Method for predicting reservoir inflow and related device

Abstract

Embodiments of the present application provide a reservoir rainfall prediction method and related equipment, which belong to the technical field of data analysis. The method comprises: obtaining reservoir measurement data of multiple sources corresponding to a first reservoir; fusing the reservoir measurement data of multiple sources corresponding to the first reservoir to obtain multi-modal fusion features, and calculating reservoir rainfall prediction data according to the multi-modal fusion features; predicting a total reservoir rainfall prediction result according to the reservoir measurement data of multiple sources corresponding to multiple reservoirs respectively through a graph neural network; and inputting the reservoir rainfall prediction data corresponding to each reservoir and the total reservoir rainfall prediction result into a digital twin biomimetic model to obtain a reservoir rainfall prediction model. Embodiments of the present application not only can dynamically display the simulation state of the rainfall of each reservoir in the current and future time periods, but also support visual deduction and multi-scenario analysis of the overall water regime evolution process of a river basin, thereby providing intuitive and reliable digital support for dispatching decisions.

Description

Technical Field

[0001] This application relates to the field of data analysis technology, and in particular to a method and related equipment for predicting the rainfall capacity of a reservoir. Background Technology

[0002] In current hydrological forecasting practices, most techniques for predicting inflow (i.e., rainfall) rely on single-mode or static models. These methods are often based on simplified assumptions and struggle to effectively capture the complex nonlinear relationships and non-stationary spatiotemporal evolution characteristics of rainfall-runoff processes, thus limiting forecast accuracy. Especially in applications involving joint scheduling of multiple reservoirs and intraday rolling scheduling, the system needs to dynamically respond to real-time meteorological changes and hydrological conditions. Traditional forecasting models, lacking adaptive capabilities and multi-scale fusion mechanisms, cannot adequately handle spatiotemporal heterogeneity, thus facing severe challenges in the coordinated optimization of multiple objectives such as flood control, water supply security, and power generation benefits. They struggle to achieve efficient real-time trade-offs and decision support among these competing objectives. Summary of the Invention

[0003] The main objective of this application is to propose a method and related equipment for predicting reservoir rainfall, which can not only dynamically display the simulated rainfall status of each reservoir in the current and future periods, but also support the visualization and multi-scenario analysis of the overall watershed hydrological evolution process, thereby providing intuitive and reliable digital support for scheduling decisions.

[0004] To achieve the above objectives, one aspect of this application proposes a method for predicting reservoir rainfall capacity, the method comprising: The system acquires reservoir measurement data from multiple sources corresponding to the first reservoir; these various reservoir measurement data are collected by corresponding sensors; the first reservoir is any one of multiple reservoirs. Multimodal fusion features are obtained by fusing reservoir measurement data from multiple sources corresponding to the first reservoir, and rainfall prediction data are calculated based on the multimodal fusion features. The total rainfall forecast is obtained by using graph neural networks to predict the total rainfall received based on reservoir measurement data from multiple sources corresponding to multiple reservoirs. The predicted rainfall data for each of the reservoirs and the predicted total rainfall are input into the digital twin bionic model to obtain the reservoir rainfall prediction model; the reservoir rainfall prediction model is used to display the simulated rainfall status of multiple reservoirs.

[0005] In some embodiments, fusing reservoir measurement data from multiple sources corresponding to the first reservoir to obtain multimodal fusion features includes: Feature extraction and alignment were performed on reservoir measurement data from different sources to obtain multiple reservoir measurement features; The multimodal fusion features are obtained by fusing the measurement features of the multiple reservoirs through a cross-modal attention mechanism and variational autoencoder.

[0006] In some embodiments, calculating the predicted rainfall data based on the multimodal fusion features includes: According to the formula The predicted rainfall data were calculated, and the For the rainfall forecast data, the For the multimodal fusion feature, the For the decoder, the This is Gaussian noise, used to describe the uncertainty in predicting rainfall.

[0007] In some embodiments, the method further includes: Based on the measurement characteristics of the multiple reservoirs and the formula The predicted rainfall data were calculated, and the For the rainfall forecast data, the As a feature of the precipitation sequence, the The upstream flow characteristics of the river, the As environmental variable characteristics, the As a characteristic of social media information, the , , , as well as For the weight parameters, the This is a nonlinear correction term.

[0008] In some embodiments, the step of predicting the total rainfall forecast result using a graph neural network based on reservoir measurement data from multiple sources corresponding to multiple reservoirs includes: The total rainfall forecast is obtained by updating the nodes of the reservoir measurement data from various sources corresponding to the multiple reservoirs through a graph neural network; the graph neural network uses each reservoir and its control node as nodes.

[0009] In some embodiments, inputting the predicted rainfall capacity data corresponding to each of the reservoirs and the predicted total rainfall capacity into a digital twin bionic model to obtain a reservoir rainfall capacity prediction model includes: Obtain the initial simulation state of the digital twin bionic model; The initial simulation state is updated based on the predicted rainfall data for each of the reservoirs and the total predicted rainfall to obtain the target simulation state. The reservoir's rainfall prediction model is established based on the target simulation state.

