Simulation verification method and system for intelligent pumped storage power station reconstruction design

Abstract

The application discloses a kind of intelligent pumped storage power station reconstruction design simulation verification method and system.The method comprises: establishing the digital twin model of the pumped storage power station, the digital twin model includes hydraulic structure model, electromechanical equipment model and control system model;The reconstruction design scheme is introduced into the digital twin model, and the model parameters are optimized based on deep learning algorithm and historical operation data;Multi-working condition simulation verification is carried out, the dynamic response characteristics of each system are analyzed, and the influence of the reconstruction design scheme on the overall performance of the power station is evaluated;Intelligent analysis is carried out on the simulation results based on machine learning, potential risks are identified, and optimization suggestions are proposed.The application also discloses a system for implementing the method.The application innovatively introduces a deep learning algorithm to intelligently optimize model parameters, reducing the workload of manual parameter adjustment and improving the efficiency of scheme optimization.

Description

Technical Field

[0001] This application belongs to the field of hydropower engineering technology, specifically involving a simulation verification method and system for the retrofit design of intelligent pumped storage hydropower stations. Background Technology

[0002] Pumped storage hydropower stations are important peak-shaving and frequency-regulating facilities in the power system, and their safe and stable operation is of great significance to the safety and stability of the power grid. As the service life of pumped storage hydropower stations increases, problems such as equipment aging and performance degradation become increasingly prominent, necessitating renovation and upgrading to maintain their good operating condition.

[0003] Currently, the retrofit design of pumped storage hydropower stations mainly relies on engineers' experience and simple numerical calculations, which has the following problems: First, the feasibility verification of the retrofit plan lacks comprehensiveness, making it impossible to accurately assess the impact of the retrofit plan on the overall performance of the hydropower station; second, traditional simulation verification methods have low computational efficiency and are difficult to handle complex multi-condition analysis needs; third, there is a lack of intelligent risk assessment tools, making it difficult to detect potential safety hazards in a timely manner. Therefore, there is an urgent need for an intelligent simulation verification method for retrofit design. Summary of the Invention

[0004] In view of this, this application provides an intelligent simulation verification method and system for the retrofit design of pumped storage hydropower stations, which solves the problems of insufficient comprehensiveness, low efficiency and outdated risk assessment methods in the existing technology for the simulation verification of retrofit design of pumped storage hydropower stations.

[0005] This application provides a simulation verification method for the retrofit design of an intelligent pumped storage hydropower station, comprising: establishing a digital twin model of the pumped storage hydropower station, the digital twin model including a hydraulic structure model, an electromechanical equipment model, and a control system model; importing the retrofit design scheme into the digital twin model, and optimizing the model parameters based on deep learning algorithms and historical operating data; performing multi-condition simulation verification through the optimized digital twin model to obtain simulation results, the simulation results including: the dynamic response characteristics of the hydraulic structures, electromechanical equipment, and control system, and the impact of the retrofit design scheme on the overall performance of the hydropower station; and performing intelligent analysis of the simulation results based on machine learning to identify potential risks and propose optimization suggestions.

[0006] Optionally, establishing a digital twin model of the pumped storage hydropower station includes: collecting actual operating data of the hydropower station, including geometric parameters of hydraulic structures, unit operating parameters, and control system parameters; establishing a model of the hydraulic structures, including upper and lower reservoirs, pressure pipelines, and tailrace system; establishing a model of the electromechanical equipment, including turbines, generators, and main transformers; and establishing a model of the control system, including governors, excitation systems, and monitoring systems.

[0007] Optionally, the modification design scheme is imported into the digital twin model, and the model parameters are optimized based on deep learning algorithms and historical operating data. This includes: acquiring and preprocessing historical operating data, organizing the historical operating data according to collection time, spatial location, parameter type, and operating state to form a high-dimensional tensor containing time, space, parameter, and state dimensions; decomposing the high-dimensional tensor using a convolutional tensor decomposition method, which includes CP decomposition and Tucker decomposition of the convolutional kernel tensor; constructing a multi-layer convolutional network structure for parameter optimization based on the decomposed high-dimensional tensor, the network structure including: an input layer for receiving input data, a feature extraction layer for extracting features, a feature fusion layer for fusing multi-dimensional features, and an output layer for outputting optimized parameters; training the multi-layer convolutional network structure using an adaptive learning rate strategy, and improving the training effect through batch normalization to obtain optimized model parameters.

[0008] Optionally, the high-dimensional tensor decomposition method using convolutional tensor decomposition further includes: using tensor ring integral decomposition to process cyclic features in the data; introducing tensor regularization constraints; applying residual connections in the network structure; adjusting tensor decomposition parameters; and determining the optimal tensor decomposition parameters based on the model performance evaluation results.

[0009] Optionally, multi-condition simulation verification is performed using the optimized digital twin model, including: designing typical operating condition test schemes including normal start-up and shutdown, accident operating conditions, and phase-shifting operation based on the digital twin model; using the digital twin model, performing multi-condition parallel simulation with distributed computing methods to analyze the dynamic response characteristics of hydraulic structures, electromechanical equipment, and control systems under different operating conditions; storing the simulation results of the digital twin model in a simulation database; and generating a simulation verification report containing key performance indicators based on the simulation results of the digital twin model.

[0010] Optionally, intelligent analysis of simulation results based on machine learning can be performed to identify potential risks and propose optimization suggestions, including: mapping key parameters of the modification scheme to a high-dimensional parameter space; constructing a Regge triangulation of the parameter space based on the correlation between parameters to obtain multiple triangular units; defining a Regge metric tensor describing local geometric features on each triangular unit; calculating the discrete curvature of the simulation results in the parameter space based on the Regge metric tensor; determining the optimal path for parameter adjustment based on the distribution characteristics of the discrete curvature, and generating specific parameter optimization suggestions based on the optimal path.

[0011] Optionally, intelligent analysis of simulation results based on machine learning to identify potential risks and propose optimization suggestions further includes: analyzing regions with large discrete curvature in the parameter space and defining them as parameter-sensitive regions; calculating the Regge metric features of the parameter-sensitive regions, including principal curvature values ​​and principal curvature directions; identifying high-curvature regions with discrete curvature anomalies based on the Regge metric features as potential risk points for the hydropower station; constructing an optimization objective function that includes constraints on parameter-sensitive regions; using a geodesic search algorithm to optimize parameters based on the optimization objective function to obtain optimized parameters that avoid high-risk regions; and verifying the stability and reliability of the optimization results by evaluating the changes in Regge metrics before and after optimization.

