Method for realizing digital twinning of offshore photovoltaic power generation system

Abstract

The invention relates to the field of offshore power generation, in particular to a method for acquiring marine environment data and motion state data of an offshore photovoltaic power generation system in real time, preprocessing the acquired data and then completing spatial-temporal feature fusion, outputting a fusion feature matrix with a timestamp and a space coordinate label, and obtaining an offshore power generation result based on the fusion feature matrix. Multi-dimensional feature extraction, mechanical property evaluation and power generation efficiency correlation analysis are achieved, a core feature set is constructed, a reliability label is added, and a high-fidelity virtual mapping model of the offshore photovoltaic power generation system is constructed based on the fusion feature matrix and the core feature set. Real-time dynamic interaction, multi-physics field coupling simulation and full-life-cycle visual management of a physical entity and a virtual model are simulated through a high-fidelity virtual mapping model.

Description

Technical Field

[0001] This invention relates to the field of offshore power generation, and more specifically to a method for realizing a digital twin of an offshore photovoltaic power generation system. Background Technology

[0002] As the global energy transition accelerates, solar photovoltaic (PV) power, as a core component of low-carbon and clean energy, is gradually expanding into the ocean. Offshore PV systems, especially deep-sea floating PV and flexible-structure PV systems, are becoming an important direction for developing marine renewable energy due to their advantages such as not being limited by land resources, stable solar radiation, and high cooling efficiency. According to the International Energy Agency, global offshore PV capacity is projected to exceed 100 GW by 2030, with flexible-structure PV systems accounting for over 60%. However, the dynamic load complexity, strong corrosive characteristics, and wide spatial distribution of the marine environment pose severe challenges to the safe operation, efficiency stability, and life-cycle economics of the system, and it is urgent to achieve systematic optimization through digital technology. To this end, a method for realizing digital twins of marine photovoltaic power generation systems is proposed. Summary of the Invention

[0003] In order to solve the technical problems existing in the prior art, the present invention provides a method for realizing digital twins of offshore photovoltaic power generation systems.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for realizing a digital twin of an offshore photovoltaic power generation system, comprising the following steps: Step S1: Real-time acquisition of marine environmental data and motion status data of offshore photovoltaic power generation system; after preprocessing the acquired data, spatiotemporal feature fusion is completed, and a fusion feature matrix with timestamp and spatial coordinate label is output. Step S2: Based on the fused feature matrix, multi-dimensional feature extraction, mechanical performance evaluation and power generation efficiency correlation analysis are realized, a core feature set is constructed and a reliability label is attached; Step S3: Based on the fused feature matrix and core feature set, construct a high-fidelity virtual mapping model of the offshore photovoltaic power generation system. Simulate the real-time dynamic interaction between physical entities and virtual models, multi-physics coupling simulation, and full life cycle visualization management through the high-fidelity virtual mapping model.

[0005] Preferably, in step S1, the marine environmental data includes meteorological data, hydrological data, and geographic data; The meteorological data includes light intensity, ambient temperature, relative humidity, wind speed, and wind direction; The hydrological data includes wave parameters, ocean current velocity, seawater temperature, and salt spray concentration. The wave parameters include wave height and period. The geographic data includes latitude and longitude coordinates and tidal levels.

[0006] Preferably, in step S1, the motion state data of the offshore photovoltaic power generation system includes kinematic parameters, structural mechanical parameters, and electrical performance parameters; The kinematic parameters include the real-time position, attitude angle, and dynamic displacement of the photovoltaic array; The structural mechanical parameters include the deformation of the flexible photovoltaic panel, the stress of the support structure, and the tension of the anchoring system; The electrical performance parameters include component output power, open-circuit voltage, short-circuit current, component temperature, and surface cleanliness.

[0007] Preferably, in step S1, the data preprocessing includes: using wavelet transform to remove high-frequency noise caused by wave impact, and smoothing the attitude angle data of the inertial measurement unit (IMU) using Kalman filtering; Outliers in the data are detected based on the 3σ criterion and the isolated forest algorithm, and missing values ​​in the data are filled using spatiotemporal interpolation. Z-score standardization is applied to data of different dimensions, and sliding window framing is used for time series data. The output is structured data containing a quadruple of ID-timestamp-spatial coordinates-feature value.

