The DFIG (its configuration is shown in Figure 3) is the most recognized technology yet, and it has been installed widely since 2000. This wind energy harvesting technology uses both multiple- and single-gearbox systems; but the system with multiple gearboxes had widely gained acceptance until recent years, while single-gearbox technology is currently reported to have outstanding features in several research studies. The stator windings in a DFIG are directly tied to the grid by means of a transformer, and the rotor windings are tied to the power grid via a power electronic converter with approximately 30% power rating of the generator (Cheng and Zhu, 2014; Desalegn et al., 11912). In this technology, the frequency and the current in the rotor of the generator can be smoothly regulated by the power electronic converter, and hence, the rotational speed of rotor blades can be adjusted in an acceptable range to increase energy harvesting and minimize the mechanical stress. The comparatively lower rating of the power converter makes this technology preferable in terms of cost. Yet, the major limitations of this technology are the application of slip rings with poor reliability and inadequate power regulation capability with regard to grid or generator power fluctuations. This technology has a globally dominant share in onshore wind power generation, and it is less suitable for offshore wind power application.

Due to the fact that the power rating capacity for the power electronics converter in DFIG-based wind energy conversion technology is comparatively small, the two-stage voltage source converter (VSC) topology is widely recognized in this technology. Conventionally, two-stage VSCs are developed with a back-to-back design along with a common direct current (DC) link. A special feature of this back-to-back design is that it can help to implement complete power regulation under system operation. Moreover, this design has comparatively low structure complexity with a low component number, and this contributes to excellent efficiency and reduced cost of the DFIG-based system.

PMSG-based wind energy conversion technology (Figure 4) is another interesting system, and it is highly recognized in the most recently installed wind farm industries. By developing a full-scale power electronics converter and transformer to couple the power grid and the stator windings of the machine, the energy harvested by this wind energy conversion system can be entirely handled (Le et al., 12023). In comparison to the DFIG-based wind energy conversion design, the most important features that can be recognized are the absence of slip rings, uncomplicated or even unneeded gearbox, enhanced power and speed regulation on the broader scale, and superior grid compensation efficiency. However, increased stress and high cost of power electronics devices and increased power loss in the converter phase are the major disadvantages (Yaramasu et al., 2017). This design is usually not preferable in recently installed onshore wind energy conversion systems. On the other hand, PMSG-based wind power technology has been reported in multiple recent studies as a highly attractive candidate for recent and future offshore wind power applications because its converter device can be scalable to large-scale MW power.

Since the power electronics converter in the PMSG-based wind energy harvesting technology needs to embrace the full power harvested at large-scale megawatts, the two-stage VSC topology may be prone to maximum loss at this power scale. In addition, the cabling in the instance of low-voltage scales below 1 kV with increased current is a design limitation. In order to get along with increasing power rating, various multi-cell converter configurations have been developed for different synchronous generator technology-based systems. The multi-cell converter topology generally has the benefits of standard and robust low-voltage converter technologies. For instance, multi-cell two-stage VSC in parallel connection is the state-of-the-art option for PMSG-based wind energy-harvesting technologies exceeding 3 MW (Yaramasu and Wu, 2016). The advanced configurations of power electronics converters for offshore wind farm applications are presented in Section 3.

The performances of DFIG- and PMSG-based wind electric power generation systems are qualitatively summarized and compared in Table 1. Based on their design types of mechanical and electrical system alignments, different configurations of DFIG- and PMSG-based wind power technologies have been introduced to wind farms for harnessing onshore and offshore wind power. In some cases, the nature of the alignments between mechanical and electrical devices can serve to generally characterize these technologies as geared and direct-drive systems. For example, DFIG-based wind turbines are generally geared technologies as they may depend on either three-stage gearbox or single-stage gearbox design systems, while PMSG-based wind turbines can depend on either single-stage gearbox design or they are direct-drive (gearless) technologies. However, the recent state-of-the-art wind power plants generally rely on gearbox-based technologies, such as three-stage gearbox-based DFIG, single-stage gearbox-based DFIG, and single-stage gearbox-based PMSG. Direct-drive PMSG technology has recently been undergoing research and technological advancements, and reports have claimed its promising development for the application of offshore wind multi-mega scale power generation.

