Yield Optimization Based on Wind Resource Abstract: Key words: yield optimization, wind energy, sustainability, maintenance. 1.

Size: px
Start display at page:

Download "Yield Optimization Based on Wind Resource Abstract: Key words: yield optimization, wind energy, sustainability, maintenance. 1."

Transcription

1 Yield Optimization Based on Wind Resource João Amaral Neto 29 June 21 Abstract: In one of the largest renewable energy sectors, the Wind, research on optimizing scheduled maintenance on wind turbines is nearly nonexistent. Unplanned maintenance is commonly synonym to large energy loss since the wind turbine must be stopped nearly throughout the whole duration of the maintenance procedure. All parties evolved in the sector are hindered by this fact. It is therefore not only in all the sectors players but also the general publics interest to optimize this process for a more sustainable world. Responsible for all the calculations is a model, which was fully developed for this study and is part of it. Having considered several programming languages the choice was Excel (VBA); being this software spread worldwide it encourages the models global implementation. Easy to use, versatility and accurate and prompt results were the guidelines for its developments. A weather forecast is the single necessary input. Running the model on two different wind farms gave conclusive results. The energy loss was reduced up to 71%. Also in some cases the time frame was cut up to 62%. Even with these promising figures energy loss must be significant in order to have a real economical impact. Nevertheless the model unveils the most convenient schedule for maintenance and its implementation is exclusively beneficial. Key words: yield optimization, wind energy, sustainability, maintenance. 1. Introduction Weather Forecast Power Curve Yield Optimization Model Algorithm Assumptions Priority Results Wind farm X Wind Farm Y Introduction Wind energy is the fastest growing renewable energy source: 27% over the last decade [1], and is currently one of the largest. Due to the inevitable perishability of fossil fuels [2] it is crucial to find sustainable, green and affordable solutions. The technological developments in the wind sector enable this energy source to compete against the traditional energy sources using fossil fuels. Wind energy offers the same output (electricity) without the heavy bourdon of any gas emissions. Wind energy may supply up to 2% of the electricity consumption of a country [3]; UK s transmission system operator (TSO) reported that the scattering of wind farms within a country produces a stable and fairly constant electricity output. Striving for an increasingly competitive sector it is vital to reduce both the operational costs and the down time of the every wind turbine generator (WTG). Reference [4] discusses the implementation of monitoring and fault diagnosis allowing early intervention. With the increasing awareness of the need to incorporate renewable electricity in the general consumption the focus turns once again to the premature fault detection in the WTG. A fault detection system (FDS) is discussed in [5,6]. Hence it is clear that the wind energy is no longer under scrutiny instead the focal point is to make it progressively more competitive. This paper is centered in the optimization of the turbines scheduled maintenances aiming towards a positive outcome. Its purpose is studying the impact of energy loss due to scheduled maintenance and finding solutions in order to mitigate the energy loss. The result may come in a reduction of the energy loss, a lighter economical weight or a reduction of the total time required for the whole process. The available literature is concentrated on fault detection or monitoring on wind turbines however not necessarily on optimizing the process from the wind resource point of view. This paper intends to fill this void. 1

2 2. Weather Forecast Most of the available predictions are not suitable for the optimization model. Since the study objects are wind turbines the wind forecast must be at hub height, which is about 8 meter high. Furthermore the predictions must also provide valuable information in the upcoming days extended at least for 3 but preferably 7 days (middle range forecasts), the final requirement is that the location where the wind is predicted must be coincident with the wind farms location. The Portuguese research group METEO-IST provides an ideal forecast that suits the optimization model. The relevant weather models used by this research group are the PSU/NCAR mesoscale model (MM5) [5], the Weather Research and Forecasting model (WRF) [6] and the Global Forecasting System (GFS) [7]. This investigation group runs both MM5 and WRF based on the coarse gridded GFS. For the MM5 they use two nested grid domains with grid sizes of 27km*27km and 9km*9km for downscaling while for the WRF the grid size is 3km*3km. The WRF model is exclusively for interior of the country. In conclusion the relevant model of the current project and conveniently available online is the MM5. The verification and convenient applicability is discussed in two papers from METEO-IST [13, 14]. 3. Power Curve The power output versus the wind speed function is called the power curve as shown in Figure 1. The cut in speed is the minimum speed which the turbine produces any energy. Cut out speed is the speed at which the wind turbine stops producing any energy and stands by. Some wind turbine generators have pitching systems that allows the blades to change the angle of attack with respect to Power [kw] Figure 1 Electrical power curve for V9 3.MW [9] Wind Speed [m/s] the wind; this system stalls the blades creating a null lift coefficient, stopping the blades rotation [8]. Rated power is the maximum power a wind turbine is designed to produce. The power curve is dependent on the air density since the torque is function of the density and the electrical output is function of the torque. 4. Yield Optimization The yield approach s goal is to minimize the production loss hence increasing the productivity, completed exclusively by optimizing the maintenance scheduling. It is outside the yield scope to improve the turbine s performance and efficiency. The call for this new approach is directly linked with the market s trend, currently leaning towards a production guaranty rather than the common availability guaranty. The production mindset stands for guarantying a given production. In order to create a model able to optimize the scheduling of the maintenance several wind farms were studied in Portugal such as: Serra de Arga, Toutiço, Pampilhosa da Serra, Serra dos Candeeiros and Pedras all scattered around the country s mainland except for the last one located in Madeira. 4.1 Model Aiming at widespread implementation the tool was developed in Excel, included in the Microsoft Office package globally used. For the present cases the run time of each optimization is nearly negligible around one tenth of a second. The model has an interface with intuitive commands, easy access and extreme flexibility in order to adapt to any case and includes vast information regarding the scheduling and costs of the desired maintenance. ρ=1,3 kg/m3 ρ=1,15 kg/m3 ρ=1,27 kg/m3 2