[0012] Optionally, the intelligent analysis of simulation results based on machine learning to identify potential risks and propose optimization suggestions also includes: using a genetic algorithm to perform preliminary optimization of key parameters of the modification scheme; establishing a modification process model describing the dynamic change process of parameters based on the preliminary optimization results; performing transient state analysis based on the modification process model to evaluate the impact of parameter changes on the hydropower station; and evaluating the reliability of parameter optimization results by calculating reconstruction errors and analyzing the impact of different decomposition orders.

[0013] Optionally, the collection of actual operating data of the hydropower station includes: setting up data collection points, including key monitoring points of hydraulic structures, monitoring points of unit operating parameters, and monitoring points of the control system; establishing a data collection system to collect power station operating data in real time; and preprocessing the collected data, including data cleaning, outlier handling, and data standardization.

[0014] This application also provides an intelligent pumped storage hydropower station retrofit design simulation verification system, comprising: a model building module for establishing a digital twin model of the pumped storage hydropower station, the digital twin model including a hydraulic structure model, an electromechanical equipment model, and a control system model; a parameter optimization module for importing the retrofit design scheme into the digital twin model and optimizing the model parameters based on deep learning algorithms and historical operating data; a simulation calculation module for performing multi-condition simulation verification through the optimized digital twin model to obtain simulation results, the simulation results including: the dynamic response characteristics of the hydraulic structures, electromechanical equipment, and control system, and the impact of the retrofit design scheme on the overall performance of the hydropower station; and an intelligent evaluation module for intelligently analyzing the simulation results based on machine learning, identifying potential risks, and proposing optimization suggestions.

[0015] This application also provides a computer device, the computer device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described method for simulation verification of intelligent pumped storage hydropower station retrofit design.

[0016] This application also provides a computer-readable storage medium that stores computer instructions for causing a computer to execute the above-described method for simulation verification of intelligent pumped storage hydropower station retrofit design.

[0017] This application also provides a computer program product, including computer instructions, characterized in that, when the computer instructions are executed by a processor, they implement the steps of the above-described method for simulation verification of intelligent pumped storage hydropower station retrofit design.

[0018] This application has the following technical effects:

[0019] By constructing a digital twin model, a comprehensive simulation verification of the design scheme for the retrofit of pumped storage hydropower stations was achieved, improving the accuracy and reliability of the retrofit design.

[0020] Deep learning algorithms were used to optimize model parameters, which improved the efficiency and accuracy of simulation verification.

[0021] Intelligent analysis based on machine learning can promptly identify potential risks and propose optimization suggestions, thereby improving the safety of the renovation design. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0024] Figure 1 This is a flowchart illustrating a simulation verification method for the retrofit design of an intelligent pumped storage hydropower station, as provided in an embodiment of this application.

[0025] Figure 2 This is a schematic diagram of the structure of a digital twin model provided in an embodiment of this application;

[0026] Figure 3This is a schematic diagram of the structure of a deep learning model based on convolutional tensor decomposition provided in an embodiment of this application;

[0027] Figure 4 This is a schematic diagram illustrating the principle of Regge metric analysis provided in an embodiment of this application. Detailed Implementation

[0028] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions of the embodiments of this application.

[0029] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this application mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when embodiments of the present invention refer to an element as being “connected” or “coupled” to another element, the one element may be directly connected or coupled to the other element, or it may mean that the one element and the other element establish a connection relationship through an intermediate element. Furthermore, “connected” or “coupled” as used herein may include wireless connection or wireless coupling. The term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” can be implemented as “A,” or as “B,” or as “A and B.”

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0031] The core of this invention is to provide a method and system for generating intelligent software test data based on publicly available information on the Internet. The invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] It should be understood that the terms "comprising" and "having," and any variations thereof, as used herein, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0033] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0034] This application provides a simulation verification method for the design of an intelligent pumped storage hydropower station retrofit. (Refer to...) Figure 1 The specific steps of this method include:

[0035] S1 establishes a digital twin model of the pumped storage hydropower station:

[0036] A digital twin model is a virtual mapping of a physical hydropower station, achieving synchronization between the physical entity and the virtual model through real-time data interaction and dynamic updates. The model comprises three sub-models: hydraulic structures, electromechanical equipment, and control systems, which achieve collaborative simulation through data interfaces and information exchange mechanisms.

[0037] Specifically, a multiphysics coupling-based modeling method is adopted to unify the description of multiple physical fields such as hydraulics, mechanical dynamics, and electrical control. For example, in the turbine model, the coupling relationship between hydraulic, mechanical, and electrical characteristics needs to be considered simultaneously, and the interaction between various physical quantities is described by establishing a fluid-structure coupling equation set.

[0038] S1.1 Collect actual operating data of the hydropower station:

[0039] This step employs a layered, distributed data acquisition architecture to achieve comprehensive data collection of hydropower station operations. At the physical layer, various sensors and data acquisition devices are deployed; at the network layer, a communication method combining industrial Ethernet and fieldbus is used; and at the application layer, a unified data acquisition and management platform is employed.

[0040] The data acquisition includes: geometric parameters of hydraulic structures such as stress, strain, seepage, and displacement; operating parameters of the generator set such as speed, power, vibration, and bearing temperature; and control parameters of the control system such as governor parameters and excitation voltage and current. For example, for acquiring the bearing temperature of the turbine, a PT100 temperature sensor is used, and data is transmitted in real time via the Modbus-TCP protocol with a sampling period of 100ms.

[0041] Specifically, it is implemented as follows:

[0042] S1.1.1 When setting up data acquisition points, the monitoring locations should be selected according to the principle of "key monitoring and reasonable layout". For example, strain gauges and displacement gauges should be installed at key parts such as the top of the dam and the foundation of the upper reservoir dam; temperature sensors and vibration sensors should be installed at key parts such as the bearings, stators and rotors of the generator unit; and data acquisition points for parameters such as voltage, current and frequency should be installed at key links of the governor and excitation system.

[0043] The placement of data acquisition points needs to consider factors such as equipment layout, environmental conditions, and construction conditions to ensure the accuracy and reliability of the collected data. For example, vibration sensors should be installed at rigid connections along the vibration transmission path, avoiding nodal points.