[0010] In some embodiments, updating the initial simulation state based on the predicted rainfall capacity of each of the reservoirs and the predicted total rainfall capacity to obtain the target simulation state includes: The initial simulation state, the predicted rainfall data for each of the reservoirs, and the total predicted rainfall are input into the formula. The target simulation state is obtained. For the target simulation state, the The initial simulation state, the The predicted rainfall capacity for each of the aforementioned reservoirs and the predicted total rainfall capacity are given. For the gate opening command, the The model parameters of the digital twin bionic model are... This represents the simulation error.

[0011] To achieve the above objectives, another aspect of this application provides a reservoir rainfall forecasting device, the device comprising: The data acquisition module is used to acquire reservoir measurement data from multiple sources corresponding to the first reservoir; the various reservoir measurement data are obtained through corresponding sensors; the first reservoir is any one of multiple reservoirs; The data fusion module is used to fuse reservoir measurement data from multiple sources corresponding to the first reservoir to obtain multimodal fusion features, and to calculate the predicted rainfall data based on the multimodal fusion features. The result prediction module is used to predict the total rainfall forecast result based on the reservoir measurement data from multiple sources corresponding to multiple reservoirs using a graph neural network. The data simulation module is used to input the predicted rainfall data corresponding to each of the reservoirs and the predicted total rainfall into the digital twin bionic model to obtain the reservoir rainfall prediction model; the reservoir rainfall prediction model is used to display the simulated rainfall status of multiple reservoirs.

[0012] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0013] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described above.

[0014] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the methods described above. The embodiments of this application include at least the following beneficial effects: This application provides a method, device, electronic device, storage medium, and program product for predicting reservoir rainfall. This solution acquires reservoir measurement data from multiple sources corresponding to a first reservoir; the various reservoir measurement data are collected by corresponding sensors; the first reservoir is any one of multiple reservoirs; the reservoir measurement data from multiple sources corresponding to the first reservoir are fused to obtain multimodal fusion features, and rainfall prediction data are calculated based on the multimodal fusion features; the total rainfall prediction result is predicted by using a graph neural network based on the reservoir measurement data from multiple sources corresponding to multiple reservoirs; the rainfall prediction data corresponding to each reservoir and the total rainfall prediction result are input into a digital twin bionic model to obtain a reservoir rainfall prediction model; the reservoir rainfall prediction model is used to display the simulated rainfall state of multiple reservoirs. By implementing the embodiments of this application, reservoir measurement data from multiple sources are fused to form a multimodal fusion feature with spatiotemporal consistency, thereby more accurately reflecting the comprehensive hydrological status of the reservoir. Simultaneously, to capture the spatial correlation and hydrological linkage effects among multiple reservoirs within the basin, graph neural network modeling is employed, treating each reservoir as a node in the graph neural network to construct a reservoir topology network. This enables collaborative prediction of the total rainfall inflow across the entire basin. Furthermore, by constructing a digital twin biomimetic model, a high-fidelity reservoir rainfall prediction model is generated. This model not only dynamically displays the simulated rainfall inflow status of each reservoir in the current and future periods but also supports the visualization and multi-scenario analysis of the overall hydrological evolution process of the basin, thus providing intuitive and reliable digital support for scheduling decisions. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of an implementation environment provided in an embodiment of this application; Figure 2 This is a flowchart of the method for predicting reservoir rainfall capacity provided in the embodiments of this application; Figure 3 This is a schematic diagram of the reservoir rainfall forecasting process in one embodiment; Figure 4 This is a schematic diagram of a multi-library topology of a graph neural network in one embodiment; Figure 5 This is a flowchart illustrating the closed loop of digital twin and deep reinforcement learning in one embodiment; Figure 6 This is a schematic diagram of the structure of the reservoir rainfall prediction device provided in the embodiments of this application; Figure 7This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0017] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0018] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0020] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0021] 1) A Graph Neural Network (GNN) is a deep learning model specifically designed for processing graph-structured data. It extracts high-level node representations by recursively applying the same set of parameters to nodes in a graph; 2) Graph Attention Network (GAT) learns the attention weights of each edge to adaptively measure the importance of different neighboring nodes to the current node.

[0022] Among related technologies, Graph Neural Networks (GNNs) are used for prediction and state estimation of water conservancy and urban water networks due to their ability to model complex spatial relationships and interactions. Research indicates that GNNs can naturally utilize the inductive bias of network topology to estimate pressure and flow in water networks from limited sensor data, and improve the model's robustness to sensor location changes through random sensor placement strategies. Compared to traditional physical models, GNNs can alleviate problems such as data scarcity, parameter uncertainty, and extensive manual calibration. Furthermore, some studies have proposed combining graph convolution with convolutional neural networks to simultaneously extract temporal and spatial features.

