Intelligent water distribution method for multi-layer water resource management

Abstract

The invention belongs to the field of water resource management, and discloses an intelligent water distribution method for multi-layer water resource management. Comprising the following steps: collecting comprehensive data, and integrating the comprehensive data in space by using an IDW-Kriging algorithm to obtain unified integrated data; a multi-module deep learning framework is adopted to deploy a water resource dynamic evolution mechanism, dynamic constraint and interaction conditions of water pollution and recovery, climate change and crop growth are introduced into the water resource dynamic evolution mechanism, and the water resource supply quantity and the water demand quantity of each field in the next cycle time are obtained; when the water resource supply quantity meets the water demand quantity of each field, full allocation is carried out, and when the water resource supply quantity does not meet the water demand quantity of each field, optimal allocation is carried out on the water demand quantity of each field in the next period by adopting an NSGA-II algorithm; and intelligent water distribution of multi-layer water resource management is realized.

Description

Technical Field

[0001] The present invention relates to the field of water resource management, and more specifically, to an intelligent water allocation method for multi-layer water resource management. Background Art

[0002] Multi-layer water resource management is an effective way to adapt to complex water resource issues. It achieves sustainable use of water resources through multi-level and multi-subject collaboration, taking into account economic, social and ecological factors. In multi-layer water resource management, intelligent water allocation is a key link in achieving efficient, equitable and sustainable use of water resources.

[0003] In the process of intelligent water allocation for multi-layer water resource management, many practical problems are faced. First of all, the sources of allocable water resources include rainfall, river water volume, groundwater level, and water storage capacity (reservoirs, lakes, etc.), and the objects of allocation include agricultural irrigation, industrial water, residential water, and ecological water. Moreover, the degree of water pollution and renewable recovery capacity will affect the allocable amount and quality of water resources. In addition, the uncertainty of climate and weather changes leads to fluctuations in water supply, which in turn increases the difficulty of allocation. Therefore, it is necessary to make intelligent and dynamic adjustments to water allocation according to actual conditions.

[0004] In view of this, the present invention proposes an intelligent water allocation method for multi-layer water resource management to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: an intelligent water allocation method for multi-layer water resource management, comprising: step S1: collecting comprehensive data, and using the IDW-Kriging algorithm to spatially integrate the comprehensive data to obtain unified integrated data; Step S2: deploy the dynamic evolution mechanism of water resources, and use the deep learning framework to set up a multi-task output module in the dynamic evolution mechanism of water resources, introduce a lightweight simplification layer and dynamic constraints and interactive auxiliary strategies to optimize the mechanism; The multi-task output module is pre-trained separately, and then the output of the multi-task output module is input into the lightweight simplified layer for feature fusion and dimensionality reduction, optimizing the entire mechanism and completing the pre-training; Input the integrated data of the current cycle time into the pre-trained water resources dynamic evolution mechanism to obtain the water resources supply and water demand in various fields in the next cycle time; Step S3: When the water supply meets the water demand of each field, full allocation is performed; when the water supply does not meet the water demand of each field, the NSGA-II algorithm is used to optimally allocate the water demand of each field in the next cycle; Step S4: Arrange the allocation process and results on a visual interface.

[0006] Preferably, the method for collecting comprehensive data includes: Comprehensive data include supply data, water quality data, demand data and environmental data; Supply data include rainfall, river flow, groundwater levels, and water storage; Water quality data include water pollution index; Demand data include agricultural irrigation, industrial water, residential water and ecological water; Environmental data include temperature, humidity, wind speed, and evaporation in the water allocation area; Comprehensive data is collected through the Internet of Things, remote sensing technology and measuring equipment, and the data is transmitted to the cloud platform for storage using 5G or LoRa networks.

[0007] Preferably, the method of using the IDW-Kriging algorithm to spatially integrate the comprehensive data to obtain unified integrated data includes: Use Python's Pandas library to load data from the cloud platform; Use the Z-score method to detect and delete outliers, and use interpolation to fill in missing values; Standardize the data format, including unified time format, unified unit, unified time resolution, unified coordinate system and normalization; Design the blank space with the water resource allocation area as the spatial boundary, divide the blank space into regular grids with a spatial resolution of 1km×1km, assign a unique ID to each grid cell, and record the longitude and latitude coordinates of its center point. Store the grid information as a DataFrame, and the columns of the DataFrame include the grid ID, the longitude of the center point, and the latitude of the center point. Identify the spatial data in the comprehensive dataset, use the IDW-Kriging algorithm for spatial interpolation, calculate the interpolation results at the center point of each grid for each data type, and store the interpolation results as a DataFrame with columns containing grid ID, timestamp, and data category; Identify non-spatial data in the synthetic dataset and merge the non-spatial data with the spatial interpolation results, using the grid ID and timestamp as primary keys; All processed comprehensive data sets are merged into a DataFrame by grid ID and timestamp. The columns of the DataFrame include: grid ID, timestamp, supply data, water quality data, demand data and environmental data to obtain integrated data; Store aggregate data in CSV, HDF5 or GeoTIFF formats.

[0008] Preferably, the method for performing spatial interpolation using the IDW-Kriging algorithm comprises: The created 1km×1km blank grid is recorded as the target point set, and the center point of each grid is the interpolation target point. The coordinates of the monitoring points and the observed values ​​in the comprehensive data set are recorded as the target point set. Recorded as monitoring point data; For each target point , calculate the Euclidean distance between it and all monitoring points, and use the inverse of the square of the Euclidean distance as the weight , calculate the IDW interpolation result of the target point ,in, is the total number of monitoring point data, is the index of the target point, is the index of the monitoring point; Calculate the Kriging interpolation result of the target point ,in, is the weight of each monitoring point; The IDW interpolation error and Kriging interpolation error of the IDW interpolation result and Kriging interpolation result are calculated using the root mean square error, and the inverse of the ratio of the IDW interpolation error to the sum of the IDW interpolation error and the Kriging interpolation error is used as the weight of the IDW interpolation. The inverse of the ratio of the Kriging interpolation error to the sum of the IDW interpolation error and the Kriging interpolation error is used as the weight of the Kriging interpolation. , the interpolation result is weightedly fused with the weight to obtain the fused interpolation result, which is recorded as the interpolation result of the IDW-Kriging algorithm.

