Water resource scheduling demand prediction method based on multi-source data analysis

Abstract

The invention discloses a water resource scheduling demand prediction method based on multi-source data analysis, and particularly relates to the field of water resource scheduling and prediction.The water resource scheduling demand prediction method comprises the steps that hydrological physics, geographic space, social economy and management rule full-dimension heterogeneous data are synchronously collected and divided into physical data subsets and social economy data subsets after standardized processing; adopting physical driving and data driving double-model parallel prediction to respectively generate predicted values of natural incoming water supply, ecological environment water demand and social economic water demand; a supply-demand balance level is judged by calculating a supply-demand ratio, and grading weight adjustment is performed on social and economic water demand by taking ecological water demand priority as a principle, so that regularized fusion of predicted values is realized; extracting a core prediction index, judging a gap level, matching a preset scheduling rule to generate a hierarchical scheduling suggestion, and finally outputting a standardized prediction report and a scheduling scheme; the method is comprehensive in data support, rigorous in prediction logic and high in scheduling suggestion operability, and provides a scientific basis for a water resource scheduling decision.

Description

Technical Field

[0001] This invention relates to the field of resource scheduling and forecasting technology, and more specifically, to a method for forecasting water resource scheduling demand based on multi-source data analysis. Background Technology

[0002] Currently, the field of water resources management has gradually established an integrated "sky-ground-hydraulic" monitoring and sensing system, integrating diverse equipment such as remote sensing satellites, drones, and ground sensors to achieve real-time collection and dynamic monitoring of multi-dimensional data including hydrological physics, geospatial data, and socio-economic data. The fusion and application of multi-source data has become a trend; the comprehensive analysis of measured hydrological data, remote sensing inversion data, and socio-economic statistical data provides rich data support for water resources allocation demand forecasting. Meanwhile, technologies such as digital twins and artificial intelligence are being rapidly adopted in water resource allocation, with digital twin river basins and smart water network systems gradually being implemented, driving the transformation of allocation management from experience-driven to data-driven. With the advancement of smart water conservancy construction, the coverage of water resource monitoring networks and the intelligent upgrading of allocation systems have been strengthened, forming a new pattern of water resource allocation development characterized by multi-technology integration, multi-data support, and multi-scenario applications. This has laid a solid foundation for the innovation and application of water resource allocation demand forecasting technology.

[0003] However, it still has some drawbacks in practical use, such as: 1. Data collection is limited in scope and lacks coordination; existing technologies focus on hydrological and physical data collection while neglecting key data such as socio-economic data and management rules. Furthermore, data from different sources suffer from inconsistent spatiotemporal benchmarks and significant differences in units, resulting in incomplete data support and difficulty in depicting the coupling relationship between water resource supply and demand. 2. The predictive model architecture is too simplistic and lacks adaptability. Most technologies only use a single physical-driven or data-driven model. Physical models lag behind changes in socioeconomic water demand, while data-driven models lack physical mechanism support and are unable to meet the complex predictive needs of both natural water inflow and socioeconomic water demand. 3. The fusion of prediction results lacks scientific rules. Existing fusion methods mostly rely on complex optimization algorithms or subjective experience assignment, which have problems such as many fitting parameters, poor reproducibility, and do not fully consider the principle of prioritizing ecological water demand. As a result, the fusion results are difficult to match actual scheduling needs. 4. Poor implementability of scheduling recommendations. The forecast results output by existing technologies are disconnected from scheduling measures. There is no precise matching mechanism between the hierarchical gaps and scheduling rules. The scheduling recommendations lack specific and operable parameter support, making it difficult to directly guide actual water resource scheduling decisions. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, the present invention provides a water resource scheduling demand prediction method based on multi-source data analysis, which solves the problems mentioned in the background art through the following scheme.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting water resource allocation demand based on multi-source data analysis, comprising: S1: Synchronous acquisition and standardization of multi-source heterogeneous data: Synchronous acquisition of real-time and historical data covering hydrophysical, geospatial, socioeconomic, and management rules; standardization operation is performed on the acquired multi-source heterogeneous raw data to construct a standardized spatiotemporal dataset, which is divided into a physical standardized data subset and a socioeconomic standardized data subset according to data type; S2: Dual-model parallel predictive analysis: Based on a standardized subset of physical data, a physics-driven predictive model is used to predict and generate forecasts of natural water supply and ecological water demand; based on a standardized subset of socio-economic data, a data-driven predictive model is used to predict and generate forecasts of socio-economic water demand. S3: Forecast Value Regularization and Fusion: The forecast values ​​output by the physical-driven forecast model and the data-driven forecast model are standardized and aligned; the supply-demand ratio of total supply to total demand within a regional time period is calculated, and the supply-demand balance level is determined based on the supply-demand ratio range; the socio-economic water demand forecast values ​​are adjusted according to the supply-demand balance level, and the fused forecast results are generated. S4: Final forecast results output and scheduling recommendations: Extract three types of forecast indicators for the future period of the region from the integrated forecast results: supply and demand balance, total water demand, and supply and demand gap; determine the gap level based on the ratio of the supply and demand gap to the total supply; match preset fixed scheduling rules for different gap levels to generate executable hierarchical scheduling recommendations; integrate the forecast indicators and hierarchical scheduling recommendations to form a water resource scheduling demand forecast report and hierarchical scheduling recommendation scheme.

