Method for generating reservoir dynamic scheduling benchmark by fusing double-drive prediction and flood control constraint

Abstract

This invention presents a method for generating dynamic reservoir scheduling benchmarks that integrates dual-drive prediction and flood control constraints. It relates to the fields of smart water conservancy, water resource optimization scheduling, and artificial intelligence. The method includes: collecting hydrological and meteorological data; employing a dual-drive hybrid prediction algorithm that combines a physical dynamics numerical model with a spatiotemporal AI large-scale model to dynamically extrapolate the inflow process curve of the reservoir for future periods; prioritizing flood control safety as the highest decision-making constraint, and combining the hybrid predicted inflow input with the reservoir capacity adjustment model for pre-adjustment calculations to generate a safe water release strategy and new reservoir capacity status; and forcibly removing rigid demand water volumes that cannot participate in allocation based on the expected safe and available reservoir capacity, generating a dynamic scheduling benchmark water volume for multi-user water resource hybrid game calculations. This invention improves system robustness and prediction accuracy; by closely integrating flood control safety with economic water resource allocation, it solves the problem of the disconnect between economic water resource allocation and flood control safety, thus avoiding the risk of dam failure.

Description

Technical Field

[0001] This invention relates to the fields of smart water conservancy, water resource optimization scheduling and artificial intelligence technology, and in particular to a method for generating dynamic scheduling benchmarks for reservoirs that integrates dual-drive prediction and flood control constraints. Background Technology

[0002] In multi-user water resource mixed game systems, especially when multiple interests such as agriculture, industry, urban water use and ecological needs are involved, the prerequisite for achieving economic allocation of water resources is that the system must accurately know "how much water is available for allocation in the reservoir in the future".

[0003] Traditional reservoir inflow forecasting and water rights allocation schemes have significant drawbacks: First, single forecasting models lack robustness. Purely physical numerical models consume a lot of computing power and are difficult to calibrate parameters; while purely AI models are prone to prediction distortion or even error drift when faced with unprecedented extreme weather events (such as torrential rains) due to the lack of constraints from the laws of conservation of mechanics. Second, there is a disconnect between economic allocation of water resources and flood control safety. Existing game theory algorithms often prioritize economic benefits and lack mandatory constraints from pre-existing flood control limits. If flood control reserves are encroached upon in actual game theory allocation, it will trigger a serious risk of dam failure.

[0004] Therefore, there is an urgent need for a high-precision dual-drive hybrid prediction mechanism that can adapt to complex climate change, and to forcibly embed flood control constraints and rigid water rights separation rules at the underlying logic level, so as to dynamically and securely generate a water quantity benchmark that can be allocated by multi-user water resource hybrid game. Summary of the Invention

[0005] To address the technical problems of poor robustness of single prediction models and the disconnect between economic allocation of water resources and flood control safety in traditional reservoir inflow prediction and water rights allocation schemes, this invention provides a method for generating dynamic reservoir scheduling benchmarks that integrates dual-drive prediction and flood control constraints.

[0006] The technical solution adopted by this invention to solve the technical problem is as follows:

[0007] This invention provides a method for generating dynamic reservoir scheduling benchmarks that integrates dual-drive prediction and flood control constraints, comprising the following steps:

[0008] S1: Collect real-time meteorological forecast data, multi-source hydrological monitoring data, and geospatial topology data of the reservoir basin;

[0009] S2: Based on the data collected in step S1, a dual-drive hybrid prediction algorithm that combines a physical dynamics numerical model and a spatiotemporal AI large model, as well as a dynamic weighted fusion algorithm, is used to dynamically extrapolate the inflow process line of the reservoir in future periods and output the hybrid predicted inflow.

[0010] S3: Taking flood control safety as the highest decision priority constraint, the mixed predicted inflow volume is substituted into the reservoir water balance differential equation to deduce the reservoir capacity evolution trajectory within the future scheduling cycle. If the predicted highest water level touches the flood control limit water level red line, the reservoir capacity adjustment model will prioritize triggering the reservoir safety water release strategy to free up flood storage capacity, and calculate the expected safe and usable reservoir capacity after the adjustment is completed under the highest decision priority constraint.