[0008] Preferably, in step S1, all sensor data are unified to UTC timestamps through the NTP network time protocol, and delayed data is corrected by timestamp backtracking. Establish a sensor location-system spatial coordinate mapping table based on the physical topology of the photovoltaic panel array; A sliding window and an LSTM neural network are used to extract temporal feature vectors, and a graph neural network (GNN) learning component is used to extract spatial feature vectors based on spatial correlation weights. An attention mechanism is used to assign dynamic weights to the temporal and spatial feature vectors, and the resulting fused feature matrix is ​​output after feature concatenation and a fully connected layer.

[0009] Preferably, in step S2, the mechanical performance evaluation includes: based on the extracted strain characteristics and material mechanical parameters, calculating the equivalent stress using the fourth strength theory and comparing it with the allowable stress to generate a strength margin index; evaluating the overturning risk of the floating photovoltaic support using the Euler critical load formula; and calculating the pull-out force safety factor of the anchoring system using the limit state design method. The fatigue damage degree was calculated using the rainflow counting method and Miner's linear cumulative damage theory on the strain time history data. A remaining life RUL prediction model was constructed based on LSTM neural network to output the remaining life of 1-5 years. Weak areas with damage degree ≥0.6 were marked by the heat map overlay method. The flexible characteristic parameters are input into the SESAM finite element model to correct the material properties and boundary conditions, and the accuracy of the mechanical performance evaluation is verified by fluid-structure interaction simulation.

[0010] Preferably, in step S2, the power generation efficiency correlation analysis includes: using random forest regression to quantify the contribution of each feature to power generation efficiency, and establishing a nonlinear mapping relationship between environmental parameters, motion parameters, and efficiency loss through gradient boosting tree (GBDT). Based on the LSTM-GRU hybrid network, the power prediction curves are output in the 1-60 minute range, and the average daily power generation is output in the 1-7 days by combining weather forecast and spatiotemporal attention mechanism. An abnormal drop in efficiency was detected by using the isolated forest algorithm. The total efficiency loss was decomposed into environmental loss, motion loss and structural loss. The potential for efficiency optimization was evaluated based on the genetic algorithm.

[0011] Preferably, in step S3, the construction of the high-fidelity virtual mapping model includes: constructing a hierarchical geometric model from the component level to the system level based on CAD drawings and laser scanning point cloud data; Flexible characteristic parameters are embedded into the model, and the Mooney-Rivlin hyperelastic constitutive equation is used to describe the mechanical behavior of the flexible film. The anisotropy of the scaffold is simulated by a fiber-reinforced composite material model. Dynamic environmental field models of wave field, ocean current field, wind field, and salt spray corrosion field are constructed based on marine environmental data. The Bayesian optimization algorithm was used to correct the model parameters with the measured data as the objective function, so that the average error between the model prediction and the measured value was ≤3%. The applicability of the model was verified by generating 200+ working conditions through Monte Carlo sampling.

[0012] Preferably, in step S3, the real-time dynamic interaction includes: deploying an edge computing gateway to access real-time data streams and feature parameter streams via the OPCUA / MTConnect protocol, establishing a control command channel from the virtual model to the physical system via the MQTT protocol, and constructing a digital thread based on blockchain technology to record the entire lifecycle data interaction log; The Kalman filter-particle filter fusion algorithm is used to achieve pose synchronization between physical entities and virtual models, realizing millimeter-level displacement synchronization at the component level and centimeter-level position synchronization at the system level. When the physical system is in a high-risk state, the virtual model prioritizes updating the state of the abnormal area and activates audible and visual alarms through an event-triggered mechanism. The virtual model and the physical system run in parallel in shadow mode and optimize the model accuracy through incremental learning.

[0013] Preferably, in step S3, the multiphysics coupling simulation and full life cycle visualization management includes: simulating the transient mechanical response under extreme wave impact from 0 to 60 seconds and the performance degradation process from 5 to 20 years based on SESAM, and optimizing the photovoltaic panel array layout and support material parameters through the NSGA-III genetic algorithm.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. In this invention, a multi-physics coupled model combining environmental load, structural response, and power generation efficiency is constructed by spatiotemporal fusion of multi-source data and deep extraction of flexible features. This reduces the mechanical performance evaluation error to within 5% and improves the power generation efficiency prediction accuracy to over 95%. 2. This invention establishes a millisecond-level dynamic synchronization mechanism between physical entities and virtual models, supporting simulation and online optimization control under extreme working conditions; 3. This invention integrates data from the entire process of design, operation and maintenance and decommissioning, and achieves predictive maintenance and fault location accuracy of <1m through a digital twin platform, reducing operation and maintenance costs by more than 30% and improving the economic efficiency of the system throughout its entire life cycle by 25%. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments, which illustrate the above and other technical features and advantages of the present invention. However, the following embodiments are merely preferred embodiments of the present invention and are not exhaustive.