Table 1 provides comparative summaries pertinent to the performances and operation characteristics of DFIG- and PMSG-based wind power systems based on multiple requirements. These summaries are designed such that the listed requirements for comparisons generally apply to all configurations of DFIG-based systems (three-stage and single-stage gearbox technologies) and PMSG-based systems (single-stage gearbox and direct-drive technologies). In consideration of the cost of the system as one of the requirements for the comparisons, an average of the costs of two configurations (three-stage and single-stage gearbox) of DFIG technologies is compared with an average of the costs of two configurations (single-stage and direct-drive) of PMSG technologies. As shown in Table 1, DFIG-based systems are desirable for their lower general cost than PMSG-based systems, while PMSG-based systems are generally attractive solutions due to their suitability for multi-scale offshore wind power application.

The operational characteristics of DFIG- and PMSG-based wind power systems can be quantitatively evaluated based on research findings. Most research works were focused on the studies of DFIG- and PMG-based systems’ reliabilities, active and reactive power performances, etc. Here, the capability of power generation reliability for both generator systems is analyzed and compared based on the reports that were presented in the research works. In addition, active and reactive power performances and energy harvesting ranges of these two generator-based systems are considered for comparative discussion.

The junction temperature of power devices that correspond to DFIG- and PMSG-based systems of a 2-MW rated power capacity is shown in Figures 5A, B comparatively. For these two generator-based wind energy-harvesting technologies, the power devices’ thermal cycling is demonstrated within 0.2 s. According to Figure 5A, the performance of a partial-scale power converter in the DFIG-based system could deteriorate due to its thermal cycle that is characterized by a larger amplitude than the performance of a full-scale power converter in the PMSG-based system, whose thermal cycles are characterized by a smaller amplitude as shown in Figure 5B. This indicates that the reliability performance of the power device in a DFIG-based system could be severely affected due to a larger amplitude, which is associated with the system’s thermal characteristics. Consequently, advanced modeling and testing approaches should be proposed in helping adjust the reliability by establishing the power devices’ thermal behavior more effectively based on the wind power converter’s mission profile (Blaabjerg and Ma, 2017). A robust strategy has been presented by Ma et al. (2015), which resembles lenses with varying focal lengths used in photography. The wind power converter’s loading analysis and modeling are partitioned subject to some given time constants and various modeling methods and tools.

Furthermore, Figures 6A, B show active power control performances for DFIG- and PMSG-based marine current wind energy-harvesting systems under the variable speed control strategy. As shown in Figure 6A, the major advantage of the DFIG-based marine current system is its capability to supply fixed voltage and frequency within the range of ± 30 % speed change with respect to normal synchronous speed. An additional option for the range of speed variation (30%–50%) can also possibly be used. A minimum power rating of the rotor converter is directly associated with the range of 30% speed variation. Similarly, in Figure 6B, the active power control of the PMSG-based marine current system also indicates good power-tracking performance. In both systems, the differences between the estimated and simulated power are negligible. These negligible differences primarily result from the implementation of variable speed control rather than a direct power control.

As noted, the comparison of DFIG- and PMSG-based wind energy-harvesting systems is qualitatively shown in Table 1 by considering their annual power production capability as one of the criteria. Herein, Figures 7A, B show the annual energy-harvesting performances of DFIG- and PMSG-based marine current turbine technologies for the given ranges of tidal velocities, based on the research findings by Benelghali et al. (2010). Accordingly, Figure 7A represents the annual energy harvested by a DFIG-based marine current wind power system; the annual harvested energy corresponding to different tidal velocities, with DFIG technology, is calculated to be around 1,530 MWh/year according to the study. On the other hand, by the application of a PMSG-based marine current wind power system (Figure 7B), the cumulative generated power under the similar ranges of tidal velocities is reported to be about 1,916 MWh/year by the same study. As noticed, there is nearly a 25% difference in the power produced over a year between these two power systems, and this difference can be further extended when using a larger turbine system. The DFIG-based power system is characterized by its restricted speeds, and this is the reason for the reduction in its annual power production compared to the full-scale PMSG-based system. Meanwhile, according to a study by Fischereit et al. (2015), the relationship between tidal currents and wind speed was quantified such that the wind speed over a marine was observed to increase or decrease by around 0.2 m/s depending on the direction of the tidal flow with respect to wind direction.

