Reservoir management optimization method and system based on digital twinning

Abstract

The invention discloses a reservoir management optimization method and system based on digital twinning. The method comprises the steps of multi-source data acquisition and virtualization mapping, digital twinning modeling, reservoir risk early warning and reservoir management decision optimization. The invention relates to the technical field of reservoir intelligent management, in particular to a reservoir management optimization method and system based on digital twinning, and the method comprises the steps: constructing a virtualization data field through multi-Bezi wavelets and fractal interpolation, and carrying out the self-learning of dynamic weight fusion through a time sequence convolution network; constructing a macroscopic twinborn body and a microcosmic twinborn body, and coupling through an asynchronous message bus; constructing multi-source risk vectors of water pressure gradient, over-overflow water level, dam body stress and seepage flow, adopting clustering and spatial diffusion, generating a dynamic risk thermodynamic diagram, and performing early warning; by taking flood control safety, water supply guarantee and ecological environment as multiple targets, a multi-target particle swarm optimization model and flow constraints are established, an optimal scheduling scheme is solved and executed in real time, and high-precision simulation, precise early warning and intelligent scheduling are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of reservoir intelligent management, and specifically refers to a reservoir management optimization method and system based on digital twin. Background Art

[0002] With the frequent occurrence of global climate change and extreme weather events, reservoir operation management faces the trade-off of multiple objectives such as flood control safety, water supply guarantee, and ecological protection.

[0003] Traditional reservoir management mostly relies on single hydrological or meteorological monitoring data, with low data update frequency and coarse spatial resolution, making it difficult to comprehensively reflect the real-time dynamics of the reservoir area. On the other hand, existing reservoir simulations mostly use macroscopic hydraulic models or empirical formulas, ignoring the microscopic processes of dam structure stress distribution and sediment transport, resulting in insufficient early warning and decision-making accuracy. In addition, the fusion of multi-source heterogeneous data, model coupling, and real-time response are still bottlenecks restricting the application of digital twin technology in reservoir management: the virtualization mapping of data lacks an adaptive weight allocation mechanism, the coupling efficiency of macroscopic and microscopic models is low, and risk early warning mostly stays at static threshold judgment, making it difficult to achieve multi-dimensional dynamic risk prediction and intelligent decision-making optimization. Summary of the Invention

[0004] In view of the above situation, to overcome the defects of the prior art, the present invention provides a reservoir management optimization method and system based on digital twin. This solution uses an adaptive multi-frequency perception network constructed by multi-Bessey wavelets to achieve fine-grained spatio-temporal feature decomposition and reconstruction, fills in the spatial dimension with a geometric interpolation operator, and introduces a temporal convolutional network to self-learn the historical mapping error and signal-to-noise ratio sequence, dynamically generating normalized weights to weighted-fuse the virtualization results of each channel. This solution constructs a macroscopic twin node by introducing a two-dimensional shallow water equation with an empirical correction term to achieve high-precision simulation of the overall water level and flow rate. At the microscopic level, a graph neural network is used for multi-granularity message passing and decoding to finely model the dam stress and sediment concentration. This solution constructs a multi-source risk index vector integrating water pressure gradient, overoverflow water level, dam stress, and leakage flow rate to achieve comprehensive risk quantification. Fuzzy C-means clustering is used to complete flexible grading, and a spatial diffusion model is introduced to smooth the risk level and enhance spatial coherence. Combining a bidirectional long short-term memory network for risk evolution prediction, a dynamic risk map is generated in advance to help the reservoir safety management transform from passive response to active pre-control.

[0005] The technical solution adopted by the present invention is as follows: A reservoir management optimization method based on digital twin, the method comprising the following steps:

[0006] Step S1: Multi-source data collection and virtualization mapping;

[0007] Step S2: Digital twin modeling;

[0008] Step S3: Reservoir risk early warning;

[0009] Step S4: Optimization of reservoir management decision-making.