3 Algorithm Firstly the 1 or 15 minute average, 3 or 7 day weather forecast data must be averaged into hourly data. Here the only variables necessary for the model are the wind speed and the air density. Once the hourly wind speed averages are calculated the next step is finding which is the working hours that have the least energy potential, in order to do so the model calculates the average wind speeds for a whole working day starting at 7, 8 or 9 am and identifies the lowest. Then within the whole forecast, either 3 or 7 day, the model identifies the day with the lowest wind speed average. At this point it is identified the day with lowest wind speed and at what time within the day that occurs. Next the model reads the chosen parameters, the number of technicians, the priority level, the price of electricity and the current date. Then using the hourly wind speed averages and the power curve the model calculates the hourly energy produced; also using the same wind speed averages and the tariff the model calculates the associated hourly price of energy. Finally the model is able to suggest the maintenance scheduling according to the priority level. This is displayed in a calendar. Thereafter the following days with the lowest wind speed averages are identified and the maintenance scheduling is equally displayed in the same calendar. In conclusion when the schedule is complete the model calculates the total energy not produced and its cost summing the previously computed hourly energy produced and hourly price of energy respectively. Also the cost of technicians is calculated according to the amount of working hours. The last calculation is the difference between the overall cost of the technicians and energy not produced Assumptions The assumptions considered should be regarded as guidelines. Often these are not made to simplify the problem they are instead a constraint. However some correspond to situations impossible to reproduce. The technicians work 9 hour days with a mandatory one-hour break (lunch). The starting hour may vary from 7, 8 or 9 a.m. Extra hours are available at the cost shown in Figure 2. HBS [ ] 2,5 2 1,5 1, Hours [h] Figure 2 Technicians Hourly Base Salary (HBS) According to the Portuguese labor law article 228º in a week day no more than 2 extra hours and during the weekly mandatory or complementary resting days no more than the regular daily working hours. The recent V9-3MW and all WTGs after it require a single yearly Major maintenance while the previous ones requires a Major and a minor both yearly maintenances typically 6 months offset. The Major maintenance requires 3 technicians on 3 non-consecutive days and the minor also requires 3 technicians for 2.5 nonconsecutive days. There is a 1 month tolerance for each maintenance. The energy production will be estimated using the wind prediction from IST and the Vestas power curves. The WTG is stopped for the whole duration of the maintenance process. The weather prediction is delivered either a 7-day or a 3-day prediction Priority In the model for a convenient visualization, the priority is color-coded. For each priority level different pre requisites (translated into wind speed) must be met since maintenance does not always have the same urgency. The requirements can be seen in Table 1. high priority lower than 2 [m/s] medium priority lower than 6.8 [m/s] low priority lower than 5.5 [m/s] Table 1 Priority requisites Week Day Week Day (extra hours) Saturday Holiday or Sunday 5. Results Two arbitrary wind farms were selected, X and Y both with V9-MW WTG and running at least since 26. For the first one all its WTG were considered for the optimization