[0009] Preferably, the method for constructing the water resources dynamic evolution mechanism includes: The dynamic evolution mechanism of water resources includes input, multi-task output module, lightweight simplification layer and output; The input of the dynamic evolution mechanism of water resources is set as integrated data, and the output is the water supply of the next cycle and the water demand in various fields; The deep learning framework of the multi-task output module includes a time series modeling module, a water quality dynamic modeling module, and a demand prediction module; The time series modeling module uses a long short-term memory network to model time series data, uses a Transformer model to capture long-term dependencies, and predicts the output as the supply data and environmental data for the next cycle. The water quality dynamic modeling module uses a state space model to describe the changes in the water pollution index, introduces water quality constraints, and uses a neural network to model the nonlinear changes in water quality recovery capacity. The output is the water pollution index prediction for the next cycle and the dynamic adjustment coefficient of the allocable water volume. The dynamic adjustment coefficient is based on the water quality constraint. The water quality constraint is to reduce the allocable proportion of the water body when the pollution index exceeds a certain threshold. The demand forecasting module uses the AquaCrop crop growth model combined with agricultural irrigation data to estimate dynamic irrigation demand, uses the ARIMA time series model combined with industrial water use data of historical periods to predict industrial water demand, uses the XGBoost regression model combined with residential water use data to predict residential water demand, and uses the minimum ecological flow constraint rule model combined with ecological water use data to estimate ecological water demand; The lightweight simplification layer includes feature fusion and dimensionality reduction. Its input is the output of the three task modules, and its output is the fused low-dimensional features, which are used for the final output prediction of the dynamic evolution mechanism of water resources. Introducing dynamic constraints and interactive auxiliary strategies in multi-task output modules, including water pollution and restoration, climate change, and crop growth; The multi-task output module uses a multi-task learning framework to share the time series characteristics of the integrated data and the characteristics of the dynamic constraints and interactive auxiliary strategies to predict the water supply and water demand in various fields respectively; The weighted sum of mean square error and mean absolute error is used as the comprehensive loss function of the dynamic evolution mechanism of water resources to impose output constraints; According to the input and output of each task module and model, the data on the historical cycle time are extracted as the training set. Each task module is trained separately first, and then the dynamic evolution mechanism of water resources is trained as a whole. The training is stopped when the loss function value is the smallest, and the pre-trained dynamic evolution mechanism of water resources is obtained.

[0010] Preferably, the method for introducing dynamic constraints and interactive auxiliary strategies in the multi-task output module includes: Use graph neural networks to model the interaction between water pollution and restoration, climate change and crop growth: define the first node as the water pollution index, the second node as the climate characteristics, and the third node as the crop growth characteristics; The first side is defined as the impact intensity of climate characteristics on water pollution index; the second side is defined as the impact intensity of water pollution index on crop growth characteristics; the third side is defined as the impact intensity of climate characteristics on crop growth characteristics; the fourth side is defined as the feedback impact of crop growth characteristics on water pollution index; The GNN update rule is used to output the dynamic adjustment coefficient of each node, among which the water pollution adjustment coefficient is used to correct the allocable water; the climate adjustment coefficient is used to correct the supply; the crop growth adjustment coefficient is used to correct the irrigation demand; the attention mechanism is used to dynamically adjust the interaction weights, and the attention mechanism is used to dynamically adjust the weights of the edges in the graph neural network.

[0011] Preferably, the method of outputting the dynamic adjustment coefficient of each node using the GNN update rule includes: Collect time series data of water pollution index, climate characteristics and crop growth characteristics, record them as original features, define the time window length as R_U, and for each node, use LSTM time series model to encode historical data, generate time embedding features, and concatenate the original features and time embedding features to obtain complete node features; For each edge, a two-layer MLP is defined to perform nonlinear transformation on the interaction features between nodes. The complete node features are input into the corresponding MLP to obtain the nonlinear interaction features. The nonlinear interaction features are concatenated with the complete node features to obtain the final node features. The final node features are used to update and replace the original features.

[0012] Preferably, the method of using the NSGA-II algorithm to optimally allocate the water demand for each field in the next cycle time includes: Step A1: Define the decision variable as the water allocation for each field in the next cycle; define the constraint condition as the water allocation must be greater than 0 and less than or equal to the maximum water allocation, and the total water allocation for all fields cannot exceed the available water supply; The optimization objectives are defined as maximizing economic benefits, maximizing ecological benefits, and minimizing social equity; Step A2: Randomly generate an initial population, set the population size N, and each individual is an allocation plan. Randomly generate allocation plans and adjust them to meet the supply constraint. If the initial solution does not meet the constraint, it can be adjusted by normalization. For each individual in the initial population, calculate the objective function value and record it; Step A3: Perform non-dominated sorting on the population, divide the Pareto frontier, assign a non-dominated level to each individual, and calculate the crowding distance for each individual in the frontier: Step A4: Use tournament selection to generate the parent population, use simulated binary crossover to generate offspring, use polynomial mutation to mutate the offspring, merge the parent population and offspring population to obtain a joint population with a size of 2N; Step A5: Repeat steps A2 to A4 until the objective function value stops changing, and output the first non-dominated frontier in the final population, which is the optimal solution set. Each solution corresponds to an allocation plan. According to actual needs, the solution that meets specific conditions is selected as the optimal allocation plan for output; Step A6: Use the optimal allocation plan to optimally allocate the water demand in each field in the next cycle.

[0013] Preferably, the visualization interface includes a terminal page of a computer or a display interface of a mobile device, and performs interface interaction with the user.

[0014] Preferably, arranging the allocation process and results on a visual interface further comprises transmitting the optimal allocation report to a user terminal via wireless transmission and issuing a warning reminder.

[0015] Technical effects and advantages of the intelligent water allocation method for multi-layer water resource management of the present invention: 1. The IDW-Kriging algorithm is used to perform spatial interpolation and integration of multi-source comprehensive data (supply data, water quality data, demand data, and environmental data), solving the problem of uneven and missing data in spatial distribution. The generated integrated data has a uniform spatial resolution (1km×1km grid) and temporal consistency, providing a high-quality data foundation for subsequent model input.

[0016] 2. A multi-module deep learning framework (time series modeling, water quality dynamic modeling, demand forecasting) is used to build a dynamic evolution mechanism for water resources, combined with dynamic constraints of water pollution and restoration, climate change, and crop growth. The model can dynamically predict the supply and demand in various fields in the next cycle, adapt to environmental changes (such as climate and pollution) and demand fluctuations (such as crop growth stages), and improve the prediction accuracy. The dynamic constraints of water pollution, climate change, and crop growth are introduced, and the interactive relationship is modeled through GNN, which enhances the adaptability and robustness of the model.