[0010] Furthermore, in step S1, the multi-source data acquisition and virtualization mapping specifically include the following steps:

[0011] Step S11: Multi-source data acquisition, collecting the flow rate and water level of hydrological stations; rainfall and wind speed of meteorological stations; pixel reflectance of remote sensing images; dam stress and seepage flow rate of Internet of Things sensors, forming a set of original data streams;

[0012] Step S12: Construction of an adaptive multi-frequency sensing network, by setting multiple layers of frequency channels, constructing scale wavelet bases; and performing wavelet decomposition and reconstruction on each data stream to extract spatio-temporal features;

[0013] Step S13: Fractal interpolation mapping, applying a geometric interpolation operator based on fractal theory to the reconstructed results of each scale wavelet, and performing fine-grained filling in the spatial dimension to obtain a preliminary virtualization field;

[0014] Step S14: Dynamic weight self-learning, first using a temporal convolutional network to process the channel mapping error sequence and signal-to-noise ratio sequence, passing through a fully connected layer mapping, and finally obtaining a normalized weight through a sigmoid function;

[0015] Step S15: Fusion of high-dimensional virtualization data fields.

[0016] Furthermore, in step S2, the digital twin modeling specifically includes the following steps:

[0017] Step S21: Construction of macroscopic twin body nodes, the macroscopic nodes output the overall water level distribution and flow field; and publish status packets to the message bus at each simulation time step; for the entire reservoir area, an empirical correction term is introduced based on the two-dimensional shallow water equation;

[0018] Step S22: Construction of microscopic twin body nodes, specifically including the following steps:

[0019] Step S221: Discretize the dam body and its surrounding area into a graph structure;

[0020] Step S222: Multi-granularity message passing model based on graph neural network;

[0021] Step S223: The microscopic nodes are decoded into stress tensors and sediment concentrations after passing through a multi-layer perceptron;

[0022] Step S224: The microscopic nodes receive the boundary condition packets sent from the macroscopic layer, and output the detailed stress distribution and deposition rate distribution based on the local model, and encapsulate them into refined status packets;

[0023] Step S23: Asynchronous message bus coupling. Design a lightweight asynchronous message bus that supports the publish - and - subscribe mode. The macro - nodes are publishers, and the micro - nodes are subscribers. At the same time, the micro - nodes send refined packets back to the bus, which are then subscribed to by the macro - layer.

[0024] Further, in step S3, for the reservoir risk early warning, after completing the digital twin modeling, based on the virtualized data field obtained from the fusion mapping and the states of the macro - and micro - twins, the reservoir risk early warning is carried out, which specifically includes the following steps:

[0025] Step S31: Construction of multi - source risk indicators;

[0026] Step S32: Risk grading. Based on fuzzy C - means clustering, each coordinate point is graded to obtain a risk - level label;

[0027] Step S33: Spatial diffusion. Introduce a spatial diffusion model, consider neighborhood effects, and generate a smooth risk level;

[0028] Step S34: Generation of a dynamic risk map;

[0029] Step S35: Risk evolution prediction. Design a bidirectional long - short - term memory network. The input sequence is the risk vector sequence of the past K steps, and the output is the risk prediction for the next steps. Update the risk heat map according to the prediction results and give an early warning in advance.

[0030] Further, in step S4, the optimization of reservoir management decisions specifically includes the following steps:

[0031] Step S41: Construction of a management objective vector, including flood control safety objectives, water supply guarantee objectives, and ecological environment objectives;

[0032] Step S42: Construction of constraint conditions, including water balance constraints, storage capacity safety constraints, discharge capacity constraints, and ecological flow constraints;

[0033] Step S43: Decision optimization. First, according to the management objective vector constructed in step S41 and the dynamic risk levels and heat maps obtained from steps S31 - S35, formalize the reservoir operation decision - making problem as a multi - objective optimization model. On the premise of meeting the constraints of water balance, upper and lower limits of storage capacity, maximum discharge capacity, and ecological flow, use a multi - objective particle swarm optimization algorithm to solve the optimal operation plan in combination with the real - time updated risk early - warning information. Execute the operation plan by controlling the reservoir gates and pumping stations in real time.