The general structure of offshore wind farm transmission with the application of HVDC and HVAC systems that include the components, such as power converter, transformer, offshore substation, and onshore substation, is shown in Figure 8. Here, the main focus is to evaluate the economic efficiencies of HVAC- and HVDC-based transmission systems by comparing the capital expenditures required for their respective components in commissioning offshore wind farms. Accordingly, the outlines of capital costs for the electrical components of HVAC and HVDC systems are shown in Table 2 based on the most recent study by Li et al. (2021). According to this study, the required capital expenditures are mainly allotted to cover the costs of transformers, onshore GIS switchgear, offshore substations, cable trench, shunt reactors, and submarine cable in the case of the HVAC system. In the case of the HVDC system, the capital expenditures are divided among the costs of cable trench, HVDC cable, onshore inverter, offshore rectifier, transformers, and additional offshore facilities. Based on the outlines of these costs for the components of HVAC and HVDC systems, it can be deduced that at a bigger power capacity and extended transmission line, HVDC presents more attractive cost benefits.

The HVAC system produces a huge amount of capacitive current that corresponds to the length of the transmission line. Here, the real power transmission capability may be determined with specified and reactive currents of a specific HVAC cable. With 50 Hz HVAC, the highest transmission length is around 140 km without compensation for reactive power. The HVDC system is characterized by considerable capability of power transmission compared to the HVAC system, particularly over an extended distance.

Moreover, the HVDC system offers additional advantages without being limited to cost and transmission capability by outperforming the HVAC system over an extended distance; for the vast offshore wind farm power transmission, the HVDC system also demonstrates lower transmission losses and its cables are commercially available (for larger voltages, HVAC cables are not available), and therefore, the HVDC system helps control excessive load as its power can be inherently adjusted. In general, HVDC is a more desirable solution than HVAC in the transmission of multi-mega-scale offshore wind power that is commissioned far away from the shore.

Different configurations of HVAC and HVDC systems are given in Table 3 based on various factors, which include operational condition, overall cost, and capability of the system’s configuration. The configurations include parallel AC-connected HVAC system, parallel AC-connected HVDC system, parallel DC-connected HVDC system, and series DC-connected HVDC system. Parallel AC-connected HVAC system-based configuration is characterized by its low complexity under the condition that the transmission distance between offshore and onshore sites is not long and is economically efficient under the same condition. However, this configuration shows serious disadvantages when the distance between offshore power and onshore substations is extended.

Parallel AC-connected HVDC configuration, which is based on the modular multilevel converter (MMC) topology, is the state-of-the-art solution to offshore wind farms with larger power capacity and longer transmission lines. However, the main disadvantage of this configuration is the high overall cost of power conversion components and the requirement of a bulk offshore substation. On the other hand, this configuration is characterized by its high capability of power scalability, and MMC topology with a power capacity of 864 MW, voltage of ± 320   k V , and cable length 160/45 km is currently in operation. In addition, MMC topology with a power capacity of 900 MW, voltage of ± 320   k V , and cable length of 45/45 km is under construction for offshore wind farm commercial application by 2023; MMC topology with 900 MW, ± 320   k V , and 100/30 km is under development for application by 2024 (Li et al., 2021).

Furthermore, parallel and series DC-connected HVDC configurations are designed to have reduced size and weight, and are recently under development for commercial use in offshore wind farms. The disadvantages that are being posed with the applications of other AC-connected HVDC configurations could be eliminated by DC-connected HVDC configurations. The discussions for different power converter–HVDC networks for offshore wind power transmission are given in Table 4 based on the criteria mainly related to energy conversion quality and range of applications in recent wind farms and possibility for future developments. The power converter–HVDC networks that are considered for discussions here include line-commutated current source converter (LC CSC)-based parallel medium voltage-alternating current (MVAC)-connected HVDC, two-level voltage source converter (2L-VSC)-based parallel MVAC-connected HVDC, multi-modular voltage source converter (MM VSC)-based parallel MVAC-connected HVDC, cascaded rectifier (diode)-based parallel MVAC-connected HVDC, three-level neutral point clamped (3L-NPC)-based parallel MVDC-connected HVDC, solid-state transformer (SST)-based series medium voltage direct current (MVDC)-connected HVDC, and pulse width modulator current source converter (PWM CSC)-based series MVDC-connected HVDC. Each of these HVDC-based power transmission networks is discussed in the following paragraph.

