The EU and UK have made ambitious commitments under the net-zero plans to decarbonise their economies by 2050. For this, offshore wind will play a major role, significantly contributing to a paradigm shift in the power generation and greater volatility of electricity prices. The operating strategy of wind farms should therefore move from power maximisation to profit maximisation which includes income from providing power system services and the reduction of maintenance costs. Wind farm flow control (WFFC) is a key enabler for this shift through mitigation of wake effects in the design and operation phases. The results of the FarmConners market showcases presented here are the first attempt to economically assess WFFC strategies with respect to electricity market prices. Here, we present a conceptual simulation study starting from individual turbine control and extend it to layouts with 10 and 32 turbines operated with WFFC based on the results of five participants. Each participant belonged to a different research group with their respective simulation environments, flow models and WFFC strategies. Via a comparative analysis of relative WFFC benefits estimated per participant, the implications of wind farm size, the applied control strategy and the overall model fidelity are discussed in zero-subsidy scenarios. For all the participants, it is seen that the income gain can differ significantly from the power gain depending on the electricity price under the same inflow, and a favourable control strategy for dominant wind directions can pay off even for low electricity prices. However, a strong correlation between income and power gain is also observed for the analysed high-electricity-price scenarios, underlining the need for additional modelling capabilities to carry out a more comprehensive value optimisation including lower prices and system requirements driven cases.
The ambitious targets “Net Zero by 2050” within the EU and UK will be driven, in large part, by a significant rise in offshore wind. The UK alone has committed to 40 GW of offshore wind by 2030, a near 4-fold increase on its current capacity. The rapid rise in offshore wind will result in a shift from electricity prices driven by fossil fuel prices to those driven by the availability of the renewable energy sources. This is expected to increase price volatility, which will challenge wind farm owners/operators to reconsider their design and operation strategy from power maximisation to revenue maximisation, particularly in zero-subsidy schemes.
Tools to support wind farm owner/operators to cope with this transition have been in development for a number of years within the field of wind farm flow control (WFFC). WFFC is the coordinated operation of wind turbines within a wind farm to serve a common goal by taking their aerodynamic interactions into account. This may include a diverse array of objectives from increased energy production and structural load alleviation to environmental and/or societal impact mitigation
The first of their kind, the FarmConners market showcases allow researchers to demonstrate the economic benefits of their control approaches in simulations using realistic environmental and market conditions based on 2020 and 2030 variable-electricity-price scenarios. Accordingly, the objectives of the showcases can be summarised as follows:
Investigate benefits of WFFC for variable electricity prices. Demonstrate the value of WFFC in existing and upcoming market scenarios. Investigate the readiness of wind farm flow control for participation in electricity markets.
Investigating the benefits when participating in the wholesale electricity markets will provide insight into how WFFC can contribute to the revenue when the power is sold for variable prices. Demonstrating the value in existing and upcoming market scenarios will help to understand the benefit of WFFC in a future with a higher share of wind energy in the energy mix. Investigating the readiness of WFFC strategies will provide a better understanding of the state of the art and reveal the gaps towards market-driven wind farm operation.
The showcases are based on the TotalControl Reference Wind Power Plant (TC-RWP) with 32 DTU 10 MW wind turbines in a staggered layout
The wind turbines and their aerodynamic effects (wakes) are neglected in the simulations, so the wind inflow can be considered as the free stream one. More information about the data generation for the FarmConners market showcases, including detailed descriptions of the tool chain and assumptions, is available in
Originally, three showcase sets were defined to reflect different market and operational situations within 2020 and 2030 scenarios – namely (1) high day-ahead electricity prices, (2) low day-ahead electricity prices, and (3) operation by the transmission system operator, where the last two include structural load alleviation as a performance indicator
The first set with high electricity prices motivates us to maximise the power output, thus resembling the current situation. The second set with low electricity prices was defined to provide incentives to not only focus on power production but include additional control objectives, in this case structural load alleviation. The third set with the objective of power tracking is an outlook to a wind farm operation that provides power system services and alleviates structural loads.