[0034] The reservoir management optimization system based on digital twins provided by the present invention includes a multi - source data acquisition and virtualization mapping module, a digital twin modeling module, a reservoir risk early - warning module, and a reservoir management decision - making optimization module;

[0035] The multi-source data acquisition and virtualization mapping module collects multi-source heterogeneous data such as the flow and water level of hydrological stations, rainfall and wind speed of meteorological stations, reflectance of remote sensing images, and dam stress and seepage flow of Internet of Things sensors in real time, and generates a spatio-temporal virtualized data field through wavelet decomposition and reconstruction of an adaptive multi-frequency sensing network, spatial interpolation based on fractal theory, and a dynamic weight self-learning mechanism, and sends the data to the digital twin modeling module;

[0036] The digital twin modeling module receives the data sent by the multi-source data acquisition and virtualization mapping module. At the macroscopic level, based on the two-dimensional shallow water equation and introducing an empirical correction term, it simulates the overall water level distribution and flow field; at the microscopic level, the dam and its surroundings are discretized into a graph structure, and multi-granularity message passing and decoding are performed based on a graph neural network to output local stress and sediment concentration; and the macroscopic and microscopic coupling is completed through a lightweight asynchronous message bus. After the state packet and refinement packet are interacted, the data is sent to the reservoir risk warning module;

[0037] The reservoir risk warning module receives the data sent by the digital twin modeling module, constructs risk indicators for pressure gradient, overspill, dam stress, and leakage based on the virtualized data field and digital twin state obtained by fusion mapping, generates a smoothed risk level using fuzzy C-means clustering and a spatial diffusion model, and predicts the risk through a bidirectional long short-term memory network to form a dynamic risk map and warning message, and sends the data to the reservoir management decision optimization module;

[0038] The reservoir management decision optimization module receives the data sent by the reservoir risk warning module, constructs a three-dimensional management objective vector for flood control safety, water supply guarantee, and ecological environment, and under the constraint conditions of water balance, safety storage capacity, maximum discharge capacity, and ecological flow, uses a multi-objective particle swarm optimization algorithm combined with real-time risk information to solve the optimal scheduling plan, and finally executes the scheduling through the reservoir gate and pump station automation control unit.

[0039] The beneficial effects achieved by the present invention using the above solution are as follows:

[0040] (1) Aiming at the problems of insufficient spatio-temporal feature extraction accuracy, rough spatial interpolation, static data source weights, and difficulty in suppressing noise in traditional multi-source data fusion methods, this solution realizes fine-grained spatio-temporal feature decomposition and reconstruction through an adaptive multi-frequency sensing network constructed by multi-Bessel wavelets, performs high-precision filling in the spatial dimension by combining a geometric interpolation operator based on fractal theory, and introduces a three-layer dilation factor temporal convolutional network to self-learn the historical mapping error and signal-to-noise ratio sequence, dynamically generates normalized weights, and weights and fuses the virtualization results of each channel, improving the spatial detail restoration ability, real-time performance, and robustness of the reservoir digital twin data field.

[0041] (2)Aiming at the problems of limited macro - simulation accuracy, insufficient capture of micro - structure details, and slow coupling response of multi - scale models in traditional reservoir digital twin modeling, this solution constructs macro - twin nodes by introducing two - dimensional shallow water equations with empirical correction terms to achieve high - precision simulation of overall water levels and flows; at the micro - level, a graph neural network is used for multi - granularity message passing and decoding to finely model dam stresses and sediment concentrations; and a lightweight asynchronous message bus is designed to realize the dynamic coupling of macro - micro structures. Through a confidence - interval error - perception mechanism, it flexibly switches between global recalculation and local incremental updates, improving the simulation accuracy, response efficiency, and structural dynamic adaptation ability of the twin system in complex hydrological scenarios.

[0042] (3)Aiming at the problems of single - dimensional risk indicators, rough grading, weak spatial correlation, and insufficient prediction of future trends in traditional reservoir risk warning methods, this solution realizes comprehensive risk quantification by constructing a multi - source risk - indicator vector that integrates water - pressure gradient, over - spill water level, dam stress, and leakage flow; uses fuzzy C - means clustering for flexible grading and introduces a spatial diffusion model to smooth risk levels and enhance spatial coherence; combines bidirectional long - short - term memory networks for risk - evolution prediction to generate dynamic risk maps in advance, thus significantly improving the accuracy, spatial perception ability, and time - efficient response ability of risk warning, and helping reservoir safety management to shift from passive response to active pre - control. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a schematic diagram of the reservoir management optimization method based on digital twin provided by the present invention;

[0044] Figure 2 is a schematic diagram of the reservoir management optimization system based on digital twin provided by the present invention;

[0045] Figure 3 is a schematic diagram of step S1;

[0046] Figure 4 is a schematic diagram of step S2;

[0047] Figure 5 is a schematic diagram of step S3.