[0017] 3. When water resources are insufficient, the NSGA-II algorithm is used for multi-objective optimization (maximizing economic benefits, maximizing ecological benefits, and minimizing social fairness) to generate the Pareto optimal solution set. This achieves a balance between economic, ecological, and social goals, ensures the fairness and efficiency of the allocation plan, and meets water demand in different fields.

[0018] This solution solves the spatial consistency, dynamic adaptability and multi-objective balance problems in water resource allocation through multi-source data integration, dynamic modeling, multi-objective optimization and user interaction. Its technical effects include improved data integration accuracy, enhanced prediction accuracy, optimized allocation fairness and improved user experience; technical advantages include efficient data processing, comprehensive dynamic modeling, flexible optimization algorithms and practical visualization functions. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of the steps of an intelligent water allocation method for multi-layer water resource management according to the present invention; Figure 2 It is a structural schematic diagram of an intelligent water distribution system for multi-layer water resource management of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Example 1

[0022] See also Figure 1 and Figure 2 As shown, this embodiment provides an intelligent water allocation method for multi-layer water resource management, including: In the process of intelligent water allocation for multi-layer water resource management, many practical problems are faced. First of all, the sources of allocable water resources include rainfall, river water volume, groundwater level, and water storage capacity (reservoirs, lakes, etc.), and the objects of allocation include agricultural irrigation, industrial water, residential water, and ecological water. Moreover, the degree of water pollution and its biorecovery capacity will affect the allocable amount and quality of water resources. In addition, the uncertainty of climate and weather changes leads to fluctuations in water supply, which in turn increases the difficulty of allocation. Therefore, it is necessary to make intelligent and dynamic adjustments to water allocation according to actual conditions.

[0023] Step S1: Collect the comprehensive data, and use the IDW-Kriging algorithm to integrate the comprehensive data in space to obtain unified integrated data; Methods for collecting comprehensive data include: Comprehensive data include supply data, water quality data, demand data and environmental data; The supply data include rainfall, river flow, groundwater level and water storage capacity; rainfall can be collected through meteorological stations and satellite remote sensing, river flow can be collected through flow meters, groundwater level can be collected through monitoring wells, and water storage capacity can be collected by placing water level meters on reservoirs and lakes; Water quality data include water pollution indexes, such as the content or percentage of COD, BOD, ammonia nitrogen and heavy metals, which can be collected through water quality monitoring stations and sensors; Demand data include agricultural irrigation, industrial water, residential water and ecological water; Agricultural irrigation generally needs to include farmland area, crop type, growth cycle and soil moisture to calculate water consumption. Similarly, industrial water use generally includes industrial type, production scale and water quota, residential water use generally includes population size, water use habits, historical water use data, and ecological water use generally includes wetland area, minimum ecological flow of rivers, and ecological restoration needs. These data also need to be collected using measuring equipment. For example, soil moisture is measured by soil moisture sensors, historical water use data can be retrieved through smart water meters, and wetland area can be measured through remote sensing technology.

[0024] Environmental data include temperature, humidity, wind speed, and evaporation in the water allocation area, which are collected through weather stations; Comprehensive data is collected through the Internet of Things, remote sensing technology and measuring equipment, and the data is transmitted to the cloud platform for storage using 5G or LoRa networks. The role of the Internet of Things is to form a sensor network with various measuring devices to collect water level, flow, water quality and other data in real time and transmit them wirelessly. Remote sensing technology uses satellites and drones to monitor rainfall distribution, land use and water coverage as an example. Measuring equipment is the equipment used for the above data collection, such as water level meters. Cloud platforms such as Alibaba Cloud and AWS support distributed storage and high efficiency.

[0025] The IDW-Kriging algorithm is used to integrate the comprehensive data spatially. The methods for obtaining unified integrated data include: Use Python's Pandas library to load data from the cloud platform in CSV, Excel, and JSON formats to gain a preliminary understanding of data characteristics and identify potential problems (such as missing values, outliers, inconsistent formats, etc.).

[0026] Use the Z-score method to detect and delete outliers, and use interpolation to fill in missing values; The task of outlier detection is to identify and process outliers caused by sensor failure, data entry errors, etc., calculate the mean and standard deviation of the data, and use the Z score method to detect outliers. Usually, points with Z > 3 or Z < -3 are considered outliers. If the proportion of outliers is small and does not affect the overall analysis, they can be deleted directly. Use the mean, median or interpolation method to fill in. You can also use Python's Pandas and Scikit-learn (KNN, regression model) for filling.

[0027] Standardize the data format to ensure uniform data format for subsequent integration and analysis, including unified time format, unified unit, unified time resolution, unified coordinate system and normalization; Unifying the time format means converting timestamps from different sources (such as "2023-01-01", "01 / 01 / 2023") into a standard format (such as ISO 8601: YYYY-MM-DD HH:MM:SS). Unifying units means, for example, unifying the rainfall unit into millimeters (mm) and the flow unit into cubic meters per second (m³ / s). Unifying the time resolution: for example, aggregating hourly data into daily data. Data normalization means scaling the data to the [0, 1] interval. Coordinate system - Ensure that all spatial data (such as monitoring station coordinates, remote sensing images) use the same geographic coordinate system (such as WGS84) or projection coordinate system (such as UTM).

[0028] Design the blank space with the water resource allocation area as the spatial boundary (e.g., longitude range: E 120°-E 122°, latitude range: N 30°-N 32°), divide the blank space into regular grids with a spatial resolution of 1km×1km, assign a unique ID to each grid cell, and record the longitude and latitude coordinates of its center point. Store the grid information as a DataFrame, where the columns of the DataFrame include the grid ID, the longitude of the center point, and the latitude of the center point. Identify spatial data in the comprehensive data set. For example, if the rainfall comes from a meteorological station, use the station coordinates; the river water volume uses the hydrological station coordinates; the groundwater level uses the monitoring well coordinates; the water storage volume uses the reservoir location coordinates. Use the IDW-Kriging algorithm for spatial interpolation. The interpolation can be implemented using Python libraries (such as pykrige, scipy) or GIS tools (such as ArcGIS's Geostatistical Analyst). For each data type (such as rainfall, COD concentration, population density, etc.), calculate the interpolation results at each grid center point separately, and store the interpolation results as a DataFrame, with columns containing grid ID, timestamp, and data category; Identify non-spatial data in comprehensive data sets, such as COD, BOD, ammonia nitrogen, and heavy metal concentrations in water quality data, and directly associate them with the coordinates of the monitoring stations. Merge non-spatial data (such as water quality indicators and water consumption) with spatial interpolation results (the value of the interpolation result at the center of each grid), using the grid ID and timestamp as the primary key; ensure the consistency of data in spatial and temporal dimensions. For example, in agricultural irrigation data, water consumption is allocated to the corresponding grid based on the farmland distribution layer, and in industrial water use, water consumption is allocated to the corresponding grid based on the industrial park distribution layer.