[0023] It is worth noting that in reservoir and river network systems, spatial relationships are not only reflected in the direct connectivity between nodes, but also include the time lag effect of flood wave propagation, the feedback effect of gate scheduling, and the impact of upstream water flow on downstream ecology. Traditional distance decay or simple adjacency relationships are insufficient to express such complex interactions, prompting researchers to explore modeling based on Graph Attention Networks (GATs). GATs adaptively measure the importance of different neighboring nodes to the current node by learning the attention weight of each edge, thus interpretably reflecting the degree of influence of key reservoirs or tributaries on the system. One literature report describes a GAT model incorporating residual connections for water network pressure estimation, utilizing a training strategy of random sensor placement and noise injection to ensure the model remains robust and accurate in the face of sensor location changes and observational noise.

[0024] Besides graph neural networks, cross-modal fusion has become a hot topic in environmental forecasting in recent years. Traditional multimodal fusion methods often simply superimpose the results after modeling different modes separately, making it difficult to handle the different temporal resolutions and semantic differences between modes. Recent research (such as XFMNet) proposes a stepwise fusion strategy, which, after aligning remote sensing precipitation images and station time series, decomposes the trend and periodic components sequentially, and dynamically fuses different modes through cross-attention gating, significantly improving long-term forecast accuracy. This suggests that in rainfall forecasting, we should not only fuse remote sensing images, but also combine GNSS PWV, meteorological texts, and social media, and use attention mechanisms to dynamically adjust weights to adapt to the differences in contributions from different weather scenarios.

[0025] Digital twins are virtual models driven by real-time data, capable of simulating, monitoring, and optimizing the operation of physical systems in a virtual environment. By combining sensor data streams with models and algorithms, digital twins can predict future states and assist in decision-making. Literature indicates that a fusion framework integrating digital twins with multiple models helps improve the efficiency of water resource forecasting and operations. Applications of digital twins in water resource management include simulating urban water networks, detecting leaks, predicting demand, and optimizing pressure; in the event of events such as pipe bursts, problems can be identified in real time, and corrective measures can be recommended.

[0026] The application of digital twins in reservoirs and hydropower systems is rapidly developing. For aging hydropower stations and complex reservoir systems, the industry is developing digital twin platforms suitable for the hydropower sector, enabling operators to simulate different operating conditions, monitor equipment health, and formulate maintenance strategies. The report points out that the "Hydropower Digital Twin System" platform, developed in collaboration with companies by the U.S. Department of Energy's National Laboratories, allows operators to adjust water levels, flow rates, and turbine speeds based on real-time data. These updates help control turbine wear, reduce downtime risks, and extend facility lifespan. By collecting real-time data on water pressure, turbine speed, and power generation, the platform supports observation of normal operating conditions and allows operators to adjust water levels and flow rates to simulate extreme weather or fluctuations in power demand, thereby assessing the impact of different scheduling schemes on equipment lifespan and power generation efficiency. The combination of digital twins and artificial intelligence is giving rise to a series of new predictive and control tools. For example, predictive frameworks integrating traditional machine learning models, artificial neural networks, and deep networks are synchronized with the digital twin to form multi-layered integrated models. These tools not only support real-time simulation but also continuously update model parameters in the feedback loop, enabling dynamic system adaptation.

[0027] Meanwhile, for aging hydropower stations and complex reservoir systems, the industry is developing digital twin platforms suitable for the hydropower sector. These platforms enable operators to simulate different operating conditions, monitor equipment health, and formulate maintenance strategies. The report points out that the "Hydropower Digital Twin System" platform, developed in collaboration with companies by the U.S. Department of Energy's national laboratories, allows operators to adjust water levels, flow rates, and turbine speeds based on real-time data. Corresponding updates help control turbine wear, reduce downtime risks, and extend facility lifespan. These examples demonstrate that digital twins provide reservoir operations with intelligent, visualized, and scalable new tools.

[0028] In view of this, this application provides a method for predicting the rainfall carrying capacity of reservoirs. This method can not only dynamically display the simulated rainfall carrying capacity of each reservoir in the current and future periods, but also support the visualization and multi-scenario analysis of the overall watershed hydrological evolution process, thereby providing intuitive and reliable digital support for scheduling decisions.

[0029] The method for predicting reservoir rainfall capacity provided in this application relates to the field of data analysis technology. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the method for predicting reservoir rainfall capacity, but is not limited to the above forms.

[0030] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0031] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0032] like Figure 1 The diagram shown is a schematic representation of an implementation environment provided in an embodiment of this application. (Refer to...) Figure 1 The implementation environment includes at least one terminal 102 and a server 101. The terminal 102 and the server 101 can be connected via a network, either wirelessly or via a wired connection, to complete data transmission and exchange.

[0033] Server 101 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0034] Additionally, server 101 can also be a node server in a blockchain network. Blockchain is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms.