[0044] The data acquisition system established in S1.1.2 adopts a three-layer architecture: the bottom layer is the field sensor network, which uses industrial Ethernet and fieldbus (such as Profibus, Modbus, etc.) to realize data transmission; the middle layer is the data acquisition server, which is responsible for data acquisition, caching and preprocessing; the upper layer is the data management platform, which realizes the functions of data storage, query and analysis.

[0045] The data acquisition system has real-time and reliability assurance mechanisms, including data caching, breakpoint resumption, and data verification functions. For example, when network communication is interrupted, field devices can temporarily store data in a local cache and automatically upload it once communication is restored.

[0046] S1.1.3 Data preprocessing includes the following steps:

[0047] Data cleaning: Median filtering, wavelet denoising, and other methods are used to remove noise and interference from the collected data. For example, the vibration signal is decomposed into four levels using the db4 wavelet to remove high-frequency noise;

[0048] Outlier handling: Outliers are identified based on the 3σ criterion and box plot method, and outlier data is processed by interpolation or deletion.

[0049] Data standardization: Using maximum-minimum standardization or Z-score standardization methods, data with different dimensions are unified to the same scale range.

[0050] S1.2 Establish the model of the hydraulic structure:

[0051] The hydraulic structure model is constructed based on the finite element method and computational fluid dynamics methods, including geometric and mechanical models of the upper and lower reservoirs, pressure pipelines, and tailrace system. This model can simulate the stress distribution, deformation characteristics, and flow properties of hydraulic structures under different operating conditions.

[0052] The upper reservoir model focuses on the following aspects:

[0053] A three-dimensional solid element model of the dam was established. The element type selected was a 20-node hexahedral isoparametric element, which can accurately describe complex geometric shapes.

[0054] A reservoir water pressure model was established, taking into account the dynamic water pressure effect caused by water level changes, and the Westergaard added mass method was used to calculate the reservoir pressure distribution.

[0055] The contact relationship between the dam body and the bedrock is simulated. The mechanical properties of the dam foundation contact surface are described by surface-to-surface contact elements, and the normal stiffness and tangential stiffness parameters are set.

[0056] Considering the seepage effect, a seepage field model is established using the saturated-unsaturated seepage theory to calculate the pore water pressure distribution.

[0057] The pressure pipeline model includes the following characteristics:

[0058] Establish a three-dimensional solid model of the pipeline, taking into account the composite structural characteristics of the steel lining and concrete lining;

[0059] The fluid-structure interaction method is used to simulate the water hammer effect in a pressure pipeline and to calculate transient pressure fluctuations.

[0060] Considering the elastic deformation of the pipeline, a stress-strain relationship model is established to analyze the structural stability of the pipeline;

[0061] Simulate the interaction between the pipeline and the surrounding rock mass, and set the contact surface parameters and boundary conditions.

[0062] The key aspects of the tailrace system model include:

[0063] Establish a three-dimensional geometric model of the tailrace tunnel, taking into account the influence of tunnel wall roughness;

[0064] Computational fluid dynamics was used to simulate the flow characteristics of the tailrace, and to calculate the velocity and pressure distributions.

[0065] Considering the influence of air intrusion, a two-phase flow model is established to describe the two-phase flow state of water and air.

[0066] The free surface flow at the tailrace outlet was simulated, and the VOF method was used to track the free surface.

[0067] The sub-models are coupled through boundary conditions and data interfaces. For example, the upper reservoir model and the pressure pipeline model are connected through the inlet water level and flow rate conditions, while the pressure pipeline model and the tailrace system model are coupled through the outlet pressure and velocity conditions.

[0068] S1.3 Establish the electromechanical equipment model:

[0069] The electromechanical equipment models mainly include the modeling of three core components: the turbine, the generator, and the main transformer. These models need to consider the coupling relationship between the electrical, mechanical, and thermal characteristics of the equipment in order to accurately reflect the dynamic response characteristics of the equipment under various operating conditions.

[0070] First, the turbine model employs a multiphysics coupling modeling approach. Regarding hydraulic characteristics, a three-dimensional flow field model of the impeller channel is established, using the Reynolds-averaged Navier-Stokes equations to describe the water flow motion, and a k-ε turbulence model to simulate the turbulent characteristics of the flow. For mechanical characteristics, the finite element method is used to establish structural models of key components such as blades and shafting, calculating stress distribution and deformation characteristics. Through fluid-structure interaction analysis, characteristic curves of the turbine under different operating conditions can be obtained, including flow-speed characteristics and efficiency-flow characteristics.

[0071] Secondly, the generator model focuses on the coupled analysis of electromagnetic and thermal fields. Regarding electromagnetic characteristics, the finite element method is used to solve Maxwell's equations, calculating electromagnetic parameters such as the air gap magnetic field distribution and stator current distribution. Equivalent circuit models of the stator and rotor windings are established to analyze the generator's voltage and current characteristics. Regarding thermal characteristics, considering multiple heat sources such as core losses and copper losses, a three-dimensional temperature field model is established to analyze the temperature distribution of each component. Furthermore, a mechanical vibration model is also required to calculate the shaft vibration characteristics and critical speed.

[0072] The main transformer model includes an electromagnetic field model and a thermal field model. In the electromagnetic field model, the equivalent circuit of the transformer is established using the magnetic circuit method, and parameters such as leakage reactance and excitation current are calculated. Through finite element analysis, the magnetic field distribution of the windings and core can be obtained. In the thermal field model, the oil circulation cooling system is considered, and a convective heat transfer model is established to analyze the temperature distribution of various components of the transformer.

[0073] It should be noted that there are close coupling relationships between the various parts of the electromechanical equipment model. For example, the output torque of the turbine directly affects the generator's speed and output power, and the electromagnetic force of the generator affects the vibration characteristics of the shaft system. Therefore, a reasonable data exchange mechanism needs to be established during model construction to ensure the coordination between the various sub-models.

[0074] S1.4 Establish the control system model:

[0075] The control system model mainly consists of three parts: a speed governor, an excitation system, and a monitoring system. It adopts a hierarchical distributed modeling architecture to achieve coordinated operation of each control loop. In this embodiment, the control system model must reflect the dynamic characteristics of each controller, as well as the transmission characteristics of the control signals and the implementation process of the control algorithm.