[0029] All processed comprehensive data sets are merged into a DataFrame by grid ID and timestamp. The columns of the DataFrame include: grid ID, timestamp, supply data, water quality data, demand data and environmental data to obtain integrated data; A more detailed description of the columns of DataFrame is as follows: Grid ID; Timestamp; Supply data: rainfall, river flow, groundwater level, water storage capacity; Water quality data: COD, BOD, ammonia nitrogen, heavy metal concentration; Demand data: agricultural irrigation water, industrial water, residential water, ecological water; Environmental data: temperature, humidity, wind speed, evaporation.

[0030] Check whether there are missing values ​​in the merged DataFrame. If so, use interpolation (such as linear interpolation) or default values ​​to fill them. Check whether the data distribution is reasonable (such as whether the rainfall is non-negative and whether the COD concentration is within a reasonable range).

[0031] Store the integrated data in CSV, HDF5 or GeoTIFF format for subsequent analysis.

[0032] If visualization is required, GIS tools or Python libraries (such as Matplotlib and Seaborn) can be used to generate spatial distribution maps. In addition, data consistency checks should be performed to check the logical relationship between supply data, demand data, water quality data, and environmental data (such as whether increased rainfall leads to increased river water volume), and whether water consumption meets the supply and demand balance (such as whether total water consumption is less than total supply).

[0033] Methods for spatial interpolation using the IDW-Kriging algorithm include: The created 1km×1km blank grid is recorded as the target point set, and the center point of each grid is the interpolation target point. The monitoring point coordinates (latitude and longitude coordinates of the rainfall observation station) and the observation value in the comprehensive data set are recorded as the target point set. (e.g. rainfall) is recorded as monitoring point data; For each target point , calculate the Euclidean distance between it and all monitoring points , using the inverse of the square of the Euclidean distance as the weight , calculate the IDW interpolation result of the target point ,in, is the total number of monitoring point data, is the index of the target point, is the index of the monitoring point, and the interpolation results of each target point are collected and stored as an IDW interpolation dataset; Calculate the Kriging interpolation result of the target point ,in, is the weight of each monitoring point; obtained by the Kriging equations, and meets the unbiased and minimum variance conditions, that is, , It is a monitoring point and target point The semivariance between is the semivariance between the target point j and all monitoring points; is the Lagrange multiplier used to satisfy the unbiasedness condition.

[0034] Use the root mean square error to calculate the IDW interpolation error and Kriging interpolation error of the IDW interpolation result and Kriging interpolation result respectively. ,in, Represents IDW interpolation error or Kriging interpolation error, For IDW interpolation results or Kriging interpolation results, the inverse of the ratio of IDW interpolation error to the sum of IDW interpolation error and Kriging interpolation error is used as the weight of IDW interpolation The inverse of the ratio of the Kriging interpolation error to the sum of the IDW interpolation error and the Kriging interpolation error is used as the weight of the Kriging interpolation. , the interpolation result is weightedly fused with the weight to obtain the fused interpolation result, which is recorded as the interpolation result of the IDW-Kriging algorithm.

[0035] By combining the IDW and Kriging interpolation methods, the local interpolation advantages of IDW and the spatial autocorrelation analysis capabilities of Kriging can be fully utilized to generate more accurate interpolation results. Finally, the integrated data is generated, and the fusion weights need to be dynamically adjusted according to the error, distance or spatial autocorrelation to ensure the accuracy and robustness of the interpolation results. This method is suitable for spatial data processing in the fields of water resources management and environmental monitoring, and has high practicality and scalability.

[0036] Step S2: deploy the dynamic evolution mechanism of water resources, and use the deep learning framework to set up a multi-task output module in the dynamic evolution mechanism of water resources, introduce a lightweight simplification layer and dynamic constraints and interactive auxiliary strategies to optimize the mechanism; The multi-task output module is pre-trained separately, and then the output of the multi-task output module is input into the lightweight simplified layer for feature fusion and dimensionality reduction, optimizing the entire mechanism and completing the pre-training; Input the integrated data of the current cycle time into the pre-trained water resources dynamic evolution mechanism to obtain the water resources supply and water demand in various fields in the next cycle time; The construction methods of the dynamic evolution mechanism of water resources include: The dynamic evolution mechanism of water resources includes input, multi-task output module, lightweight simplification layer and output; The input of the dynamic evolution mechanism of water resources is set as integrated data, and the output is the water supply of the next cycle and the water demand of various fields. The fields here refer to agricultural irrigation, industrial water, residential water, and ecological water. The deep learning framework of the multi-task output module includes a time series modeling module, a water quality dynamic modeling module, and a demand prediction module; Among them, the time series modeling module uses long short-term memory networks to model time series data, that is, it processes the time series characteristics of rainfall, river water volume, groundwater level, water storage capacity, temperature, humidity, etc. respectively, and uses the Transformer model to capture long-term dependencies. It is particularly suitable for long-term trend prediction of rainfall and climate data. The prediction output is the prediction of supply data and environmental data for the next cycle; that is, it includes the predicted values ​​of rainfall, river water volume, groundwater level and water storage capacity, and the prediction of environmental data includes the predicted values ​​of temperature, humidity, wind speed and evaporation.

[0037] The function of the water quality dynamic modeling module is to model the dynamic changes of the water pollution index based on water quality data and pollution source data, and estimate the distributable water volume in combination with the water quality recovery capacity. The state space model is used to describe the changes in the water pollution index (COD, BOD, ammonia nitrogen, etc.), considering the input of the pollution source and the output of the water quality recovery capacity, introducing water quality constraints, and using neural networks (such as MLP or RNN) to model the nonlinear changes in water quality recovery capacity, combined with the dynamic adjustment of ecological restoration measures. The output is the prediction of the water pollution index for the next cycle and the dynamic adjustment coefficient of the distributable water volume. The dynamic adjustment coefficient is based on the water quality constraint. The water quality constraint is to reduce the distributable proportion of the water body when the pollution index exceeds a certain threshold. The threshold is pre-set; The function of the demand forecasting module is to predict the demand for agricultural irrigation, industrial water, domestic water and ecological water in the next cycle based on demand data and environmental data. The AquaCrop crop growth model is used in combination with agricultural irrigation data, i.e. soil moisture, crop type and growth cycle, to estimate dynamic irrigation demand. Environmental data (such as temperature and evaporation) are introduced to adjust irrigation demand to reflect the impact of climate change. The ARIMA time series model is used in combination with industrial water data of historical cycle time to predict industrial water demand. The XGBoost regression model is used in combination with domestic water data, i.e. population size, water use habits and historical data, to predict residential water demand. The minimum ecological flow constraint rule model is used in combination with ecological water data, i.e. wetland area and ecological restoration needs, to estimate ecological water demand; therefore, the output is the agricultural irrigation demand, industrial water demand, domestic water demand and ecological water demand in the next cycle.

