Agricultural water right distribution method based on dual-stage stochastic programming and neural network

Abstract

The invention relates to an agricultural water right distribution method based on dual-stage stochastic programming and a neural network. The method comprises the following steps: collecting historical data to construct a data set; the irrigation water demand and the available water supply are predicted by inputting the data set into the long-short-term memory neural network; establishing a two-stage stochastic programming model, introducing a decision variable to convert the model into a deterministic sub-model, and solving an optimal water distribution target and configuration water quantity through an interactive algorithm; and according to the optimal water distribution target and the configured water quantity, generating water right intelligent distribution schemes under different water inflow scenes, and dynamically adjusting a water right distribution threshold value between the regions. Through deep fusion of double-stage stochastic programming and the neural network, a data-driven and model-driven water right intelligent distribution framework is constructed, a long-short-term memory neural network is introduced to predict a future irrigation water demand and an available water supply interval value, the adaptability of the model to dynamic climate is improved, and the water right distribution efficiency is improved. And the optimal distribution threshold values under different inflow water levels are quantified, so that the water demand of a high marginal benefit area is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agricultural water resource management, and particularly relates to an agricultural water right allocation method based on two-stage stochastic programming and neural networks. BACKGROUND

[0002] Agricultural water resource management is the core link to ensure food security and ecological sustainability, especially in economic crop planting areas, the water supply and demand contradiction is intensified due to climate fluctuations, terrain fragmentation and multi-water source coordinated scheduling problems. Traditional water right allocation models, including linear programming and fuzzy programming, are difficult to cope with the dynamic influence of climate uncertainty on irrigation demand, lack the ability to quantitatively analyze the spatial and temporal heterogeneity of regional water resources, and rely on static supply and demand balance calculation, ignoring the impact of drought and flood and other extreme climate events on the irrigation system, resulting in significant adaptability defects in the practical application of water right allocation strategies.

[0003] The linear programming, nonlinear programming, dynamic programming and other methods commonly used in current water resource planning are difficult to balance local interests and global efficiency if the terrain of the planting area is fragmented and the water source is unevenly distributed. At the same time, the existing models respond to seasonal precipitation changes with a lag, making it difficult to generate real-time adjustment allocation schemes, and the robustness of the allocation strategy is insufficient. Moreover, the cost difference between local reservoirs and external water transfer is not fully quantified, resulting in an imbalance between the economy and ecology of the allocation strategy. SUMMARY

[0004] The technical problem of the present application is to dynamically balance economic benefits and water shortage penalties through a two-stage decision mechanism, set pre-allocation targets and adjust strategies according to actual water supply, solve the robustness deficiency of traditional models under climate uncertainty, improve the spatial and temporal accuracy and economic benefits of water resource allocation, dynamically balance regional water supply and demand contradictions, reduce drought and flood risks, and support multi-water source coordinated scheduling and water right transaction decisions under complex climate scenarios.

[0005] The purpose of the present application is to solve the above problems, and provide an agricultural water right allocation method based on two-stage stochastic programming and neural networks, comprising the following steps: S1: Collect historical meteorological data, hydrological data, crop growth parameters and planting area information to form a data set D1, and calculate the net irrigation water requirement IR during the crop growth period; S2: Divide the planting subarea according to the annual water supply of the water source; S3: Construct a long short-term memory neural network, input the data set D1 into the long short-term memory neural network, output the predicted irrigation water requirement and available water supply interval value, and divide the water requirement and supply interval under high, medium and low three levels of inflow; S4: An interval two-stage stochastic programming model is established, taking the pre-water distribution target as the first-stage decision variable, the water shortage as the second-stage decision factor, combining the probability distribution of different water inflow levels, and taking the maximum system net income as the objective function; S5: The decision variable is introduced to convert the interval model into a deterministic sub-model, and the optimal water distribution target and allocation amount are solved by an interactive algorithm; S6: According to the optimal water distribution target and allocation amount, the intelligent allocation scheme of water rights under different water inflow scenarios is generated, and the inter-regional water rights allocation threshold is dynamically adjusted.

