input time step of 7 days, a spatial scale of 32x32, 6 channels, 2 ConvLSTM layers, 64 and 32 convolutional kernels, 3×3×3 and 5×5×5 convolutional kernel sizes, a stride of 1 for both, a learning rate of 1e-4, a batch size of 8, and a dropout rate of 0.3.

[0088] Furthermore, the predicted water storage and utilization includes the predicted transpiration, and the irrigation district data prediction module further includes:

[0089] The transpiration loss function training unit is used to construct a transpiration loss function based on the predicted transpiration, the actual transpiration of the sample, the standard crop coefficient, and the parameters of the irrigation district prediction model, and to train and optimize the irrigation district prediction model based on the transpiration loss function.

[0090] Specifically, the transpiration loss function can be expressed as:

[0091]

[0092] In the formula, L represents the transpiration loss function, T represents the number of prediction time steps, preferably 24 hours in the future, with 1 step per hour, and H and W represent the spatial dimensions, preferably corresponding to the number of rows and columns of farmland in the irrigation area. ET represents the predicted transpiration at time step t and spatial location (i,j). 0,t,i,j This represents the true transpiration of the sample at time step t and spatial location (i,j). The reference crop coefficient representing time step t and spatial location (i,j) is a table lookup value for the standard crop coefficient. λ represents the regularization coefficient, preferably 1e-5. W ConvLSTM W represents the set of gating parameters for the ConvLSTM layer. γ W δ These represent the convolution kernel weights of the feature mapping layer and the translation correction factor in the growth stage, respectively. This indicates L2 regularization.

[0093] In particular, the ConvLSTM layer can effectively capture the dynamic correlation characteristics of water demand in irrigation areas in time and space. The gated fusion unit updates the state by weighting, filters redundant information and retains key features, making the generated spatiotemporal features of water demand more in line with the actual change pattern. This provides higher quality basic data for subsequent prediction of water storage and utilization. It incorporates the characteristics of crop growth stages to achieve dynamic adaptation of water demand prediction. Furthermore, it performs special optimization through the transpiration loss function to ensure that the model can maintain stable prediction accuracy under different crop types and complex environments in different irrigation areas.

[0094] like Figure 4As shown, the water demand fusion prediction module further includes:

[0095] The gate weight calculation unit is used to generate gate weights by passing the concatenated vector of the soil latent heat flux, the basic evapotranspiration and the predicted water storage and utilization through a feature fusion gate mechanism mapping convolution.

[0096] The gated fusion unit is used to perform a weighted fusion of the spliced ​​vector and the predicted water storage and utilization based on the gate weights using a feature fusion gate mechanism to generate fused water demand features.

[0097] A multi-scale convolutional mapping unit is used to aggregate the fused water demand features through multi-scale convolution to generate the comprehensive predicted water demand.

[0098] Specifically, the process of generating fused water demand features using the feature fusion gating mechanism can be represented as follows:

[0099]

[0100] G t =σ(W g *P t +b g )

[0101]

[0102] In the formula, P t This indicates the concatenation of vectors. The model output representing time step t includes predictions of water storage and utilization. t ET represents the soil latent heat flux at time step t. t G represents the basal transpiration at time step t. t W represents the gating weights, σ ​​represents the Sigmoid activation function, and W represents the gate weights. g Let b represent the gated convolution matrix. g Indicates the gated bias term, * indicates the convolution operation, U t ⊙ represents the water requirement characteristic of fusion, and ⊙ represents the Hadamard product.

[0103] In particular, the gating weight calculation unit generates gating weights by mapping and convolving the spliced ​​vectors of soil latent heat flux, basal transpiration, and predicted water storage and utilization through a gating mechanism. The gating fusion unit then performs weighted fusion of relevant data based on these weights, which can identify the importance of different input data in water demand prediction. For example, when soil moisture is low, the predicted soil moisture of predicted water storage and utilization is given a higher weight. When crop transpiration is vigorous, the weights of basal transpiration and predicted transpiration of predicted water storage and utilization are increased. This effectively avoids interference from irrelevant or low-quality information on the fusion results, making the generated fused water demand characteristics more reflective of the actual water demand in the irrigation area.

