Hydropower station generator rotor hoisting intelligent positioning system based on multi-modal data
Through multimodal data fusion and intelligent decision-making architecture, the problems of single data, rigid decision-making and inefficient interaction of the traditional hydropower station generator rotor lifting system are solved, and high-precision and safe rotor lifting are achieved, lifting efficiency and equipment safety are improved, and immersive human-computer interaction is provided.
Patent Information
- Application Number
- CN202510600735.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-12
- Publication Date
- 2025-07-18
AI Technical Summary
The rotor lifting system of traditional hydropower station generators relies on manual experience and single modal data, and has problems such as insufficient data perception capabilities, rigid decision-making, lack of coordination, and inefficient interaction, making it difficult to achieve precise positioning and safe and efficient lifting.
The architecture of multimodal data acquisition layer, multimodal data fusion layer, intelligent decision-making layer, dynamic execution layer and multimodal interaction layer is adopted, and binocular vision sensors, six-axis inertial measurement units, distributed fiber strain sensors, etc. is integrated, and data fusion and dynamic decision-making are carried out through deep learning and group intelligent algorithms, combined with augmented reality display and voice interaction, and full-dimensional data perception and real-time collaborative control are realized.
It realizes high-precision and safe generator rotor lifting, reduces manual intervention, improves lifting efficiency and equipment safety, reduces energy consumption, and provides immersive human-computer interaction and hierarchical safety protection.
Smart Images
Figure CN120328387A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of hydropower stations, and particularly to an intelligent positioning system for hoisting a generator rotor of a hydropower station based on multimodal data. Background Art
[0002] A hydropower station is a comprehensive engineering facility that converts water energy into electrical energy. Generally, it includes a reservoir formed by water retaining and water discharging buildings, a water diversion system of the hydropower station, a power generation plant, and electromechanical equipment, etc. The high-level water in the reservoir flows into the plant through the water diversion system to drive the water turbine to rotate, converting water energy into mechanical energy. The water turbine drives the generator to rotate, converting mechanical energy into electrical energy. Positioning during the hoisting of the generator rotor of a hydropower station is because precise positioning can ensure a uniform air gap between the rotor and the stator, which is crucial for the stable operation of the generator. If the air gap is uneven, it will cause uneven magnetic field distribution, resulting in increased vibration and noise of the generator, reducing power generation efficiency and equipment service life; accurate positioning can make the center of the rotor coincide with the rotation center of the unit, ensuring dynamic balance during high-speed rotation of the unit and avoiding damage to components such as bearings caused by additional stress due to imbalance; correct positioning can make the components of the generator connect tightly and cooperate precisely, ensuring good electrical performance of the generator, reducing power consumption, improving power generation quality, making parameters such as output voltage and frequency meet the requirements of the power grid, and also avoiding difficulties in later maintenance and potential safety hazards caused by improper installation.
[0003] Traditional hydropower station generator rotor hoisting systems generally rely on manual experience and single-modal data drive, with significant limitations, specifically manifested as follows: insufficient data perception ability, mostly using single sensors, lacking collaborative acquisition and fusion of multimodal data, visual data and mechanical data being separated, unable to identify stress concentration areas of cables in real time, not fully utilizing the dynamic correlation between environmental parameters and equipment status, and prone to sudden risks; rigid decision-making models, existing path planning algorithms rely on static environment assumptions, difficult to cope with dynamic obstacle interference during hoisting, control strategies mostly use fixed PID parameters, unable to adaptively adjust speed and posture, and frequent manual intervention is required when the rotor swing amplitude is too large; lack of multi-objective coordination, often taking a single objective as the optimization direction, failing to establish a quantitative game relationship among efficiency, safety, and energy consumption. For example, forcibly increasing the speed will exacerbate equipment vibration, and overly conservative safety strategies will lead to a sharp increase in energy consumption; defects in real-time performance and reliability, data fusion delay generally exceeds 200 ms, dynamic path replanning response lags, difficult to meet the hoisting requirements with millimeter-level accuracy, the hardware architecture lacks redundant design, and a single-point failure of a sensor may cause a chain shutdown; inefficient human-computer interaction, the operation interface mostly relies on two-dimensional plane charts, lacking spatial situation awareness ability, and prone to misjudgment due to information overload during manual intervention, and the risk warning mechanism only provides simple threshold alarms, without implementing a hierarchical response strategy. Summary of the Invention
[0004] Aiming at the deficiencies of the existing technology, the present invention provides an intelligent positioning system for the hoisting of a hydroelectric generator rotor based on multi-modal data, which solves the problems of single data, rigid decision-making, lack of collaboration, and low-efficiency interaction existing in the existing technology.