[0035] Terminal 102 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. Terminal 102 and server 101 can be directly or indirectly connected via wired or wireless communication, and this embodiment of the application does not impose any limitations.

[0036] For example, based on Figure 1 The implementation environment shown in this application embodiment provides a method for predicting reservoir rainfall capacity. The following description uses the application of this method in server 101 as an example. It can be understood that this method can also be applied in terminal 102.

[0037] Figure 2 This is an optional flowchart of the method for predicting the rainfall carrying capacity of a reservoir provided in the embodiments of this application. The subject executing the method for predicting the rainfall carrying capacity of a reservoir can be any of the aforementioned electronic devices (including servers or terminals). Figure 2 The method may include, but is not limited to, steps S201 to S204.

[0038] Step S201: Obtain reservoir measurement data from multiple sources corresponding to the first reservoir; the various reservoir measurement data are obtained through corresponding sensors; the first reservoir is any one of multiple reservoirs.

[0039] In some embodiments, a network of multiple sensors deployed in the reservoir area and catchment area is used to simultaneously collect measured data from different sources. These sensor types include water level gauges, rain gauges, current meters, and weather stations, which are used to monitor the reservoir's own water level changes, rainfall intensity on the reservoir surface and in the upstream area, inflow, and related meteorological elements. The aim is to comprehensively capture the dynamic hydrological state of the reservoir from different dimensions, providing a rich and reliable input foundation for subsequent data fusion and model calculation.

[0040] As an optional implementation method, this approach integrates multi-source heterogeneous data, including remote sensing imagery, radar precipitation products, precipitable water retrieved from GNSS (Global Navigation Satellite System), time-series monitoring data from hydrological stations, soil moisture information, meteorological text forecasts, and social media content. It also leverages a distributed storage and computing platform based on HDFS (a core component of the Hadoop ecosystem) and Spark to systematically clean, denoise, format-calibrate, spatiotemporally aggregate, and perform batch-stream integrated processing on this data, thereby constructing a unified, standardized, efficient, and usable data warehouse system.

[0041] Furthermore, remote sensing imagery, radar precipitation products, GNSS precipitable water data, rainfall and water level time series from hydrological stations, soil moisture, meteorological text reports, and social media sentiment are used as data sources. This data is accessed through a sensor network to edge computing nodes, and after initial screening, it is aggregated to a distributed storage platform, such as HDFS. Spark is used to integrate batch and stream processing, performing time synchronization, missing value imputation, coordinate alignment, noise filtering, and anomaly detection on the data. For remote sensing data, convolutional neural networks are used to extract spatial features; for text, pre-trained models are used to generate embeddings; and for time series data, standardization, sliding windowing, and periodic decomposition are performed.

[0042] Step S202: The reservoir measurement data from multiple sources corresponding to the first reservoir are fused to obtain multimodal fusion features, and the rainfall prediction data are calculated based on the multimodal fusion features.

[0043] In some embodiments, heterogeneous data from different sensors are integrated and complemented through feature alignment, spatiotemporal registration, and deep learning processes to eliminate potential information conflicts and scale differences, thereby generating a multimodal fusion feature that can comprehensively and consistently characterize the current integrated state of the reservoir.

[0044] As an optional implementation, feature extraction and alignment are performed on reservoir measurement data from different sources to obtain multiple reservoir measurement features. These features are then fused using a cross-modal attention mechanism and a variational autoencoder to obtain multimodal fused features. Physically meaningful feature vectors are extracted using methods such as sliding window statistics and frequency domain transformation, and their timestamps and data formats are standardized to form a multi-source feature set that is strictly aligned in the spatiotemporal dimension. Subsequently, the cross-modal attention mechanism and the variational autoencoder are introduced as core algorithms for deep feature fusion. The cross-modal attention mechanism dynamically evaluates and weights the interaction importance of features from different sources, capturing their synergistic or inhibitory relationships on rainfall prediction. The variational autoencoder further learns the deep probability distribution structure from these weighted features, extracting a compact and robust multimodal fused feature representation through encoding and decoding processes.

[0045] Feature extraction and alignment were performed on reservoir measurement data from different sources to obtain multiple reservoir measurement features. These features were then fused using a cross-modal attention mechanism and variational autoencoder to obtain a multimodal fusion feature. This not only preserved the original multi-source information to the maximum extent but also eliminated observation noise and redundancy through nonlinear modeling. Ultimately, a high-dimensional feature that can comprehensively and stably represent the complex hydrological state of reservoirs was formed, laying a solid data foundation for subsequent accurate prediction tasks.

[0046] In some embodiments, it can be based on the formula The predicted rainfall data were obtained through calculation. For rainfall forecast data, For multimodal fusion features, For decoder, The noise is Gaussian, used to describe the uncertainty in predicting rainfall. This formula can not only output rainfall prediction data... More importantly, by sampling the noise distribution, a probabilistic estimate of the rainfall forecast interval can be achieved, thereby providing an important basis for subsequent reservoir scheduling decisions at the risk perception level and effectively improving the system's robustness and decision reliability in uncertain environments.