The three original showcases enable us to compare different operational strategies with control objectives beyond power maximisation. Researchers worldwide were invited to participate by simulating their code using one or more showcase sets.
Five participants have submitted their results for this FarmConners market showcase study. These participants are members in different research groups and thus used different simulation environments, different models for the wind farm flow and different control strategies.
Due to the limited number of participating models capable of providing the relevant outputs for the alleviation of structural loads, the results presented here are mostly limited to the potential benefits of WFFC during the first showcase set, i.e. high day-ahead electricity prices. For that showcase set, the highest 25 % of the electricity prices (both in 2020 and 2030 scenarios) and their respective binned wind inflow data are used
Distribution of day-ahead electricity prices per binned wind sector in 2020
Nevertheless, all three showcase sets are demonstrated for a single turbine with much lower computational cost for aeroelastic analyses, and the extension of the prospects for full-scale WFFC is discussed. Additionally, the differences of the expected benefits between smaller and larger wind farms are highlighted through the analyses of the subset and full layout of TC-RWP, i.e. 10 and 32 turbines. The considered setups for all participants are summarised in Table
Overview of the participants (participant IDs P1–P5) per considered layout. The layout of the subset and full TotalControl Reference Wind Power Plant (TC-RWP) with 10 and 32 wind turbines, respectively, is indicated when the results are presented in Sect.
The structure of the article is as follows. Sect.
The analysis of a single wind turbine, presented in this section, was performed by participant P1. Control objectives of both optimised revenue and reduced structural loads were analysed. It therefore provides deeper insight into the revenue-based control paradigm. An overview of this approach has been presented previously
The new method proposed here by P1 is to include the electricity price in the decision-making of the turbine’s operational mode. Using downregulation, power boosting, and individual blade control (IBC) flexibly according to the instantaneous weather and price conditions, the damage and revenue accumulation of the turbine over time can be managed. Currently, downregulation (also referred to as curtailment or derating) is used commercially to follow reference power levels according to the grid system requirements. As previous numerical
Proposed methodology for flexible control operation of a single wind turbine developed and applied by participant P1. IBC: individual blade pitch control.
The procedure of implementing and evaluating a flexible control method at the wind turbine level is summarised in Fig.
For this study, the baseline proportional integral (PI) controller of the DTU 10 MW reference wind turbine is modified to allow varying nominal power ratings from 5 to 13 MW in downregulation and power-boost modes. This is achieved by changing the set points for power coefficient (
The approach chosen here was to keep the tip speed ratio constant for all power levels and vary the set points by changing the fixed pitch angle below rated. As other studies have shown
The output of the described procedure is a controller with a required rated value as input and the option to apply IBC. The next step is to create a surrogate model of the response that can be used for evaluation and optimisation. As shown in the literature
The mean of the results of the three turbulence seeds for each operating point was then tabulated in a 3D matrix for each quantity of interest (mean of power, DEL of blade root moment, etc.), forming the basis of the surrogate used here. To ensure the smoothness of the model and avoid local fluctuations due to controller tuning or seed-to-seed variability, filtering is applied using a Gaussian convolution kernel. This structure was probed using a spline-based interpolation to produce the surrogate response of the system for any of the quantities of interest.
The evaluation part of the process was implemented similar to a time marching simulation, where at each time step the inputs are wind speed, TI, electricity price, and controller mode and the outputs are instantaneous and cumulative responses. The accumulation of quantities such as energy production was calculated by integrating the values over time. For quantities such as the standard deviation of the rotor speed, the accumulation was done by averaging, and the cumulative load was calculated using Eq. (
The baseline single-turbine response was identified by using the entire 1-year time series of the FarmConners market showcase data set, with the baseline rating of 10 MW without IBC. The optimisation objective is then to reduce the loads and/or increase the revenue over the whole period of each data set (2020 and 2030 time series). This problem is not trivial as there are multiple conflicting objectives in an online optimisation setup. Some of these include the high variation of the sensitivity of the different loads to the controller state and wind speed, the correlation of probability of wind speeds and prices, and the uncertainty of future inputs. For this preliminary application of the method, the problem was simplified and treated as an offline single-objective optimisation problem with a perfect preview. The blade root out-of-plane moment (BROop) was chosen as a representative load to be used to correlate power output level to loads for all conditions.