[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: An intelligent positioning system for the hoisting of a hydroelectric generator rotor based on multi-modal data, comprising a multi-modal data acquisition layer, the multi-modal data acquisition layer is connected to a multi-modal data fusion layer, the multi-modal data fusion layer is connected to an intelligent decision-making layer, the intelligent decision-making layer is connected to a dynamic execution layer, and the dynamic execution layer is connected to a multi-modal interaction layer;
[0006] The multi-modal data acquisition layer is used to collect visual data, mechanical sensor data, environmental parameter data, and equipment status data of the hoisting environment in real time. The multi-modal data fusion layer fuses heterogeneous data through a deep learning model and the Kalman filtering algorithm to generate dynamic characteristics of the hoisting process with high confidence. The intelligent decision-making layer generates optimal hoisting path, speed, and attitude control instructions based on a swarm intelligence algorithm and a multi-objective game model. The dynamic execution layer adjusts the motion parameters of the hoisting equipment in real time according to the decision-making instructions and integrates a redundant safety control mechanism. The multi-modal interaction layer provides a three-dimensional visualization interface, a multi-source data cockpit, and a voice interaction interface.
[0007] Preferably, the multi-modal data acquisition layer includes a binocular vision sensor array, a six-axis inertial measurement unit, a distributed fiber optic strain sensor, and a multi-spectral environmental sensor.
[0008] The binocular vision sensor array is used for the stereoscopic depth map of the component hoisting area. The six-axis inertial measurement unit monitors the angular velocity and acceleration of the rotor in real time. The distributed fiber optic strain sensor measures the stress distribution of the cable. The multi-spectral environmental sensor collects wind speed, temperature, and humidity parameters.
[0009] Preferably, the multi-modal data fusion layer includes a multi-modal feature alignment module based on Transformer, a dynamic weight allocation mechanism, and an abnormal data detection module.
[0010] The multi-modal feature alignment module based on Transformer extracts the spatio-temporal correlation features of visual data and mechanical data. The dynamic weight allocation mechanism adjusts the fusion weight according to the sensor confidence. The abnormal data detection module uses the isolation forest algorithm to identify and eliminate noise data.
[0011] Preferably, the intelligent decision-making layer includes an improved ant colony optimization algorithm, a multi-agent game model, and an online reinforcement learning module;
[0012] The improved ant colony optimization algorithm introduces the cable stress distribution characteristics as the pheromone update weight through a dynamic pheromone update mechanism. The higher the pheromone concentration of the path with smaller stress. A curvature penalty term is added to the path selection objective function using the spatial curvature constraint term to ensure path smoothness. Through the combination of local search optimization and the simulated annealing algorithm, random perturbations are introduced during the ant colony iteration process to avoid falling into local optima.
[0013] The multi-agent game model takes equipment energy consumption, path efficiency, and safety margin as game participants. The online reinforcement learning module dynamically adjusts the game strategy weights through Q-learning.
[0014] Preferably, the safety margin game participant is modeled using the potential field method, with the dynamic balance between the obstacle repulsive force and the target attractive force. The reinforcement learning reward function includes a negative feedback term of the equipment vibration amplitude.
[0015] Preferably, the dynamic execution layer includes a double-loop PID controller, a virtual commissioning module based on digital twin, and a safety interlock mechanism.
[0016] The inner loop of the double-loop PID controller controls the motor torque, and the outer loop adjusts the lifting speed. The virtual commissioning module based on digital twin pre-acts the lifting process and verifies the feasibility of the control instructions. The safety interlock mechanism automatically switches to the safety mode when the cable stress exceeds the threshold.
[0017] Preferably, the multi-modal interaction layer includes an augmented reality display module, a multi-dimensional data dashboard, and a voice command recognition module.
[0018] The augmented reality display module superimposes virtual path planning lines and real-time stress nephograms. The multi-dimensional data dashboard synchronously displays the vibration spectrum, energy consumption curve, and environmental parameter heat map. The voice command recognition module supports the semantic parsing and execution of natural language commands.