[0006] The technical effects and advantages of this invention are as follows: Multi-source data collection and standardization across all dimensions: This solution covers hydrological, physical, and geospatial data across all dimensions. Through synchronous collection and unified standardization processing, it eliminates data heterogeneity and constructs a unified spatiotemporal dataset, providing comprehensive data support for accurate prediction. The dual-model parallel forecasting approach balances accuracy and mechanism. It adopts a dual-model parallel architecture driven by both physics and data. The physical model ensures the physical rationality of natural water inflow forecasting, while the data model captures the dynamic changes in social water demand. The two complement each other to improve forecasting accuracy and adaptability. The rule-based fusion is scientific and reproducible. With ecological water demand as the core, fixed weights are set based on the supply-demand ratio, and the predicted values ​​are fused through simple linear operations. There are no complex fitting parameters, and the fusion process is transparent and reproducible, which meets the core requirements of actual scheduling. The hierarchical scheduling recommendations are precise; a one-to-one correspondence mechanism between gap levels and scheduling rules is established, clarifying the specific parameters and execution standards of scheduling measures at each level. The output scheduling recommendations are directly linked to actual parameters such as engineering design values ​​and historical data, and can be directly used for scheduling decisions. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0008] Figure 2 This is a schematic diagram of the S1 process of the present invention.

[0009] Figure 3 This is a schematic diagram of the S2 process of the present invention.

[0010] Figure 4 This is a schematic diagram of the S3 process of the present invention.

[0011] Figure 5 This is a schematic diagram of the S4 process of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] refer to Figures 1-5 The method for predicting water resource allocation demand based on multi-source data analysis, as shown, includes: S1: Synchronous Acquisition and Standardization of Multi-Source Heterogeneous Data: Through two sub-steps—data acquisition and data standardization—the entire process from multi-source heterogeneous raw data to a standardized basic spatiotemporal dataset is achieved. First, for the comprehensive data required for water resource scheduling demand forecasting, including natural hydrology, geospatial data, socioeconomic data, and management rules, continuous raw data covering historical and real-time periods are acquired synchronously using different acquisition devices. The range of the time series index t is defined to distinguish historical and real-time data. Then, through four standardization operations—cleaning, interpolation, spatiotemporal alignment, and normalization—data quality is gradually improved, eliminating data noise, filling data gaps, unifying the spatiotemporal benchmark, and standardizing data units, ultimately forming a standardized spatiotemporal dataset with a unified structure and standardized format. The specific steps are as follows: S101: Data Acquisition: For the comprehensive data required for water resource allocation demand forecasting, real-time and historical data covering hydrophysical, geospatial, socioeconomic, and management rules are simultaneously collected using four different acquisition methods; details are as follows: The timestamp is denoted as 't'; meteorological and hydrological data and engineering status data are collected through a network of physical sensors deployed in the monitoring area, including real-time data such as real-time rainfall. Real-time runoff Real-time gate opening Real-time water level Historical data includes historical rainfall time series datasets, historical runoff time series datasets, historical gate opening time series datasets, and historical water level time series datasets. Geospatial data is collected through satellite and aerial remote sensing platforms, including real-time data such as real-time land use monitoring data, real-time soil moisture data, and real-time vegetation index. Historical data includes time-series datasets of historical land use types. Historical soil type time series dataset Historical vegetation index time series dataset; Socioeconomic data is collected through the socioeconomic databases of government and industry management departments, including real-time data on water consumption by district. Real-time water price Real-time industry output value Historical data includes historical water consumption time series datasets for different zones, historical water price time series datasets, and historical industry output time series datasets. Through the dispatch management backend of the water resources management department, dispatch-related data is collected. Real-time data includes real-time dispatch instructions, and historical data includes dispatch rule datasets and historical decision record datasets. The dispatch rule datasets include water allocation rules, engineering operation rules, and emergency dispatch rules. The historical decision record datasets include historical dispatch instructions, historical supply and demand balance results, and historical water rights trading records.