[0076] The governor model adopts a proportional-integral-derivative (PID) control structure, including an electro-hydraulic conversion system, a hydraulic system, and a mechanical actuator. First, a guide vane opening control loop model is established, including a speed feedback loop, a PID controller, an electro-hydraulic servo valve, and a hydraulic actuator. The electro-hydraulic servo valve's dynamic characteristics are described using a second-order oscillatory element, while the hydraulic actuator considers the effects of hydraulic pressure changes and piston movement.

[0077] The excitation system model includes an automatic voltage regulator (AVR) and limiters. The AVR employs a multi-loop feedback control structure, including an outer voltage loop and an inner current loop. The model incorporates a power angle stabilizer (PSS) to improve the system's dynamic stability by introducing compensation signals for power variations. The limiter model includes protection functions such as overexcitation limiting, underexcitation limiting, and V / Hz limiting.

[0078] The monitoring system model adopts a distributed control system (DCS) architecture, including a field control layer, a process control layer, and a monitoring layer. The modeling focuses on the implementation of functions such as data acquisition, signal processing, control algorithm execution, and human-machine interaction. A communication network model is established to simulate the data transmission process between different layers, including the impact of network characteristics such as communication latency and data packet loss.

[0079] In summary, by establishing a complete digital twin model (the structure of a digital twin is as follows) Figure 2 This allows for a comprehensive mapping of the physical entities of pumped storage hydropower stations, providing fundamental support for subsequent simulation verification.

[0080] S2 imports the renovation design scheme into the digital twin model and optimizes it:

[0081] The core of this step is to integrate the transformation design scheme with the digital twin model and optimize the model parameters through deep learning algorithms to improve the accuracy and reliability of the model.

[0082] S2.1 Acquire and preprocess historical runtime data:

[0083] In this embodiment, operational data is first extracted from the power plant's historical operation database, including multi-dimensional information such as equipment operating parameters, control system status, and environmental parameters. This data is systematically organized according to acquisition time, spatial location, parameter type, and operating status, forming a four-dimensional tensor structure. The time dimension reflects the changing patterns of parameters over time, the spatial dimension describes the distribution characteristics of parameters at different monitoring points, the parameter dimension contains different types of monitoring data, and the status dimension corresponds to different operating conditions.

[0084] A sliding time window method is used to segment the data, with the window length determined based on the parameter's variation characteristics. For example, a smaller time window (e.g., 1 second) is used for rapidly changing parameters (such as pressure pulsation), while a larger time window (e.g., 1 minute) is used for slowly changing parameters (such as bearing temperature). Furthermore, Fourier transform analysis is employed to analyze the data's spectral characteristics, identifying and filtering out interference signals.

[0085] S2.2 employs the convolution tensor decomposition method:

[0086] In the embodiments of this application, such as Figure 3 As shown, a multi-level convolutional tensor decomposition strategy is employed, combining CP decomposition and Tucker decomposition to extract essential features from the data. CP decomposition decomposes the original tensor into the sum of several rank-1 tensors, with each component representing a basic pattern. Tucker decomposition introduces a kernel tensor, which can capture the interaction relationships between different dimensions.

[0087] To handle the cyclic features in the data, a tensor ring integral solution is introduced. Specifically, the data in the time dimension is treated as a periodic signal, and a cyclic convolution kernel is constructed through cyclic shift operations to extract periodic change patterns. Simultaneously, tensor regularization constraints, including L1 norm and kernel norm constraints, are introduced to suppress overfitting.

[0088] Residual connections are incorporated into the network structure to mitigate the vanishing gradient problem in deep networks. Optimal tensor decomposition parameters, including the rank of the CP decomposition and the kernel tensor dimension of the Tucker decomposition, are determined using cross-validation. For example, when processing turbine vibration data, decomposition parameters that balance computational complexity and accuracy requirements can be selected by analyzing the variation curve of the reconstruction error.

[0089] S2.3 Constructing a multi-layer convolutional network structure:

[0090] The multi-layer convolutional network in this embodiment employs an encoder-decoder structure. The input layer receives multi-dimensional features after tensor decomposition and extracts local features through multi-layer convolution operations. The feature extraction layer uses convolutional kernels of different scales to capture multi-scale spatiotemporal correlations. The feature fusion layer uses an attention mechanism to adaptively weight features according to their importance.

[0091] The convolutional layers employ a depthwise separable convolutional structure, reducing the number of parameters while maintaining the model's expressive power. Skip connections are established between feature maps to fuse low-level and high-level features. Finally, the output layer reconstructs the optimized model parameters through deconvolution operations.

[0092] S2.4 employs an adaptive learning rate strategy:

[0093] This application employs an adaptive learning rate adjustment strategy based on the Adam optimizer. The initial value of the learning rate is determined through grid search and dynamically adjusted during training based on changes in the loss function. When the rate of decrease in the loss function slows down, a learning rate decay mechanism is triggered, achieving fine-tuning of parameters.

[0094] Batch normalization is employed during training to standardize the input of each layer, accelerating training convergence. Simultaneously, a dropout mechanism is introduced to prevent overfitting, with the dropout rate dynamically adjusted based on the network layer depth. Furthermore, an early stopping strategy is used, automatically terminating the training process if performance metrics on the validation set fail to improve for several consecutive epochs.

[0095] Through the above steps, intelligent optimization of model parameters was achieved, improving the accuracy of the digital twin model. The optimized model can more accurately reflect the actual operating status of the hydropower station, providing a reliable foundation for subsequent simulation verification.

[0096] S3 uses an optimized digital twin model for multi-condition simulation verification:

[0097] This application adopts a systematic simulation verification method, and comprehensively evaluates the feasibility and effectiveness of the modification design scheme through simulation analysis of multiple working conditions and multiple scenarios.

[0098] When designing typical operating condition test schemes, the key operating states of the hydropower station are considered. First, normal start-up and shutdown conditions, including pumping start-up, power generation start-up, and normal shutdown, are examined, with a focus on analyzing the dynamic response characteristics of the equipment. Second, accident conditions, such as load shedding, emergency shutdown, and grid failures, are tested to verify the system's protection functions and safety performance. Third, phase-shifting operation conditions are tested to analyze the stability and regulation characteristics of the unit at different speeds. In addition, special operating conditions, such as sudden changes in reservoir water level and extreme weather conditions, must also be considered.