[0038] However, the complexity of multi-model fusion and multi-task learning may lead to high computational cost, difficulty in training and risk of overfitting. To address this problem, a lightweight simplification layer is designed, including feature fusion and dimensionality reduction. Its input is the output of the three task modules, and its output is the fused low-dimensional features, which are used for the final output prediction of the dynamic evolution mechanism of water resources; The structure of the lightweight simplified layer can be set according to actual needs, for example, fully connected layer + dimensionality reduction (the output features of the three task modules are concatenated into a high-dimensional feature vector, and then the dimensionality is reduced through a fully connected layer), or principal component analysis (PCA) or autoencoder dimensionality reduction (PCA or autoencoder is used to reduce the dimensionality of the concatenated features and extract the most important features).

[0039] By designing a lightweight simplified layer, the computational cost of the entire model is reduced and the training speed of the model is accelerated. It can also be used as a "task integration layer" in the multi-task learning framework to fuse or reduce the dimension of the output of each task module and extract shared features or task-specific features. Its goal is to compress the high-dimensional features from the three task modules into low-dimensional features while retaining key information for the final prediction task.

[0040] Introducing dynamic constraints and interactive auxiliary strategies in multi-task output modules, including water pollution and restoration, climate change, and crop growth; The multi-task output module uses a multi-task learning framework to share the time series characteristics of the integrated data and the characteristics of the dynamic constraints and interactive auxiliary strategies to predict the water supply and water demand in various fields respectively; The function of the multi-task output module is to integrate the outputs of the time series modeling module, the water quality dynamic modeling module, the demand forecasting module and the dynamic constraint module to generate the final forecast results. It uses a multi-task learning framework to share time series features and dynamic constraint features to predict the water resource supply and water demand in various fields respectively. The weighted sum of mean square error and mean absolute error is used as the loss function to impose output constraints, balance the accuracy of supply forecast and demand forecast, and consider the priority of ecological water use; According to the input and output of each module and model, the historical cycle data is extracted as the training set. Each module is trained separately first, and then the model is trained as a whole. The training is stopped when the loss function value is the smallest, and the pre-trained dynamic evolution mechanism of water resources is obtained.

[0041] During the training process, the lightweight simplified layer needs to match the training strategy of the entire mechanism. First, the three task modules (time series modeling module, water quality dynamic modeling module, and demand forecasting module) are trained separately to optimize their respective loss functions. At this stage, the lightweight simplified layer can temporarily not participate in the training, or only use simple splicing and dimensionality reduction (such as the fully connected layer).

[0042] In the overall training phase, the outputs of the three task modules are input into the lightweight layer, which performs feature fusion and dimensionality reduction. The comprehensive loss function of the entire mechanism is optimized, in which the parameters of some task modules (such as LSTM and Transformer of the time series modeling module) can be frozen, and only the lightweight layer and task-specific layers are fine-tuned to reduce the training difficulty and computational cost.

[0043] During the overall training phase, the validation set is used to monitor the comprehensive loss function, and training is stopped when the loss function value no longer decreases to avoid overfitting.

[0044] This paper designs a dynamic evolution mechanism of water resources to meet the needs of multi-task forecasting (supply and demand forecasting), and introduces dynamic constraints of water pollution and restoration, climate change, and crop growth and their mutual influence modeling. The model adopts a multi-module structure, including a time series modeling module, a water quality dynamic modeling module, a demand forecasting module, and a dynamic constraint and interactive auxiliary strategy. The final forecast results and allocation recommendations are generated through a multi-task learning framework. Comprehensively consider supply data, water quality data, demand data, and environmental data to achieve comprehensive forecasting. Introduce dynamic constraints and interactive modeling to reflect the complex relationship between water pollution, climate change, and crop growth. Support real-time updates and dynamic adjustments to adapt to climate fluctuations and demand changes. Provide scientific support for the sustainable use of water resources.

[0045] The interaction between water pollution and restoration, climate change, and crop growth is a complex nonlinear system, including: 1. Interaction between water pollution and climate change: Concentrated rainfall (such as heavy rain) may lead to increased surface runoff, washing pollutants such as industrial wastewater, agricultural fertilizers and pesticides into water bodies, thereby exacerbating water pollution. This phenomenon is particularly evident in areas with high rainfall concentration and low vegetation coverage. High temperatures and high evaporation will accelerate the concentration of pollutants in water bodies (such as heavy metals and ammonia nitrogen) and reduce the ability of water quality to recover. For example, high temperatures may accelerate the decomposition of organic matter and increase the concentration of BOD and COD. Reduced rainfall (such as drought) may reduce the dilution capacity of water bodies, leading to an increase in the pollution index and indirectly reducing the amount of water that can be allocated.

[0046] 2. Interaction between water pollution and crop growth: The use of polluted water (such as excessive heavy metals and ammonia nitrogen) for irrigation may have toxic effects on crop growth and reduce crop yields. For example, heavy metal accumulation in the soil may inhibit root growth and affect the absorption of water and nutrients by crops. Ecological water use (such as wetland restoration and river ecological flow) improves water quality and may reduce the negative impact of polluted water on crops. For example, wetland restoration can filter pollutants and reduce the heavy metal content in irrigation water. Long-term use of polluted water for irrigation may lead to soil pollution, reduce the irrigable area of ​​farmland, and indirectly reduce the demand for agricultural water.

[0047] 3. Interaction between climate change and crop growth: High temperatures and droughts directly increase crop transpiration and significantly increase crop water requirements. For example, rice water requirements at high temperatures may be 20%-30% higher than normal temperatures. Reduced rainfall reduces soil moisture and increases irrigation demand. For example, when soil moisture is below the crop water requirement threshold, irrigation demand may double. Climate fluctuations (such as concentrated rainfall) may hinder crop growth. For example, heavy rains may cause waterlogging in farmland, affecting crop root respiration and reducing yields.