[0048] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0050] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation of the present invention.

[0051] Embodiment 1, referring to Figure 1 , the method for optimizing reservoir management based on digital twin provided by the present invention includes the following steps:

[0052] Step S1: Multi-source data collection and virtualization mapping, performing multi-source data collection and virtualization mapping through multi-source data collection, construction of an adaptive multi-frequency sensing network, fractal interpolation mapping, dynamic weight self-learning, and fusion of a high-dimensional virtual data field;

[0053] Step S2: Digital twin modeling, performing digital twin modeling through construction of macroscopic twin body nodes, construction of microscopic twin body nodes, and coupling of an asynchronous message bus;

[0054] Step S3: Reservoir risk early warning, performing reservoir risk early warning through construction of multi-source risk indicators, risk grading, spatial diffusion, generation of a dynamic risk map, and prediction of risk evolution;

[0055] Step S4: Optimization of reservoir management decision-making, constructing a management target vector and constraint conditions, and performing optimization of reservoir management decision-making.

[0056] Embodiment 2, referring to Figure 1 and Figure 3 , based on the above embodiment, in step S1, the multi-source data collection and virtualization mapping specifically includes the following steps:

[0057] Step S11: Multi-source data collection, collecting the flow rate and water level of hydrological stations; rainfall and wind speed of meteorological stations; pixel reflectance of remote sensing images; stress and seepage flow of the dam of Internet of Things sensors to form a set of original data streams , where x and y respectively represent the abscissa and ordinate of the spatial coordinates, t represents the time point, i represents the index of the sample, and N represents the total number of samples, represents the i-th data sample;

[0058] Step S12: Adaptive multi-frequency perception network construction, by setting up multiple layers of frequency channels , constructing the scale wavelet basis ,in, Adopt Dobesi wavelet; perform wavelet decomposition and reconstruction on each data stream to extract spatiotemporal features;

[0059] Step S13: Fractal interpolation mapping: Apply the geometric interpolation operator F based on fractal theory to the wavelet reconstruction results of each scale, perform fine-grained filling in the spatial dimension, and obtain a preliminary virtualized field: ;

[0060] Step S14: Dynamic weight self-learning, using a three-layer temporal convolutional network to process the past T-step mapping error sequence of each channel SNR Each layer contains causal convolution with a kernel size of 3 and dilation rates of 1, 2, and 4 respectively; the output dimension is 64, which is mapped to an N-dimensional vector through a fully connected layer and finally normalized by an exponential function. Processing to obtain normalized weights ;

[0061] Step S15: high-dimensional virtualized data field fusion, using the following formula to complete multi-source fusion mapping:

[0062] ;

[0063] in, The output of the network W is generated by the weight, satisfying , It represents the virtualization result of the i-th channel obtained by wavelet and fractal interpolation.

[0064] By performing the above operations, this solution addresses the problems of insufficient spatiotemporal feature extraction accuracy, rough spatial interpolation, static data source weights and difficulty in suppressing noise in traditional multi-source data fusion methods. It uses an adaptive multi-frequency perception network constructed by Dobesi wavelets to achieve fine-grained spatiotemporal feature decomposition and reconstruction, combines geometric interpolation operators based on fractal theory to perform high-precision filling in the spatial dimension, and introduces a three-layer expansion factor temporal convolutional network to self-learn historical mapping errors and signal-to-noise ratio sequences, dynamically generate normalized weights, and weightedly fuse the virtualization results of each channel, thereby improving the spatial detail restoration capability, real-time performance and robustness of the reservoir digital twin data field.