[0006] Further, in step S1, the hydrological data includes surface water resource quantity, groundwater resource quantity and river runoff data; the surface water resource data includes reservoir storage data and weir pond water supply data.

[0007] Preferably, in step S1, evapotranspiration, precipitation and irrigation data are also collected, and effective rainfall is calculated by soil moisture balance method, and the calculation formula is: ; In the formula, P e represents the daily effective rainfall, P represents the daily rainfall.

[0008] Further, in step S3, the long short-term memory neural network has the calculation formula:

[0009] In the formula, , , , respectively represent the forget gate, the input gate, the output gate and the output of the candidate cell state, represents the input at time t, i represents the serial number of the input feature, represents the hidden state at the previous time of time t, W , b respectively represent the weight matrix and the bias vector corresponding to each gating unit, p and q respectively represent the network input dimension and the number of hidden layer nodes, represents a sigmoid nonlinear activation function, represents a hyperbolic tangent activation function.

[0010] Preferably, the updated cell state, cell output and network output have the calculation formula: ; ; ; wherein, denotes vector dot product operation, denotes updated cell state, denotes unit output, denotes network output.

[0011] Further, in step S3, the water demand and supply intervals under high, medium and low water levels are divided, including using the local external water source proportion control distribution risk.

[0012] Preferably, an interval two-stage stochastic programming model is established, and the expression is: ; wherein, f denotes system net income, B ij denotes water source i to the citrus planting area j at the time of water distribution, W ij denotes water source i to the citrus planting area j pre-watering target, C ij denotes water source i unmet planting area j water shortage penalty coefficient when the pre-watering target is not met, S ij denotes water source i unmet citrus planting area j water shortage amount when the pre-watering target is not met, i denotes different water source count units, with a maximum value of I, j denotes each citrus planting area count unit, with a maximum value of J.

[0013] Preferably, the constraint conditions include water source available water quantity constraint, crop water demand constraint and non-negative constraint, and the expression is: (1) Water source available water quantity constraint: ; (2) Crop water demand constraint: ; (3) Non-negative constraint ; wherein, denotes the maximum water supply capacity of the water source i denotes the minimum water demand for normal growth of citrus in the th planting area, j denotes the minimum water demand for normal growth of citrus in the th planting area, jThe maximum water requirement for normal growth of citrus in a planting area.

[0014] Further, in step S5, the introduction of the decision variable converts the interval model into a deterministic sub-model, and the calculation formula is: ; In the formula, represents the interval value of the pre-allocation water, when = 1, the upper limit of the value is reached, and the corresponding pre-allocation water target is the maximum value, and the risk borne is also the largest; when = 0, the lower limit of the value is reached, the allocation water target is the minimum value, the total benefit of water resources is the smallest, and the risk borne is the smallest.

[0015] Preferably, in step S6, the dynamic adjustment of the regional water right allocation threshold value is associated with the certain correlation between the benefit generated by unit water consumption and the water shortage penalty coefficient, the water right allocation is dynamically adjusted, the economic benefit is improved, and the penalty is reduced, and then the coefficient benefit is taken as the adjustment target, and the calculation formula is: ; In the formula, S represents the system benefit coefficient, P represents the purchase price, L represents the unit area yield, Q represents the unit area irrigation water consumption.

[0016] Compared with the prior art, the beneficial effects of the present application include: 1) The agricultural water right allocation method based on two-stage stochastic programming and neural network provided by the present application, through deep fusion of two-stage stochastic programming and neural network, constructs a data-driven and model-driven water right intelligent allocation framework, and introduces a long short-term memory neural network to predict future irrigation water requirement and available water interval value, and improves the adaptability of the model to dynamic climate input.

[0017] 2) The agricultural water right allocation method based on two-stage stochastic programming and neural network provided by the present application, constructs an interval two-stage stochastic programming model, quantifies the optimal allocation threshold value under different water inflow levels, and guarantees the water demand of the high marginal benefit area.