[0047] Furthermore, based on the measurement characteristics of multiple reservoirs and formulas The predicted rainfall data were obtained through calculation. For rainfall forecast data, As a feature of the precipitation sequence, For the upstream flow characteristics of the river, For environmental variable characteristics, Characteristics of social media information , , , as well as For weight parameters, This is a nonlinear correction term. This is the temporal decay weight, used to adjust the contribution of features at different time steps to the current prediction. , , as well as These are the influence coefficients of the corresponding features at each time step, which can be learned through model training. As an introduced nonlinear correction term, it is typically implemented using lightweight neural networks to capture complex patterns in unstructured information such as social media that are difficult to represent by linear models. This approach not only integrates physical monitoring data and environmental factors but also innovatively incorporates social media signals as auxiliary input. It can further capture the indirect impacts of human activities and public perception on watershed water conditions beyond traditional hydrological characteristics, thereby enhancing the model's adaptability to complex real-world scenarios and its predictive robustness while maintaining model interpretability.

[0048] Step S203: Based on the reservoir measurement data from multiple sources corresponding to multiple reservoirs, the total rainfall forecast result is obtained by using a graph neural network.

[0049] In some embodiments, multiple reservoirs and their connecting channels within a watershed are abstracted into a topological graph structure, where each reservoir is represented as a node in the graph, and the spatial relationships between reservoirs, such as hydrological connections and river flow directions, are defined as edges in the graph. The features of each node are jointly constituted by the measurement data of its corresponding multi-source reservoirs. Through its unique message passing and neighborhood aggregation mechanisms, the graph neural network can iteratively learn and update the hidden state of each node, thereby effectively capturing the inflow impact of upstream reservoirs, the lag effect of downstream reservoirs, and the complex hydrological spatiotemporal interaction processes within the entire watershed.

[0050] As an optional implementation, a graph neural network (Graph Neural Network) is used to update the nodes of the reservoir measurement data from multiple sources corresponding to the various reservoirs, thereby predicting the total rainfall. The Graph Neural Network uses each reservoir and its control nodes as nodes. In the task of predicting the total rainfall of the watershed, a Graph Neural Network is used as the core modeling framework to construct a hydrological topology graph structure, where the spatially distributed multi-reservoir system and its associated control nodes are collectively constructed. In this graph, each reservoir and its key control node are abstracted as an independent node, and its node features are composed of the multi-source reservoir measurement data corresponding to that location. The Graph Neural Network, through its unique message passing mechanism, performs multiple rounds of iterative calculations along predefined edges such as river channel connections and confluence relationships. In each iteration, each node aggregates hydrological state information from its upstream neighboring nodes and integrates its own multimodal features, thereby dynamically updating its hidden state representation. This node update process enables each reservoir node not only to perceive changes in its own measurement data but also to understand its spatial location and hydrological response logic within the entire watershed. After multiple rounds of neighborhood information propagation and node state updates, the graph neural network finally aggregates the final states of all nodes into an embedding vector that can represent the overall hydrological conditions of the entire watershed through a global pooling layer or readout function, and outputs an accurate prediction result of the total rainfall based on this vector.

[0051] Furthermore, for the graph neural network part, a graph attention network is used to adaptively learn the adjacency weights, and its node update formula is:

[0052] in :node Input or update features, For learnable weighted attention vectors, For attention weights, For vector concatenation, It is a non-linear activation. This node update formula allows the model to dynamically focus on upstream nodes that are more important for prediction.

[0053] For multi-reservoir systems, a directed graph is constructed with reservoirs and their control nodes as nodes and river network flow directions as edges. A spatiotemporal graph convolutional network (ST-GCN) is employed to propagate information across the topology, capturing cross-site overflow, cascading, and storage-discharge coupling relationships. The update rule for the graph convolutional layer is as follows:

[0054] in,( = A + I) is the adjacency matrix plus the identity matrix, ( ) for( The degree matrix of ) This represents the hidden state of the (k-1)th layer. Let be the weight matrix, ( ) is the activation function, which can also be set as an identity mapping. This model enables predictions to take into account the cascading effects of upstream confluence, inter-reservoir scheduling, and local extreme events.

[0055] For the topology of a multi-reservoir system, reservoirs are treated as nodes. Each node's attributes include water level, reservoir capacity, inflow, and outflow, while edge attributes include distance, river slope, and control relationships. In the time dimension, the state of the past (T) steps is used as input via a sliding window. A spatiotemporal graph convolutional network is used for updates.

[0056] The first term implements spatial convolution, and the second term captures temporal dependencies. The weights are for temporal convolution. This structure can model inter-repository chaining and feedback relationships, capturing the coupling effects of flood wave propagation and scheduling.