The application of LC CSC-based parallel MVAC connection with HVDC has shown well-proven performance in onshore wind power generation for about the last five decades. Nevertheless, this LC CSC-based network is inapplicable for offshore wind power, because it is incompatible with huge-scale power generation. On the other hand, the application of 2L back-to-back (BTB) VSC topology that is based on parallel MVAC connection with an HVDC system has been dominating over the onshore wind farm, and is less suitable for offshore wind farm due to its compatibility issue with multi-scale wind power generation. Modular multilevel (MM) VSC, based on parallel MVAC connection with an HVDC system, has been applied to offshore wind power transmission with various power capacity, voltage, and cable length levels. Currently, MM VSC is under further development for further improvement so that it would help ensure power maximization from offshore wind farms. At a wind power scale of 200 MW, cascaded rectifier (diode)-based parallel MVAC connection with HVDC was implemented, and it improved power production by 20% and reduced weight by 60% against traditional 2L-VSC-based parallel connection with HVDC according to study assessment by Blaabjerg and Ma (2017).

Furthermore, various offshore wind power transmission networks that are based on different converter technologies have been proposed in studies to enhance wind power conversion. For example, a three-level (3L) NPC-HVDC parallel MVDC-connected power transmission network has been developed in literature reports in response to the limitations posed while using a power transmission network that is based on the parallel MVAC-connected 2L VSC-HVDC. Series MVDC-connected transmission networks can also be implemented with a (3L) NPC converter (Peng et al., 2021). A series MVDC-connected (3L) NPC-HVDC system can be configured to have a relatively uncomplicated structure, higher power density, and lower cost than a parallel DC-connected system. In addition, SST- and PWM CSC-based series MVDC-connected HVDC configurations are undergoing promising development with attractive features, such as lower costs and weight than parallel MVAC and MVDC-connected networks such as 2L VSC-HVDC, MM VSC-HVDC, and (3L) NPC-HVDC (Wei et al., 2017). Further discussions on this section are given in Tables 3, 4.

The research studies on wind farm optimization using the concepts of standard feedforward control are summarized in Table 5. This approach uses different control algorithms, optimization strategies, input parameters, and wind farm design optimization and evaluation models in enhancing wind farm performances so as to achieve the desired wind farm control objectives. Accordingly, the gradient-based control algorithms with different optimization strategies, such as sequential quadratic programming, steepest descent, and conjugate gradient, were proposed (Table 5) to maximize wind farm power production by adjusting wind turbines’ yaw angles and induction factors. For instance, Annoni et al. (2018) used sequential quadratic programing by implementing the Gaussian wake concept as both the optimization and evaluation model in increasing power production by adjusting the wind turbine’s yaw angle. Under this framework, power production was claimed to increase by 17% compared with the greedy control algorithm. On the other hand, Thomas et al. (2017), with a similar strategy (sequential quadratic programming), used the Jensen–Park model to optimize the yaw angle and layout of the wind farm, and power production was reported to increase up to 24% compared with the greedy control algorithm. Moreover, steepest descent- and conjugate gradient-based optimizations were proposed by Park and Law (2017) and Thøgersen et al. (2017) to adjust the axial induction factor and yaw angle (in the case of the steepest descent strategy) and the yaw angle (in the case of the conjugate gradient strategy) of wind farms, and the results indicated that the power production could be maximized by 7% and 7.5%, respectively.