[0104] like Figure 4 As shown, the multi-scale convolutional mapping unit further includes:

[0105] A dilated convolutional mapping subunit is used to convolve the fused water demand features through multi-drillion-rate convolution to generate water demand at multiple time scales.

[0106] A cross-time-step aggregation subunit is used to aggregate the multi-time-scale water demand across time steps using attention weights to generate the comprehensive predicted water demand.

[0107] Specifically, the process of generating the comprehensive predicted water demand through multi-scale convolution can be represented as follows:

[0108]

[0109] In the formula, Z t Z k Represents water demand at multiple timescales, including time t and time k. U represents a multi-hole convolution with hole rates of 1, 3, and 5. t Indicating the water requirement characteristics of fusion, α t,k This represents the attention weight of time step t to historical time step k, used to measure the degree of attention paid to historical time step k. K represents the transpose of the query vector at time step t, generated from the model output. j The key vector representing historical time step k is generated from water demand at multiple time scales. This represents the comprehensive predicted water demand at a future time t+τ, where τ represents the predicted lead time step. Preferably, τ=1 indicates one hour in the future. * represents the convolution operation. W y b y This represents the learnable parameters of the convolution kernel.

[0110] Specifically, the parameters of the fusion mapping model based on feature fusion gating mechanism and multi-scale convolution include: the number of convolution kernels in the feature fusion gating mechanism is 3, the kernel size is 3×3, the number of convolution kernels per channel in the multi-scale convolution is 32, the kernel size is 3×3, and the ReLU activation function is used.

[0111] In particular, the dilated convolutional mapping subunit in the multi-scale convolutional mapping unit uses multi-diffraction rate convolution processing to integrate water demand features, which can capture the water demand variation patterns at different time scales, such as water demand fluctuations caused by sudden weather changes in a short period of time, and water demand trends generated by crop growth and seasonal changes over a longer period of time, thus improving the comprehensiveness and adaptability of the prediction.

[0112] Furthermore, the water demand fusion prediction module also includes:

[0113] The comprehensive loss function training unit is used to construct a data fitting term based on the regularization calculation of the comprehensive predicted water demand and the actual water demand of the sample, construct a water vapor conservation term based on the water vapor change rate of the irrigation area, construct a comprehensive loss function based on the weighted sum of the data fitting term and the water vapor conservation term, and train the fusion mapping model through the comprehensive loss function.

[0114] Specifically, the comprehensive loss function can be expressed as:

[0115]

[0116] In the formula, L data L moisture Representing the data fitting term and the water vapor conservation term respectively, τ represents the prediction lead time step, and H and W represent the spatial dimensions. Y represents the comprehensive forecast of water demand. true This represents the actual water requirement of the sample. Denotes the square of the Frobenius norm. The local variation rate of the water vapor mixing ratio is expressed as qv, where v is the detected wind speed, q is the water vapor mixing ratio (i.e., the mass of water vapor contained in a unit mass of moist air), q is the local meteorological station monitoring data in the irrigation area, E represents the evaporation rate, P represents the precipitation rate, and E and P are both local setpoints in the irrigation area. |||1 represents the L1 norm, L... total Let λ1 and λ2 represent the comprehensive loss function, and λ1 and λ2 are two weighting coefficients, preferably 0.6 and 0.4.

[0117] In particular, the water vapor conservation term is forced to satisfy the water vapor balance during model training, which is suitable for scenarios that need to simulate the water cycle, namely the comprehensive predicted water demand of irrigation areas that includes farmland soil moisture prediction. This can make up for the problem that pure data fitting may ignore the actual situation of the irrigation area and improve the rationality of the prediction.

[0118] In particular, by using a gating mechanism for precise weighting, the importance of different data in water demand forecasting is identified, avoiding interference from low-quality information on the fusion results. This makes the generated fused water demand features more reflective of the actual water demand in the irrigation area. Through dilated convolution mapping, the water demand variation patterns at different time scales can be captured, enabling the comprehensive water demand forecast to accurately respond to short-term water demand changes and grasp long-term water demand trends. Through comprehensive loss function optimization, the forecast accuracy and physical rationality are guaranteed.