[0057] Figure 3 This is a schematic diagram of the reservoir rainfall forecasting process in one embodiment, such as... Figure 3 As shown, the process begins with the collection of multi-source data, which is then transmitted to a big data platform. Next, data from the big data platform is fused and predicted. The results of this fusion prediction are used for both spatiotemporal modeling and direct transmission to the scheduling stage. After spatiotemporal modeling is completed, the simulation stage begins, and the simulation results are also input into the scheduling process. The scheduling stage comprehensively processes the information from the fusion prediction and simulation before proceeding to the execution stage. During execution, real-time monitoring is performed, and the data obtained from this real-time monitoring is fed back to the big data platform, thus forming a complete data processing and business execution loop.

[0058] Figure 4 This is a schematic diagram of a multi-library topology of a graph neural network in one embodiment, such as... Figure 4 As shown, Reservoir 1 delivers water to Reservoir 2 through discharge, and there is a flood overflow coupling relationship between Reservoir 1 and Reservoir 2. Reservoir 2 supplies water to Reservoir 3 through discharge, and Reservoir 3 also discharges water. Reservoir 2 also has a bidirectional relationship of scheduling dependence and feedback with Reservoir 5. In addition, Reservoir 1 and Reservoir 4 are connected through ecological compensation, presenting a complex connection and operation mode of water scheduling, interaction, and ecological compensation among the reservoirs.

[0059] By using graph neural networks to update the nodes of reservoir measurement data from various sources corresponding to the multiple reservoirs, the total rainfall forecast is obtained. This can explicitly model the spatial dependence and hydrological connectivity between reservoirs, and elevate the traditional isolated single-point prediction to a system-level collaborative inference, thereby significantly improving the ability to characterize the overall behavior of complex watershed systems and the accuracy of prediction.

[0060] Step S204: Input the predicted rainfall data for each reservoir and the predicted total rainfall into the digital twin bionic model to obtain the reservoir rainfall prediction model; the reservoir rainfall prediction model is used to display the simulated rainfall status of multiple reservoirs.

[0061] In some embodiments, an initial simulation state of the digital twin bionic model can be obtained; the initial simulation state is updated based on the rainfall forecast data corresponding to each reservoir and the total rainfall forecast result to obtain the target simulation state; and a reservoir rainfall forecast model is established based on the target simulation state. The initial simulation state is obtained from the established digital twin bionic platform, reflecting a digital mapping of the hydrodynamic properties and topological relationships between reservoirs in the basin at a certain reference time. Subsequently, the independent rainfall forecast data for each reservoir generated in the aforementioned steps, along with the total rainfall forecast result obtained from the graph neural network, are input into the digital twin platform. Through its embedded multi-source data assimilation mechanism and state update algorithm, the initial simulation state is dynamically corrected and continuously updated. This update process essentially involves deep integration of real-time forecast information and the simulation model to generate a target simulation state that more closely resembles the actual future hydrological scenario. Based on this optimized target state, the digital twin platform, by calling its built-in hydrodynamic model and multi-objective simulation inference engine, ultimately constructs a high-fidelity reservoir rainfall forecast model.

[0062] It can not only accurately reflect the dynamic changes in rainfall carrying capacity of each reservoir in the future, but also support collaborative simulation and forward-looking analysis of multi-objective competitive scenarios, thus providing a forward-looking and operational decision-making basis for intelligent watershed scheduling.

[0063] As an optional implementation, the initial simulation state, the predicted rainfall data for each reservoir, and the total predicted rainfall are input into the formula. The target simulation state is obtained. For the target simulation state, This is the initial simulation state. This provides the predicted rainfall capacity for each reservoir and the predicted total rainfall capacity. This is the gate opening command. For the model parameters of the digital twin bionic model, To simulate errors, real-time prediction data and control strategies can be integrated into the simulation process to generate a target simulation state X_(t+1) that conforms to physical laws and reflects the latest prediction information. This state update method based on mechanism and data fusion significantly improves the accuracy and reliability of digital twin systems in spatiotemporal extrapolation, providing a more accurate dynamic simulation environment for subsequent scheduling decisions.

[0064] Figure 5 This is a flowchart illustrating the closed loop of digital twin and deep reinforcement learning in one embodiment, as shown below. Figure 5 As shown, sensor and remote sensing data are used as inputs and provided to the prediction model (multimodal + GNN). The prediction model transmits data to the digital twin simulation on the one hand, and receives information fed back from the digital twin simulation on the other hand. The digital twin simulation transmits the processed results to the reinforcement learning agent. The reinforcement learning agent generates control actions and applies them to the physical reservoir system. The state data of the physical reservoir system is then collected and fed back to the sensor and remote sensing data link, thus forming a closed-loop process from data acquisition, model prediction, simulation analysis, intelligent decision-making to actual system control and back to data acquisition.