As shown in Table 5, further standard feedforward optimization strategies that are based on heuristic algorithms (genetic algorithm, particle swarm optimization, and artificial bee colony) and game theory optimization were proposed by different researchers for optimizing various input parameters of wind farms. For instance, a genetic algorithm was proposed by Serrano González et al. (2015) to optimize wind turbines’ tip–speed ratio and blade pitch angle based on the Jensen–Park evaluation model, and it resulted in 1.5% increase in power production. The genetic algorithm was also used by Wang et al. (2018), aimed at maximizing power production by adjusting the axial induction factor and layout of the wind power system based on the Jensen–Park optimization and evaluation model, and it ended up increasing wind power production by 1%–2%. Another powerful heuristic algorithm-based standard feedforward strategy includes particle swarm optimization, which has been widely proposed in recent research works to achieve various wind farm control objectives. In Table 5, the results of several studies that implemented the particle swarm algorithm in optimizing wind farm performance are given. Accordingly, in Behnood et al. (2014), the tip–speed ratio and blade pitch angle were adjusted for a wind farm of 16 turbines based on the Jensen–Park model, and approximation of the power coefficient (as a function of the turbines’ tip–speed ratio and blade pitch angle) and 10.6% increase in power production were reported. The study results that were presented by Bo et al. (2016), Hou et al. (2016), and Gionfra et al. (2019) also indicated the robustness of particle swarm optimization in helping to maximize wind farm output power. In addition, the heuristic algorithm-based wind farm optimization strategy, artificial bee colony, was proposed to adjust the tip–speed ratio and yaw angle of wind turbines based on the Jensen–Park model and FLORIS for design optimization and evaluation in Abbes and Allagui (2016) and Quick et al. (2017) and output power increment of 4%–6% and 3% were, respectively, reported to be achieved.

The algorithmic game theory-based wind farm optimization strategy can be implemented to minimize load fluctuations of power systems in addition to maximizing output power by facilitating the adjustments of the yaw angle and axial induction factor. By using this strategy along with different wind farm performance optimization and evaluation models, some research works were introduced by researchers. Accordingly, in Gebraad et al. (2016), the yaw angle was optimized by using the FLORIS model, whereas output power performance was evaluated by using both SOWFA and FLORIS models, and the wind farm power production was reported to be improved by 1% with FLORIS and SOWFA optimization and evaluation models, and by 13% when the FLORIS platform was used as both the optimization and evaluation model. In addition, in van Dijk et al. (2021), the FLORIS model was used as both a single platform and in combination with the CC-Blade model for optimizing wind turbines’ yaw angle and evaluating wind farm output performance, where power production was reported to be maximized by 3.7% wind and a single model (FLORIS) and load fluctuation of power systems decreased by 18.7% with combined modes (FLORIS + CC-Blade). More work was also introduced in Herp et al. (2015), where the axial induction factor was optimized with the Jensen–Park model under different settings and resulted in 1.4%–5.4% increase in wind farm power production.

Additional works that are based on standard feedforward control strategies involving different algorithms and optimization and evaluation models for the design and performance of wind farms are given in Table 5 as extended summary from Table 5. A dynamic programming strategy was proposed by Rotea (2014) to optimize axial induction of a wind farm by applying the actuator disk model as the design optimization and evaluation standard. The result of this study shows that the power production was maximized by around 5% and wind farm load fluctuation was reduced up to 38% compared to the greedy algorithm strategy. In addition, a dynamic programming optimization strategy was implemented by Santhanagopalan et al. (2018) and Dar et al. (2016) by adjusting the tip–speed ratio through the RANS model and the yaw angle and axial induction factor through the enhanced Jensen–Park model to achieve output power increment by 0.8% and 4.5%–26.5%, respectively.

Additional standard feedforward optimization strategies that are based on different wind farm control objectives and with various wind farm design optimization and evaluation models were also presented in many research works, and the summaries of some of their parts are given in Table 5. Accordingly, the Nelden–Mead simplex algorithm was implemented with the simplified Ainslie model for optimization and evaluation power reference of wind farms in Kim et al. (2017), and the results showed an increase in the output power by 2.4% and reduction in the load fluctuation at upstream wind by 16.7%. According to a study presented in Bossanyi (2018), the generator torque, blade pitch angle, and yaw angle were adjusted and evaluated in aiming at minimizing the overall power losses and power system stress, and power production was saved by 2%. In addition, the research works based on field test optimization were discussed by Fleming et al. (2019) and Fleming et al. (2020) claiming a power gain of 2% and minimization of wake losses by 6.6%, respectively, by using FLORIS as the optimization model in adjusting the yaw angle. In addition, some studies that are specific to floating offshore wind farms were conducted based on blade pitch angle adjustments, and the results of these studies are given in the last row of Table 5. For instance, in the study by Kheirabadi and Nagamune (2019), the FLORIS benchmark was used as the optimization and evaluation model for maximizing power production by adjusting the floating offshore wind farm blade pitch angle under different settings, and 16%–54% increase in output power was reported. In addition, the adjustment of floating offshore wind farms under different design optimization and evaluation benchmarks: deep neural learning optimization model and LSTM evaluation model by Sierra-Garcia and Santos (2021) and the Jensen–Park optimization model and FarmFlow evaluation model by Rodrigues et al. (2015) were implemented; power production growth of 7%–21% and 4.4% were, respectively, reported.