[0011] S4: Based on the expected safe and available reservoir capacity, forcibly remove the rigid demand water volume that cannot participate in allocation, and generate a dynamic scheduling benchmark water volume for multi-user water resource mixed game calculation.

[0012] In a preferred embodiment, in step S2, the physical dynamics numerical model uses physical equations to rigorously calculate the precipitation generation and confluence process based on real-time weather forecast data and geospatial topology data.

[0013] As a preferred embodiment, the physical dynamics numerical model adopts a one-dimensional / two-dimensional coupled distributed hydrodynamic model, which solves the Saint-Venant equations based on the theorems of mass conservation and momentum conservation, and outputs a predicted inflow sequence with strict physical boundary constraints.

[0014] In a preferred implementation, in step S2, the spatiotemporal AI big data model predicts flow by mining the nonlinear evolution patterns in historical long-sequence hydrological and meteorological data.

[0015] As a preferred implementation, the spatiotemporal AI large model employs a spatiotemporal graph convolutional network combined with a long short-term memory network. The spatiotemporal graph convolutional network is used to extract the spatial topological features of the watershed where the reservoir is located, and the long short-term memory network is used to process the temporal dependency between historical weather forecast data and flow, outputting a data-driven predicted inflow sequence.

[0016] As a preferred implementation, the loss function of the spatiotemporal AI large model is as follows:

[0017] ;

[0018] in, Mean square error, The spatiotemporal AI large model output for the spatiotemporal AI large model. This is a sequence of actual measured inflow values ​​from the reservoir; the second term... This is a penalty for a single-sided flood peak, used to prevent underestimation of flood levels; and All of these are weight parameters.

[0019] As a preferred implementation, the specific implementation process of the dynamic weighted fusion algorithm in step S2 is as follows:

[0020] Calculate the mixed predicted water inflow ;in, To predict the inflow sequence The weighting coefficients, To predict the inflow sequence Weighting coefficients;

[0021] The weight dynamic adaptive update strategy is as follows: ; ;in, This is an extreme weather index, ranging from [0,1].

[0022] In a preferred implementation, step S3 involves calculating and extrapolating the reservoir capacity evolution trajectory within future scheduling cycles using the reservoir water balance differential equation. The mathematical expression for this equation is as follows:

[0023] ;

[0024] in, for Storage capacity at any time For the current storage capacity, To predict the mixed water inflow, To ensure the safe release of water from the reservoir, This refers to the amount of loss due to evaporation and leakage. To calculate the time step.

[0025] As a preferred implementation, in step S3, a hard constraint of flood control veto is injected into the reservoir capacity adjustment model, which applies to any time within a future scheduling cycle. System-mandated verification <= , for Storage capacity at any time The current statutory flood control limit reservoir capacity is used; if the highest water level is predicted to reach the flood control limit, the reservoir capacity regulation model will suspend all economic water supply requests and prioritize generating a safe water release strategy for the reservoir. To free up flood storage capacity, until a safe and usable reservoir capacity that meets the highest decision-priority constraint of flood control safety is derived throughout the entire scheduling cycle. .

[0026] In a preferred embodiment, the specific calculation formula for the dynamic scheduling reference water volume in step S4 is as follows:

[0027] ;

[0028] in, To ensure the expected safe and usable storage capacity, Physical dead storage capacity to maintain the structural stability of the dam. This is to meet the basic water needs of residents. This is the basic ecological base flow demand.

[0029] The beneficial effects of this invention are:

[0030] This invention provides a method for generating a dynamic reservoir scheduling benchmark that integrates dual-drive prediction and flood control constraints. It uses a dual-drive hybrid prediction algorithm that combines a physical dynamics numerical model and a spatiotemporal AI large model to dynamically extrapolate the inflow process line of the reservoir in future periods. At the output end, a reservoir capacity adjustment model is forcibly introduced, and flood control safety is taken as the highest decision priority constraint. Under the premise of absolutely ensuring physical flood control safety, a safe water release strategy and a new reservoir capacity state are generated. Based on the expected safe and available reservoir capacity, the rigid demand water volume that cannot participate in the allocation is forcibly removed, generating a dynamic scheduling benchmark water volume for multi-user water resource hybrid game calculation.