[0014] S102: Data Standardization: Cleaning, imputing, spatiotemporal alignment, and normalization of the raw data to form a standardized spatiotemporal dataset. The specific steps are as follows: Cleaning: using 3 The criterion for removing outliers is that for a data point x, if... If it is, then it is considered an outlier. The mean of the data. The standard deviation of the data; Interpolation: For missing values, spatiotemporal kriging interpolation is used. For spatial points... The interpolation formula at point is: ,in, The weighting coefficients are determined by the variogram model and satisfy the properties of unbiasedness and minimum variance. For spatial points The measured value at point n, where n is the number of measured points; Spatiotemporal alignment: Spatial alignment: Unify all data to a preset spatial grid. Each spatial grid corresponds to an administrative or hydrological region. Time alignment: unify the time scale of all data to the daily scale t to ensure that data of different carriers and types can be directly compared and correlated in the spatiotemporal dimension. Normalization: Max-min normalization is used, and the formula is: ,in The minimum value of the data sequence. The maximum value of the data sequence is [0, 1]. Output standardized spatiotemporal dataset The dataset is divided into two subsets based on data type, which serve as the inputs to the dual model: Physical Class Standardized Data Subset ,Include: Standardized subset of socio-economic data ,Include: ; S2: Dual-track parallel data processing and feature construction: A dual-track parallel architecture of a physics-driven model and a data-driven model is adopted, with independent predictions based on a standardized spatiotemporal dataset subset output by S1. The physics-driven model is based on a distributed hydrological model and utilizes a physically-based standardized data subset. Through a complete process of runoff calculation, runoff calculation, and regional aggregation, the natural water supply capacity is predicted; the data-driven model uses an LSTM temporal neural network as its core and is based on a standardized subset of socio-economic data. Through a complete process of data preprocessing, model building, training and validation, and rolling forecasting, the socio-economic water demand is predicted; the details are as follows: S201: Model A—Physics-Driven Model: Based on Physics-Classified Data Subsets A simplified distributed hydrological model employing the SCS curve number method for runoff calculation, confluence calculation, and regional water volume aggregation is used to achieve the entire calculation process from grid-scale runoff to regional-scale natural water supply prediction; the specific process is as follows: Standardized subset of physical data As input; SCS curve number method for grid-based runoff calculation: realizes the conversion of rainfall into grid runoff and directly correlates with potential retention parameters. With production flow rate; details are as follows: The curve number CN is determined by looking up a table: based on the grid. Land use types within and soil type To obtain the corresponding curve number CN, consult the SCS curve number manual. CN is a fixed industry classification lookup value, ranging from 0 to 100, reflecting the flow generation capacity of the underlying surface. The larger the CN value, the stronger the flow generation capacity.

[0015] Potential retention parameters Conversion: Curve Number CN and Potential Retention Parameter The conversion formula is: ,in The unit is inches. To convert to millimeters, multiply by a fixed conversion factor of 25.4. The converted unit will correspond to the rainfall amount. Maintain consistency.

[0016] The grid-based runoff calculation uses the core equations of the SCS curve number method to calculate the grid. Daily runoff , Applicable conditions: When the rainfall is less than the initial loss value, the runoff is 0; among which For grid Daily rainfall, from Standardized measured data.

[0017] Grid flow calculus: calculating the flow generated by the grid The calculation extends to the regional outlet section, yielding the grid's contribution to the regional water inflow; details are as follows: Convergence time confirmed: Mesh flow path length extracted based on digital elevation model (DEM) and constant flow rate Computational grid Convergence time to the outlet section : ,in The fixed flow velocities are classified according to land use type (0.5 m / s for cultivated land, 0.3 m / s for forest land, and 1.0 m / s for construction land), and are fixed values ​​based on industry experience. The flow calculation uses a linear reservoir flow model to calculate the grid flow rate. The calculation extends to the outlet section, yielding the outlet runoff contributed by the grid. .

[0018] Regional Natural Water Supply Aggregation: Spatially aggregate the outflow from all grids within the region to generate regional-scale natural water supply forecasts. The details are as follows: Spatial summation of regional runoff: The sum of the outlet runoff of all grids within region k is the regional runoff. : ; Future time period forecast generation: For future time periods ( =1, 2, ..., T), and the predicted future rainfall values, based on historical rainfall time-series statistical predictions, are again substituted into the calculation process of grid runoff generation calculation using the SCS curve number method, grid confluence calculation, and regional natural water supply aggregation to obtain the predicted natural water supply value for the region in the k-th future time period. : ; Ecological and environmental water demand calculation: The Tennant method is used to calculate the predicted ecological and environmental water demand for region k time period t. As a basic demand item in supply and demand balance analysis: Where X is the ecological base flow ratio coefficient, which is the industry standard recommended value, based on manual preset; The multi-year average flow of region k in the current month is given by... The data were obtained from historical runoff time series data.