[0019] The present invention provides an intelligent positioning system for the hoisting of a hydroelectric generator rotor based on multi-modal data, having the following beneficial effects:
[0020] The present invention provides an intelligent positioning system for hoisting rotors of hydropower station generators based on multimodal data. The technology of the present invention systematically solves traditional hoisting problems through innovative multimodal data fusion and intelligent decision-making architecture, realizes full-dimensional data perception, integrates multimodal sensors, realizes cross-modal feature alignment through the Transformer model, constructs a digital twin to identify potential risks in real time, makes dynamic intelligent decisions, generates a global path using an improved ant colony algorithm, quantifies conflicting targets and dynamically adjusts strategy weights using a multi-agent game model, implements real-time collaborative control, reduces data fusion and path planning response time based on a 5G-MEC edge cloud architecture, and adjusts rotor swing in combination with a dual closed-loop PID controller in advance without manual intervention. A hierarchical safety fault-tolerant mechanism is established, and multiple protections are provided by three-level safety interlocks and redundant hardware design. The cable stress exceeds the limit and can automatically switch to a safe mode, creating an immersive human-computer interaction. The AR interface allows people to intuitively perceive the spatial situation, and the multi-dimensional data dashboard assists in decision-making. The natural language interaction module can instantly parse and execute complex instructions, realizing the transition of hoisting operations from "experience-driven" to "data-intelligent-driven", providing a new generation of highly reliable and high-precision solutions for the precision hoisting of large equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 Schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION
[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0023] like Figure 1 As shown, an embodiment of the present invention provides a hydropower station generator rotor hoisting intelligent positioning system based on multimodal data, including a multimodal data acquisition layer, the multimodal data acquisition layer is connected to a multimodal data fusion layer, the multimodal data fusion layer is connected to an intelligent decision layer, the intelligent decision layer is connected to a dynamic execution layer, and the dynamic execution layer is connected to a multimodal interaction layer;
[0024] The multi-modal data acquisition layer is used to collect visual data, mechanical sensor data, environmental parameter data, and equipment status data of the hoisting environment in real time. The multi-modal data fusion layer fuses heterogeneous data through a deep learning model and the Kalman filtering algorithm to generate dynamic features of the hoisting process with high confidence. The intelligent decision-making layer generates optimal hoisting path, speed, and attitude control instructions based on swarm intelligence algorithms and multi-objective game models. The dynamic execution layer adjusts the motion parameters of the hoisting equipment in real time according to the decision-making instructions and integrates a redundant safety control mechanism. The multi-modal interaction layer provides a three-dimensional visualization interface, a multi-source data cockpit, and a voice interaction interface.
[0025] The multi-modal data acquisition layer includes a binocular vision sensor array, a six-axis inertial measurement unit, a distributed fiber optic strain sensor, and a multi-spectral environmental sensor.
[0026] The binocular vision sensor array is used to construct a stereo depth map of the hoisting area. The six-axis inertial measurement unit monitors the angular velocity and acceleration of the rotor in real time. The distributed fiber optic strain sensor measures the stress distribution of the cable. The multi-spectral environmental sensor collects wind speed, temperature, and humidity parameters.
[0027] The multi-modal data fusion layer includes a multi-modal feature alignment module based on Transformer, a dynamic weight allocation mechanism, and an abnormal data detection module.
[0028] The multi-modal feature alignment module based on Transformer extracts the spatio-temporal correlation features of visual data and mechanical data. The dynamic weight allocation mechanism adjusts the fusion weight according to the sensor confidence. The abnormal data detection module uses the Isolation Forest algorithm to identify and remove noise data.
[0029] The intelligent decision-making layer includes an improved ant colony optimization algorithm, a multi-agent game model, and an online reinforcement learning module;
[0030] The improved ant colony optimization algorithm introduces the cable stress distribution feature as the pheromone update weight through a dynamic pheromone update mechanism. The pheromone concentration of the path with smaller stress is higher. A curvature penalty term is added to the path selection objective function using the spatial curvature constraint term to ensure path smoothness. Through local search optimization combined with the simulated annealing algorithm, random perturbations are introduced during the ant colony iteration process to avoid falling into local optima.
[0031] The multi-agent game model takes equipment energy consumption, path efficiency, and safety margin as game participants. The online reinforcement learning module dynamically adjusts the game strategy weight through Q-learning.
[0032] The safety margin game participant is modeled using the potential field method, and the dynamic balance between the obstacle repulsive force and the target attractive force is achieved. The reinforcement learning reward function includes a negative feedback term of the equipment vibration amplitude.
[0033] The dynamic execution layer includes a double-loop PID controller, a virtual commissioning module based on digital twin, and a safety interlock mechanism.