[0031] This invention provides a method for generating a reservoir dynamic scheduling benchmark that integrates dual-drive prediction and flood control constraints. It adopts a dual-drive hybrid prediction algorithm that uses a physical dynamics numerical model and a spatiotemporal AI large model in parallel. This avoids the problems of poor robustness of a single prediction model, as well as prediction distortion, error drift, etc., thereby improving the robustness and prediction accuracy of the system.

[0032] This invention provides a method for generating a reservoir dynamic scheduling benchmark that integrates dual-drive prediction and flood control constraints. It takes flood control safety as the highest decision-making priority constraint. Under the premise of absolutely ensuring physical flood control safety, it generates a safe water release strategy and a new reservoir capacity state through a reservoir capacity adjustment model. By closely integrating flood control safety with the economic allocation of water resources, it solves the problem of the disconnect between the economic allocation of water resources and flood control safety and avoids the risk of dam failure.

[0033] This invention provides a method for generating a reservoir dynamic scheduling benchmark that integrates dual-drive prediction and flood control constraints. By forcibly embedding flood control constraints and rigid water rights separation rules at the logical level, it can dynamically and securely generate a water quantity benchmark that can be allocated in a mixed game among multiple users. Attached Figure Description

[0034] Figure 1 The present invention provides a flowchart of a method for generating a reservoir dynamic scheduling benchmark that integrates dual-drive prediction and flood control constraints. Detailed Implementation

[0035] The present invention will be further described in detail below with reference to the accompanying drawings.

[0036] In a first aspect, the present invention provides a method for generating a reservoir dynamic scheduling benchmark that integrates dual-drive prediction and flood control constraints.

[0037] See Figure 1 As shown, the present invention provides a method for generating a reservoir dynamic scheduling benchmark that integrates dual-drive prediction and flood control constraints. The specific implementation process is as follows:

[0038] Step S1: Collect real-time meteorological forecast data, multi-source hydrological monitoring data, and geospatial topology data for the reservoir basin;

[0039] After the system starts up, it frequently retrieves real-time meteorological forecast data (including rainfall, current evaporation, such as the Meteo gridded rainfall / evaporation forecast matrix for the next 24 hours) from the reservoir's basin via its built-in data acquisition interface and IoT gateway. Grid ), soil anterior moisture content, and real-time water level and flow data of upstream tributaries (such as the reservoir inflow sequence over the past 72 hours). Q Simultaneously, load the digital elevation model (DEM) of the watershed where the reservoir is located and static spatial parameters such as land cover type (e.g., static watershed digital elevation and infiltration rate characteristic spatial map DEM). Mask This provides a multi-dimensional data matrix for the dual-drive hybrid prediction algorithm. This process primarily relies on the underlying dynamic data management platform to ensure the efficient and unified aggregation of heterogeneous sensing data.

[0040] Step S2: Employ a dual-drive hybrid prediction algorithm that combines a physical dynamics numerical model with a spatiotemporal AI large model to dynamically extrapolate the inflow process curve of the reservoir for future periods.

[0041] The real-time meteorological forecast data, multi-source hydrological monitoring data, and geospatial topology data collected in step S1 are input into the dual-drive hybrid prediction algorithm. The inflow sequence is calculated and predicted through two parallel channels: a physical dynamics numerical model and a spatiotemporal AI large model. Finally, the predicted inflow sequence output from the two parallel channels is fused through a dynamic weighted fusion algorithm to obtain the hybrid predicted water volume.