[0099] In the simulation process, a distributed parallel computing architecture is employed to improve computational efficiency. Complex simulation tasks are decomposed into multiple sub-tasks, which are then executed in parallel on the computing cluster. For example, for the flow field calculation of a water turbine, a region decomposition method is used, dividing the computational domain into multiple sub-regions, with each computing node responsible for the calculation of one sub-region. Data exchange and synchronization between nodes are achieved through a message passing interface (MPI).

[0100] The simulation results are stored using a distributed database architecture, supporting efficient access to massive amounts of data. The database design includes time-series tables, spatial tables, and relational tables, enabling unified management of multi-dimensional data. A data indexing mechanism is established to improve query efficiency. Simultaneously, data compression storage is implemented to reduce storage space usage.

[0101] The simulation verification report generation adopts a combination of template-based and intelligent methods. Standardized report templates are established, including performance indicator analysis, dynamic characteristic evaluation, and safety performance verification. Through data mining and feature extraction, abnormal changes in key performance indicators are automatically identified, generating analytical conclusions and optimization suggestions.

[0102] This application employs a systematic simulation verification method, comprehensively evaluating the feasibility and effectiveness of the retrofit design scheme through simulation analysis under multiple operating conditions and scenarios. An example of a pumped storage power station unit retrofit project is used for illustration:

[0103] In the verification of normal start-up and shutdown conditions, the pumping start-up process was analyzed first. The simulation focused on the following parameters: the dynamic response curve of the unit speed increasing from 0 rpm to 500 rpm as the guide vane opening gradually increases from 0% to 85%; changes in turbine inlet and outlet pressures, especially pressure fluctuations that may occur during the initial start-up phase; bearing vibration and temperature changes, etc. For example, the simulation revealed that when the speed reached 300 rpm, the bearing vibration exceeded 0.12 mm due to the excessively rapid increase in guide vane opening, necessitating optimization of the guide vane opening adjustment curve.

[0104] For verification of accident operating conditions, an emergency shutdown scenario was used as an example. The simulation involved a sudden fault requiring an emergency shutdown of the unit while it was operating at rated load (300MW). Simulation analysis of the water hammer pressure during the rapid closing of the guide vanes (closing time 3 seconds) revealed that the maximum pressure rise in the pressure pipeline reached 1.8MPa, exceeding the design limit of 1.5MPa. Based on this, a pressure relief channel was added to the modification plan during the guide vane closing process, controlling the maximum pressure rise to within 1.3MPa.

[0105] For phase-shifting operation, the focus was on verifying the unit's operating characteristics at both low speed (approximately 350 rpm) and high speed (approximately 550 rpm). Simulation analysis revealed that during high-speed phase-shifting operation, the bearing temperature rose rapidly, from the normal operating temperature of 45℃ to 65℃. Further analysis indicated insufficient flow in the bearing cooling system; therefore, an automatic adjustment function for the bearing cooling water flow was added to the modification plan.

[0106] In the special operating condition verification, a sudden drop in the reservoir water level was simulated. For example, the dynamic response characteristics of the unit were analyzed when the water level dropped from the normal high water level of 732m to 722m within 10 minutes. The simulation results showed that the water level change caused vortices at the inlet, resulting in flow fluctuations and unit vibration. To address this issue, an inlet flow guide device was added to the modification plan.

[0107] When performing parallel simulations of the above operating conditions using a distributed computing approach, the computational tasks are distributed across 64 computing nodes. For example, for the turbine flow field calculation, the flow channel region is divided into 2 million grid cells, with each computing node responsible for the computation of approximately 30,000 grid cells. Data exchange between nodes is achieved through MPI, improving computational efficiency by approximately 15 times.

[0108] After the simulation results are stored in a distributed database, a time-series-based indexing mechanism is established. For example, for rotational speed data, a data block is created every 5 minutes, with each block containing 300 sampling points. This storage structure improves query efficiency by approximately 8 times and achieves a data compression rate of 70%.

[0109] When generating the simulation verification report, the system automatically extracts key performance indicators. For example, for bearing vibration data, it calculates the root mean square value, peak value of the spectrum, and envelope characteristics, identifying the second harmonic vibration anomaly present during startup. Based on these analysis results, the report automatically generates optimization suggestions for bearing modification, including specific measures such as adjusting bearing clearance and optimizing lubrication circuits.

[0110] The simulation verification process described above not only validated the feasibility of the modification plan but also identified potential problems through multi-condition analysis, providing crucial information for optimizing the plan. Distributed computing and intelligent analysis significantly improved verification efficiency and ensured the reliability of the verification results.

[0111] S4 uses machine learning to intelligently analyze simulation results:

[0112] This application uses geometric learning theory to construct a parameter space analysis framework, and uses the Regge metric tensor to describe the geometric characteristics of parameter changes, thereby enabling intelligent identification and assessment of potential risks.

[0113] like Figure 4 In the parameter mapping process, a high-dimensional parameter space is first constructed. Key parameters of the modification scheme, such as turbine guide vane opening, generator excitation current, and governor PID parameters, are used as basis vectors in this space. Principal component analysis (PCA) is used to reduce the dimensionality, preserving the main characteristic directions. In the reduced parameter space, each point represents a specific set of parameter configurations.

[0114] Based on the correlation between parameters, the Delaunay triangulation algorithm is used to construct the Regge triangulation. For each triangular element, a Regge metric tensor reflecting local geometric features is defined. This metric tensor contains directional and amplitude information about parameter variations, capable of describing the local curvature characteristics of the parameter space. The discrete curvature distribution in the parameter space is calculated by solving the Einstein field equations.

[0115] During parameter optimization, a geodesic search algorithm is employed to find the optimal path. This algorithm, based on the geometry of the parameter space, avoids high-risk areas and finds a stable parameter adjustment path. For example, when adjusting the guide vane opening of a water turbine, the algorithm analyzes the curvature distribution of the parameter space to avoid dangerous parameter areas that could lead to water hammer effects.