[0065] Example 3, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S2, the digital twin modeling specifically includes the following steps:

[0066] Step S21: Construct the macro twin node, and the macro node outputs the overall water level distribution and the flow field , where represents the variation of the water surface height with time and space, represents the volume of water flowing through a unit cross-section per unit time; and a status packet is published to the message bus at each simulation time step ; For the entire reservoir area, an empirical correction term is introduced based on the two-dimensional shallow water equations:

[0067] ;

[0068] where represents the density of the fluid in the reservoir, u represents the velocity field in the reservoir, describing the water flow velocity at each position in the reservoir; represents the rate of change of the water flow velocity, represents the convective acceleration factor, used to describe the inertial force of the water flow; p represents the water pressure, represents the gradient of the water pressure, represents the dynamic viscosity coefficient, with a value range of [0.0001, 0.01], represents the Laplace operator, represents the external force term, including the driving forces of gravity, reservoir operation, and boundary inflow and outflow, represents the model correction factor, with a value range of [-1, 1];

[0069] Step S22: Construct the micro twin node, which specifically includes the following steps:

[0070] Step S221: Discretize the dam body and its surrounding area into a graph structure , the node represents the grid point, and the edge represents the spatial adjacency relationship;

[0071] Step S222: The multi-granularity message passing model based on the graph neural network:

[0072] ;

[0073] where represents the feature of the i-th node in the (l + 1)-th layer, represents the hyperbolic tangent function, j represents the index of the node, represents all neighbors directly connected to represents the feature of the i-th node in the l-th layer, represents the feature of the j-th node in the l-th layer, represents the multi-layer perceptron;

[0074] Step S223: The microscopic nodes are decoded into stress tensors after passing through a multi-layer perceptron and sediment concentration ;

[0075] Step S224: The microscopic nodes receive the boundary condition packets sent from the macroscopic layer , and based on the local model, output the detailed stress distribution and deposition rate distribution, which are encapsulated into refined state packets ;

[0076] Step S23: Asynchronous message bus coupling. Design a lightweight asynchronous message bus that supports the publish-subscribe mode, with the macroscopic nodes as publishers and the microscopic nodes as subscribers; at the same time, the microscopic nodes send back refined packets to the bus, which are then subscribed to by the macroscopic layer; coupling logic: First, set the error threshold based on the 95% confidence interval of historical operation statistics. If the macroscopic prediction error is greater than the error threshold, immediately trigger a global recalculation; otherwise, only update the boundary conditions and perform a local incremental micro-simulation.

[0077] By performing the above operations, for the problems of limited macroscopic simulation accuracy, insufficient capture of microscopic structural details, and slow coupling response of multi-scale models in traditional reservoir digital twin modeling, this solution constructs macroscopic twin nodes by introducing a two-dimensional shallow water equation with an empirical correction term to achieve high-precision simulation of the overall water level and flow rate; in the microscopic layer, a graph neural network is used for multi-granularity message passing and decoding to finely model the dam stress and sediment concentration; and a lightweight asynchronous message bus is designed to achieve dynamic coupling of macro and micro structures. Through the confidence interval error perception mechanism, it flexibly switches between global recalculation and local incremental update, improving the simulation accuracy, response efficiency, and structural dynamic adaptation ability of the twin system in complex hydrological scenarios.

[0078] Example 4, refer to Figure 1 and Figure 5 , this example is based on the above example. In step S3, for the reservoir risk warning, after completing the digital twin modeling, according to the virtualized data field obtained by fusion mapping and the states of the macroscopic and microscopic twins, perform reservoir risk warning, which specifically includes the following steps:

[0079] Step S31: Construction of multi-source risk indicators, constructing a risk indicator vector:

[0080] ;

[0081] Among them, , represents the water pressure gradient risk indicator, represents the water pressure gradient risk threshold, represents the gradient vector of water pressure, represents taking the modulus; , represents the overspill risk indicator, denotes taking the maximum value, denotes the water level height, denotes the maximum safety water level; , denotes the dam body stress risk index, denotes the ultimate stress of the dam body; , denotes the leakage risk index, denotes the leakage threshold;

[0082] Step S32: Risk classification. Based on fuzzy C - means clustering, each coordinate point of is classified to obtain the risk level label ;

[0083] Step S33: Spatial diffusion. Introduce a spatial diffusion model, consider the neighborhood influence, and generate a smoothed risk level, which is expressed as follows:

[0084] ;

[0085] Among them, denotes the smoothed risk level after spatial diffusion, denotes the diffusion coefficient, and its value range is , denotes the neighborhood set of the coordinate point , denotes the number of coordinate points in the neighborhood set of the coordinate point , and respectively denote the abscissa and ordinate of the coordinate points in the neighborhood set;

[0086] Step S34: Generation of dynamic risk map. When any point , a warning message is triggered;

[0087] Step S35: Risk evolution prediction. Design a bidirectional long - short - term memory network. The input sequence is the risk vector sequence of the past K steps , and the output is the risk prediction for the next steps ; Update the risk heat map according to the prediction result and give an early warning, which is expressed as follows:

[0088] ;

[0089] Among them, denotes the risk level at the future moment obtained by prediction, denotes the bidirectional long - short - term memory network.