The method was applied for both data sets using the hourly resolution; i.e. the control strategy of the turbine is changing every hour according to the wind and price conditions for the time steps. In Fig.
Example time series of instantaneous values for a subset of 2030
The sensitivity of the method for different objectives is examined with three cases. Case 1 aims to increase the revenue, case 2 aims to increase the revenue in a load-neutral manner, and case 3 aims to decrease the loads. In addition, case 2 was run with IBC and without IBC. In the lower-right plot, each small change contributes more to the DEL according to the number of events (see Eq.
Cumulative values of damage equivalent loads (DELs) and revenue for operation of a single wind turbine with different objectives, in relation to the baseline operation. TBMx: tower bottom side-to-side moment, TBMy: tower bottom fore–aft moment, TBMz: tower bottom torsion, BRMx: blade root edgewise moment, BRMy: blade root flapwise moment, BRMz: blade root torsion, BROop: blade root out-of-plane moment, BRIp: blade root in-plane moment, TTMx: tower top roll moment, TTMy: tower top pitch moment, TTMz: tower top yaw moment, LSSMy: non-rotating low-speed shaft bending moment about the
The presented results are based on a basic optimisation logic including user-defined thresholds and trial and error methods. Nevertheless, the results show that there is sensitivity of structural loads and revenue to the method used here, and in other tested cases, not shown here, different trade-off levels could be achieved. Moreover, a different behaviour was observed between the two years. Finding an effective tuning for the 2020 data set was much harder to achieve than for the 2030 data set. More threshold values had to be tested, and fine adjustments could result in larger changes. On the other hand, for 2030, even with rough estimations, the optimisation objectives were achieved without much tuning.
Since the weather time series are the same for both years, the difference lies in the price patterns. Fig.
Contribution of price and wind speed bins to revenue and damage equivalent load (DEL) of blade root out-of-plane moment (BROop), obtained by multiplying the counts of each bin with the response obtained with the baseline wind turbine controller in simulations with the full data sets for 2020 and 2030. All values are normalised to the maximum value for the specific year and metric. Note that regions with high contributions to revenue on the right are strongly correlated to the bins with (below rated) wind speed and direction in Fig.
The results presented here for a single turbine are an initial application of the proposed method that was first published by
The next step would be to add the dimensions of wind direction and wind farm (flow) control to evaluate the holistic potential of flexible control in wind energy in the future electricity markets. In that regard, the rest of the paper focuses on the most beneficial price and wind speed bins for wind farm flow control as identified in Fig.
The analysis of the single-turbine control for increased revenue by P1 is extended in this section to wind farm level to explore the potential power and income gains of different WFFC-oriented models with varying control strategies, both for the 10-turbine layout (participant P2) and the full TC-RWP with 32 turbines (participants P3–P5).
Table
Overview of participating models.
The wind farm model used by participant P2 is FLORIS velocity deficit – legacy Gauss (Gaussian model) by wake-added turbulence – Crespo-Hernández model wake superposition – modelled with SOSFS (sum of squares freestream superposition) to combine the wake velocity deficits to the base flow field wake steering – wake deflection model by
Additionally, power loss due to yaw misalignment in the controlled upstream turbine is modelled by scaling the effective wind speed as suggested by
The values for the corresponding model parameters utilised in the sub-models listed above can be found in Table
Model parameters for FLORIS used by P2 and P4. P2 used a set of in-house calibrated parameters; P4 used the default values. Parameters description
The wind farm optimisation has been executed using the algorithms available in the FLORIS library, using sequential least squares programming (SLSQP).
The control strategy at the wind farm level was aimed at maximising the total energy production of the wind farm per bin through yaw steering.
