How Weather Forecasting Will Drive Down Electricity Bills in the Renewable Energy Future

Renewable energy sources like wind and solar power are inherently variable due to their dependence on natural phenomena. For renewable systems to be integrated into our energy mix, comprehensive system management must be in place, where energy sources are accurately scheduled ahead of time to ensure electricity demand is always matched by supply. This therefore requires precise forecasting of renewable generation. The dependence of these systems on the weather means that accurate meteorological data is essential as an input into renewables forecast modelling. As renewables increase in the energy mix globally, the need for accurate and valid weather data is becoming increasingly more important.

The Rise of Renewable Energy

The renewables revolution is well under way. Today, renewable energy (excluding hydro) only accounts for 8% of global electricity generation [1], but its current growth is rapid - almost 40% of the additional global power generation in 2016 was from renewables. The fastest rises compared to 1990 levels have been from solar photovoltaic (PV) (45.5%) and wind power (24%) [2], due to national and international policies, and sharply falling costs. Record-low solar PV auction prices below 3 cents per kilowatt-hour have occurred across the globe, including in Argentina, Chile, India, Jordan, Saudi Arabia and the United States. Similar records have been seen for electricity from wind developments, spearheaded in Chile, India, Mexico, Morocco, Denmark and the Netherlands.

The Rise of Electricity Bills

But does lower renewable kWh prices result in cheaper electricity bills for everyday consumers? End-user electricity prices have been on a continuous upward trend in many countries. In the UK, there was a 63% rise in electricity prices between 2004 and 2014 [3]. A similar picture has been seen across Europe and further afield. According to the US Energy Information Administration, US utility residential electricity rates have increased by around 20% in the last 10 years [4].  The most influential driver of this has been the rising price of natural gas, and this trend is expected to continue. For electricity bills to come down, cheaper energy sources must be realised – can renewables provide this?

The Renewable Energy Future

The growth of renewables is expected to continue. The International Energy Agency (IEA) predict that by 2040, the installed capacity of solar and wind will increase by a factor of 14 and 4, respectively. This will result in the share of global electricity generation accounted for by solar and wind rising from 5% today to 34% in 2040 (figure 1) [1].

Figure 1. Global electricity generation mix to 2040. Source: Bloomberg New Energy Finance, New Energy Outlook 2017. 

But as governments remove feed-in-tariffs and other supportive incentives, renewables need to be able to stand on their own feet. The only way this can happen is if they are cheaper than conventional sources. Ever more efficient solar PV modules and wind turbines will reduce costs, but price decreases must also come from more efficient operating and maintenance procedures. Yet due to the uncontrollable variability of wind and solar PV systems, their integration into power systems poses challenges.

Balancing Supply and Demand

Supply must always balance demand. This is achieved by grid operators scheduling power generation in advance to match the expected demand at all times. The limited ability to anticipate variable energy sources on a minute-by-minute basis results in supply and demand deviating from each other in real-time and as a result, grid operators must have additional power plants (usually fossil-based) on continual standby.  These balancing costs can be substantial, and ultimately, greater imbalancing costs lead to higher electricity prices for consumers. Ofgem, the UK’s electricity regulator, notes that balancing costs of around £850,000 per year adds £9 to the average UK electricity bill [5]. Worse still, getting the balance wrong leads to blackouts or power surges, further impacting energy costs.

Power system flexibility is therefore required – the capability to maintain continuous grid balance in the face of rapid and large swings in supply or demand. Without it, wholesale prices of renewable energy become depressed, particularly when wind and solar production exceeds the demand on sunny, windy days.

Renewable Energy Forecasting

To achieve power system flexibility, comprehensive forecasting of renewable systems is essential. Traditional forecasting approaches in the energy sector have mainly been statistical techniques, comprising of time-series methods or regressions. However, with the advances in big data, processing capacity and tools to manipulate these datasets, artificial intelligence-driven modelling is starting to take more precedence. This is due to its flexible nature and ability to handle non-linearity, which is inherent in many energy forecasting problems, especially those required at a granular time-scale. Figure 2 displays some of the inputs for solar generation forecasting using artificial intelligence-driven modelling.

Figure 2. System architecture of solar PV generation forecasting using artificial intelligence-driven modelling.

The Need for Weather Data

The power produced by a solar plant is strongly dependent on the global horizontal irradiance (GHI) – the total amount of shortwave radiation received from above by a horizontal surface, at an instantaneous point in time. GHI is measured by on-site pyranometers and expressed in Watt per square metre (Wm^-2). GHI values are then adjusted as per the tilt and orientation of the solar panels – the global tilted irradiance, GTI (figure 3).

Figure 3. Global Horizontal Irradiance (GHI) and Global Tilted Irradiance (GTI).