[0042] According to this invention, the dual-drive hybrid prediction algorithm mainly includes two parallel computing channels:

[0043] Channel 1 is a physical dynamics numerical model, which uses physical equations to rigorously solve the precipitation generation and confluence process based on real-time weather forecast data combined with geospatial topological data.

[0044] Specifically, the physical dynamics numerical model can adopt a one-dimensional / two-dimensional coupled distributed hydrodynamic model, which solves the Saint-Venant equations based on the theorems of mass and momentum conservation, and outputs a predicted inflow sequence with strict physical boundary constraints. .

[0045] Channel 2 is the Spatiotemporal AI Big Data Model (Deep Learning Model), which is a big data prediction model based on deep learning networks. It predicts flow by mining the nonlinear evolution patterns in historical long-sequence hydrological and meteorological data (referring to multi-dimensional monitoring data that is continuous on the time axis and has topological correlation in space, specifically including: hydrology: inflow, river water level, soil moisture content; meteorology: rainfall, evaporation).

[0046] Specifically, the spatiotemporal AI large-scale model can employ a combination of a spatiotemporal graph convolutional network (GCN) and a long short-term memory network (LSTM). The GCN is used to extract the spatial topological features of the watershed where the reservoir is located, while the LSTM is used to process the temporal dependency between historical weather forecast data and flow, outputting a data-driven predicted inflow sequence. ( ).

[0047] Specifically, the loss function of the spatiotemporal AI large model adopts a customized combination of mean squared error and flood peak penalty to enhance the ability to capture extreme flood peaks. Its mathematical expression is as follows:

[0048] ;

[0049] in, For the loss function of spatiotemporal AI large-scale models, Mean square error, The spatiotemporal AI large model output for the spatiotemporal AI large model. This is a sequence of actual measured inflow values ​​from the reservoir; the second term... This is a single-sided flood peak penalty term used to prevent underestimation of flood levels; the weighting parameter is set as follows. =1.0, =2.0.

[0050] According to this invention, the specific implementation process of the Dynamic Weighting Fusion algorithm is as follows:

[0051] The system defines the extreme weather index E. idx The range is [0,1]. This extreme weather index E idx The range can be calculated based on the variance of the real-time rainfall forecast intensity deviating from the historical mean, with the extreme weather index E during extreme torrential rain. idx Approaching 1.

[0052] Calculate the predicted mixed inflow using the following formula. :

[0053] ;

[0054] in, To predict the inflow sequence The weighting coefficients, To predict the inflow sequence The weighting coefficients.

[0055] The weight dynamic adaptive update strategy is as follows:

[0056] ;

[0057] .

[0058] The input and output formats are shown in Table 1:

[0059] Table 1

[0060]

[0061] Step S3: Taking flood control safety as the highest decision priority constraint, and combining the mixed predicted inflow input reservoir capacity adjustment model to perform pre-adjustment calculations, generating reservoir safe water release strategies and new reservoir capacity status;

[0062] The reservoir capacity regulation model forcibly assigns "flood control demand" the highest decision priority (one-vote veto power). The system substitutes the mixed predicted inflow from step S2 into the reservoir water balance differential equation to deduce the reservoir capacity evolution trajectory within future scheduling cycles. If the predicted highest water level touches the flood control limit, the reservoir capacity regulation model prioritizes triggering the reservoir's safe water release strategy to free up flood storage capacity, and calculates the expected safe and usable reservoir capacity after regulation is completed, provided that flood control constraints are not violated.

[0063] Specifically, a hard constraint of "flood control veto" is injected into the reservoir capacity regulation model. The reservoir capacity evolution trajectory within future scheduling cycles is calculated and extrapolated using the reservoir water balance differential equation; its mathematical expression is as follows:

[0064] ;

[0065] in, for Storage capacity at any time For the current storage capacity, To ensure the safe release of water from the reservoir, This refers to the amount of loss due to evaporation and leakage. To calculate the time step.