[0018] 3) The agricultural water right allocation method based on two-stage stochastic programming and neural network provided by the present application, through the two-stage decision mechanism, dynamically balances the economic benefit and the water shortage penalty, and improves the robustness under climate uncertainty. BRIEF DESCRIPTION OF DRAWINGS

[0019] The present application will be further described below in combination with the drawings and examples.

[0020] Figure 1 A flowchart of an embodiment of the agricultural water right allocation method based on two-stage stochastic programming and neural networks of the present application.

[0021] Figure 2 A zoning diagram of the research area for the agricultural water right allocation method of the present application. DETAILED DESCRIPTION

[0022] In order for those skilled in the art to better understand the technical solutions of the present application, the specific implementation schemes proposed by the present application will be described clearly and completely below in conjunction with the drawings in the application.

[0023] The present application selects Anfusi Town of Zhijiang City, Yichang City, Hubei Province as the research area, and takes the main economic crop of citrus in Anfusi Town as the research object.

[0024] As shown in the agricultural water right allocation method based on two-stage stochastic programming and neural networks, the method comprises the following steps: Figure 1 S1: Collect historical meteorological data, hydrological data, crop growth parameters and planting area information to form a data set D1, and calculate the net irrigation water requirement IR of the crop growth period.

[0025] In step S1, the hydrological data includes surface water resources, groundwater resources and river runoff data; the surface water resources data includes reservoir storage data and weir pond water supply data.

[0026] As shown in the agricultural water right allocation method based on two-stage stochastic programming and neural networks, the method comprises the following steps: Figure 2 The data of the research area is collected from the meteorological data of Yichang Meteorological Bureau, including daily observation data of daily rainfall, relative humidity, maximum temperature, minimum temperature, wind speed and evaporation.

[0027] The data of the research area is collected from the water resources data of Zhijiang City Water Resources and Lakes Bureau, including surface water resources, reservoir storage and weir pond water supply, groundwater resources and river runoff.

[0028] The data of the research area is collected from the crop planting situation and village-level industrial development plan of Zhijiang City Anfusi Government.

[0029] The data of the research area is collected from the market supervision administration bureau of Yichang City, and the crop growth parameters of the research area include growth period and water requirement coefficient.

[0030] The potential evapotranspiration is calculated by using the Penman-Monteith formula, and the calculation formula is: ; In the formula, ET c represents the daily water requirement of citrus, K c ​represents the coefficient of citrus crop.

[0031] ET 0Penman-Monteith model is used to calculate the daily potential evapotranspiration ET 0, the calculation formula is: ; In the formula, represents the slope of the saturated water vapor pressure and temperature curve, represents the net radiation of the canopy surface, represents the soil heat flux, represents the psychrometer constant, represents the daily average temperature, represents the wind speed at a height of 2m, represents the saturated water vapor pressure, is the actual water vapor pressure.

[0032] According to the water quota standard for citrus irrigation in Yichang City, and referring to the fruit growth and development period time period and temperature index in the southern citrus planting area, the citrus crop coefficient Kc value in different partitions in Yichang is determined, and the calculation formula is: ; ; In the formula, IR represents the crop net irrigation water requirement, Pe represents the effective rainfall during the crop growth period, and IRI represents the crop irrigation water requirement index.

[0033] The effective rainfall is calculated by the method recommended by the USDA Soil Conservation Service of the United States Department of Agriculture, using the soil water balance method, considering factors such as evapotranspiration, precipitation and irrigation, and the calculation formula is: ; In the formula, P e represents the daily effective rainfall, P represents the daily rainfall.

[0034] Using the above calculation formula, combined with meteorological data including daily observation data such as daily rainfall, relative humidity, maximum temperature, minimum temperature, wind speed, sunshine, evaporation, etc., the crop planting net irrigation water requirement, effective rainfall and evapotranspiration are calculated.