[0017] Example: like Figure 1 As shown, this invention proposes a method for realizing a digital twin of an offshore photovoltaic power generation system, which is used to perform a collaborative data acquisition step for offshore photovoltaic systems. This method collects two types of core data in real time: marine environmental data, which includes meteorological data such as light intensity, ambient temperature, relative humidity, wind speed and wind direction; hydrological data such as wave parameters, ocean current speed, seawater temperature and salt spray concentration; and geographical data such as latitude and longitude coordinates and tidal levels; and motion state data of the offshore photovoltaic power generation system, which includes kinematic parameters, structural mechanical parameters and electrical performance parameters.

[0018] The collected data is preprocessed by using wavelet transform to remove high-frequency noise caused by wave impact, smoothing IMU attitude angle data by Kalman filtering, detecting outliers based on the 3σ criterion and isolated forest algorithm and filling missing values ​​by spatiotemporal interpolation, and dividing the data of different dimensions into frames by sliding window after Z-score standardization, outputting structured data of "sensor ID-timestamp-spatial coordinates-feature value" quadruple; Subsequently, spatiotemporal feature fusion was carried out. All sensor data were unified to UTC timestamps through the NTP network time protocol and timestamp backtracking correction was performed on delayed data. A "sensor location-system spatial coordinate" mapping table was established based on the physical topology of the photovoltaic array. A sliding window and LSTM neural network were used to extract time feature vectors, and a graph neural network (GNN) was used to learn the spatial correlation weights between components. Then, an attention mechanism was used to assign dynamic weights to the spatiotemporal feature vectors. After feature splicing and a fully connected layer, a fused feature matrix with timestamps and spatial coordinate labels was output.

[0019] We conducted a feature analysis of offshore photovoltaic systems, extracting multi-dimensional features based on the aforementioned fused feature matrix. These features encompass core environmental, motion, mechanical, and electrical characteristics, and we then performed a mechanical performance assessment. Based on the extracted strain characteristics and material mechanical parameters, the equivalent stress is calculated using the fourth strength theory and compared with the allowable stress to generate a strength margin index. The overturning risk of the floating photovoltaic support is assessed by the Euler critical load formula, and the pull-out force safety factor of the anchoring system is calculated by the limit state design method. The fatigue damage degree was calculated using the rainflow counting method and Miner's linear cumulative damage theory on the strain time history data. A remaining life RUL prediction model was constructed based on LSTM neural network to output the remaining life of 1-5 years. Weak areas with damage degree ≥0.6 were marked by the heat map overlay method. Flexible characteristic parameters are input into the SESAM finite element model to correct material properties and boundary conditions, and the accuracy of mechanical performance evaluation is verified through fluid-structure interaction simulation. Simultaneously, a correlation analysis of power generation efficiency is conducted, using random forest regression to quantify the contribution of each feature to power generation efficiency, and establishing a nonlinear mapping relationship between environmental parameters, motion parameters, and efficiency loss through gradient boosting tree (GBDT). The system outputs 1-60 minute power prediction curves based on an LSTM-GRU hybrid network, and outputs the average daily power generation for 1-7 days by combining weather forecasts and a spatiotemporal attention mechanism. It monitors abnormal efficiency drops through the isolated forest algorithm, decomposes the total efficiency loss into environmental loss, motion loss and structural loss, evaluates the efficiency optimization potential based on a genetic algorithm, and finally constructs a core feature set and attaches a reliability label.