[0019] Output predictions: Physics-driven prediction results , : Predicted natural water supply for region k during time period t, the sequence is as follows Its generation logic is as follows: future rainfall forecast → SCS curve number method grid runoff generation → runoff calculation → regional spatial aggregation; The predicted ecological water demand for region k at time t is as follows: .

[0020] S202: Model B—Data-Driven Predictive Model: Based on Standardized Subsets of Socioeconomic Data This paper describes a complete LSTM time-series neural network model process, encompassing data preprocessing, model building, training and validation, and rolling forecasting, to achieve time-series forecasting of socioeconomic water demand. The model learns from historical time-series data to capture the dynamic correlation between water consumption, water prices, and industrial output, and uses a rolling forecasting method to generate water demand forecasts for future periods. The specific steps are as follows: The model input is a standardized subset of socioeconomic data. ; Input data preprocessing: Time series window reconstruction: Reconstructing the original time series data into a supervised learning format, and setting the time series window length. This was determined through 5-fold cross-validation. That is, using the previous... Feature data of the day Predict water consumption on day t .

[0021] Dataset partitioning: Historical data is divided into training set (70%), validation set (20%) and test set (10%) in chronological order to ensure the continuity of time series data and avoid information leakage caused by random partitioning.

[0022] Normalization: Perform max-min normalization on all input features, consistent with the normalization method in S102; LSTM model construction: A three-layer temporal neural network structure is adopted, with the following specific parameters: Input layer: The input layer has 3 dimensions, corresponding to 3 features: water consumption, water price, and industry output value; the time series sequence length of each sample is... That is, using the previous Three features at each time step can be used to predict water consumption at the next time step.

[0023] Hidden layers: 2 LSTM layers, with 64 hidden units in the first layer and 32 hidden units in the second layer (both determined by 5-fold cross-validation, selecting the parameter combination with the minimum loss on the validation set); each layer has a dropout layer (Dropout=0.2) to prevent the model from overfitting.

[0024] Output layer: Fully connected layer with dimension 1, outputting the predicted water consumption for day t. .

[0025] Activation function: The hidden layer uses the tanh activation function, and the output layer uses the linear activation function (the water consumption is a continuous value, which belongs to the regression prediction problem).

[0026] Model training and validation: Loss function: Mean squared error (MSE) is used, and the formula is as follows: ,in To measure the actual water consumption, To predict water consumption, N is the total number of samples participating in the MSE calculation.

[0027] Optimizer: The Adam optimizer is used with a learning rate of 0.001 (determined through cross-validation).

[0028] Early stopping strategy: Stop training when the validation set loss does not decrease for 10 consecutive epochs to prevent the model from overfitting.

[0029] Model evaluation: Calculate the mean absolute error (MAE) and coefficient of determination on the test set. ),Require Only then can it be put into use.

[0030] Rolling prediction generation The rolling forward forecasting method is used to generate the predicted socioeconomic water demand for region k over the next T time periods. The specific process is as follows: Utilizing the last of history Feature data of the day Input the trained LSTM model to predict the water consumption on day t+1. .

[0031] Will As a known value, and , (The forecasts for water prices and output value are based on historical time-series statistical forecasts for the same period) together to construct a new time-series window. Predict water consumption on day t+2 .

[0032] Repeat the above process to predict the water consumption sequence for the next T time periods. .

[0033] Regional aggregation: The grid-scale forecasts are aggregated to region k to obtain the socioeconomic water demand forecast for region k at time t. .

[0034] Output Predicted Values: Data-Driven Prediction Results ,in, The predicted socioeconomic water demand for region k at time t is generated by the following logic: historical time series data preprocessing → LSTM model training and validation → rolling forward prediction → regional spatial aggregation.

[0035] S3: Predicted Value Regularization Fusion: Based on the physics-driven prediction results output by the S2 dual model. and data-driven prediction results A three-stage rule-based fusion method is adopted, which involves standardized alignment of predicted values, hierarchical calculation of supply-demand ratios, and adjustment of ecological priority weights, to achieve dynamic integration of predicted values ​​for natural supply and social demand. The specific process is as follows: S301: Standardization and Alignment of Predicted Values: Eliminates differences in spatiotemporal references and dimensions between the two models' predicted values, ensuring that the three types of predicted values—natural water inflow, ecological water demand, and socioeconomic water demand—can be directly used for mathematical calculations and comparative analysis; the alignment rules reuse the spatiotemporal references and dimensional standards from S1; specifically as follows: Spacetime reference alignment: Spatial alignment: Aligning the k-scale predictions of the region output by the physics-driven model. , The regional k-scale predicted values ​​output by the data-driven model The data is uniformly bound to the administrative or hydrological region k defined in S102 to ensure that the fused objects are prediction data of the same spatial unit. Time alignment: Predicting future time-period sequences from the outputs of the two models. , , ( =1, 2, ..., T), uniformly calibrated to the daily time scale t defined in S102, to ensure that the fusion objects are prediction data of the same time unit; Unit Alignment and Standardization: All water volume forecasts will be uniformly converted to 10,000 cubic meters per day. The specific conversion rules are as follows: If the original forecast value is in cubic meters per day, then divide by the conversion factor of 10. 4 If the original forecast value is in cubic meters per second, multiply it by the number of days and seconds (86400) and then divide by 10. 4 ; Verification rule: The dimensional error of all predicted values ​​after alignment must be less than 0.1% to ensure the accuracy of mathematical calculations; Output the standardized and aligned set of predictions The superscript align indicates a predicted value that has been aligned in terms of time, space and dimensions.