[0065] Steps S201 to S204 of this embodiment involve acquiring reservoir measurement data from multiple sources corresponding to a first reservoir; these various reservoir measurement data are collected by corresponding sensors; the first reservoir is any one of multiple reservoirs; the reservoir measurement data from multiple sources corresponding to the first reservoir are fused to obtain multimodal fusion features, and rainfall forecast data are calculated based on these features; a graph neural network is used to predict the total rainfall forecast result based on the reservoir measurement data from multiple sources corresponding to each of the multiple reservoirs; the rainfall forecast data for each reservoir and the total rainfall forecast result are input into a digital twin bionic model to obtain a reservoir rainfall forecast model; the reservoir rainfall forecast model is used to demonstrate the rainfall capacity of multiple reservoirs. The simulation of reservoir rainfall is achieved by fusing reservoir measurement data from multiple sources to form a multimodal fusion feature with spatiotemporal consistency, thus more accurately reflecting the comprehensive hydrological status of the reservoir. Furthermore, to capture the spatial correlation and hydrological linkage effects among multiple reservoirs within the basin, graph neural network modeling is employed, treating each reservoir as a node in the graph neural network to construct a reservoir topology network. This enables collaborative prediction of the total rainfall inflow across the entire basin. By constructing a digital twin biomimetic model, a high-fidelity reservoir rainfall prediction model is generated. This model not only dynamically displays the simulated rainfall inflow status of each reservoir in the current and future periods but also supports the visualization and multi-scenario analysis of the overall hydrological evolution process of the basin, providing intuitive and reliable digital support for scheduling decisions.

[0066] Please see Figure 6This application also provides a reservoir rainfall forecasting device that can implement the above-described method. The device includes: The data acquisition module 601 is used to acquire reservoir measurement data from multiple sources corresponding to the first reservoir; the various reservoir measurement data are acquired through corresponding sensors; the first reservoir is any one of multiple reservoirs; The data fusion module 602 is used to fuse reservoir measurement data from multiple sources corresponding to the first reservoir to obtain multimodal fusion features, and calculate the predicted rainfall data based on the multimodal fusion features. The result prediction module 603 is used to predict the total rainfall forecast result based on the reservoir measurement data from multiple sources corresponding to multiple reservoirs through a graph neural network. The data simulation module 604 is used to input the rainfall forecast data corresponding to each reservoir and the total rainfall forecast result into the digital twin bionic model to obtain the reservoir rainfall forecast model; the reservoir rainfall forecast model is used to display the rainfall simulation status of multiple reservoirs.

[0067] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0068] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0069] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0070] Please see Figure 7 , Figure 7 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 701 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 702 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 702 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 702 and is called and executed by the processor 701 using the methods described in the embodiments of this application. The input / output interface 703 is used to implement information input and output; The communication interface 704 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 705 transmits information between various components of the device (e.g., processor 701, memory 702, input / output interface 703, and communication interface 704); The processor 701, memory 702, input / output interface 703, and communication interface 704 are connected to each other within the device via bus 705.

[0071] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0072] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0073] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0074] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0075] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0076] The method, apparatus, electronic device, storage medium, and program product for predicting reservoir rainfall capacity provided in this application embodiment acquire reservoir measurement data from multiple sources corresponding to a first reservoir. These various reservoir measurement data are collected by corresponding sensors. The first reservoir can be any one of multiple reservoirs. The multi-source reservoir measurement data corresponding to the first reservoir are fused to obtain multimodal fusion features, and rainfall capacity prediction data are calculated based on these features. A graph neural network is used to predict the total rainfall capacity based on the multi-source reservoir measurement data corresponding to each of the multiple reservoirs. The rainfall capacity prediction data for each reservoir and the total rainfall capacity prediction result are input into a digital twin bionic model to obtain a reservoir rainfall capacity prediction model. This model is used to display the simulated rainfall capacity status of multiple reservoirs. It not only dynamically displays the simulated rainfall capacity status of each reservoir in the current and future periods but also supports the visualization and multi-scenario analysis of the overall watershed hydrological evolution process, thus providing intuitive and reliable digital support for scheduling decisions.