The wind rose has been divided into 144 sectors (2.5
Yaw misalignment for wake steering is limited within the range [0
Participant P3 used PyWake velocity deficit model – Fuga wake superposition – linear superposition, which is one of the basic assumptions in Fuga added turbulence – Fuga assumes uniform turbulence represented by vertical shear and surface roughness height; no added turbulence is needed for estimation of wake expansion and deficit decrease over distance; wake yaw deficit model – Fuga yaw deficit and Fuga deflection models; Fuga yaw deficit models implicitly the deficit and deflection models and calculates a set of lookup tables with the (yaw) longitudinal deficits UT and UL, due to transverse and longitudinal unit forces, respectively; Fuga deflection, like the Fuga yaw deficit, estimates tables for VT and VL, the transverse deficits generated by unit forces in the transverse and longitudinal directions; since Fuga is a linear model, the final deflected deficit is calculated by linear superposition of UT, UL, VT and VL.
The control strategy was applied to wind turbine and wind farm levels.
The control strategy at the wind farm level aimed to maximise the total wind farm power production by individual control of wind turbine's derating and wake steering. The set point of each turbine was determined using TOPFARM, the Python package developed by DTU for wind farm optimisation. The routine built a lookup table of set points (derating and yaw angle per turbine) dividing the wind rose in 120 sectors of 3
Wind turbine loads were not estimated by P3, but each individual turbine curve was constructed to operate at maximum power coefficient,
Participant P4 used FLORIS velocity deficit model – legacy Gauss (Gaussian model) by wake-added turbulence model – Crespo-Hernández model wake superposition – modelled with SOSFS (sum of squares freestream superposition) to combine the wake velocity deficits to the base flow field wake steering – wake deflection model by
Default parameters of the FLORIS sub-models were used in P4 simulation, which are also listed in Table
Wake steering was chosen by P4 as the control strategy. Considering the range of wind directions per inflow bin and that different wind directions have different potential for wake steering, P4 took an averaging approach when simulating the wind farm flow and power production, both for the non-yawed cased (normal operation) and the yawed case (with WFFC).
For a given inflow bin, 31 flow cases were considered, each representing a flow case with a wind direction in the range of [
For each of the 31 flow cases, the optimal yaw angle of all turbines was found using the SLSQP (sequential least squares programming) algorithm included in FLORIS, with the maximal iterations set to 200. The objective of this optimisation problem is to maximise the total power output of the wind farm:
Thus, for each inflow bin, there are 31 sets of optimal yaw angles, each with a different total power output
By considering 31 flow cases with different wind directions and solving the yaw optimisation problem separately for each inflow bin, P4 took an idealised or “greedy” approach that tends to explore the full potential of wake steering, since the effectiveness of wake steering can be quite sensitive to the wind direction. However, in real-life implementation, limits on the speed and accuracy of the yawing system, uncertainty of the measured inflow wind direction, and other factors can make the reported energy gain hard to be fully realised.
Participant P5 used an in-house analytical wake model which combines a Gaussian wake model with a recursive wake merging methodology velocity deficit model – legacy Gauss (Gaussian model) by wake-added turbulence model – Crespo-Hernández model wake superposition – modelled using a recursive wake merging methodology wake steering – wake deflection model of
The control strategy applied by P5 was a combination of wake steering and axial induction control of individual wind turbines within the farm to achieve the optimisation objective. For power maximisation, this included finding the optimal yaw and thrust set points of all the turbines within the farm to achieve maximum power production according to the equation
The FarmConners market showcases define only the layout of the wind farm and as simulation input the wind speed and direction as well as electricity prices. Any other detail of the implementation is left to the participants. Possible reasons for different results include (1) different models for wind farm flow and power production or different parameters for the same model, (2) different control strategies, (3) different optimisation problem formulation and bounds/constraints on design variables, and (4) different optimisation methods or different settings of the same algorithm. The percentage wake losses summarised in Table
Percentage wake loss for participants P2–P5 in the high-price scenario for 2020 and 2030, i.e.