[0090] By performing the above operations, in view of the problems of single risk index dimension, rough grading, weak spatial correlation, and insufficient prediction of future trends in traditional reservoir risk early warning methods, this solution constructs a multi-source risk index vector integrating water pressure gradient, overflow water level, dam body stress, and leakage flow to achieve comprehensive risk quantification; uses fuzzy C-means clustering to complete flexible grading, and introduces a spatial diffusion model to smooth the risk level and enhance spatial coherence; combines a bidirectional long short-term memory network for risk evolution prediction and generates a dynamic risk map in advance, thus significantly improving the accuracy, spatial perception ability, and time-lapse response ability of risk early warning, and helping the reservoir safety management to shift from passive response to active pre-control.

[0091] Example 5, refer to Figure 1 , based on the above example, in step S4, the optimization of the reservoir management decision specifically includes the following steps:

[0092] Step S41: Construct a management objective vector, including flood control safety objectives, water supply guarantee objectives, and ecological environment objectives, which are expressed as follows:

[0093] ;

[0094] Among them, represents the management objective vector, represents the flood control safety objective, defined as the sum of the water level drops in the risk areas, and is expressed as: ; among them, represents the set of all reservoir domain grids, represents the water level change amount after scheduling, represents the early warning level threshold, with a value of 3; represents the water supply guarantee objective, defined as the opposite of the absolute value of the difference between the outflow and the demand flow, and is expressed as: ; among them, represents the outflow, represents the demand flow; represents the ecological environment objective, defined in the form of an ecological flow deviation penalty term, and is expressed as: ; among them, represents the ecological flow;

[0095] Step S42: Construct constraint conditions, including water balance constraints, reservoir capacity safety constraints, water discharge capacity constraints, and ecological flow constraints, which are expressed as follows:

[0096] ;

[0097] Among them, represents the reservoir capacity at time t + 1, represents the reservoir capacity at time t, represents the inflow, represents the evaporation loss volume, and respectively represent the upper and lower limits of the reservoir capacity, represents the maximum water discharge capacity;

[0098] Step S43: Decision optimization. First, based on the management objective vector constructed in Step S41 and the dynamic risk level and heat map obtained in Steps S31 - S35, the reservoir scheduling decision problem is formalized as a multi-objective optimization model; on the premise of satisfying the water balance, upper and lower limits of the reservoir capacity, maximum water discharge capacity, and ecological flow constraint conditions, a multi-objective particle swarm optimization algorithm is used to solve the optimal scheduling scheme in combination with the real-time updated risk warning information; the scheduling scheme is executed in real time by controlling the reservoir gates and pump stations.

[0099] Embodiment Six. Refer to Figure 1 and Figure 2 . Based on the above embodiments, the reservoir management optimization system based on digital twin provided by the present invention includes a multi-source data acquisition and virtualization mapping module, a digital twin modeling module, a reservoir risk warning module, and a reservoir management decision optimization module;

[0100] The multi-source data acquisition and virtualization mapping module collects multi-source heterogeneous data of the flow rate and water level of hydrological stations, rainfall and wind speed of meteorological stations, reflectance of remote sensing images, and dam stress and seepage flow rate of Internet of Things sensors in real time, and generates a spatio-temporal virtualized data field through wavelet decomposition and reconstruction of an adaptive multi-frequency sensing network, spatial interpolation based on fractal theory, and a dynamic weight self-learning mechanism, and sends the data to the digital twin modeling module;

[0101] The digital twin modeling module receives the data sent by the multi-source data acquisition and virtualization mapping module. At the macroscopic level, based on the two-dimensional shallow water equation and introducing an empirical correction term, it simulates the overall water level distribution and flow field; at the microscopic level, the dam and its surroundings are discretized into a graph structure, and multi-granularity message passing and decoding are performed based on a graph neural network to output local stress and sediment concentration; and through a lightweight asynchronous message bus, macroscopic and microscopic coupling is completed. After the state packet and refinement packet are interacted, the data is sent to the reservoir risk warning module;