[0020] Implement the construction of a digital twin for marine photovoltaic systems, and build a high-fidelity virtual mapping model based on the fusion feature matrix and core feature set: build a hierarchical geometric model from the component level to the system level based on CAD drawings and laser scanning point cloud data, embed flexible feature parameters into the model and use the Mooney-Rivlin hyperelastic constitutive equation to describe the mechanical behavior of the flexible film, simulate the anisotropy of the support through a fiber-reinforced composite material model, and build a dynamic environmental field model of wave field, ocean current field, wind field and salt spray corrosion field based on marine environmental data; The Bayesian optimization algorithm is used to correct the model parameters with the measured data as the objective function. The applicability of the model is verified by generating 200 working conditions through Monte Carlo sampling, and the real-time dynamic interaction between the physical entity and the virtual model is realized. An edge computing gateway is deployed to access real-time data streams and feature parameter streams via the OPCUA / MTConnect protocol. A control command channel from the virtual model to the physical system is established via the MQTT protocol. A digital thread based on blockchain technology is constructed to record the entire lifecycle data interaction log. A Kalman filter-particle filter fusion algorithm is used to achieve pose synchronization. When the physical system is in a high-risk state, the virtual model prioritizes updating the state of the abnormal area and activates audible and visual alarms through an event triggering mechanism. At the same time, the virtual model and the physical system run in parallel in shadow mode, and the model accuracy is optimized through incremental learning. Multiphysics coupling simulation and full life cycle visualization management were carried out. Based on SESAM, transient mechanical response and medium- and long-term performance degradation process under extreme wave impact were simulated. The photovoltaic panel array layout and support material parameters were optimized by NSGA-III genetic algorithm. Based on modal parameters, resonance frequency points were identified and the resonance amplitude was reduced by more than 40% through topology optimization. Combined with historical typhoon data, storm surge impact was simulated to assess the failure risk of the anchoring system. During the design phase, manufacturability analysis (DFM) is performed using virtual models; during the operation and maintenance phase, precise operation and maintenance work orders are generated; and during the decommissioning phase, the optimal decommissioning plan is simulated. By connecting to SCADA, ERP and weather forecasting systems through API gateways, embedding reinforcement learning modules to train control strategies, and following the ISO / IEC 11135 standard to open model call interfaces, digital management of the entire lifecycle of offshore photovoltaic power generation systems can be realized.

[0021] The above description is merely a preferred embodiment of the present invention and is illustrative rather than restrictive. Those skilled in the art will understand that many changes, modifications, and even equivalents can be made within the spirit and scope defined by the claims of the present invention, all of which will fall within the protection scope of the present invention.

Claims

1. A method for realizing a digital twin of an offshore photovoltaic power generation system, characterized in that, Includes the following steps: Step S1: Real-time acquisition of marine environmental data and motion status data of offshore photovoltaic power generation system; after preprocessing the acquired data, spatiotemporal feature fusion is completed, and a fusion feature matrix with timestamp and spatial coordinate label is output. Step S2: Based on the fused feature matrix, multi-dimensional feature extraction, mechanical performance evaluation and power generation efficiency correlation analysis are realized, a core feature set is constructed and a reliability label is attached; Step S3: Based on the fused feature matrix and core feature set, construct a high-fidelity virtual mapping model of the offshore photovoltaic power generation system. Simulate the real-time dynamic interaction between physical entities and virtual models, multi-physics coupling simulation, and full life cycle visualization management through the high-fidelity virtual mapping model.

2. The method for realizing a digital twin of an offshore photovoltaic power generation system according to claim 1, characterized in that, In step S1, the marine environmental data includes meteorological data, hydrological data, and geographic data; The meteorological data includes light intensity, ambient temperature, relative humidity, wind speed, and wind direction; The hydrological data includes wave parameters, ocean current velocity, seawater temperature, and salt spray concentration. The wave parameters include wave height and period. The geographic data includes latitude and longitude coordinates and tidal levels.

3. The method for realizing a digital twin of an offshore photovoltaic power generation system according to claim 1, characterized in that, In step S1, the motion state data of the offshore photovoltaic power generation system includes kinematic parameters, structural mechanical parameters, and electrical performance parameters. The kinematic parameters include the real-time position, attitude angle, and dynamic displacement of the photovoltaic array; The structural mechanical parameters include the deformation of the flexible photovoltaic panel, the stress of the support structure, and the tension of the anchoring system; The electrical performance parameters include component output power, open-circuit voltage, short-circuit current, component temperature, and surface cleanliness.

4. The method for realizing a digital twin of an offshore photovoltaic power generation system according to claim 1, characterized in that, In step S1, data preprocessing includes: using wavelet transform to remove high-frequency noise caused by wave impact, and using Kalman filtering to smooth the attitude angle data of the IMU inertial measurement unit. Outliers in the data are detected based on the 3σ criterion and the isolated forest algorithm, and missing values ​​in the data are filled using spatiotemporal interpolation. Z-score standardization is applied to data of different dimensions, and sliding window framing is used for time series data. The output is structured data containing a quadruple of ID-timestamp-spatial coordinates-feature value.

