Wind Power Generation

Installed capacity

In 2020, onshore wind electricity generation increased annually by 144 TWh (+11%) and capacity by 108 GW, twice as much as in 2019. China’s onshore wind capacity tripled from 2019 to 69 GW, whereas the United States’ capacity doubled to 17 GW: these two countries together accounted for 79% of global wind deployment (cf. Figure 1). Offshore wind generation growth amounted to 25 TWh (+29%) in 2020, with capacity additions of 6 GW, the same as in 2019. Overall, 1 592 TWh of electricity were generated from wind installations in 2020, 12% more than in 2019 (Bojek and Bahar, 2021)

Net Zero emissions by 2050 Scenario, generation must increase on average 18% per year during 2021-2030. It is also necessary to raise annual capacity additions to 310 GW of onshore wind and 80 GW of offshore wind. The most critical areas for improvement to boost wind electricity generation are cost reductions and technology improvements for offshore wind and facilitating permits for onshore wind (Bojek and Bahar, 2021). As with the solar resource, wind resource is also very abundant (Figure 2).

Figure 1. Global wind energy installed capacity for 2020. Source: Our World in Data

Figure 2. Global mean power density for wind generation. Source: Global Wind Atlas.

Wind generation and grid management

As distributed energy mix becomes more complex, there are increasing challenges to maintain the stability of the grid. There can also be cascading effects if generators simultaneously disconnect due to a wind drought. Both grid managers and the wind industry realized that wind turbines needed to be modified and actively managed.

Grid operators have been issuing new rules requiring every wind farm to stay online during grid disruptions and to regulate its output power to keep its characteristics within narrowly specified ranges. These requirements govern the voltage, frequency (cycles per second), and shape of the oscillations of the alternating current (AC) electricity (Acee, 2019).

Wind turbine developers are responding to these new rules by equipping the turbine with new control capabilities and operating procedures. For example, to be able to provide extra power quickly on demand, wind generators must be generating below their full capacity, thereby creating the “headroom” to respond for a call from the grid for extra power. The most recently installed wind turbines can also contribute extra electricity to the grid to compensate for a falling frequency by using power electronics to reduce the rotational speed of the blades and other rotating components (Acee, 2019).

However, when wind speed is high and air temperature is low, more power can transit in the lines, which can allow transmission managers to accept more power from wind farms. This is known as dynamic line rating (DLR), but in order to do this high spatial resolution data and forecasts, mainly for wind speed and air temperature are needed (Troccoli et al., 2022).

Wind variability

Wind power production exhibits variations on all timescales, however there are distinct peaks of wind speed variability (Gouzden et al., 2020):

Turbulent peak in the sub-second to minute range
Diurnal peak, driven by the heating and cooling of the earth surface
Synoptic peak which depends on changing weather patterns with a scale of variation which ranges from days to weeks
Annual and inter-annual (or decadal) variability

Variability of wind power production might be classified into regular cycles (diurnal and seasonal/annual), and irregular cycles (synoptic, inter-annual).

However, the power generated by wind turbines varies rapidly due to the fluctuation of wind speed and wind direction. It is also dependent on terrain, humidity, date and time of the day, making grid management challenging for distribution networks. And of course, there is also geographic variations, with high wind power along most coastal areas (see Global Wind Atlas).

Climate change is also affecting atmospheric dynamics and changing wind patterns, making it essential to assess how this affects the wind speed and other factors for a given utility. Wind turbines are increasing not only in capacity but also size, with some offshore turbines currently reaching 100 m heights, which makes them even more sensitive to extreme weather events.

Wind forecasting

Depending on different functional requirements, predictive horizons could be divided into four major time scales (Table 1, Hanifi et al., 2020).

Models can be further divided into persistence methods, physical methods, time series models and machine learning models. Their differences are in the required input data, the accuracy at different time scales and the complexity of the process (Nazir et al., 2020). It is worth noting that the same model categorization applies to other energy conversion models, including solar PV power and hydropower.

Persistence Methods

Used as a reference output to identify improvements in other methods. It assumes that short-term wind power forecasting in the future will be equal to present measured power. The accuracy of this method quickly deteriorates with increasing time span. However, it is a simple and economical method, as it does not require parameter evaluation nor external variables, aside from the local measurements (Nazir et al., 2020).

Physical Methods

Physical methods use detailed physical characterisation of the area, such as obstacle avoidance, local roughness and surface changes, terrain effects, wake effect, daily change of the thermal stratification, weather forecasting data of temperature, pressure, acceleration or descent, the scale of local wind speeds within a windfarm, the fan power curves, and wind farm design, etc. These variables are typically used in complex mathematical models, such as Numerical Weather Prediction (NWP), to determine wind speed and direction (as well as many other meteorological data). Then, the predicted wind speed is fitted to the wind turbine power curve (normally provided by the turbine manufacturer) to forecast wind power (Nazir et al., 2020).

Statistical Methods

These methods are based on non-linear and linear relationships between meteorological data (such as wind speed, wind direction and temperature from observations and/or NWP models) and the generated power. Historical data are needed to train the model, which is later tuned by comparing the model prediction and the on-line measured power. This method is generally easy to model and relatively inexpensive. It is typically for short term time spans. Statistical methods can be divided into two main subclassifications: time series based and machine learning models (Nazir et al., 2020).

Hybrid Approach

The combinations of different forecasting methods, such as machine learning and fuzzy logic models, are called hybrid approaches. The main aim of this method is to retain the merits of each technique and improve the overall accuracy. In statistics and machine learning, diverse predictive models are often developed by using multiple algorithms and different training datasets. This process is often named as ensemble modelling, which is a more advanced type of hybrid forecasting (Nazir et al., 2020).