[0102] The reservoir risk warning module receives the data sent by the digital twin modeling module, constructs risk indicators of pressure gradient, overspill, dam stress, and leakage based on the virtualized data field and digital twin state obtained by fusion mapping, generates a smoothed risk level using fuzzy C-means clustering and a spatial diffusion model, and predicts the risk through a bidirectional long short-term memory network to form a dynamic risk map and warning message, and sends the data to the reservoir management decision optimization module;

[0103] The reservoir management decision optimization module receives the data sent by the reservoir risk early warning module, constructs a three-dimensional management objective vector of flood control safety, water supply guarantee and ecological environment, and under the constraint conditions of water balance, safety storage capacity, maximum water discharge capacity and ecological flow, uses a multi-objective particle swarm optimization algorithm combined with real-time risk information to solve the optimal scheduling plan, and finally executes the scheduling through the automatic control unit of the reservoir gate and pumping station.

[0104] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0105] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention.

[0106] The above describes the present invention and its implementation manners. Such description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A reservoir management optimization method based on digital twins, characterized in that, The method includes the following steps: Step S1: Multi-source data collection and virtualization mapping, which is carried out through multi-source data collection, construction of an adaptive multi-frequency sensing network, fractal interpolation mapping, dynamic weight self-learning, and fusion of high-dimensional virtual data fields for multi-source data collection and virtualization mapping; Step S2: Digital twin modeling, which is carried out through construction of macro twin body nodes, construction of micro twin body nodes, and coupling of asynchronous message buses for digital twin modeling; Step S3: Reservoir risk early warning, which is carried out through construction of multi-source risk indicators, risk grading, spatial diffusion, generation of dynamic risk maps, and prediction of risk evolution for reservoir risk early warning; Step S4: Optimization of reservoir management decision-making, constructing a management target vector and constraint conditions for optimization of reservoir management decision-making.

2. The method for optimizing reservoir management based on digital twin according to claim 1, characterized in that: In step S1, the multi-source data collection and virtualization mapping specifically includes the following steps: Step S11: Multi-source data collection, collecting the flow rate and water level of hydrological stations; rainfall and wind speed of meteorological stations; pixel reflectivity of remote sensing images; dam stress and seepage flow rate of Internet of Things sensors to form a set of original data streams; Step S12: Construction of an adaptive multi-frequency sensing network, by setting multiple layers of frequency channels to construct scale wavelet bases; and performing wavelet decomposition and reconstruction on each data stream to extract spatio-temporal features; Step S13: Fractal interpolation mapping, applying a geometric interpolation operator based on fractal theory to the wavelet reconstruction results at each scale for fine-grained filling in the spatial dimension to obtain a preliminary virtualization field; Step S14: Dynamic weight self-learning, first using a temporal convolutional network to process the channel mapping error sequence and signal-to-noise ratio sequence, passing through a fully connected layer for mapping, and finally passing through a normalized exponential function for processing to obtain a normalized weight; Step S15: Fusion of high-dimensional virtual data fields.

3. The method for optimizing reservoir management based on digital twin according to claim 1, wherein: In step S2, the digital twin modeling specifically includes the following steps: Step S21: Construction of macro twin body nodes, where the macro nodes output the overall water level distribution and flow field; and publish status packets to the message bus at each simulation time step; for the entire reservoir area, an empirical correction term is introduced based on the two-dimensional shallow water equation; Step S22: Construction of micro twin body nodes, which specifically includes the following steps: Step S221: Discretize the dam body and its surrounding areas into a graph structure; Step S222: A multi-granularity message passing model based on graph neural networks; Step S223: The micro nodes are decoded into stress tensors and sediment concentrations after passing through a multi-layer perceptron; Step S224: The micro nodes receive the boundary condition packets sent from the macro layer, output the detailed stress distribution and sedimentation rate distribution based on the local model, and encapsulate them into refined status packets; Step S23: Asynchronous message bus coupling, designing a lightweight asynchronous message bus to support the publish and subscribe modes, with the macro nodes as publishers and the micro nodes as subscribers; at the same time, the micro nodes send back refined packets to the bus, which are then subscribed to by the macro layer.