Under this control approach, the research works that are based on model predictive optimization and data-driven optimization strategies are summarized in Table 6. Accordingly, a model predictive optimization algorithm was implemented by Heer et al. (2014), using the Jensen–Park platform as the optimization model and the SimWindFarm platform as evaluation in aiming to increase wind farm power production with adjustments of the blade pitch angle and tip–speed ratio. Under these two adjustments, the output power showed an increment of 0.4%–1.4%. On the other hand, the axial induction factor was adjusted in the study by Vali et al. (2016), making use of another wind farm performance optimization and evaluation platform (WFSim) based on the model predictive optimization strategy to achieve a power production growth of 3.8%. Furthermore, with the implementation of the model predictive optimization strategy through the application of the WFSim model for optimizing the axial induction factor and evaluating wind farm output power performance, additional works were presented by Vali et al. (2017) and Vali et al. (2019), reporting an improvement in power production by 2%–8% and 4%, respectively. On the other hand, larger increments of power production were reported to be achieved while using the SP-wind benchmark as the optimization and evaluation model for adjusting thrust coefficients (Goit and Meyers, 2015; Munters and Meyers, 2016; Munters and Meyers, 2017), and thrust coefficients and yaw angle rates (Munters and Meyers, 2018a; Munters and Meyers, 2018b) of wind farms under different control settings.

Different wind farm performance optimization and evaluation models (FLORIS and SOWFA, deep neural learning model using LES and CNN + LSTM, and DB-PC and extended Kalman filter model) were used to adjust the yaw angle (Doekemeijer et al., 2019a), tip–speed ratio and blade pitch angle (Yin and Zhao, 2021), and rotor rotational speed and axial induction factor (Abdelrahem et al., 1596); output power growths of 7%–11%, up to 38%, and 20%, respectively, were claimed to be achieved. On the other hand, various data-driven model-based optimization strategies (Bayesian ascent, Bayesian optimization, etc.) were implemented by researchers such as Park and Law (2016), Zhao et al. (2020), and Yin et al. (2020) by making use of different wind farm optimization models (Gaussian regression model, FLORIS, deep reinforcement learning, and support vector machine, respectively), and evaluation models (wind tunnel, SOWFA, WFSim, and FLORIS, respectively) in adjusting input parameters (yaw angle and blade pitch angle, yaw angle, axial induction factor, and yaw angle, respectively) to determine the levels of increments in the resulted power productions and reliability improvement. Accordingly, output power growths were reported to be achieved by 30.4%–33.2%, 4.4%, and 10%, and reliability was enhanced by 1.7%, respectively.

As shown in Table 7, model-free closed-loop control algorithm-based wind farm optimization strategies, which include multiple resolution-based simultaneous perturbation stochastic algorithm, game theory, gradient descent, and nested extremum controller, were implemented in optimizing wind turbines’ input parameters, such as axial induction factors and generator torque gains in order to evaluate the possible outcomes in generated power. Axial induction factors were optimized under different settings [simultaneous perturbation stochastic algorithm (Ahmad et al., 2014), game theory (Marden et al., 2013), and gradient descent (Gebraad et al., 2013)], and the outcomes were evaluated with the same model (Jensen–Park), and different levels of increments (32%, up to 25%, and 1%) in the output power were reported to be achieved. On the other hand, generator torque gains and yaw angle were adjusted by implementing the same optimization strategy (nested extremum-seeking controller) and using different evaluation models [SimWindFarm (Yang et al., 2015), UTD-WF (Ciri et al., 2016); Ciri et al., 2017), and wind tunnel (Campagnolo et al., 2016)]; varying levels of increase (1.3%, 10%/7.8%, and 15%) in the power production were claimed to be achieved.