[0036] S302: Rule-based Weighted Fusion: This method employs a linear, rule-based process of supply-demand ratio calculation, tiered judgment, and weight adjustment to achieve dynamic fusion of forecast values. The core logic is as follows: First, the ratio of total supply to total demand in region k at time period t is calculated to determine the supply-demand balance. Then, based on the supply-demand balance level, preset fixed weights are used to adjust the socio-economic water demand forecast. Finally, the fused supply-demand forecast result is generated. Details are as follows: Basic supply and demand parameter definitions: For region k and time period t, define the core parameters required for fusion: Total supply The sum of the predicted natural water supply and the initial water storage is given by the formula: ,in The initial water storage in region k at time t is the measured data from the S101 physical sensor network, not the predicted value. Prioritize ensuring demand The predicted water demand for the ecological environment is calculated using the following formula: Core rule: This part of the demand will be given full priority and will not be subject to any weighting adjustments; Adjustable demand The predicted socioeconomic water demand is calculated using the following formula: Core rule: Only this part of the requirements will be subject to hierarchical weight adjustments; Calculation of supply-demand ratio and determination of equilibrium status before integration: Formula for calculating the supply-demand ratio before integration: Supply-demand ratio before integration for region k in time period t. The ratio of aggregate supply to aggregate demand is given by the formula: ; Equilibrium state classification: based on supply and demand ratio The value range of is used to divide the supply and demand balance state of region k in time period t into 5 levels. The grading threshold is an empirical value obtained from historical statistics. The specific grading standard is as follows: when The supply-demand balance level is 1, indicating a supply surplus; when The supply and demand balance level is 2, indicating that supply and demand are in balance; when The supply and demand balance level is 3, indicating a slight shortage; when The supply and demand balance level is 4, indicating a moderate shortage; when The supply and demand balance level is 5, indicating a severe shortage. Socioeconomic Water Demand Weighting Adjustment: Weighting Rules: Based on the supply-demand balance level, fixed adjustment weights are assigned to the socioeconomic water demand forecast. The weights are constants between 0 and 1, and the specific assignment criteria are as follows: When the supply and demand balance level is Level 1 and Level 2, the social water demand is fully met, and the weight is not adjusted. When the supply and demand balance level is 3, social water demand is reduced by 10%, and the weight is adjusted to 0.9; When the supply and demand balance level is 4, social water demand is reduced by 30%, and the weight is adjusted to 0.7; When the supply and demand balance level is 5, social water demand is reduced by 50%, and the weight is adjusted to 0.5; Adjusted social water demand calculation formula: Adjusted socioeconomic water demand forecast for region k time period t for: ; Validation of fusion results: Validation rule: Adjusted total demand Must meet This ensures that the integration results conform to the basic logic of supply and demand balance; Exception handling: If the validation fails, the weights will be adjusted. Further reduce by 0.1 until the verification rules are met, with the minimum weight not lower than 0.3; Output fusion prediction results: Integrated supply and demand forecast results The parameters are defined as follows: : Predicted natural water supply for region k during time period t (consistent with the output of the physical driving model, no adjustment) : The predicted ecological water demand for region k at time t (consistent with the output of the physical driving model, with full priority guarantee); : Adjusted socioeconomic water demand forecast for region k during time period t (final social demand forecast after weight adjustment). The predicted sequence covers T future time periods, i.e. , , ; S4: Final prediction results output and scheduling suggestions: based on the fused prediction results from S3. The system employs a three-stage process: core indicator extraction, gap level determination, and tiered scheduling rules. It directly outputs core forecast indicators for water resource allocation demand and actionable tiered scheduling recommendations. The specific steps are as follows: S401: Forecast Indicator Output: Three types of forecast indicators are directly extracted from the fusion forecast results. These indicators cover three core dimensions: supply and demand balance, total water demand, and supply and demand gap, comprehensively representing the region's future water resource allocation demand situation, as detailed below: Based on fusion prediction results For region k in the future T time periods (t+1, t+2, ..., t+T), calculate the following core indicators: Regional supply and demand balance forecasting indicators: Indicator definition: Supply-demand ratio in scheduling analysis The mathematical function reflecting the degree of matching between the natural supply capacity and aggregate demand in region k during time period t is as follows: ; Indicator sequence: Generates the supply-demand ratio sequence for region k over the next T time periods. ; Total water demand forecast indicators: Indicator definition: Total water demand This reflects the sum of the ecological water demand and the adjusted socio-economic water demand that need to be prioritized in region k during time period t; its specific mathematical function is: ; Indicator sequence: Generates the total water demand sequence for region k over the next T time periods. The dimensions are consistent with S3, which is 10,000 cubic meters per day.