4. The reservoir management optimization method based on digital twin according to claim 1, wherein: In step S3, for the reservoir risk early warning, after completing the digital twin modeling, based on the virtualization data field obtained through fusion mapping and the states of the macro and micro twins, reservoir risk early warning is carried out, which specifically includes the following steps: Step S31: Construction of multi-source risk indicators; Step S32: Risk grading. Based on fuzzy C-means clustering, each coordinate point is graded to obtain a risk level label. Step S33: Spatial diffusion. Introduce a spatial diffusion model, consider neighborhood influence, and generate a smoothed risk level. Step S34: Generate a dynamic risk map. Step S35: Risk evolution prediction. Design a bidirectional long short-term memory network. The input sequence is the risk vector sequence of the past K steps, and the output is the risk prediction for the next step. Update the risk heat map according to the prediction result and give an early warning.

5. The method for optimizing reservoir management based on digital twin according to claim 1, wherein: In step S4, the optimization of the reservoir management decision specifically includes the following steps: Step S41: Construct a management objective vector, including flood control safety objectives, water supply guarantee objectives, and ecological environment objectives; Step S42: Construct constraint conditions, including water balance constraints, reservoir capacity safety constraints, water discharge capacity constraints, and ecological flow constraints; Step S43: Decision optimization. First, according to the management objective vector constructed in step S41 and the dynamic risk level and heat map obtained in steps S31–S35, formalize the reservoir scheduling decision problem as a multi-objective optimization model; on the premise of satisfying the water balance, upper and lower limits of the reservoir capacity, maximum water discharge capacity, and ecological flow constraint conditions, use a multi-objective particle swarm optimization algorithm, combined with real-time updated risk warning information, to solve for the optimal scheduling plan; execute the scheduling plan in real time by controlling the reservoir gates and pump stations.

6. A reservoir management optimization system based on digital twins, which is used to implement the reservoir management optimization method based on digital twins described in any one of claims 1-5, characterized in that: It includes a multi-source data acquisition and virtualization mapping module, a digital twin modeling module, a reservoir risk warning module, and a reservoir management decision optimization module.

7. The digital twin-based reservoir management optimization system according to claim 6, characterized in that: The multi-source data acquisition and virtualization mapping module collects multi-source heterogeneous data of hydrological station flow, water level, meteorological station rainfall, wind speed, remote sensing image reflectivity, and dam body stress and seepage flow of IoT sensors in real time, and generates a spatio-temporal virtualized data field through wavelet decomposition and reconstruction of an adaptive multi-frequency sensing network, spatial interpolation based on fractal theory, and a dynamic weight self-learning mechanism, and sends the data to the digital twin modeling module; The digital twin modeling module receives the data sent by the multi-source data acquisition and virtualization mapping module. At the macroscopic level, based on the two-dimensional shallow water equation and introducing an empirical correction term, it simulates the overall water level distribution and flow field; at the microscopic level, the dam body and its surroundings are discretized into a graph structure, and multi-granularity message passing and decoding are performed based on a graph neural network to output local stress and sediment concentration; and through a lightweight asynchronous message bus, macroscopic and microscopic coupling is completed. After the state packet and refinement packet are exchanged, the data is sent to the reservoir risk warning module; The reservoir risk warning module receives the data sent by the digital twin modeling module, constructs risk indicators of pressure gradient, overspill, dam body stress, and leakage based on the virtualized data field and digital twin state obtained by fusion mapping, generates a smoothed risk level using fuzzy C-means clustering and a spatial diffusion model, and predicts the risk through a bidirectional long short-term memory network to form a dynamic risk map and warning message, and sends the data to the reservoir management decision optimization module; The reservoir management decision optimization module receives the data sent by the reservoir risk early warning module, constructs a three-dimensional management objective vector for flood control safety, water supply guarantee and ecological environment, and under the constraint conditions of water balance, safe storage capacity, maximum discharge capacity and ecological flow, uses a multi-objective particle swarm optimization algorithm combined with real-time risk information to solve the optimal scheduling plan, and finally executes the scheduling through the reservoir gate and pump station automatic control unit.

Images