[0048] 4. The feedback mechanism of the three factors: Climate change (such as high temperature and drought) may simultaneously aggravate water pollution and crop water demand. For example, drought reduces water flow and increases pollutant concentration; at the same time, crop water demand increases, leading to an intensified contradiction between irrigation demand and available water. Ecological water use (such as wetland water replenishment) may improve water quality and crop growth environment at the same time. For example, wetland water replenishment can reduce the water pollution index and reduce the toxic effects of irrigation water; but increased ecological water use may reduce the available water for agriculture and industry, leading to allocation conflicts. Water pollution (such as excessive heavy metals) may affect crop growth through soil pollution, indirectly increasing the pressure of climate change (such as drought) on agricultural water demand. For example, polluted soil reduces crop yields, and farmers may increase irrigation to make up for yield losses.

[0049] Therefore, in order to capture the complexity and dynamics of the above-mentioned interactions, the model needs to combine graph neural networks (GNNs), attention mechanisms, and multi-task loss functions for modeling.

[0050] Introducing dynamic constraints and interactive assistance strategies for water pollution and restoration, climate change, and crop growth. Specific methods include: Use graph neural networks to model the interaction between water pollution and restoration, climate change, and crop growth: define the first node as the water pollution index, including COD, BOD, ammonia nitrogen, heavy metal content, etc. The second node is the climate characteristics, including temperature, evaporation, rainfall, rainfall concentration, etc. The third node is the crop growth characteristics, including crop water demand, crop yield, soil moisture, etc.; The first side is defined as the intensity of the impact of climate characteristics on the water pollution index; for example, the positive impact of rainfall concentration on the pollution index, and the negative impact of high temperature on the water quality recovery rate. The second side is the intensity of the impact of the water pollution index on crop growth characteristics; for example, the negative impact of the pollution index on crop yield. The third side is the intensity of the impact of climate characteristics on crop growth characteristics. For example, the positive impact of temperature and evaporation on crop water demand; the fourth side is the feedback effect of crop growth characteristics on the water pollution index. For example, increased irrigation demand may aggravate water pollution (through the washout of fertilizers and pesticides); The GNN update rule is used to output the dynamic adjustment coefficient of each node, among which the water pollution adjustment coefficient is used to correct the allocable water; the climate adjustment coefficient is used to correct the supply; the crop growth adjustment coefficient is used to correct the irrigation demand; the attention mechanism is used to dynamically adjust the interaction weights, and the attention mechanism is used to dynamically adjust the weights of the edges in the graph neural network to reflect the real-time impact intensity of the current time step. For example, during droughts, the weight of the impact of climate characteristics (such as high temperature and reduced rainfall) on crop water demand should be higher; during heavy rains, the weight of the impact of concentrated rainfall on water pollution should be higher.

[0051] The specific method of using the attention mechanism to dynamically adjust the weight of edges in the graph neural network can be: input the pollution index, climate characteristics, and crop growth characteristics of the current time step; calculate the query, key, and value; calculate the attention weight and obtain the weighted interaction characteristics; the output of the attention mechanism is the weight of each interaction relationship, which is used to dynamically adjust the edge weights in the GNN. For example: the weight of the impact of climate characteristics on the water pollution index; the weight of the impact of water pollution index on crop growth characteristics; the weight of the impact of climate characteristics on crop growth characteristics.

[0052] Methods for using GNN update rules to output dynamic adjustment coefficients of each node include: Collect time series data of water pollution index, climate characteristics and crop growth characteristics, record them as original features, define the time window length as R_U, which means the historical information of the first R_U time steps is considered at the current cycle time. For example, R_U=5 means considering the data of the 5 time steps before the current time step. For each node, use the LSTM time series model to encode the historical data, generate time embedding features, and concatenate the original features and time embedding features to obtain complete node features; For each edge, a two-layer MLP is defined to perform nonlinear transformation on the interaction features between nodes. The complete node features are input into the corresponding MLP to obtain the nonlinear interaction features. The nonlinear interaction features are concatenated with the complete node features to obtain the final node features. The final node features are used to update and replace the original features.

[0053] LSTM is introduced to capture the dynamic dependencies in the time dimension. Multi-layer perceptron (MLP) is introduced to perform nonlinear modeling of the interaction features between nodes, enhancing the model's ability to express complex feedback mechanisms.

[0054] Step S3: When the water supply meets the water demand of each field, full allocation is performed; when the water supply does not meet the water demand of each field, the NSGA-II algorithm is used to optimally allocate the water demand of each field in the next cycle; The method of using NSGA-II algorithm to optimally allocate water demand in various fields in the next cycle time includes: Step A1: Define the decision variable as the water allocation for each field in the next cycle; define the constraint condition as the water allocation must be greater than 0 and less than or equal to the maximum water allocation, that is, the water allocation for each field cannot exceed the water demand for the field in the t+1 cycle, and the total water allocation for all fields cannot exceed the available water supply; The optimization objectives are defined as maximizing economic benefits, maximizing ecological benefits, and minimizing social equity (the goal is to minimize the gap in satisfaction rates in each field); Step A2: Randomly generate an initial population, set the population size N (for example, N = 100), each individual (solution) is an allocation plan, randomly generate allocation plans, and adjust them to meet the supply constraint. If the initial solution does not meet the constraint, it can be adjusted by normalization; ensure that each allocation plan meets the constraint, and calculate and record the objective function value for each individual in the initial population; the objective function value here is calculated by the defined optimization goal; Step A3: Perform non-dominated sorting on the population, divide the Pareto frontier, and assign a non-dominated rank to each individual. The rank of the first frontier is 1, the rank of the second frontier is 2, and so on. For the individuals in each frontier, calculate the crowding distance to measure the distribution density of the solution: the crowding distance of the boundary individuals is set to infinity to retain the boundary solution. The larger the crowding distance, the sparser the individual distribution, and the more priority it has to retain; Step A4: Use tournament selection to generate the parent population, use simulated binary crossover to generate offspring, use polynomial mutation to mutate the offspring, merge the parent population and offspring population to obtain a joint population with a size of 2N; Step A5: Repeat steps A2 to A4 until the objective function value stops changing. Output the first non-dominated frontier in the final population, which is the optimal solution set. Each solution corresponds to an allocation plan. According to actual needs (such as giving priority to ecological water use or residential water use), select the solution that meets specific conditions as the optimal allocation plan for output; Step A6: Use the optimal allocation plan to optimally allocate the water demand in each field in the next cycle.