[0034] The inner loop of the double-loop PID controller controls the motor torque, and the outer loop adjusts the lifting speed. The virtual commissioning module based on digital twin pre-acts the lifting process and verifies the feasibility of control instructions. When the cable stress exceeds the threshold, the safety interlock mechanism automatically switches to the safety mode.
[0035] The multi-modal interaction layer includes an augmented reality display module, a multi-dimensional data dashboard, and a voice command recognition module.
[0036] The augmented reality display module superimposes virtual path planning lines and real-time stress cloud maps. The multi-dimensional data dashboard synchronously displays vibration spectra, energy consumption curves, and environmental parameter heat maps. The voice command recognition module supports semantic parsing and execution of natural language commands.
[0037] The safety interlock mechanism includes a three-level safety response strategy, specifically:
[0038] First-level warning (stress threshold 70%): Reduce the lifting speed and issue an audible and visual alarm;
[0039] Second-level intervention (stress threshold 85%): Activate the auxiliary support device;
[0040] Third-level emergency stop (stress threshold 95%): Trigger the electromagnetic brake and release the emergency anchoring mechanism.
[0041] The hardware architecture includes:
[0042] The edge computing node cluster deploys NVIDIA Jetson Orin and FPGA acceleration cards, and a 5G-MEC edge cloud platform to achieve time-synchronized transmission of multi-modal data. The anti-electromagnetic interference sensor bus adopts a fiber optic communication protocol.
[0043] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. An intelligent positioning system for hoisting the generator rotor of a hydropower station based on multi-modal data, including a multi-modal data acquisition layer, characterized in that: The multi-modal data acquisition layer is connected to the multi-modal data fusion layer, the multi-modal data fusion layer is connected to the intelligent decision-making layer, the intelligent decision-making layer is connected to the dynamic execution layer, and the dynamic execution layer is connected to the multi-modal interaction layer; The multi-modal data acquisition layer is used to collect visual data, mechanical sensor data, environmental parameter data, and equipment status data of the hoisting environment in real time. The multi-modal data fusion layer fuses heterogeneous data through a deep learning model and the Kalman filtering algorithm to generate dynamic features of the hoisting process with high confidence. The intelligent decision-making layer generates optimal hoisting path, speed, and attitude control instructions based on the swarm intelligence algorithm and the multi-objective game model. The dynamic execution layer adjusts the motion parameters of the hoisting equipment in real time according to the decision-making instructions and integrates a redundant safety control mechanism. The multi-modal interaction layer provides a three-dimensional visualization interface, a multi-source data cockpit, and a voice interaction interface.
2. The intelligent positioning system for hoisting the generator rotor of a hydropower station based on multimodal data according to claim 1, characterized in that: The multi-modal data acquisition layer includes a binocular vision sensor array, a six-axis inertial measurement unit, a distributed fiber optic strain sensor, and a multi-spectral environmental sensor.
3. The intelligent positioning system for hoisting the generator rotor of a hydropower station based on multimodal data according to claim 1, characterized in that: The multi-modal data fusion layer includes a multi-modal feature alignment module based on Transformer, a dynamic weight allocation mechanism, and an abnormal data detection module.
4. The intelligent positioning system for hoisting the generator rotor of a hydropower station based on multi-modal data according to claim 1, characterized in that: The intelligent decision-making layer includes an improved ant colony optimization algorithm, a multi-agent game model, and an online reinforcement learning module; The improved ant colony optimization algorithm introduces the cable stress distribution feature as the pheromone update weight through the dynamic pheromone update mechanism. The pheromone concentration of the path with smaller stress is higher. The curvature penalty term is added to the path selection objective function using the spatial curvature constraint term to ensure path smoothness. Through local search optimization combined with the simulated annealing algorithm, random perturbations are introduced during the ant colony iteration process to avoid falling into local optima.
5. The intelligent positioning system for hoisting the generator rotor of a hydropower station based on multimodal data according to claim 4, characterized in that: The safety margin game participant is modeled using the potential field method, and the dynamic balance between the obstacle repulsive force and the target attractive force is achieved. The reinforcement learning reward function includes a negative feedback term for the vibration amplitude of the equipment.
6. The intelligent positioning system for hoisting the generator rotor of a hydropower station based on multi-modal data according to claim 1, wherein: The dynamic execution layer includes a double closed-loop PID controller, a virtual commissioning module based on digital twin, and a safety interlock mechanism.
7. An intelligent positioning system for hoisting the generator rotor of a hydropower station based on multimodal data according to claim 1, characterized in that: The multi-modal interaction layer includes an augmented reality display module, a multi-dimensional data dashboard, and a voice command recognition module.