Therefore, the WFFC algorithms detailed in Sect.
Before presenting the specific results per participant, a comparative analysis between 2020 and 2030 is performed for the weather and price conditions to support the subsequent discussions. The price time series for 2020 and 2030 were simulated for the same meteorological year assuming different energy scenarios at the systems level
As shown in Fig.
In conclusion, the price distribution among wind direction and wind speed bins is not equal in both years despite being based on the same meteorological time series. This reflects how the specific combination of site wind conditions and market prices shall impact planned operation and profitability assessments for a particular wind farm as the market evolves during its lifetime.
Figures
Power gain when using wind farm flow control (WFFC) in simulations with the wind inflow and price information for 2020 as shown in Fig.
Power gain when using wind farm flow control (WFFC) in simulations with the wind inflow and price information for 2030 as shown in Fig.
The results from P2 for 2020 are shown in the upper-left polar plot in Fig.
The discussed Figs.
There are, however, differences between the participants. First, the maximum power gain ranges from 2 % for P2 to 10 % for P5. Participant P5 reports also the largest range of power gains. The low levels of power gain estimated by P2 might simply be related to the smaller layout with overall lower wake losses. In the larger layout considered by the rest of the participants, higher wake losses are observed, which are expected to result in potentially higher WFFC gains. The polar plots for P2 and P4 for both years show local maxima for wind directions from the north (with bin centres of 15
This pattern of power gains across sectors does not appear for P3 and P5, which applied both wake steering and induction control to the TC-RWP layout. Having applied combined control strategies, their results show more variation between wind speed bins in the same sector. Instead of single or two adjacent sectors, advantageous wind directions seem to cover a broader range of wind directions.
An interesting observation is that P3 has the lowest and almost no power gain for western wind (sectors with 255
Figure
Energy gain (
The results for 2020 and 2030 are consistent for each participant reaching similar percentages in both years. Participant P2 achieves in both years a normalised energy gain of 1 %, P3 of 1.7 %, and P4 of 5.5 %. The energy gain reported by P5 in 2030 is also about the same as that in 2020. However, P5 is the only participant with somewhat different energy gains in 2020 and 2030. Participant P5 was also the one with the largest variance of power gains across the bins, reaching up to 10 %.
The normalised energy gains from P2, P3 and P5 range from 1 % to 2 %. Participant P4, reaching a much higher energy gain of 5 %, was the one with the highest wake loss (see Table
All participants achieve a minimum of 1 % energy gain. A recent expert elicitation revealed that wind farm operators and turbine manufacturers consider already an increase of the annual energy production of less than 1 % as sufficient to justify the field implementation of WFFC
The total income gain per participant during high prices for both years is shown in Fig.
Total income gain (
The income gain in 2030 is in general much higher than in 2020 because of the higher price level in that showcase set. Comparing the income gains per participant for both years, the numbers are consistent with the energy gains shown in Fig.
Income gain when using wind farm flow control (WFFC) in simulations with the wind inflow and price information for 2020 as shown in Fig.
Income gain when using wind farm flow control (WFFC) in simulations with the wind inflow and price information for 2030 as shown in Fig.
The income gain per bin shown in Figs.
Consequently for both 2020 and 2030 simulations, the sectors with estimated power gain are seen beneficial for the additional income via WFFC. However, the energy production is higher for the higher wind speed bins below rated, and, together with a larger number of samples per bin, the income gain for 9 and 11 m s
Most significant across all participants and both years is the bin with a wind direction of 315
Income gain per wind turbine when using wind farm flow control (WFFC) in simulations with the wind inflow and price information for 2020 as shown in Fig.
In order to investigate the contribution of the individual turbines to the total income gain, the reported values for 2020 and 2030 scenarios are broken down into the contribution of single wind turbines for the considered layouts in Figs.
Income gain per wind turbine when using wind farm flow control (WFFC) in simulations with the wind inflow and price information for 2030 as shown in Fig.