[0066] Flood control veto constraint verification: for any time within a future scheduling cycle System-mandated verification <= ,in This represents the current statutory flood control limit for the reservoir. If the predicted highest water level reaches the flood control limit, indicating a risk of exceeding the limit, the reservoir capacity regulation model will suspend all economic water supply requests and prioritize generating a safe water release strategy. To free up flood storage capacity, until a safe and usable reservoir capacity that meets the highest decision-priority constraint of flood control safety is derived throughout the entire scheduling cycle. .

[0067] Step S4: Based on the expected safe and available reservoir capacity, forcibly remove the rigid demand water volume that cannot participate in allocation, and generate a dynamic scheduling benchmark water volume for multi-user water resource mixed game calculation;

[0068] The system deducts from the expected safe and available reservoir capacity determined in step S3 the physical dead reservoir capacity for maintaining dam structural stability, the basic water consumption required for residential use (which can be obtained through the demand acquisition system), and the basic ecological base flow demand (the minimum ecological base flow required to maintain the basic ecological health of the downstream river channel). The net water volume remaining after forced deduction is the dynamic scheduling benchmark water volume. This dynamic scheduling benchmark water volume serves as the highest available water rights pool and is distributed to the downstream multi-user water resource hybrid game system at a preset frequency, serving as the absolute upper limit of competition and cooperation among the parties.

[0069] According to the present invention, the specific implementation process of the multi-user water resource mixed game system is as follows:

[0070] Before the multi-user water resource mixed game scenario (reservoir cooperative game and water plant non-cooperative game) is launched, the rigid demand water volume is forcibly separated from the adjusted expected safe and available reservoir capacity.

[0071] Specifically, the dynamic scheduling benchmark water volume that can be used for market-based allocation The specific calculation formula is as follows:

[0072] ;

[0073] in, Physical dead storage capacity to maintain the structural stability of the dam; This is the basic water consumption that must be met for residential life; This is the basic ecological base flow demand.

[0074] Calculated dynamic scheduling benchmark water volume This represents the final maximum available water rights. The system pushes this dynamic scheduling baseline water volume value to the multi-user water resource hybrid game engine at a high refresh rate of 0.05Hz. The competition for water quotas among various parties (agriculture, industry, etc.) is strictly confined to this baseline for Nash equilibrium solution, completely eliminating the risk of flooding and water shortages for residents due to blindly pursuing economic benefits from the system's bottom layer.

[0075] Secondly, the present invention provides a multi-user water resource hybrid game system that can realize the reservoir dynamic scheduling benchmark generation method that integrates dual-drive prediction and flood control constraints provided in the first aspect.

[0076] The first aspect of this invention provides a method for generating a reservoir dynamic scheduling benchmark that integrates dual-drive prediction and flood control constraints, which operates in a multi-user water resources hybrid game system. The hardware of the multi-user water resources hybrid game system includes at least IoT devices and communication gateways for collecting hydrological and meteorological data, as well as a core server cluster containing CPU and GPU heterogeneous computing power nodes, for executing computational and inference processes such as dual-drive hybrid prediction algorithms and dynamic weighted fusion algorithms that run in parallel with physical dynamic numerical models and spatiotemporal AI large models.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. However, these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a reservoir dynamic scheduling benchmark that fuses double drive prediction and flood control constraints, characterized in that, Includes the following steps: S1: Collect real-time meteorological forecast data, multi-source hydrological monitoring data, and geospatial topology data of the reservoir basin; S2: Based on the data collected in step S1, a dual-drive hybrid prediction algorithm that combines a physical dynamics numerical model and a spatiotemporal AI large model, as well as a dynamic weighted fusion algorithm, is used to dynamically extrapolate the inflow process line of the reservoir in future periods and output the hybrid predicted inflow. S3: Taking flood control safety as the highest decision priority constraint, the mixed predicted inflow volume is substituted into the reservoir water balance differential equation to deduce the reservoir capacity evolution trajectory within the future scheduling cycle. If the predicted highest water level touches the flood control limit water level red line, the reservoir capacity adjustment model will prioritize triggering the reservoir safety water release strategy to free up flood storage capacity, and calculate the expected safe and usable reservoir capacity after the adjustment is completed under the highest decision priority constraint. S4: Based on the expected safe and available reservoir capacity, forcibly remove the rigid demand water volume that cannot participate in allocation, and generate a dynamic scheduling benchmark water volume for multi-user water resource mixed game calculation.