5. The method for realizing a digital twin of an offshore photovoltaic power generation system according to claim 1, characterized in that, In step S1, all sensor data are unified to UTC timestamps through the NTP network time protocol, and delayed data is corrected by timestamp backtracking. Establish a sensor location-system spatial coordinate mapping table based on the physical topology of the photovoltaic panel array; A sliding window and an LSTM neural network are used to extract temporal feature vectors, and a graph neural network (GNN) learning component is used to extract spatial feature vectors based on spatial correlation weights. An attention mechanism is used to assign dynamic weights to the temporal and spatial feature vectors, and the resulting fused feature matrix is ​​output after feature concatenation and a fully connected layer.

6. The method for realizing a digital twin of an offshore photovoltaic power generation system according to claim 1, characterized in that, In step S2, the mechanical performance evaluation includes: based on the extracted strain characteristics and material mechanical parameters, calculating the equivalent stress using the fourth strength theory and comparing it with the allowable stress to generate a strength margin index; evaluating the overturning risk of the floating photovoltaic support using the Euler critical load formula; and calculating the pull-out force safety factor of the anchoring system using the limit state design method. The fatigue damage degree was calculated using the rainflow counting method and Miner's linear cumulative damage theory on the strain time history data. A remaining life RUL prediction model was constructed based on LSTM neural network to output the remaining life of 1-5 years. Weak areas with damage degree ≥0.6 were marked by the heat map overlay method. The flexible characteristic parameters are input into the SESAM finite element model to correct the material properties and boundary conditions, and the accuracy of the mechanical performance evaluation is verified by fluid-structure interaction simulation.

7. The method for realizing a digital twin of an offshore photovoltaic power generation system according to claim 1, characterized in that, In step S2, the power generation efficiency correlation analysis includes: using random forest regression to quantify the contribution of each feature to power generation efficiency, and establishing a nonlinear mapping relationship between environmental parameters, motion parameters, and efficiency loss through gradient boosting tree (GBDT). Based on the LSTM-GRU hybrid network, the power prediction curves are output in the 1-60 minute range, and the average daily power generation is output in the 1-7 days by combining weather forecast and spatiotemporal attention mechanism. An abnormal drop in efficiency was detected by using the isolated forest algorithm. The total efficiency loss was decomposed into environmental loss, motion loss and structural loss. The potential for efficiency optimization was evaluated based on the genetic algorithm.

8. The method for realizing a digital twin of an offshore photovoltaic power generation system according to claim 1, characterized in that, In step S3, the construction of the high-fidelity virtual mapping model includes: constructing a hierarchical geometric model from the component level to the system level based on CAD drawings and laser scanning point cloud data; Flexible characteristic parameters are embedded into the model, and the Mooney-Rivlin hyperelastic constitutive equation is used to describe the mechanical behavior of the flexible film. The anisotropy of the scaffold is simulated by a fiber-reinforced composite material model. Dynamic environmental field models of wave field, ocean current field, wind field, and salt spray corrosion field are constructed based on marine environmental data. The Bayesian optimization algorithm was used to correct the model parameters with the measured data as the objective function, so that the average error between the model prediction and the measured value was ≤3%. The applicability of the model was verified by generating 200+ working conditions through Monte Carlo sampling.

9. The method for realizing a digital twin of an offshore photovoltaic power generation system according to claim 1, characterized in that, In step S3, real-time dynamic interaction includes: deploying an edge computing gateway to access real-time data streams and feature parameter streams via the OPCUA / MTConnect protocol, establishing a control command channel from the virtual model to the physical system via the MQTT protocol, and constructing a digital thread based on blockchain technology to record the entire lifecycle data interaction log; The Kalman filter-particle filter fusion algorithm is used to achieve pose synchronization between physical entities and virtual models, realizing millimeter-level displacement synchronization at the component level and centimeter-level position synchronization at the system level. When the physical system is in a high-risk state, the virtual model prioritizes updating the state of the abnormal area and activates audible and visual alarms through an event-triggered mechanism. The virtual model and the physical system run in parallel in shadow mode and optimize the model accuracy through incremental learning.

10. The method for realizing a digital twin of an offshore photovoltaic power generation system according to claim 1, characterized in that, In step S3, multiphysics coupling simulation and full life cycle visualization management include: simulating the transient mechanical response under extreme wave impact from 0 to 60 seconds and the performance degradation process from 5 to 20 years based on SESAM, and optimizing the photovoltaic panel array layout and support material parameters through the NSGA-III genetic algorithm.

Images