[0035] The crop irrigation water requirement calculation formula for each region is: ; In the formula, Q represents the crop planting water requirement, S represents the crop planting area, represents the crop planting unit area net irrigation water requirement.

[0036] S2: Divide the planting areas according to the annual water supply of the water source.

[0037] S3: Construct a long short-term memory neural network and input the data set D1 into the long short-term memory neural network to output the predicted irrigation water demand and water supply interval values, and divide the water demand and water supply intervals under three water inflow levels: high, medium and low.

[0038] In step S3, the long short-term memory neural network is calculated as follows:

[0039] Where, 、 、 、 Represent the outputs of the forget gate, input gate, output gate, and candidate cell state, respectively. Represents the input at time t, i represents the sequence number of the input feature, represents the hidden state at the moment before time t, W 、 b Respectively represent the weight matrix and bias vector corresponding to each gating unit, p and q Represent the network input dimension and the number of hidden layer nodes, represents the sigmoid nonlinear activation function, represents the hyperbolic tangent activation function.

[0040] The updated cell state, unit output, and network output are calculated as follows: ; ; ; Where, Represents vector dot multiplication operation, represents the updated cell state, represents the unit output, Represents network output.

[0041] The Z-Score method is used to standardize the input data of the LSTM model. The original data is subtracted from its mean and then divided by its standard deviation to convert the data into data that conforms to the standard normal distribution. The calculation formula is: ; Where μ represents the data mean; σ represents the data standard deviation; and Z represents the distance between the citrus irrigation water requirement and the ensemble mean.

[0042] The first 45 years of data were used as training input, and the remaining 7 years of data were used for validation and comparison. The three custom parameters that affect the prediction were determined as follows: the number of hidden layer neurons was 200, the learning rate reduction cycle was 800, the learning rate reduction factor was 0.1, and the number of training iterations was 1000. Because the LSTM model may produce different results in each prediction, the average of each indicator obtained from the 6 experimental results was used as the final indicator. IR is 330.53 mm, and the water requirement per unit area is 3300.00 m 3 / hm 2 .

[0043] Based on the current water extraction capacity of the Anfusi water diversion project in Zhijiang City and years of water supply data from various water supply projects, an LSTM model was used to predict the available water volume in the citrus irrigation water supply area. Of the surface water citrus irrigation water supply data from the past 15 years, the first 12 years were used as training input, and the remaining three years were used for validation and comparison. Multiple experiments were conducted, and the average of each indicator obtained from six experiments was used as the final indicator. After averaging these results, the predicted water volume available for citrus irrigation reservoirs and ponds in Anfusi Village, Zhijiang City, in wet and dry years in 2025 was 18.7507 million to 20.353 million m, respectively. 3 and 1.7493~1.9597 million m 3 .

[0044] S4: Establish an interval two-stage stochastic programming model, with the pre-allocation target as the first-stage decision variable and the water shortage as the second-stage decision factor, combined with the probability distribution of different water inflow levels, and taking the maximization of the system net benefit as the objective function.

[0045] Based on data on citrus water demand in different areas of Anfusi, Zhijiang City, and local agricultural irrigation water availability, water costs and water shortage penalty coefficients were introduced to determine the optimal allocation of irrigation water resources for citrus in Anfusi in a two-stage process. First, in the first stage, a pre-allocation target for citrus irrigation was determined based on the normal annual irrigation water demand of citrus, which served as the decision variable in the first stage. Due to differences in citrus planting volume and water supply capacity across regions, the actual water availability may be less than the pre-allocation target. To compensate for the water shortage, adjustments to regional water availability or the introduction of external water sources can be used. However, reducing water availability would negatively impact citrus yield, while introducing external water sources would increase water costs, resulting in an economic penalty. To reduce water costs, the water allocation in the first stage was adjusted, and water shortage was used as a decision factor in the second stage.