[0116] Specifically, in this application embodiment, in order to better understand the geometric characteristics of the parameter space and its application in hydropower station parameter optimization, the specific implementation process of Regge triangulation and geodesic search is described in detail below:

[0117] First, let's take the optimization of turbine guide vane opening as an example. The key parameters involved include guide vane opening θ (0-100%), turbine speed n (0-500 rpm), and pipeline pressure p (0-2 MPa). These three parameters constitute a three-dimensional parameter space. In practical applications, the parameter space is discretized into a finite number of sampling points, each representing a specific set of parameter configurations. For example, a point is taken every 5% for guide vane opening, every 20 rpm for speed, and every 0.1 MPa for pressure.

[0118] The Delaunay triangulation algorithm is used to partition the parameter space. The core of this algorithm is to ensure that the circumcircle of any triangle does not contain any other sampling points. The specific steps are as follows:

[0119] 1. First, construct an initial triangle that covers all sampling points;

[0120] 2. Insert sampling points one by one to find the triangle containing each point;

[0121] 3. Divide the triangle into three new triangles;

[0122] 4. Check if the newly generated triangle satisfies the Delaunay criterion; if not, perform an edge swap operation.

[0123] 5. Repeat steps 2-4 until all sampling points have been processed.

[0124] Define a Regge metric tensor G on each triangular unit, with the expression:

[0125] G = [g11 g12 g13]

[0126] [g21 g22 g23]

[0127] [g31 g32 g33]

[0128] Here, gij represents the metric relationship between parameters. For example, g11 represents the degree of influence of guide vane opening change on system response, and g12 represents the coupling relationship between guide vane opening and rotational speed. Specifically, by analyzing historical operating data and simulation results, the impact of parameter changes on system performance is statistically analyzed, and the components of the metric tensor are constructed.

[0129] Based on the Regge metric tensor, discrete curvature in the parameter space can be calculated. The formula for calculating the discrete curvature R on each triangular element is:

[0130] R = det(G) / A

[0131] Where det(G) is the determinant of the metric tensor, and A is the area of ​​the triangle. A larger curvature value indicates that parameter changes in that region have a significant impact on system performance. For example, during the rapid decrease of the guide vane opening from 80% to 20%, the curvature value in this region is significantly higher than in other regions due to the water hammer effect.

[0132] During parameter optimization, a geodesic search algorithm is used to find the optimal path. A geodesic is the shortest path between two points in the parameter space.

[0133] The specific implementation steps of the algorithm are as follows:

[0134] 1. Determine the starting point (current parameter configuration) and the ending point (target parameter configuration);

[0135] 2. Construct a path search grid in the parameter space;

[0136] 3. For each possible path in the grid, calculate its geodesic length;

[0137] 4. Simultaneously consider the maximum curvature value on the path to avoid high-risk areas;

[0138] 5. Select the path with the shorter geodesic length and the maximum curvature value within a safe range as the optimal path.

[0139] Taking guide vane opening adjustment as an example, suppose we need to adjust from the current opening of 85% to the target opening of 30%. Traditional methods might directly and linearly decrease the opening, but this could lead to severe water hammer effects. Using a geodesic search algorithm, the optimal path is:

[0140] 1. First, slowly reduce to 70%, at which point the curvature value is small and the system response is stable;

[0141] 2. Use a smaller adjustment step size in the 60%-40% region because the curvature value is larger in this region;

[0142] 3. Finally, adjust smoothly to 30%.

[0143] This method ensures the smoothness of parameter adjustments while effectively avoiding risks such as water hammer. Practice has shown that using this method reduces the maximum pressure fluctuation during adjustment by approximately 40%, significantly improving the safety of the adjustment process.

[0144] For the analysis of parameter-sensitive regions, the focus is on areas with large discrete curvature. The principal curvature values ​​and directions in these regions are calculated to identify the directions in which parameter changes have the greatest impact on system performance. For example, when analyzing governor parameters, it can be found that the integral time constant in the PID parameters has a significant impact on system stability.

[0145] Furthermore, this application employs a genetic algorithm for parameter optimization. Multiple performance indicators are considered when designing the fitness function, such as conditioning quality, energy conversion efficiency, and equipment lifespan. Parameter combinations are continuously optimized through crossover and mutation operations. During the optimization process, a state transition model describing the dynamic changes in parameters is established to evaluate the transient characteristics during parameter adjustment.

[0146] Specifically, in the embodiments of this application, parameter-sensitive region analysis and genetic algorithm optimization are two complementary and important steps. The following detailed explanation uses speed controller PID parameter optimization as a specific example:

[0147] The parameter-sensitive region analysis process is as follows:

[0148] First, a PID parameter space is constructed, including the proportional gain Kp (0.5–5), the integral time constant Ti (0.1–2 s), and the derivative time constant Td (0–0.5 s). Within this three-dimensional parameter space, local geometric features have been described using Regge triangulation. For each parameter-sensitive region, the principal curvature values ​​κ1, κ2, and κ3, and their corresponding principal curvature directions v1, v2, and v3 are calculated.

[0149] Taking a typical working condition as an example, in the region near Ti = 0.2s, the principal curvature values ​​are as follows:

[0150] κ1 = 2.35 (Ti direction)

[0151] κ2 = 0.82 (Kp direction)

[0152] κ3 = 0.31 (Td direction)

[0153] This indicates that in this region, the variation of the integral time constant Ti has the greatest impact on system stability.

[0154] Specific analysis results show:

[0155] 1. When Ti decreases from 0.2s to 0.15s, the system overshoot increases rapidly from 15% to 35%;

[0156] 2. When Ti increases from 0.2s to 0.25s, the system settling time increases from 2.5s to 4.8s;

[0157] 3. Changes in Kp and Td of the same magnitude have a relatively small impact on system performance.

[0158] Genetic algorithm optimization design is more comprehensive and systematic. First, the chromosome encoding scheme is designed:

[0159] Chromosome structure = [Kp_bit, Ti_bit, Td_bit]

[0160] in:

[0161] Kp_bit: 10-bit binary, mapping range 0.5-5

[0162] Ti_bit: 8-bit binary, mapping range 0.1-2s

[0163] Td_bit: 6-bit binary, mapping range 0-0.5s

[0164] The design of the fitness function considers multiple performance metrics:

[0165] F=w1*(1-Os / Os_max)+w2*(1-Ts / Ts_max)+w3*Ef+w4*L

[0166] in:

[0167] Os: Overshoot

[0168] Ts: Adjustment time

[0169] Ef: Energy conversion efficiency

[0170] L: Factor affecting expected lifespan of equipment

[0171] w1-w4: Weighting coefficients

[0172] Specific evaluation index calculation methods:

[0173] 1. The overshoot Os is obtained by simulating the system's step response;

[0174] 2. The settling time Ts is defined as the time it takes for the system response to enter the ±2% band;

[0175] 3. Energy conversion efficiency Ef considers the combined effects of turbine efficiency and generator efficiency;

[0176] 4. The equipment lifespan impact factor L is based on the assessment of equipment stress and vibration levels.

