1. Introduction
According to the statistics of the International Energy Agency (IEA), the share of new and renewable energy consumption is steadily increasing, and wind power accounts for 36% of the total increase. It is growing much faster than other renewable power generations, such as solar power (27%), hydro (22%), and biomass (12%) [
1,
2]. In particular, there is an increasing proportion of offshore wind turbines that have better wind conditions and can mitigate problems such as noise pollution during operation. However, offshore wind turbines are exposed to harsher weather conditions than onshore due to sea salinity and strong winds [
3,
4,
5]. With the recent rapid increase in variable renewable electricity (VRE), there are growing concerns about the deterioration of the flexibility of the electric power system [
6]. Grid flexibility is the ability of a system to manage fluctuations and uncertainties in electricity supply and demand stably. The VRE will play an essential role in future energy systems, but integrating large-scale VREs into energy systems requires additional flexibility enhancement [
7,
8]. In order to respond cost-effectively to the spread of the VRE, it is necessary to develop an accurate prediction method for renewable energy generation along with strategies such as introducing energy storage systems and sector-coupling technologies [
9].
In general, forecasted weather environment data can be used to predict the output of a wind turbine in advance. However, because offshore wind turbines are exposed to very variable weather conditions, such as rapid wind speed fluctuations and strong winds, tropical heat, hail, and snow, the current environment may differ from the predicted data. Furthermore, unlike stationary wind turbines that only consider wind data, the output of floating offshore wind power is immediately affected by waves, so the height and period of waves must also be considered. In the case of marine environment data, it is more accurate to predict the weather data one hour ago than one month ago, and it is more accurate to predict the weather data one minute ago. Therefore, it is necessary to develop a real-time prediction method that can effectively predict output using data predicted at a closer time. The output prediction of offshore wind turbines requires high-speed calculation time and accuracy. However, the existing simulation tools take a long calculation time and are difficult to calculate close to reality in consideration of the dynamics and control values of the wind turbine [
10,
11]. In particular, floating offshore wind turbines (FOWT) are exposed to harsher weather fluctuations than stationary ones, and the output of FOWT is affected by the complex dynamics related to moorings, floats, towers, and nacelles. For the optimization of these complex structures, a digital twin system can be an effective alternative.
A digital twin is a virtual model that accurately reflects a physical object. Digital twin utilizes IoT communication, AI-based machine learning, and real-time analytic software to implement virtual models to analyze the state of real systems and diagnose causality. In addition, the virtual models can predict outcomes by applying various input parameters that will occur in real systems in the near future [
12,
13,
14,
15,
16,
17,
18,
19]. In recent years, digital twins have been studied as an alternative and new method for predicting the output of wind turbines [
20,
21]. Recent studies on the digital twin technology of wind turbines contributed to the improvement of operational conditions by developing fault diagnosis, condition monitoring, and residual life prediction techniques for offshore fixed and floating wind turbines based on data collected from SCADA devices [
22,
23,
24]. Among them, Fahim, M. et al. [
24] used machine learning and SCADA data to develop a digital twin-based model capable of generator output prediction and real-time condition monitoring. When predicting the output of a wind turbine based on machine learning, once the model is trained, it is stable in prediction and inference. However, this modeling method has several drawbacks. Developing models with machine learning is quite tricky without long-term measured SCADA data. In addition, when the capacity, size, and controller of the wind turbine are changed, it is difficult to modify and verify the system [
13]. Therefore, there is a need for a physics-based digital twin that can predict the output without the historical and SCADA data of the wind turbine. If a physics-based model is used, the digital twin can be developed without SCADA data.
Once a physics-based model is created for the digital twin, it is easy to analyze and validate a variety of input data and compute complex systems such as wind turbines in near real-time as a reduced order model (ROM). The ROM system uses the correlation of input and output data to simplify various models in full 3D simulations, system simulations, and characteristics of historical data and is effective in implementing them as close to real-time as possible. The mechanism of ROM is similar to that of artificial neural networks, but it is more advantageous to analyze the correlation between input and output within an operational wind speed range [
25]. It is possible to calculate the output of a wind turbine using real-time field data through an IoT sensor connected to a physics-based model with ROM or to predict the output of a generator using weather forecast data.