[0046] Establish an interval two-stage stochastic programming model, the expression is: ; In the formula, f represents the system net benefit, B ij represents the water source i to the citrus planting area j system benefit at the time of water distribution, W ij represents the water source i to the citrus planting area j pre-watering target, C ij represents the water source i unmet planting area j water shortage penalty coefficient when the pre-watering target is not met, S ij represents the water source i unmet citrus planting area j water shortage amount when the pre-watering target is not met, i represents the different water source count units, with a maximum value of I, j represents the count units of each citrus planting area, with a maximum value of J.

[0047] Because the amount of incoming water has a significant impact on the amount of water shortage, the amount of water shortage at different incoming water levels is treated as a discrete function, and the probability of different levels of incoming water is assumed to be P h , 0 P h <1, where h =1, 2, 3; h =1 represents the minimum incoming water, which is the low incoming water level, and the maximum water shortage; h =2 indicates that the incoming water is more, which is the medium incoming water level; h =3 represents the maximum incoming water, which is the high incoming water level, and the minimum water shortage; and , so the two-stage stochastic programming model can be represented as: ; In the formula, S ijh represents the water source h to the citrus planting area i when the incoming water level is j , and the amount of water shortage that does not meet the pre-watering target.

[0048] In order to solve the problem of uncertainty of the pre-watering target value, benefit and penalty coefficient of the model, interval parameters are introduced to construct the model, which can more accurately consider the possible value range and calculate and analyze according to the specific situation. + represents the upper limit value of the parameter, - represents the lower limit value of the parameter, and the calculation formula of the two-stage stochastic programming model of this interval is: ; wherein: is the system net benefit, yuan, is the water source i to the citrus planting area j at the time of water distribution; is the water source i to the citrus planting area j at the time of pre-water distribution; C ij is the water source i to the citrus planting area j at the time of pre-water distribution target not being met, wherein C > B; P h is the probability of different levels of incoming water, and 0 P h <1, wherein h = 1, 2, 3. h = 1 represents that the incoming water is the least, is a low water level, and the water deficit is the largest; h = 2 indicates that the incoming water is more, is a medium water level; h = 3 represents that the incoming water is the most, is a high water level, and the water deficit is the smallest; represents the water deficit of the water source h to the citrus planting area i at the time of pre-water distribution target not being met; j represents different water sources, i ; represents each citrus planting area, j .

[0049] The constraint conditions include water source available water constraints, crop water demand constraints, and non-negative constraints, expressed as: (1) Water source available water constraints: ; (2) Crop water demand constraints: ; (3) Non-negative constraints ; wherein, represents the maximum available water of the water source i , represents the minimum water demand for normal growth of citrus in the i-th planting area, represents the maximum water demand for normal growth of citrus in the i-th planting area. j j

[0050] ​​​​S5: Introducing decision variables to transform the interval model into a deterministic sub-model, and solving the optimal water allocation target and water allocation through an interactive algorithm.

[0051] In step S5, the decision variables are introduced to transform the interval model into a deterministic sub-model, and the calculation formula is: ; In the formula, represents the interval value of the pre-allocation water, when = 1, reaches the upper limit of the value, and the corresponding pre-allocation target is the maximum value, and the risk is also the largest; when = 0, reaches the lower limit of the value, and the allocation target is the minimum value, and the total water resource benefit is the minimum, and the risk is the smallest.

[0052] S6: According to the optimal water allocation target and water allocation, generate water right intelligent allocation scheme under different inflow scenarios, and dynamically adjust the inter-regional water right allocation threshold.

[0053] According to the relevant data of Anfusi citrus planting, the ratio of local water supply and external water supply is 2:1, and the pre-allocation water target of the first stage of citrus planting is calculated, and the calculation formula is: ; In the formula, Q represents the pre-allocation water target, T is the planting area, C represents the unit area irrigation water consumption, N is the water supply ratio; In step S6, the inter-regional water right allocation threshold is dynamically adjusted, and there is a certain correlation between the yield of unit water consumption and the water shortage penalty coefficient. Dynamic adjustment of water right allocation can improve economic benefit and reduce penalty, and then take the maximum coefficient yield as the adjustment target, and the calculation formula is: S=(P×L) / Q; In the formula, S represents the system yield coefficient, P represents the purchase price, L represents the unit area yield, and Q represents the unit area irrigation water consumption.