[0077] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0078] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0079] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0080] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0081] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0082] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0083] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0084] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0085] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0086] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0087] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for predicting the rainfall carrying capacity of a reservoir, characterized in that, The method includes the following steps: The system acquires reservoir measurement data from multiple sources corresponding to the first reservoir; these various reservoir measurement data are obtained through corresponding sensors; the first reservoir is any one of multiple reservoirs. Multimodal fusion features are obtained by fusing reservoir measurement data from multiple sources corresponding to the first reservoir, and rainfall prediction data are calculated based on the multimodal fusion features. The total rainfall forecast is obtained by using graph neural networks to predict the total rainfall received based on reservoir measurement data from multiple sources corresponding to multiple reservoirs. The predicted rainfall capacity of each of the aforementioned reservoirs and the total predicted rainfall capacity are input into a digital twin bionic model to obtain a reservoir rainfall capacity prediction model; the reservoir rainfall capacity prediction model is used to display the simulated rainfall capacity status of multiple reservoirs; Based on the measurement characteristics of the multiple reservoirs and the formula The predicted rainfall data were calculated, and the For the rainfall forecast data, the As a feature of the precipitation sequence, the The upstream flow characteristics of the river, the As environmental variable characteristics, the As a characteristic of social media information, the , , , as well as For the weight parameters, the This is a nonlinear correction term; The calculation of the rainfall prediction data based on the multimodal fusion features includes: According to the formula The predicted rainfall data were calculated, and the For the rainfall forecast data, the For the multimodal fusion feature, the For the decoder, the Gaussian noise is used to describe the uncertainty in predicting rainfall. The step of inputting the predicted rainfall capacity of each of the reservoirs and the predicted total rainfall capacity into a digital twin bionic model to obtain a reservoir rainfall capacity prediction model includes: Obtain the initial simulation state of the digital twin bionic model; The initial simulation state is updated based on the predicted rainfall data for each of the reservoirs and the total predicted rainfall to obtain the target simulation state. Establish the reservoir rainfall prediction model based on the target simulation state; The step of updating the initial simulation state based on the predicted rainfall capacity of each of the reservoirs and the predicted total rainfall capacity to obtain the target simulation state includes: The initial simulation state, the predicted rainfall data for each of the reservoirs, and the total predicted rainfall are input into the formula. The target simulation state is obtained. For the target simulation state, the The initial simulation state, the The predicted rainfall capacity for each of the aforementioned reservoirs and the predicted total rainfall capacity are given. For the gate opening command, the The model parameters of the digital twin bionic model are... This represents the simulation error.

2. The method according to claim 1, characterized in that, The process of fusing reservoir measurement data from multiple sources corresponding to the first reservoir to obtain multimodal fusion features includes: Feature extraction and alignment were performed on reservoir measurement data from different sources to obtain multiple reservoir measurement features; The multimodal fusion features are obtained by fusing the measurement features of the multiple reservoirs through a cross-modal attention mechanism and variational autoencoder.

3. The method according to claim 1, characterized in that, The method of using a graph neural network to predict total rainfall load based on reservoir measurement data from multiple sources corresponding to multiple reservoirs includes: The total rainfall forecast is obtained by updating the nodes of the reservoir measurement data from various sources corresponding to the multiple reservoirs through a graph neural network; the graph neural network uses each reservoir and its control node as nodes.

4. A device for predicting the rainfall carrying capacity of a reservoir, characterized in that, The device includes: The data acquisition module is used to acquire reservoir measurement data from multiple sources corresponding to the first reservoir; the various reservoir measurement data are acquired through corresponding sensors; the first reservoir is any one of multiple reservoirs; The data fusion module is used to fuse reservoir measurement data from multiple sources corresponding to the first reservoir to obtain multimodal fusion features, and to calculate the predicted rainfall data based on the multimodal fusion features. The result prediction module is used to predict the total rainfall forecast result based on the reservoir measurement data from multiple sources corresponding to multiple reservoirs using a graph neural network. The data simulation module is used to input the predicted rainfall data for each of the reservoirs and the predicted total rainfall into the digital twin bionic model to obtain a reservoir rainfall prediction model; the reservoir rainfall prediction model is used to display the simulated rainfall status of multiple reservoirs; based on the measurement characteristics of the multiple reservoirs and formulas... The predicted rainfall data were calculated, and the For the rainfall forecast data, the As a feature of the precipitation sequence, the The upstream flow characteristics of the river, the As environmental variable characteristics, the As a characteristic of social media information, the , , , as well as For the weight parameters, the This is a nonlinear correction term; The calculation of the rainfall prediction data based on the multimodal fusion features includes: According to the formula The predicted rainfall data were calculated, and the For the rainfall forecast data, the For the multimodal fusion feature, the For the decoder, the Gaussian noise is used to describe the uncertainty in predicting rainfall. The step of inputting the predicted rainfall capacity of each of the reservoirs and the predicted total rainfall capacity into a digital twin bionic model to obtain a reservoir rainfall capacity prediction model includes: Obtain the initial simulation state of the digital twin bionic model; The initial simulation state is updated based on the predicted rainfall data for each of the reservoirs and the total predicted rainfall to obtain the target simulation state. Establish the reservoir rainfall prediction model based on the target simulation state; The step of updating the initial simulation state based on the predicted rainfall capacity of each of the reservoirs and the predicted total rainfall capacity to obtain the target simulation state includes: The initial simulation state, the predicted rainfall data for each of the reservoirs, and the total predicted rainfall are input into the formula. The target simulation state is obtained. For the target simulation state, the The initial simulation state, the The predicted rainfall capacity for each of the aforementioned reservoirs and the predicted total rainfall capacity are given. For the gate opening command, the The model parameters of the digital twin bionic model are... This represents the simulation error.

5. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 3.

Images