2. The method according to claim 1, wherein, In step S2, the physical dynamics numerical model uses physical equations to rigorously calculate the precipitation generation and confluence process based on real-time weather forecast data and geospatial topology data.

3. The method according to claim 2, wherein, The physical dynamics numerical model adopts a one-dimensional / two-dimensional coupled distributed hydrodynamic model, and solves the Saint-Venant equations based on the theorems of mass conservation and momentum conservation to output a predicted inflow sequence with strict physical boundary constraints.

4. The method of claim 1, wherein the method further comprises: In step S2, the spatiotemporal AI big data model predicts flow by mining the nonlinear evolution patterns in historical long-sequence hydrological and meteorological data.

5. The method for generating reservoir dynamic scheduling benchmarks that integrates dual-drive prediction and flood control constraints according to claim 4, characterized in that, The spatiotemporal AI large model uses a spatiotemporal graph convolutional network combined with a long short-term memory network. The spatiotemporal graph convolutional network is used to extract the spatial topological features of the watershed where the reservoir is located, and the long short-term memory network is used to process the time dependency between historical weather forecast data and flow, and output a data-driven predicted inflow sequence.

6. The method for generating reservoir dynamic scheduling benchmarks integrating dual-drive prediction and flood control constraints according to claim 1, characterized in that, The loss function of the spatiotemporal AI large model is as follows: ； in, Mean square error, The spatiotemporal AI large model output for the spatiotemporal AI large model. This is a sequence of actual measured inflow values ​​from the reservoir; the second term... This is a penalty for a single-sided flood peak, used to prevent underestimation of flood levels; and All of these are weight parameters.

7. The method for generating reservoir dynamic scheduling benchmarks integrating dual-drive prediction and flood control constraints according to claim 1, characterized in that, In step S2, the specific implementation process of the dynamic weighted fusion algorithm is as follows: Calculate the mixed predicted water inflow ;in, To predict the inflow sequence The weighting coefficients, To predict the inflow sequence Weighting coefficients; The weight dynamic adaptive update strategy is as follows: ; ;in, This is an extreme weather index, ranging from [0,1].

8. The method for generating reservoir dynamic scheduling benchmarks integrating dual-drive prediction and flood control constraints according to claim 1, characterized in that, In step S3, the reservoir capacity evolution trajectory within the future scheduling cycle is calculated and deduced using the reservoir water balance differential equation. Its mathematical expression is as follows: ； in, for Storage capacity at any time For the current storage capacity, To predict the mixed water inflow, To ensure the safe release of water from the reservoir, This refers to the amount of loss due to evaporation and leakage. To calculate the time step.

9. The method for generating reservoir dynamic scheduling benchmarks integrating dual-drive prediction and flood control constraints according to claim 1, characterized in that, In step S3, a hard constraint of flood control veto is injected into the reservoir capacity adjustment model, which applies to any time within a future scheduling cycle. System-mandated verification <= , for Storage capacity at any time The current statutory flood control limit reservoir capacity is used; if the highest water level is predicted to reach the flood control limit, the reservoir capacity regulation model will suspend all economic water supply requests and prioritize generating a safe water release strategy for the reservoir. To free up flood storage capacity, until a safe and usable reservoir capacity that meets the highest decision-priority constraint of flood control safety is derived throughout the entire scheduling cycle. .

10. The method for generating reservoir dynamic scheduling benchmarks integrating dual-drive prediction and flood control constraints according to claim 1, characterized in that, In step S4, the specific calculation formula for the dynamic scheduling benchmark water volume is as follows: ； in, To ensure the expected safe and usable storage capacity, Physical dead storage capacity to maintain the structural stability of the dam. This is to meet the basic water needs of residents. This is the basic ecological base flow demand.

Images