[0055] Step S4: Arranging the allocation process and results on a visual interface; The visual interface includes the terminal page of the computer or the display interface of the mobile device, and interacts with the user.

[0056] Arranging the allocation process and results on a visual interface also includes transmitting the optimal allocation report to the user terminal via wireless transmission and issuing a warning reminder.

[0057] Users perform interactive operations such as clicking, searching and downloading through the computer page (computer display screen, where the computer refers to a smart device, and the mobile device can be a mobile phone or tablet).

[0058] Example 2

[0059] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides an intelligent water allocation system for multi-layer water resource management, including: Data collection and integration module: used to collect comprehensive data, use IDW-Kriging algorithm to integrate the comprehensive data in space, and obtain unified integrated data; Model construction and use module: A multi-module deep learning framework is used to deploy the dynamic evolution mechanism of water resources, and the dynamic constraints and interaction conditions of water pollution and restoration, climate change and crop growth are introduced into the dynamic evolution mechanism of water resources. The integrated data of the current cycle time is input into the pre-trained dynamic evolution mechanism of water resources to obtain the water supply of the next cycle time and the water demand of various fields; Solution optimization module: When the water supply meets the water demand of each field, full allocation is carried out; when the water supply does not meet the water demand of each field, the NSGA-II algorithm is used to optimally allocate the water demand of each field in the next cycle; Visualization module: Arrange the allocation process and results on a visual interface.

[0060] Combining the advantages of IDW and Kriging, the spatial interpolation accuracy is improved through weighted fusion, and the complexity of multi-source data integration is solved. The time series modeling, water quality dynamic modeling and demand forecasting modules work together, combined with GNN and attention mechanism to capture dynamic constraints and interactive relationships. Through non-dominated sorting and crowding distance calculation, the Pareto optimal solution set is generated, which balances economic, ecological and social goals and adapts to complex allocation scenarios. The dynamic constraints of water pollution, climate change and crop growth are introduced, and the interactive relationships are modeled through GNN, which enhances the adaptability and robustness of the model. The visual interface and warning reminder function improve the transparency and timeliness of decision-making and meet the diverse needs of users.

[0061] Example 3

[0062] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the intelligent water allocation method for multi-layer water resource management provided above is implemented.

[0063] Since the electronic device introduced in this embodiment is an electronic device used to implement a multi-layer water resource management intelligent water allocation method in the embodiment of this application, based on the multi-layer water resource management intelligent water allocation method introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the electronic device used by the technical personnel of this field to implement the multi-layer water resource management intelligent water allocation method in the embodiment of this application, it belongs to the scope of protection of this application.

[0064] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0065] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent water allocation method for multi-layer water resource management, characterized in that: include: Step S1: Collect the comprehensive data, and use the IDW-Kriging algorithm to integrate the comprehensive data in space to obtain unified integrated data; Step S2: deploy the dynamic evolution mechanism of water resources, and use the deep learning framework to set up a multi-task output module in the dynamic evolution mechanism of water resources, introduce a lightweight simplification layer and dynamic constraints and interactive auxiliary strategies to optimize the mechanism; The multi-task output module is pre-trained separately, and then the output of the multi-task output module is input into the lightweight simplified layer for feature fusion and dimensionality reduction, optimizing the entire mechanism and completing the pre-training; Input the integrated data of the current cycle time into the pre-trained water resources dynamic evolution mechanism to obtain the water resources supply and water demand in various fields in the next cycle time; Step S3: When the water supply meets the water demand of each field, full allocation is performed; when the water supply does not meet the water demand of each field, the NSGA-II algorithm is used to optimally allocate the water demand of each field in the next cycle; Step S4: Arrange the allocation process and results on a visual interface.

2. The intelligent water allocation method for multi-layer water resource management according to claim 1, characterized in that: The method for collecting comprehensive data includes: Comprehensive data include supply data, water quality data, demand data and environmental data; Supply data include rainfall, river flow, groundwater levels, and water storage; Water quality data include water pollution index; Demand data include agricultural irrigation, industrial water, residential water and ecological water; Environmental data include temperature, humidity, wind speed, and evaporation in the water allocation area; Comprehensive data is collected through the Internet of Things, remote sensing technology and measuring equipment, and the data is transmitted to the cloud platform for storage using 5G or LoRa networks.

3. The intelligent water allocation method for multi-layer water resource management according to claim 2, characterized in that: The method of using the IDW-Kriging algorithm to spatially integrate the comprehensive data to obtain unified integrated data includes: Use Python's Pandas library to load data from the cloud platform; Use the Z-score method to detect and delete outliers, and use interpolation to fill in missing values; Standardize the data format, including unified time format, unified unit, unified time resolution, unified coordinate system and normalization; Design the blank space with the water resource allocation area as the spatial boundary, divide the blank space into regular grids with a spatial resolution of 1km×1km, assign a unique ID to each grid cell, and record the longitude and latitude coordinates of its center point. Store the grid information as a DataFrame, and the columns of the DataFrame include the grid ID, the longitude of the center point, and the latitude of the center point. Identify the spatial data in the comprehensive dataset, use the IDW-Kriging algorithm for spatial interpolation, calculate the interpolation results at the center point of each grid for each data type, and store the interpolation results as a DataFrame with columns containing grid ID, timestamp, and data category; Identify non-spatial data in the synthetic dataset and merge the non-spatial data with the spatial interpolation results, using the grid ID and timestamp as primary keys; All processed comprehensive data sets are merged into a DataFrame by grid ID and timestamp. The columns of the DataFrame include: grid ID, timestamp, supply data, water quality data, demand data and environmental data to obtain integrated data; Store aggregate data in CSV, HDF5 or GeoTIFF formats.

4. The intelligent water allocation method for multi-layer water resource management according to claim 3, characterized in that: The method for performing spatial interpolation using the IDW-Kriging algorithm includes: The created 1km×1km blank grid is recorded as the target point set, and the center point of each grid is the interpolation target point. The coordinates of the monitoring points and the observed values ​​in the comprehensive data set are recorded as the target point set. Recorded as monitoring point data; For each target point , calculate the Euclidean distance between it and all monitoring points, and use the inverse of the square of the Euclidean distance as the weight , calculate the IDW interpolation result of the target point ,in, is the total number of monitoring point data, is the index of the target point, is the index of the monitoring point; Calculate the Kriging interpolation result of the target point ,in, is the weight of each monitoring point; The IDW interpolation error and Kriging interpolation error of the IDW interpolation result and Kriging interpolation result are calculated using the root mean square error, and the inverse of the ratio of the IDW interpolation error to the sum of the IDW interpolation error and the Kriging interpolation error is used as the weight of the IDW interpolation. The inverse of the ratio of the Kriging interpolation error to the sum of the IDW interpolation error and the Kriging interpolation error is used as the weight of the Kriging interpolation. , the interpolation result is weightedly fused with the weight to obtain the fused interpolation result, which is recorded as the interpolation result of the IDW-Kriging algorithm.