[0119] Furthermore, the irrigation district basic data calculation module includes:

[0120] The soil latent heat flux calculation unit is used to determine the saturated water vapor pressure difference based on the air temperature conversion, determine the air pressure specific heat capacity at constant pressure based on the air pressure conversion, and calculate the soil latent heat flux using the improved Penman-Monteith formula based on the water vapor pressure difference, the solar radiation, the air specific heat capacity at constant pressure, and the wind speed.

[0121] Specifically, the improved Penman-Monteith formula for calculating soil latent heat flux is as follows:

[0122]

[0123] In the formula, LE t The latent heat flux of the soil at time step t is represented by Δ, which represents the saturated water vapor pressure difference. The conversion is determined, where e s R represents saturated vapor pressure, T is air temperature, and R is the saturated vapor pressure. n,t Let t represent the solar radiation at time t, and ρ represent the air density, preferably 1.225 kg / m³. 3 c p This represents the specific heat capacity of air at constant pressure, determined by conversion using the current air pressure. e s,t e a,t Let r represent the saturated vapor pressure and actual vapor pressure at time t, respectively, which are determined by conversion from the detected air temperature and by monitoring at the local meteorological station in the irrigation area. s r a These represent wind resistance and crop resistance, respectively. The observation height z is 2 meters, the zero plane displacement d = 0.67h, and the momentum roughness length z 0m The water vapor roughness length z is 0.1h. 0h The value is 0.01h, where h is the current average crop height, the von Kármán constant k is 0.41, and u z For the wind speed to be detected, r s The preferred value is 70 s / m, where γ represents the wet / dry constant, which is approximately 0.0665 kPa / ℃ under standard atmospheric pressure.

[0124] In particular, it has enabled the inverse calculation of water demand (irrigation volume) using soil latent heat flux.

[0125] Furthermore, the irrigation district basic data calculation module also includes:

[0126] The basal transpiration calculation unit is used to multiply the saturated vapor pressure difference by solar radiation and divide by the sum of the saturated vapor pressure difference and the ecliptic constant to generate an energy term. The energy term is then used to calculate the basal transpiration using the improved and extended Penman-Monteith formula.

[0127] Specifically, the improved Penman-Monteith formula for calculating basal transpiration is as follows:

[0128]

[0129] In the formula, ET t R represents the baseline transpiration at time step t, Δ represents the saturated vapor pressure difference, and R represents the basal transpiration at time step t. n Represents solar radiation, γ represents the tween constant, ρ represents air density, and c represents the density of air. p The specific heat capacity of air at constant pressure, r a Representing crop resistance, γ and ρ are constants as mentioned above, and c p r a For the conversion value, κ represents the extinction coefficient, which is preferably 0.5, and LAI represents the leaf area index, which is determined based on crop data stored in the Internet of Things.

[0130] In particular, it enables the calculation of water demand and irrigation volume by back-calculating the soil moisture balance equation based on actual evapotranspiration.

[0131] Understandably, the standard form of the Penman-Monteith formula is:

[0132]

[0133] This embodiment improves and extends the standard Penman-Monteith formula and integrates it with prediction data to accurately predict the soil water demand and crop water demand in the irrigation area. This allows for a more accurate assessment of irrigation water volume and water delivery power, ensuring that the power generated by photovoltaic water pumps and photovoltaic valves is minimized while meeting the soil and crop water demand, thereby achieving zero-carbon water resource management in the irrigation area.

[0134] Furthermore, the zero-carbon water resource management module includes:

[0135] The net irrigation volume prediction calculation unit is used to calculate the net irrigation volume prediction of the irrigation district based on the detected irrigation district water holding capacity, irrigation efficiency, effective precipitation and the comprehensive predicted water demand.

[0136] A photovoltaic power prediction calculation unit is used to determine the hydraulic power demand based on the predicted net irrigation volume, and to set the sum of the operating power of the photovoltaic water pump and the photovoltaic valve to be less than the hydraulic power demand.