[0054] For the study area, the ratio of yield coefficient to penalty coefficient is set to 1:1.3, and the system yield and water shortage penalty coefficient under different water supply conditions in each region are shown in Table 1:

[0055] Table 1 The water supply capacity of water sources, crop water requirement and other related basic data of citrus planting areas in Anfusi Town of Zhijiang City in 2023 are substituted into the model, the maximum system benefit is taken as the objective function, the water supply capacity of water sources, crop water requirement, variable non-negative and the like are taken as constraint conditions, the linear programming model is solved by lingo, and the optimal water distribution scheme of Anfusi Town of Zhijiang City is shown in Table 2:

[0056] Table 2 The model constructed is verified by using the present situation data of the research area in 2023. The upper limit of the benefit of the citrus planting area A after optimization is 4958.70 million yuan, the original benefit is 4470.96 million yuan, and the benefit is increased by 10.91%; the upper limit of the benefit of the area B after optimization is 6672.70 million yuan, the original benefit is 6267.05 million yuan, and the benefit is increased by 6.47%; the upper limit of the benefit of the area C after optimization is 5993.95 million yuan, the original benefit is 5481.67 million yuan, and the benefit is increased by 9.35%. The overall benefit of Anfusi Town of Zhijiang City is increased by 8.67%. By comparing the original benefit and the optimization result, it is found that the model considers the water inflow under different water inflow levels, and effectively improves the comprehensive benefit of water resources to a certain extent, which can indicate the effectiveness of the model.

[0057] In addition, the agricultural water right allocation method according to the embodiment of the application can be implemented as an agricultural water right allocation system based on two-stage stochastic programming and neural network, comprising: A data acquisition module is configured to acquire meteorological data, hydrological data, crop growth parameters and planting area information. An LSTM prediction module is configured to generate time series prediction of irrigation demand and water supply capacity, and output water demand and water supply intervals under three water inflow levels of high, medium and low. A two-stage stochastic programming module is configured to solve a multi-stage water right allocation strategy containing interval parameters and probability scenarios, and quantify the optimal allocation scheme under climate uncertainty. A dynamic adjustment module is configured to update the water right allocation threshold according to real-time climate feedback, preferentially call local water sources and control the calling proportion of external water sources.

[0058] The above-mentioned embodiments are only preferred technical solutions of the application, and should not be regarded as limitations of the application. The protection scope of the application should be based on the technical solutions recited in the claims, including equivalent replacement solutions of the technical features recited in the claims. That is, equivalent replacement improvements within this range are also within the protection scope of the application.

Claims

1. The agricultural water rights allocation method based on two-stage stochastic programming and neural network is characterized by: The following steps are involved: S1: Collect historical meteorological data, hydrological data, crop growth parameters and planting area information to form data set D1, and calculate the net irrigation water requirement IR during the crop growth period; S2: Divide the planting areas according to the annual water supply of the water source; S3: Construct a long short-term memory neural network and input the data set D1 into the long short-term memory neural network to output the predicted irrigation water demand and water supply interval values, and divide the water demand and water supply intervals under three water inflow levels: high, medium and low; S4: Establish an interval two-stage stochastic programming model, with the pre-allocation target as the first-stage decision variable and the water shortage as the second-stage decision factor, combined with the probability distribution of different water inflow levels, and taking the maximization of the system net benefit as the objective function; S5: Introduce decision variables to transform the interval model into a deterministic sub-model, and solve the optimal water allocation target and allocated water volume through an interactive algorithm; S6: Based on the optimal water distribution target and allocated water volume, generate intelligent water rights allocation plans under different water inflow scenarios, and dynamically adjust the water rights allocation thresholds between regions.