[0037] Supply and demand gap forecasting indicators: Indicator definition: Supply-demand gap This reflects the difference between aggregate demand and aggregate supply in region k during time period t, with a positive value indicating a deficit and a negative value indicating a surplus; its specific mathematical function is: ; Indicator Sequence: Generates the supply and demand gap sequence for region k over the next T time periods. ; illustrate: A value greater than 0 indicates a supply-demand gap, requiring the initiation of scheduling measures. This indicates a balance between supply and demand or a supply surplus, and the implementation of routine scheduling.

[0038] Output predictive index set It includes the supply-demand ratio sequence, total water demand sequence, and supply-demand gap sequence for region k over the next T time periods.

[0039] S402: Generation of Tiered Scheduling Suggestions: Based on Supply and Demand Gap in Core Forecasting Indicators First, the gap level in region k during time period t is determined. Then, for different gap levels, preset fixed scheduling rules are matched to generate executable hierarchical scheduling suggestions; specifically as follows: Gap Level Determination: Based on the Supply-Demand Gap Quantity With total supply The ratio is used to classify the gap level of region k in time period t into 4 levels. The classification threshold is an empirical value obtained from historical statistics. The specific judgment criteria are as follows: when A gap level of 0 indicates that there is no gap. when The gap level is 1, indicating a slight gap; when The gap level is 2, indicating a moderate gap; when The gap level is 3, indicating a severe gap; Tiered scheduling rules and suggestions generation: For each gap level, the corresponding fixed scheduling rules are executed to generate scheduling suggestions for region k time period t. The specific rules are as follows: Level 0 (No Gap): Routine scheduling will be implemented: Scheduling measures: The reservoir will release water according to the historical average for the same period. Execute the water release. The statistical average of the historical water release of reservoir r during the same period is derived from the S101 historical decision record dataset; the existing water resource allocation status is maintained without the need for additional scheduling measures.

[0040] Level 1 (Mild Shortage): Implement reservoir water replenishment scheduling: Scheduling measures: The reservoir will increase the shortfall amount based on the historical water release volume for the same period. That is, the amount of water discharged. The discharge volume shall not exceed the maximum allowable discharge volume of the reservoir. , The design values ​​for the reservoir project are derived from the S101 scheduling rule dataset; the water resources replenished by the reservoir's water reserves fully cover the minor shortage and ensure water demand.

[0041] Level 2 (Moderate Shortage): Implement reservoir water replenishment and water rights trading scheduling. Dispatch measure 1 (reservoir water replenishment): The reservoir will release water according to the maximum allowable discharge volume. Perform water release; Dispatch Measure 2 (Water Rights Trading): Remaining portion of the gap Water rights trading is used to supplement water supply; the trading partners are regions where the supply-demand gap is negative (with surplus water rights); the trading volume does not exceed the surplus gap. By releasing water at full capacity from reservoirs and engaging in cross-regional water rights trading, the moderate water shortage can be addressed in a coordinated manner.

[0042] Level 3 (Severe Water Shortage): Implement reservoir water replenishment, water rights trading, and emergency water restriction scheduling; Dispatch measure 1 (reservoir water replenishment): The reservoir will release water according to the maximum allowable discharge volume. Perform water release; Dispatch measure 2 (water rights trading): executed according to the maximum tradable water volume (the maximum tradable water volume is the region's historical maximum water rights trading volume, from the S101 historical decision record dataset). Dispatch Measure 3 (Emergency Water Restriction): The remaining water shortage will be addressed through emergency water restriction; the water restriction will apply to areas with low water use efficiency, calculated based on S101 socio-economic data, i.e., the output value per unit of water consumption. Prioritize water restriction The smallest area; the water restriction ratio is determined based on the remaining shortfall, with the maximum water restriction ratio not exceeding 30%; Output hierarchical scheduling suggestion set It includes the gap level determination results for region k in the next T time periods and corresponding scheduling measures recommendations.