[0137] Specifically, the process of calculating the predicted net irrigation volume for the irrigation district is as follows:

[0138]

[0139] In the formula, I irr θ represents the predicted net irrigation volume of the irrigation district. fc Z represents the irrigation district's water holding capacity. r η represents root depth, P represents irrigation efficiency. e Indicates effective precipitation. It represents the sum of the predicted net irrigation volume of the irrigation district over a predicted period of time.

[0140] Specifically, the process of calculating hydraulic power demand is as follows:

[0141]

[0142] In the formula, P hyd ρ represents the hydraulic power demand. water H represents the density of water, g represents the acceleration due to gravity, and H represents the acceleration due to gravity. tot Indicates the total head, determined based on historical measurement data for different irrigation volumes. irr This represents the predicted net irrigation volume for the irrigation district, where A represents the irrigated area, and t. irr Indicates the planned irrigation time, ρ water g, A, t irr All are given or set values.

[0143] In particular, by dynamically adjusting the maximum operating power of photovoltaic water pumps and valves based on the predicted net irrigation volume, clear scheduling and management control is achieved.

[0144] In this embodiment, a ConvLSTM model is used to fuse soil latent heat flux and basal transpiration to predict crop water demand, thereby improving the accuracy of water resource management in irrigation districts, reducing waste and imbalance, optimizing the scheduling energy consumption of zero-carbon emission photovoltaic water pumps and photovoltaic valves, and promoting the zero-carbon transformation of agricultural irrigation districts from the perspective of irrigation energy consumption optimization. The ConvLSTM layer effectively captures the dynamic correlation characteristics of irrigation district water demand in time and space. The gated fusion unit updates the state with weights, filters redundant information, and retains key features, making the generated spatiotemporal characteristics of water demand more consistent with actual changes. This provides higher-quality basic data for subsequent prediction of water storage and utilization. Integrating crop growth stage characteristics enables dynamic adaptation of water demand prediction, and specific optimization using the transpiration loss function ensures that the model maintains stable prediction accuracy under different crop types and complex irrigation district environments. By using a gating mechanism for precise weighting, the importance of different data in water demand forecasting is identified, avoiding interference from low-quality information on the fusion results. This makes the generated fused water demand features more reflective of the actual water demand in the irrigation area. Through dilated convolution mapping, the water demand variation patterns at different time scales can be captured, enabling the comprehensive water demand forecast to accurately respond to short-term water demand changes and grasp long-term water demand trends. Through comprehensive loss function optimization, the forecast accuracy and physical rationality are guaranteed.

[0145] Those skilled in the art will recognize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0146] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A zero-carbon water resource integrated management system for agricultural irrigation areas, characterized in that, include: The irrigation district basic data calculation module is used to calculate the soil latent heat flux and the basic transpiration of crops in the irrigation district based on the detected air temperature, air pressure, solar radiation and wind speed, respectively. The irrigation district data prediction module is used to generate predicted water storage and utilization of the irrigation district based on multidimensional meteorological data and crop data and an irrigation district prediction model based on the ConvLSTM architecture. The water demand fusion prediction module is used to generate the comprehensive predicted water demand of the irrigation area by using the soil latent heat flux, the basic evapotranspiration and the predicted water storage and utilization through a fusion mapping model based on feature fusion gating mechanism and multi-scale convolution. The zero-carbon water resource management module is used to calculate the predicted net irrigation volume of the irrigation area based on the comprehensive predicted water demand, and to manage the photovoltaic water pumps and photovoltaic valves of the irrigation area based on the predicted net irrigation volume of the irrigation area. The irrigation district prediction model includes a ConvLSTM layer and a feature mapping layer, and the irrigation district data prediction module includes: The convolutional gating mechanism generation unit is used to take the multidimensional meteorological data and the crop data as input data, and generate the input gate, forget gate, output gate and candidate state of the updated gating mechanism through the convolution operation of the ConvLSTM layer according to the input data and the hidden state of the previous time step. The gated fusion unit is used to generate state updates through an update gating mechanism of the ConvLSTM layer that weights the candidate states based on the input gate and the forget gate weights the states of the previous time step. ConvLSTM output unit, used to generate spatiotemporal characteristics of water demand based on the state update and the output gate; A crop growth stage enhancement unit is used to encode the spatiotemporal characteristics of water demand and the crop growth stage through a feature mapping layer to generate the predicted water storage and utilization. The crop growth stage enhancement unit includes: The water requirement feature mapping subunit is used to map the crop growth stages in the irrigation area to the crop growth stage codes through convolutional coding; The factor mapping subunit is used to encode the crop growth stages through linear convolution to generate mapping factors and growth stage translation correction factors. The prediction subunit is used to adjust the spatiotemporal characteristics of water demand based on the mapping factor and the growth stage translation correction factor in order to generate the predicted water storage and utilization. The water demand fusion prediction module includes: The gate weight calculation unit is used to generate gate weights by passing the concatenated vector of the soil latent heat flux, the basic evapotranspiration and the predicted water storage and utilization through a feature fusion gate mechanism mapping convolution. The gated fusion unit is used to perform a weighted fusion of the spliced ​​vector and the predicted water storage and utilization based on the gate weights using a feature fusion gate mechanism to generate fused water demand features. A multi-scale convolutional mapping unit is used to aggregate the fused water demand features through multi-scale convolution to generate the comprehensive predicted water demand.