[0177] Specific parameter settings for the genetic algorithm:

[0178] Population size: 100

[0179] Crossover probability: 0.85

[0180] Mutation probability: 0.01

[0181] Maximum number of algebras: 200

[0182] During the optimization process, a state transition model is established to describe the parameter change process:

[0183] x(k+1)=Ax(k)+Bu(k)

[0184] y(k)=Cx(k)+Du(k)

[0185] Among them, the state variable x includes speed deviation, power deviation, etc.; the input u is the change of PID parameters; and the output y is the system performance index.

[0186] Here is an example of a typical iteration in the optimization process:

[0187] 1. Initial PID parameters: Kp = 2.0, Ti = 0.5s, Td = 0.1s

[0188] 2. Calculate fitness: F = 0.72

[0189] 3. After cross-operation, a new parameter combination is obtained: Kp = 2.2, Ti = 0.3s, Td = 0.1s

[0190] 4. State transition model prediction:

[0191] Speed ​​fluctuation reduced by 12%

[0192] Power fluctuations decreased by 8%

[0193] Adjustment time reduced by 15%

[0194] 5. Recalculate fitness: F = 0.85

[0195] 6. Due to improved adaptability, the new parameter combinations are retained.

[0196] After 200 iterations of optimization, the optimal PID parameters were finally obtained as follows:

[0197] Kp=2.5, Ti=0.28s, Td=0.08s

[0198] Performance metrics for this set of parameters:

[0199] Overshoot: 12.5% ​​(meets the requirement of <15%)

[0200] Adjustment time: 2.2s (meets the requirement of <3s)

[0201] Energy conversion efficiency: 92.8% (an increase of 1.2 percentage points)

[0202] Equipment vibration level: reduced by 25%

[0203] By combining parameter sensitivity analysis and genetic algorithm optimization, the accuracy of the optimization direction is ensured while simultaneously achieving the search for the global optimum. The resulting PID parameters not only meet dynamic performance requirements but also improve equipment lifespan and operating efficiency. In practical applications, this method significantly improves the speed governor's regulation quality and substantially enhances the overall system performance.

[0204] Finally, by calculating the model reconstruction error, the impact of different decomposition orders on the optimization results is analyzed to verify the reliability of parameter optimization. An evaluation index system is established, including stability, economic, and safety indicators, to comprehensively evaluate the effectiveness of the optimization results.

[0205] Through the above intelligent analysis process, potential risks in the renovation plan can be identified in a timely manner, and targeted optimization suggestions can be given, providing an important guarantee for the safe and stable operation of the hydropower station.

[0206] This application also provides an intelligent pumped storage hydropower station retrofit design simulation verification system, comprising: a model building module for establishing a digital twin model of the pumped storage hydropower station, the digital twin model including a hydraulic structure model, an electromechanical equipment model, and a control system model; a parameter optimization module for importing the retrofit design scheme into the digital twin model and optimizing the model parameters based on deep learning algorithms and historical operating data; a simulation calculation module for performing multi-condition simulation verification through the optimized digital twin model to obtain simulation results, the simulation results including: the dynamic response characteristics of the hydraulic structures, electromechanical equipment, and control system, and the impact of the retrofit design scheme on the overall performance of the hydropower station; and an intelligent evaluation module for intelligently analyzing the simulation results based on machine learning, identifying potential risks, and proposing optimization suggestions.

[0207] This application also provides a computer device, the computer device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described method for simulation verification of intelligent pumped storage hydropower station retrofit design.

[0208] This application also provides a computer-readable storage medium that stores computer instructions for causing a computer to execute the above-described method for simulation verification of intelligent pumped storage hydropower station retrofit design.

[0209] This application also provides a computer program product, including computer instructions, characterized in that, when the computer instructions are executed by a processor, they implement the steps of the above-described method for simulation verification of intelligent pumped storage hydropower station retrofit design.

[0210] The above description is merely a preferred embodiment of this application and does not limit the patent scope of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the patent protection scope of this application. It should be understood that although the flowcharts of the embodiments of this application indicate various operation steps with arrows, the execution order of these steps is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of the embodiments of this application, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where the execution times are different, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and the embodiments of this application do not limit this.

[0211] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application without departing from the technical concept of this application also fall within the protection scope of the embodiments of this application.

Claims

1. A simulation verification method for the retrofit design of an intelligent pumped storage hydropower station, characterized in that, include: A digital twin model of the pumped storage hydropower station is established, the digital twin model including a hydraulic structure model, an electromechanical equipment model, and a control system model; The transformation design scheme is imported into the digital twin model, and the model parameters are optimized based on deep learning algorithms and historical operating data; Multi-condition simulation verification was performed using the optimized digital twin model to obtain simulation results, which include the dynamic response characteristics of hydraulic structures, electromechanical equipment and control systems, as well as the impact of the modification design scheme on the overall performance of the hydropower station. Intelligent analysis of simulation results is performed based on machine learning to identify potential risks and propose optimization suggestions; Establishing a digital twin model of the pumped storage hydropower station includes: Collect actual operating data of the hydropower station, including geometric parameters of hydraulic structures, unit operating parameters, and control system parameters; Establish a model of the hydraulic structure, including the upper and lower reservoirs, pressure pipelines, and tailrace system; Establish a model of the aforementioned electromechanical equipment, including the water turbine, generator, and main transformer; Establish the control system model, including the speed governor, excitation system, and monitoring system; The modification design scheme is imported into the digital twin model, and the model parameters are optimized based on deep learning algorithms and historical operating data, including: Historical operation data is acquired and preprocessed, and the historical operation data is organized according to the acquisition time, spatial location, parameter type and operation status to form a high-dimensional tensor containing time dimension, spatial dimension, parameter dimension and state dimension; A convolutional tensor decomposition method is used to decompose high-dimensional tensors, which includes CP decomposition and Tucker decomposition of convolutional kernel tensors. Based on the decomposed high-dimensional tensor, a multi-layer convolutional network structure for parameter optimization is constructed. The network structure includes: an input layer for receiving input data, a feature extraction layer for extracting features, a feature fusion layer for fusing multi-dimensional features, and an output layer for outputting optimized parameters. The multi-layer convolutional network structure is trained using an adaptive learning rate strategy, and the training effect is improved by batch normalization to obtain optimized model parameters. Intelligent analysis of simulation results based on machine learning identifies potential risks and proposes optimization suggestions, including: Map the key parameters of the renovation plan to a high-dimensional parameter space; Based on the correlation between parameters, a Regge triangulation of the parameter space is constructed to obtain multiple triangular elements. Define a Regge metric tensor describing the local geometric features on each triangular unit; The discrete curvature of the simulation results in the parameter space is calculated based on the Regge metric tensor. Based on the distribution characteristics of the discrete curvature, the optimal path for parameter adjustment is determined, and specific parameter optimization suggestions are generated based on the optimal path.