Although lower gains are reported due to the smaller layout investigated by P2, the normalised standard deviation (or the coefficient of variance) among the income gain per turbine reaches up to 60 %, which is equivalent to P4 results with the 32-turbine layout. The highest variance is observed in P3 results, reaching more than 67 %, and the lowest is reported by P5, with less than 40 %. This behaviour is the same for both 2020 and 2030, and it follows the trends presented in Fig.
For the 2020 scenario, wind turbines with negative income gain are WT1 for P3 and P4 and WT13 for P2. These wind turbines are located in the upwind direction for inflow from the west where the bins contain many samples and relatively high prices (see Fig.
As northwest winds become more frequent for high prices in 2030, the profile of the waked turbines changes slightly and only P3 reports an income loss for WT1 in Fig.
During both 2020 and 2030 scenarios, it is very interesting to compare P3 and P5 results as both have implemented combined flow control strategies for their optimisation in full TC-RWP layout. Figures
This article presents the results of the FarmConners market showcases, which are the first to study WFFC in simulations with variable electricity prices. The results from five participants are analysed and compared to demonstrate the potential benefit of WFFC in electricity market scenarios. The analysis starts at the individual turbine level with the examination of a method that applies different control strategies depending on the electricity price and finishes at the wind farm level with a comparative study of four different implementations of WFFC strategies simulating scenarios with high electricity prices in 2020 and 2030.
The main outcomes and observations of the FarmConners market showcases are summarised in the list below. They are given as qualitative statements due to the associated uncertainties in the reported benefits and sorted according to the steps followed in the analysis.
The reported numerical results are deterministic values for particular simulation environments. The uncertainties of the actual price signals and especially of wind forecasts can be in the same range as the reported gains here. While this is out of the scope of this conceptual study, an uncertainty quantification should be included in future investigations for a comprehensive evaluation of the estimated benefits per participating model and to identify the true value of the technology in the variable market scenarios. For the former, further reading is encouraged on studies investigating the sensitivity and optimisation of widely used WFFC-oriented models under input uncertainties as well as the embedded uncertainties in the model parameters
Moreover, the lack of participation in the showcase sets that also consider structural load alleviation motivates further research towards developing WFFC beyond power maximisation. These algorithms should be tested in extensive simulation studies covering a variety of future energy scenarios to evaluate their performance.
This will also help to quantify the benefit of maximising the income instead of the power gain as this study indicates. Unfavourable wind conditions in which the operational strategy only slightly increases the power gains can still result in high income gains if the electricity prices are high at the same time.
The notebooks for the market showcase results, including data snippets, can be achieved via the public repository of FarmConners market showcases (
All the data used in FarmConners market showcases are available for non-commercial purposes. Please contact us using the details provided under the wiki page of the FarmConners market showcases
KK, TG and IE organised the FarmConners market showcases, disseminated the showcases, prepared and shared the data sets, analysed the results, and drafted most of the manuscript. LAAR, MAS, JF, JM, VP and IS participated in the FarmConners market showcases, prepared and ran the simulations, and provided descriptions of their participating algorithms in Sects. 2 and 3. They are listed in alphabetical order of the last names.
At least one of the (co-)authors is a member of the editorial board of
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The authors would like to thank Michael Smailes from ORE Catapult for his support on the organisation and dissemination of the FarmConners market showcases. Additionally, the authors thank Kaushik Das, Matti Juhani Koivisto, Juan Pablo Murcia Leon and Polyneikis Kanellas from DTU Wind and Energy Systems for the data generation by simulating the energy scenarios using the DTU Balancing Tool Chain and their contributions to the definition of the FarmConners market showcases.
The FarmConners market showcases are organised and conducted under the FarmConners project, funded by the European Union’s Horizon 2020 research and innovation programme with grant agreement no. 857844. The contribution by Vasilis Pettas is funded nationally by Stiftung Energieforschung Baden-Württemberg (project name WOOP, FKZ: A341 21).
This paper was edited by Sara C. Pryor and reviewed by two anonymous referees.