[0043] S403: Final Output: The core prediction metrics and hierarchical scheduling recommendations are integrated into two standardized documents, which serve as the final output of the entire technical solution; the details are as follows: Generate a water resource allocation demand forecast report that includes regional overview, core indicators, and situation analysis. The regional overview includes the administrative / hydrological area k covered by the forecast and the time interval T of the forecast period. The core indicators include the supply-demand ratio sequence, total water demand sequence, and supply-demand gap sequence for each region over the next T periods. The situation analysis includes an analysis of the water resource supply-demand balance status of each region in the future periods based on the core indicators, and identification of key areas and periods with gaps.

[0044] The system generates a tiered dispatching recommendation scheme that includes the distribution of gap levels, dispatching measures, implementation standards, and risk warnings. The gap level distribution includes the gap level determination results for each region over the next T time periods. The dispatching measures include specific dispatching measures for each region corresponding to the gap level, including reservoir water release volume, water rights trading partners and water volume, and emergency water restriction areas and proportions. The implementation standards include the implementation time, responsible parties, and monitoring requirements for the dispatching measures. The risk warnings include warnings of potential water use risks and contingency plans for areas with moderate and severe gaps.

[0045] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting water resource allocation demand based on multi-source data analysis, characterized in that, include: S1: Synchronous acquisition and standardization of multi-source heterogeneous data: Synchronous acquisition of real-time and historical data covering hydrophysical, geospatial, socioeconomic, and management rules; standardization operation is performed on the acquired multi-source heterogeneous raw data to construct a standardized spatiotemporal dataset, which is divided into a physical standardized data subset and a socioeconomic standardized data subset according to data type; S2: Dual-model parallel predictive analysis: Based on a standardized subset of physical data, a physics-driven predictive model is used to predict and generate forecasts of natural water supply and ecological water demand; based on a standardized subset of socio-economic data, a data-driven predictive model is used to predict and generate forecasts of socio-economic water demand. S3: Predictive value regularization fusion: Standardize and align the predicted values ​​output by the physics-driven prediction model and the data-driven prediction model; Calculate the supply-demand ratio of total supply to total demand within a regional time period, determine the supply-demand balance level based on the supply-demand ratio range, and adjust the socio-economic water demand forecast values ​​according to the supply-demand balance level to generate integrated forecast results. S4: Final forecast results output and scheduling recommendations: Extract three types of forecast indicators for the future period of the region from the integrated forecast results: supply and demand balance, total water demand, and supply and demand gap; determine the gap level based on the ratio of the supply and demand gap to the total supply; match preset fixed scheduling rules for different gap levels to generate executable hierarchical scheduling recommendations; integrate the forecast indicators and hierarchical scheduling recommendations to form a water resource scheduling demand forecast report and hierarchical scheduling recommendation scheme.

2. The water resource allocation demand forecasting method based on multi-source data analysis according to claim 1, characterized in that: The construction of the standardized spatiotemporal dataset includes: Statistical criteria are used to clean the collected multi-source heterogeneous raw data to remove outliers. Interpolation methods are used to fill in missing data values. The processed data is then aligned to a preset spatial grid and time scale for spatiotemporal alignment. Finally, a unified dimensional standardization method is used for normalization to form a standardized spatiotemporal dataset.

3. The water resource allocation demand forecasting method based on multi-source data analysis according to claim 1, characterized in that: The physics-driven prediction model includes: Using a standardized subset of physical data as input, and selecting daily rainfall, land use type, and soil type at the grid scale as input data, a three-stage distributed hydrological model architecture is adopted, consisting of SCS curve number method for runoff calculation, runoff calculation, and regional water volume aggregation. The process sequentially executes grid runoff calculation, grid runoff calculation, and regional natural water supply aggregation to generate predicted values ​​of natural water supply for the region in the future. At the same time, the Tennant method is used to calculate predicted values ​​of ecological and environmental water demand for the region in the future. Finally, the physical-driven prediction results of the two types of prediction values ​​are output: predicted values ​​of natural water supply and predicted values ​​of ecological and environmental water demand for the region in the future.

4. The water resource allocation demand forecasting method based on multi-source data analysis according to claim 1, characterized in that: The data-driven prediction model includes: Using a standardized subset of socioeconomic data as input, this study selects daily water consumption, daily water price, and daily industrial output as input data. Employing an LSTM temporal neural network model architecture, the study executes a complete process encompassing data preprocessing, model building, training and validation, and rolling forecasting. The data preprocessing stage involves temporal window reconstruction, temporal dataset partitioning, and normalization to convert the input data format. The model building stage uses a three-layer temporal neural network structure with a dropout layer to prevent overfitting. The model training and validation stage uses cross-validation to determine the temporal window length, number of hidden units, and learning rate parameters, employs an early stopping strategy to prevent overfitting, and verifies model effectiveness using preset evaluation indicators. The rolling forecasting stage uses a rolling forward forecasting method, combining historical statistical forecasts of water price and industrial output to sequentially generate future water consumption forecasts. These forecasts are then spatially aggregated to obtain the future socioeconomic water demand forecast for the region, ultimately outputting a data-driven forecast result containing these forecasts.