2. The method according to claim 1, characterized in that, The convolutional tensor decomposition method is used to decompose high-dimensional tensors, and also includes: Use tensor ring integral solutions to handle cyclic features in the data; Introduce tensor regularization constraints; Apply residual connections to the network structure; Adjust the tensor decomposition parameters; Based on the model performance evaluation results, the optimal tensor decomposition parameters are determined.

3. The method according to claim 1, characterized in that, Multi-condition simulation verification was performed using the optimized digital twin model, including: Based on the aforementioned digital twin model, a test scheme for typical operating conditions, including normal start-stop, accident conditions, and phase-shifting operation, was designed. Using the aforementioned digital twin model, a distributed computing approach is employed to perform parallel simulations under multiple operating conditions, analyzing the dynamic response characteristics of hydraulic structures, electromechanical equipment, and control systems under different operating conditions. The simulation results of the digital twin model are stored in the simulation database; Based on the simulation results of the digital twin model, a simulation verification report containing key performance indicators is generated.

4. The method according to claim 1, characterized in that, Intelligent analysis of simulation results based on machine learning, identifying potential risks and proposing optimization suggestions, also includes: The region with a large discrete curvature in the parameter space is analyzed and defined as the parameter-sensitive region. Calculate the Regge metric features of the parameter-sensitive region, including principal curvature values ​​and principal curvature directions; Based on the Regge metric features, high curvature regions with discrete curvature anomalies are identified as potential risk points for hydropower stations. Construct an optimization objective function that includes constraints on parameter-sensitive regions; The geodesic search algorithm is used to optimize parameters based on the objective function to obtain optimized parameters that avoid high-risk areas. The stability and reliability of the optimization results are verified by evaluating the changes in the Regge metric before and after optimization.

5. The method according to claim 1, characterized in that, Intelligent analysis of simulation results based on machine learning, identifying potential risks and proposing optimization suggestions, also includes: Genetic algorithms were used to preliminarily optimize the key parameters of the modification scheme; Based on the preliminary optimization results, a transformation process model describing the dynamic changes of parameters is established; Transient state analysis was conducted based on the aforementioned modification process model to assess the impact of parameter changes on the hydropower station. The reliability of the parameter optimization results is evaluated by calculating the reconstruction error and analyzing the impact of different decomposition orders.

6. The method according to claim 1, characterized in that, The collected actual operating data of the hydropower station includes: Set up data collection points, including key monitoring points for hydraulic structures, monitoring points for unit operating parameters, and monitoring points for the control system; Establish a data acquisition system to collect power plant operation data in real time; The collected data is preprocessed, including data cleaning, outlier handling, and data standardization.

7. A simulation verification system for the retrofit design of an intelligent pumped storage hydropower station, characterized in that, include: The model building module is used to establish a digital twin model of the pumped storage hydropower station, which includes a hydraulic structure model, an electromechanical equipment model, and a control system model. The parameter optimization module is used to import the transformation design scheme into the digital twin model and optimize the model parameters based on deep learning algorithms and historical operating data. The simulation calculation module is used to perform multi-condition simulation verification through the optimized digital twin model to obtain simulation results. The simulation results include: the dynamic response characteristics of hydraulic structures, electromechanical equipment and control systems, as well as the impact of the modification design scheme on the overall performance of the hydropower station. The intelligent evaluation module is used to perform intelligent analysis of simulation results based on machine learning, identify potential risks, and propose optimization suggestions. Establishing a digital twin model of the pumped storage hydropower station includes: Collect actual operating data of the hydropower station, including geometric parameters of hydraulic structures, unit operating parameters, and control system parameters; Establish a model of the hydraulic structure, including the upper and lower reservoirs, pressure pipelines, and tailrace system; Establish a model of the aforementioned electromechanical equipment, including the water turbine, generator, and main transformer; Establish the control system model, including the speed governor, excitation system, and monitoring system; The modification design scheme is imported into the digital twin model, and the model parameters are optimized based on deep learning algorithms and historical operating data, including: Historical operation data is acquired and preprocessed, and the historical operation data is organized according to the acquisition time, spatial location, parameter type and operation status to form a high-dimensional tensor containing time dimension, spatial dimension, parameter dimension and state dimension; A convolutional tensor decomposition method is used to decompose high-dimensional tensors, which includes CP decomposition and Tucker decomposition of convolutional kernel tensors. Based on the decomposed high-dimensional tensor, a multi-layer convolutional network structure for parameter optimization is constructed. The network structure includes: an input layer for receiving input data, a feature extraction layer for extracting features, a feature fusion layer for fusing multi-dimensional features, and an output layer for outputting optimized parameters. The multi-layer convolutional network structure is trained using an adaptive learning rate strategy, and the training effect is improved by batch normalization to obtain optimized model parameters. Intelligent analysis of simulation results based on machine learning identifies potential risks and proposes optimization suggestions, including: Map the key parameters of the renovation plan to a high-dimensional parameter space; Based on the correlation between parameters, a Regge triangulation of the parameter space is constructed to obtain multiple triangular elements. Define a Regge metric tensor describing the local geometric features on each triangular unit; The discrete curvature of the simulation results in the parameter space is calculated based on the Regge metric tensor. Based on the distribution characteristics of the discrete curvature, the optimal path for parameter adjustment is determined, and specific parameter optimization suggestions are generated based on the optimal path.

Images