2. The agricultural irrigation area zero-carbon water resource integrated management system according to claim 1, characterized in that, The predicted water storage and utilization includes the predicted evapotranspiration, and the irrigation district data prediction module further includes: The transpiration loss function training unit is used to construct a transpiration loss function based on the predicted transpiration, the actual transpiration of the sample, the standard crop coefficient, and the parameters of the irrigation district prediction model, and to train and optimize the irrigation district prediction model based on the transpiration loss function.

3. The agricultural irrigation area zero-carbon water resource integrated management system according to claim 1, characterized in that, The multi-scale convolutional mapping unit includes: A dilated convolutional mapping subunit is used to convolve the fused water demand features through multi-drillion-rate convolution to generate water demand at multiple time scales. A cross-time-step aggregation subunit is used to aggregate the multi-time-scale water demand across time steps using attention weights to generate the comprehensive predicted water demand.

4. The agricultural irrigation area zero-carbon water resource integrated management system according to claim 1, characterized in that, The water demand fusion prediction module also includes: The comprehensive loss function training unit is used to construct a data fitting term based on the regularization calculation of the comprehensive predicted water demand and the actual water demand of the sample, construct a water vapor conservation term based on the water vapor change rate of the irrigation area, construct a comprehensive loss function based on the weighted sum of the data fitting term and the water vapor conservation term, and train the fusion mapping model through the comprehensive loss function.

5. The agricultural irrigation area zero-carbon water resource integrated management system according to claim 1, characterized in that, The irrigation district basic data calculation module includes: The soil latent heat flux calculation unit is used to determine the saturated water vapor pressure difference based on the air temperature conversion, determine the air pressure specific heat capacity at constant pressure based on the air pressure conversion, and calculate the soil latent heat flux using the improved Penman-Monteith formula based on the water vapor pressure difference, the solar radiation, the air specific heat capacity at constant pressure, and the wind speed.

6. The agricultural irrigation area zero-carbon water resource integrated management system according to claim 5, characterized in that, The irrigation district basic data calculation module also includes: The basal transpiration calculation unit is used to multiply the saturated vapor pressure difference by solar radiation and divide by the sum of the saturated vapor pressure difference and the ecliptic constant to generate an energy term. The energy term is then used to calculate the basal transpiration using the modified Penman-Monteith formula.

7. The agricultural irrigation area zero-carbon water resource integrated management system according to any one of claims 1 to 6, characterized in that, The zero-carbon water resource management module includes: The net irrigation volume prediction calculation unit is used to calculate the net irrigation volume prediction of the irrigation district based on the detected irrigation district water holding capacity, irrigation efficiency, effective precipitation and the comprehensive predicted water demand. A photovoltaic power prediction calculation unit is used to determine the hydraulic power demand based on the predicted net irrigation volume, and to set the sum of the operating power of the photovoltaic water pump and the photovoltaic valve to be less than the hydraulic power demand.

Images