2. The agricultural water rights allocation method based on two-stage stochastic programming and neural network according to claim 1 is characterized in that: In step S1, the hydrological data includes surface water resources, groundwater resources and river runoff data; the surface water resources data includes reservoir water storage data and weir pond water supply data.

3. The agricultural water rights allocation method based on two-stage stochastic programming and neural network according to claim 1 is characterized in that: Step S1 also includes collecting evapotranspiration, precipitation and irrigation data, and calculating effective rainfall using the soil water balance method. The calculation formula is: ； Where, P e represents the daily effective precipitation, P Indicates daily precipitation.

4. The agricultural water rights allocation method based on two-stage stochastic programming and neural network according to claim 1 is characterized in that: In step S3, the long short-term memory neural network is calculated as follows: ； ； ； ； Where, 、 、 and Represent the outputs of the forget gate, input gate, output gate, and candidate cell state, respectively. Represents the input at time t, i represents the sequence number of the input feature, represents the hidden state at the moment before time t, W 、 b Respectively represent the weight matrix and bias vector corresponding to each gating unit, p and q Represent the network input dimension and the number of hidden layer nodes, represents the sigmoid nonlinear activation function, represents the hyperbolic tangent activation function.

5. The agricultural water rights allocation method based on two-stage stochastic programming and neural network according to claim 4 is characterized in that: The updated cell state, unit output, and network output are calculated as follows: ； ； ； Where, Represents vector dot multiplication operation, represents the updated cell state, represents the unit output, Represents network output.

6. The agricultural water rights allocation method based on two-stage stochastic programming and neural network according to claim 1 is characterized in that: In step S3, the division of water demand and water supply intervals under three water inflow levels, namely high, medium and low, includes controlling the allocation risk according to the ratio of local external water sources.

7. The agricultural water rights allocation method based on two-stage stochastic programming and neural network according to claim 1 is characterized in that: The interval two-stage stochastic programming model is established, and the expression is: ； Where, f represents the net benefit of the system, B ij Indicates water source i To citrus growing areas j System benefits when distributing water, W ij Indicates water source i To citrus growing areas j Pre-allocation target, C ij Indicates water source i Unsatisfied planting area j Water shortage penalty coefficient when pre-allocating water target, S ij Indicates water source i Unmet citrus growing areas j Water shortage at the pre-allocation target, i Indicates different water source counting units, the maximum value is 1, j Represents the counting unit of each citrus growing area, with the maximum value being J.

8. The agricultural water rights allocation method based on two-stage stochastic programming and neural network according to claim 7 is characterized in that: The constraints include the available water quantity constraint, the crop water requirement constraint and the non-negative constraint, and the expression is: (1) Constraints on available water volume of water sources: ； (2) Crop water requirement constraints: ； (3) Non-negative constraints ； Where, Indicates water source i The maximum water supply, Indicates the j The minimum water requirement for normal growth of citrus in each planting area is: Indicates the j The maximum water requirement for normal growth of citrus in each planting area.

9. The agricultural water rights allocation method based on two-stage stochastic programming and neural network according to claim 1 is characterized in that: In step S5, the decision variables are introduced to transform the interval model into a deterministic sub-model, and the calculation formula is: ； Where, Indicates the interval value of pre-allocated water. =1, When the upper limit is reached, the corresponding pre-allocation target is the maximum value, and the risk is also the greatest; when =0, When the lower limit of the value is reached, the water allocation target is the minimum, the total benefit of water resources is the minimum, and the risk is the minimum.

10. The agricultural water rights allocation method based on two-stage stochastic programming and neural network according to claim 1 is characterized in that: In step S6, the dynamic adjustment of the inter-regional water rights allocation threshold is based on the certain correlation between the income generated by unit water consumption and the water shortage penalty coefficient. The water rights allocation is dynamically adjusted to improve economic benefits and reduce penalties, and the maximum coefficient benefit is set as the adjustment goal. The calculation formula is: ； Where, S represents the system profit coefficient, P represents the purchase price, L represents the yield per unit area, Q It indicates the amount of irrigation water used per unit area.

Images