GHI, and therefore solar PV output, has seasonal and diurnal cycles. GHI also dramatically fluctuates on much shorter timescales due to sharp changes in cloud cover and atmospheric conditions.

The power produced by wind turbines is dependent on the cube of wind speed. Turbines have a lower limit at which no power is generated, and an upper limit at which turbines are shut down to avoid damage and failure. Although less pronounced compared to solar, wind power may fluctuate on seasonal timescales, down to very short-term fluctuations due to localised wind speed changes.

All fluctuations in GHI and wind speed must be captured to high precisions – renewable energy forecasting is only good if the weather data feeding into the forecast models is good.

Different Weather Sources for Different Forecast Lead Time

The best source of weather data for forecasting of renewables depends upon the forecast lead time required, particularly for solar power forecasting (figure 4).

Figure 4. Conceptual plot of forecast lead time vs. forecast skill for different sources of weather data: ground observations, satellite imagery or NWP model output.

At sub-hourly lead times, the highest solar forecast skill is provided by methods based on on-site ground observations, such as from real-time infrared imagery.

But forecasts based on these local observations become less skilful as the forecast lead time increases, which is when satellite-based methods for computing GHI become useful. Geostationary satellite imagery (e.g. from Meteosat-10 over Europe and Africa) allows clouds further afield to be captured, and cloud motion vectors enables the future positions of clouds over the site of interest to be inferred.

Obviously, clouds evolve, so the use of satellite imagery to compute solar radiation is limited to around 7 hours ahead. After this, weather forecasts from numerical weather prediction (NWP) models become most effective for solar forecasting. NWP models incorporate our physical and dynamical understanding of the atmosphere to forecast the future state of weather using mathematical equations and huge computational power.

NWP Forecast Uncertainty

However, NWP forecasts are not perfect. It is impossible to predict with certainty for every point in space and time, but there is still plenty of room for improvement.

The main limit to their accuracy is the spatial resolution of the NWP model. The finer the spatial resolution, the smaller the grid cells that divide up the atmosphere, and therefore the more representative the predicted variables will be to every point in that cell. This itself is limited by computational power, despite the NWP models like the GFS and ECMWF running on the most super of supercomputers. Consequently, there’s a lot of information the models struggle to infinitely resolve. This includes all atmospheric phenomena that are smaller in size than the grid resolution, such as turbulence and small cloud features like cumulus clouds. Surface conditions are also not fully resolvable: is the surface cover vegetation or water or concrete? What is the topography of the surface?

Anything that cannot be resolved are parameterised in the model, which means that they are approximated using over-simplified assumptions. A cumulus cloud, which has an average width of about a 1km, must be parameterised, and hence may be poorly represented by the model. This provides a large source of uncertainty in solar forecasting; solar plant output can dramatically fall within seconds as a cloud moves over and blocks out the Sun’s radiation.

Good observational data is also essential – it’s a case of garbage in, garbage out. Where observational data is not available, interpolation is used to fill the grid with initial conditions. But any errors in the initial observational data will result in errors in the interpolated data too, and this may lead to large uncertainties creeping in. Chaos theory (the butterfly effect) means that the forecast output is highly sensitive to the initial and boundary conditions and so may have very different outcomes with just slight differences to the model inputs. Chaos will increase with time, so forecasts become increasingly more uncertain the further out they are made for.

High model generation frequency (how often the model runs) helps to eliminate this issue by re-running the model with the latest observational data. For example, the GFS runs every 6 hours, whilst the Met Office’s UKV model runs every 3 hours. As with the spatial resolution, model run frequency is limited by computer power.

Reducing Overall Costs

The savings due to good energy forecasting can be substantial - five-fold reductions in the costs of hourly imbalance penalties levied on generation have been seen in the past. The UK's Central Electricity Generating Board claimed that if perfect wind power prediction was possible, it would add 10% to the value of wind power on the system. 

Conclusion

With an increasing share of wind and solar in the generation mix across the globe, the overall generation variability is set to rise, and hence the challenges of balancing supply and demand will become greater. But while wind and solar power cannot be dispatched, they are schedulable, with the support of accurate weather forecasting. Ground instruments, satellite imagery and NWP models are all essential when forecasting renewable generation in real-time. As accurate forecasting of renewables becomes more prominent, renewable energy prices will fall, and this should lead to lower electricity bills for consumers.

References

[1]	Bloomberg New Energy Finance, “New Energy Outlook,” 2017.
[2]	International Energy Agency, “Renewables information 2017,” 2017.
[3]	Ofgem, “Annual Report 2015-2016,” 2016.
[4]	U.S. Energy Information Administration, “Annual Energy Outlook,” 2017.
[5]	Ofgem, “Balancing and Settlement Code (BSC) P333,” 2016.