4 while in Y only the two most energy producing ones were taken into account. Firstly the energy not produced was calculated knowing before hand the exact dates of maintenance and the historical wind speeds. Once this was completed the first optimization was computed considering the same time frame as the actual maintenance but progressively choosing the lowest wind speed days. Then feeding the historical wind speed data to the model it ran on the three priority levels (always considering the initial time frame from the actual maintenance). Unfortunately due to software updates 28 data is not available. Leaving 27 and 29 to be considered. 5.1 Wind farm X In 27 and 29 the complete duration of the entire maintenance process was roughly 13 weeks. An average intervention is shown in Figure 3. All WTGs have similar plots when maintenance is being performed. All variables except for Estimated Possible Production are available in the SCADA system. Possible power is an estimation of the production, completed from an internal routine in the wind turbine. Estimated possible production is determined directly from the power curve with the wind speed and the air density. Wind speed is measured with an ultra sonic anemometer placed on the nacelle. Grid production is the actual power production of the wind turbine. The results from the actual and the 1º optimization are described in Table 2. For the 27 maintenance the priority level low, medium and high concluded the whole process in 11, 8 and 5 week respectively. The results from the calculation of the energy loss in all optimizations are shown in Table 3. The duration of the maintenance was 16, 1 and 6 week for high, medium and low priority respectively. For 29 the results from the whole maintenance process are presented in Table 4. Plotting the results graphically clearly show the gap between the actual maintenance and all other optimization cases, seen in Figure Wind Farm Y The same procedure was completed for this wind farm. First the actual energy loss was computed and later the first optimization. The results are shown in Table 5. The duration for the 27 maintenance in low, medium and high priority was 4, 3 and 2 weeks respectively. Hence now computing the results with the model yields Table 6 for 27. On the other hand for 29 the maintenance duration was 5, 2 and 2 weeks for low, medium and high priority respectively. The Table 7 shows the results from all the optimizations in this year. And finally the results for both 27 and 29 in all priority cases is can be seen in Figure 5. Power [kw] :1 1:2 1:3 1:4 1:5 11: 11:1 11:2 11:3 11:4 11:5 12: 12:1 12:2 12:3 12:4 12: Wind Speed [m/s] Possible Power Grid Production Estimated Possible Production Wind Speed Total Energy not produced: kwh Time Figure 3 15/11/27 Wind farm X maintenance Year # of Energy not produced [kwh] hours [h] actual 1 st optimization Difference [kwh] % difference % % Table 2 First optimization in 27 and 29 in wind farm X 4

5 27 optimized % low % medium % high % Table 3 Total energy loss 27 in wind farm X 29 optimized % low % medium % high % Table 4 Total energy loss 29 in wind farm X Energy [MWh] 15, 1, 5,, Actual 1º Optimization Low Priority Medium Priority High Priority Figure 4 Energy loss 27 & 29 in wind farm X Year # of hours [h] Energy not produced [kwh] actual 1 st optimization Difference [kwh] % difference % % Table 5 First optimization in 27 and 29 in wind farm Y 27 optimized % low % medium % high % Table 6 Total energy loss 27 in wind farm Y 29 optimized % low % medium % high % Table 7 Total energy loss 29 in wind farm Y Energy [MWh] Actual 1º Optimization Low Priority Medium Priority High Priority Figure 5 Energy loss 27 & 29 by year in wind farm Y

6 6. Conclusions The whole scope of this paper is based on a single question, whether it is possible to optimize the maintenance procedures, which ended with an affirmative conclusion. Some results show maintenance was done in days where the wind resource was abundant. Also it seems large number of extra hours were necessary. Both previous requirements need to be mitigated in order to lower energy loss optimizing the process. Once this it is kept in mind optimization is in progress and finding the optimal point is the drive force. In 29 in Wind farm X there is a maximum reduction in energy loss was in low priority, grater than 75%. It results in an energy loss drop of over 1MWh, better result than the first optimization. For medium and high priority levels the whole procedure in under the time frame initially required. In 27 in the same wind farm low and medium priority show a decrease in energy loss of 53% and 38%. The two priority levels conclude the maintenance work under the initial time. Still considering wind farm X the energy loss from 27 are lower than from 29. In the latter year the wind speeds during the 13- week initial maintenance plan were significantly higher then for the first year. This leads to a higher energy loss in 29 however it introduced more options for optimizing. For wind farm Y in 27 the model in low priority lowered the energy loss by 71%. Medium and high priority both managed to keep the energy loss below the actual figure. On the other hand in 29 also all priority levels reduced the energy loss comparing to the actual case. The power curve proved to be an accurate tool to estimate the energy production when considering short wind speed averages, which was the case. Increasingly larger errors are introduced with longer wind speed averages, ultimately underestimating the energy production. Even though the economical benefit is minor the broad implementation of this model may lead to the sum of several minor benefits eventually adding to a significant one. Nevertheless its implementation is exclusively beneficial both for the promoters and the entity providing the maintenance service. Currently sustainability in all its forms is a greatly appreciated concept to be associated with. In order to be coupled with it one must proactively seek better, greener, mutually beneficial solutions and the implementation of this model is a step in that direction. References [1] G.M. Joselin Herbert, S. Iniyan, Ranko Goic, (21) Performance, reliability and failure analysis of wind farm in a developing Country. Renewable Energy, doi:1.116/j.renene [2] Meng QY, Bentley RW (28). Global oil peaking: responding to the case for abundant supplies of oil. Energy 33 p [3] (28 June 21) [4] Y. Amirat, M.E.H. Benbouzid, E. Al-Ahmar, B. Bensaker, S. Turri (29). A brief status on condition monitoring and fault diagnosis in wind energy conversion systems. Renewable and Sustainable Energy Reviews 13 p [5] (28 June 21) [6] (29 June 21) [7] (29 June 21) [8] The British Wind Energy Association. Wind Turbine Technology. (Sep 25) [9] Vestas Wind Systems A/S. General Specification V9-3MW VCS 5Hz. (1 October 28) 6