This paper is the first attempt to implement and evaluate a physics-based output prediction model (P-bOPM) for digital twin construction on a 10 MW FOWT. A P-bOPM of 10 MW was implemented, and its performance was confirmed through a comparative evaluation of the reduced model test results and the developed digital twin system. The P-bOPM for a digital twin was designed considering the geographical characteristic of Korea and simulated using ANSYS Twin Builder. The wind turbine was scaled by considering the DTU 10 MW and the national renewable energy laboratory (NREL) 5 MW standard wind turbine equipped with a semi-submersible type platform for the marine environment of Korea. The average wind speed of the site was 8.5 m/s, and the rated wind speed of the wind turbine was 11.3 m/s. The blade pitch and nacelle yaw control systems of the 10 MW FOWT were designed taking into account the operating characteristics of the wind turbine simulator developed by NREL called the fatigue, aerodynamics, structures, and turbulence code (FAST). Since the motion of the floater in the offshore wind turbine continuously changes the output power and the control value, the six degrees of freedom (6-dof) of the FOWT were implemented using the ROM for real-time status updating. Through the FAST simulation, the correlation of the 6-dof of floater in various marine environments was analyzed and applied to learning the ROM system. The wind speed and sea level were used as input data, and 6-dof was generated as output data for the ROM system. In the equation of the output power of 10 MW FOWT, the wind speed was replaced by the effective wind speed determined by the motion equation of the 6-dof. In the P-bOPM proposed in this paper, the output power was predicted by calculating the effective wind speed reflecting the dynamic ROM system. Additionally, for verification of the P-bOPM of the 10 MW FOWT, the reduced model test was conducted by the Korea Research Institute of Ships and Ocean Engineering (KRISO). As a result, the testing of the 1/35 reduced model was performed with a variable marine environment and showed good accuracy with the P-bOPM of the 10 MW FOWT. The error occurred in the output power due to the error of the developed ROM system for the real-time prediction of 6-dof, and the accuracy was 92% when comparing the result of the reduced model test and the P-bOPM for the digital twin. Through this, the accuracy and reliability of the P-bOPM were confirmed. The digital twin integrated with P-dOPM, considering the offshore wind farm’s variability, can greatly help improve the power system’s flexibility.
5. Conclusions
This paper dealt with the development of a P-bOPM of a 10 MW FOWT for a digital twin. The 10 MW wind turbine system was modeled taking into account the offshore environment of Korea, and the volume and weight of the turbine and floater were calculated using the scaling method from NREL. A direct-driven PMSG was modeled for the 10 MW FOWT. In order to confirm the dynamic characteristics of the floater affecting the output power, the 4-dof according to the wind speed and sea level was analyzed using the FAST. Based on these data, the ROM was designed, and it is used to predict the dynamic characteristics of the floater according to the offshore environmental conditions. As a result, the ROM accuracy of the pitch was the lowest at 93.1%, and the ROM accuracy of the yaw was the highest at 98.9%.
The P-bOPM of a 10 MW FOWT for a digital twin was implemented, and its performance was confirmed through a comparative evaluation of the reduced model test results and the developed digital twin system. The RTDS plays the role of an environment sensor and transmits signals through TCP/IP communication with the P-dOPM using python code. The wind speed and sea level data transmitted to the P-bOPM and the output power and blade pitch control of the 10 MW FOWT are calculated in near real-time. Through the model test, the 6-dof motion of the floater was captured and linked with the FAST simulator to analyze the output power and thrust of the wind turbine. The output power of the 10 MW FOWT, considering the effective wind speed by the thrust and the motion of the floater, was calculated close to real-time. The accuracy of the digital twin equipped with P-bOPM of 10 MW FOWT was 92.3%, showing satisfactory results with the target accuracy of 90% or more. The digital twin integrated with P-dOPM, considering the variability of the offshore wind farm, can be of great help in improving the flexibility of the power system.