5. The water resource allocation demand forecasting method based on multi-source data analysis according to claim 1, characterized in that: The determination of the supply-demand balance level includes: First, the total supply and total demand within the region are defined. The total supply is the sum of the predicted natural water supply and the initial water storage in the region. The total demand is the sum of the predicted ecological water demand and the unadjusted socio-economic water demand before integration. Then, the ratio of total supply to total demand is calculated as the supply-demand ratio before integration. Finally, based on historical statistical experience, a five-level threshold for the supply-demand ratio is set, dividing the supply-demand balance within the region into five levels: a supply-demand ratio greater than or equal to 1.2 indicates a supply surplus; a supply-demand ratio greater than or equal to 1.0 and less than 1.2 indicates a supply-demand balance; a supply-demand ratio greater than or equal to 0.8 and less than 1.0 indicates a slight deficit; a supply-demand ratio greater than or equal to 0.6 and less than 0.8 indicates a moderate deficit; and a supply-demand ratio less than 0.6 indicates a severe deficit.

6. The water resource allocation demand forecasting method based on multi-source data analysis according to claim 5, characterized in that: The adjustment of the hierarchical weights includes: Only the socio-economic water demand forecast is adjusted in terms of weight; the ecological environment water demand forecast is not included in any weight adjustment. The weight assignment corresponds one-to-one with the five levels of supply-demand balance determined before fusion, and a fixed adjustment weight is preset. The adjustment weight corresponding to the supply surplus level and the supply-demand balance level is 1.0, the adjustment weight corresponding to the slight deficit level is 0.9, the adjustment weight corresponding to the moderate deficit level is 0.7, and the adjustment weight corresponding to the severe deficit level is 0.

5. The adjusted socio-economic water demand forecast is obtained by multiplying the socio-economic water demand forecast by the corresponding level's adjustment weight. After adjustment, the matching relationship between the adjusted total demand and total supply needs to be verified. The adjusted total demand is the sum of the ecological environment water demand forecast and the adjusted socio-economic water demand forecast. It is required that the adjusted total demand does not exceed 1.05 times the total supply. If the verification fails, the adjustment weight is reduced by 0.1 until the verification rule is met, and the adjustment weight is not lower than 0.

3. Finally, a fused forecast result is generated that includes the natural water supply forecast, the ecological environment water demand forecast, and the adjusted socio-economic water demand forecast.

7. The water resource allocation demand forecasting method based on multi-source data analysis according to claim 1, characterized in that: The determination of the gap level includes: First, define the merged total supply and merged total demand within the regional time period. The merged total supply is the sum of the predicted natural water supply and the initial regional water storage, while the merged total demand is the sum of the predicted ecological and environmental water demand and the adjusted socio-economic water demand. Calculate the ratio of the merged total supply to the merged total demand as the supply-demand ratio for scheduling analysis. Simultaneously, calculate the supply-demand gap within the regional time period, which is the difference between the merged total demand and the merged total supply. A positive supply-demand gap indicates the existence of a supply-demand gap. Then, using the merged total supply as a benchmark... Based on historical statistical experience, a four-level threshold for the supply-demand gap is set, dividing the gap level within a region into four levels. When the supply-demand gap is less than or equal to 0, it is level 0 with no gap, corresponding to the supply-demand balance or supply surplus state represented by the supply-demand ratio in scheduling analysis. When the supply-demand gap is greater than 0 and less than or equal to 0.1 times the total supply after merging, it is level 1 with a slight gap. When the supply-demand gap is greater than 0.1 times the total supply after merging and less than or equal to 0.3 times the total supply after merging, it is level 2 with a moderate gap. When the supply-demand gap is greater than 0.3 times the total supply after merging, it is level 3 with a severe gap.

8. The water resource allocation demand forecasting method based on multi-source data analysis according to claim 7, characterized in that: The generation of executable hierarchical scheduling recommendations includes: Based on four levels, each corresponding to a preset fixed scheduling rule, Level 0 (no gap) executes regular scheduling, with the reservoir releasing water according to the historical release volume for the same period; Level 1 (slight gap) executes reservoir water replenishment scheduling, increasing the gap amount based on the historical release volume for the same period, but not exceeding the maximum allowable release volume; Level 2 (moderate gap) executes joint scheduling of reservoir water replenishment and cross-regional water rights trading; Level 3 (severe gap) executes comprehensive scheduling of reservoir water replenishment, cross-regional water rights trading, and emergency water restriction, with water restriction targeting areas with low water use efficiency. The final result is a graded scheduling recommendation that includes the gap level and corresponding scheduling measures.

Images