5. The intelligent water allocation method for multi-layer water resource management according to claim 4, characterized in that: The method for constructing the water resources dynamic evolution mechanism includes: The dynamic evolution mechanism of water resources includes input, multi-task output module, lightweight simplification layer and output; The input of the dynamic evolution mechanism of water resources is set as integrated data, and the output is the water supply of the next cycle and the water demand in various fields; The deep learning framework of the multi-task output module includes a time series modeling module, a water quality dynamic modeling module, and a demand prediction module; The time series modeling module uses a long short-term memory network to model time series data, uses a Transformer model to capture long-term dependencies, and predicts the output as the supply data and environmental data for the next cycle. The water quality dynamic modeling module uses a state space model to describe the changes in the water pollution index, introduces water quality constraints, and uses a neural network to model the nonlinear changes in water quality recovery capacity. The output is the water pollution index prediction for the next cycle and the dynamic adjustment coefficient of the allocable water volume. The dynamic adjustment coefficient is based on the water quality constraint. The water quality constraint is to reduce the allocable proportion of the water body when the pollution index exceeds a certain threshold. The demand forecasting module uses the AquaCrop crop growth model combined with agricultural irrigation data to estimate dynamic irrigation demand, uses the ARIMA time series model combined with industrial water use data of historical periods to predict industrial water demand, uses the XGBoost regression model combined with residential water use data to predict residential water demand, and uses the minimum ecological flow constraint rule model combined with ecological water use data to estimate ecological water demand; The lightweight simplification layer includes feature fusion and dimensionality reduction. Its input is the output of the three task modules, and its output is the fused low-dimensional features, which are used for the final output prediction of the dynamic evolution mechanism of water resources. Introducing dynamic constraints and interactive auxiliary strategies in multi-task output modules, including water pollution and restoration, climate change, and crop growth; The multi-task output module uses a multi-task learning framework to share the time series characteristics of the integrated data and the characteristics of the dynamic constraints and interactive auxiliary strategies to predict the water supply and water demand in various fields respectively; The weighted sum of mean square error and mean absolute error is used as the comprehensive loss function of the dynamic evolution mechanism of water resources to impose output constraints; According to the input and output of each task module and model, the data on the historical cycle time are extracted as the training set. Each task module is trained separately first, and then the dynamic evolution mechanism of water resources is trained as a whole. The training is stopped when the loss function value is the smallest, and the pre-trained dynamic evolution mechanism of water resources is obtained.

6. The intelligent water allocation method for multi-layer water resource management according to claim 5, characterized in that: The method of introducing dynamic constraints and interactive auxiliary strategies in the multi-task output module includes: Use graph neural networks to model the interaction between water pollution and restoration, climate change, and crop growth: define the first node as the water pollution index, the second node as the climate characteristics, and the third node as the crop growth characteristics; The first side is defined as the impact intensity of climate characteristics on water pollution index; the second side is defined as the impact intensity of water pollution index on crop growth characteristics; the third side is defined as the impact intensity of climate characteristics on crop growth characteristics; the fourth side is defined as the feedback impact of crop growth characteristics on water pollution index; The GNN update rule is used to output the dynamic adjustment coefficient of each node, among which the water pollution adjustment coefficient is used to correct the allocable water; the climate adjustment coefficient is used to correct the supply; the crop growth adjustment coefficient is used to correct the irrigation demand; the attention mechanism is used to dynamically adjust the interaction weights, and the attention mechanism is used to dynamically adjust the weights of the edges in the graph neural network.

7. The intelligent water allocation method for multi-layer water resource management according to claim 6, characterized in that: The method of using the GNN update rule to output the dynamic adjustment coefficient of each node includes: Collect time series data of water pollution index, climate characteristics and crop growth characteristics, record them as original features, define the time window length as R_U, and for each node, use LSTM time series model to encode historical data, generate time embedding features, and concatenate the original features and time embedding features to obtain complete node features; For each edge, a two-layer MLP is defined to perform nonlinear transformation on the interaction features between nodes. The complete node features are input into the corresponding MLP to obtain the nonlinear interaction features. The nonlinear interaction features are concatenated with the complete node features to obtain the final node features. The final node features are used to update and replace the original features.

8. The intelligent water allocation method for multi-layer water resource management according to claim 7, characterized in that: The method of using the NSGA-II algorithm to optimally allocate the water demand in each field in the next cycle time includes: Step A1: Define the decision variable as the water allocation for each field in the next cycle; define the constraint condition as the water allocation must be greater than 0 and less than or equal to the maximum water allocation, and the total water allocation for all fields cannot exceed the available water supply; The optimization objectives are defined as maximizing economic benefits, maximizing ecological benefits, and minimizing social equity; Step A2: Randomly generate an initial population, set the population size N, and each individual is an allocation plan. Randomly generate allocation plans and adjust them to meet the supply constraint. If the initial solution does not meet the constraint, it can be adjusted by normalization. For each individual in the initial population, calculate the objective function value and record it; Step A3: Perform non-dominated sorting on the population, divide the Pareto frontier, assign a non-dominated level to each individual, and calculate the crowding distance for each individual in the frontier: Step A4: Use tournament selection to generate the parent population, use simulated binary crossover to generate offspring, use polynomial mutation to mutate the offspring, merge the parent population and offspring population to obtain a joint population with a size of 2N; Step A5: Repeat steps A2 to A4 until the objective function value stops changing, and output the first non-dominated frontier in the final population, which is the optimal solution set. Each solution corresponds to an allocation plan. According to actual needs, the solution that meets specific conditions is selected as the optimal allocation plan for output; Step A6: Use the optimal allocation plan to optimally allocate the water demand in each field in the next cycle.

9. The intelligent water allocation method for multi-layer water resource management according to claim 8, characterized in that: The visualization interface includes a terminal page of a computer or a display interface of a mobile device, and performs interface interaction with the user.

10. The intelligent water allocation method for multi-layer water resource management according to claim 9, characterized in that: Arranging the allocation process and results on the visual interface also includes transmitting the optimal allocation report to the user terminal via wireless transmission and issuing a warning reminder.

Images