Renewable energy forecasting is critical for integrating variable renewable energy resources like wind and solar power into the grid. Wind and solar power are heavily dependent on weather conditions, and, hence, their energy generation patterns are quite erratic. This creates problems not only for grid operators, who must account for these frequent fluctuations in their load management schedules, but also for power generators who have to pay hefty charges for deviation from scheduled energy generation. Accurate renewable energy forecasts help generators plan their operations better. For grid operators, forecasts help them predict the ramp up and down in generation and manage load accordingly.
This helps in reducing fuel costs, improving system reliability and minimising energy curtailment. For a particular region or state, renewable energy forecasting can be carried out in a centralised as well as decentralised manner. Many in the industry believe that centralised forecasting is more suitable for ensuring overall grid stability as it provides energy forecasts for all the projects in a specific region. Decentralised forecasts provide only plant level information. Thus, for grid operators, centralised forecasting would offer better consistency since it uses uniform methodology for all forecasts. However, even in centralised forecasts, it is vital to use a mix of techniques to improve accuracy and prevent errors.
Time horizon is often a key factor in determining forecasting accuracy, which tends to decrease over a longer time as per a 2020 report by the International Renewable Energy Agency (IRENA) titled “Advanced Forecasting of Variable Renewable Power Generation”.Very short-term forecasting that ranges from a few seconds to minutes to an hour helps in real-time power despatch and maintenance of smooth grid operations. Grid load management requires short-term forecasting ranging from 1 hour to 24 hours. Mediumterm forecasting, ranging from one week to one month, is useful for planning grid maintenance, and long-term forecasting helps authorities in overall generation and transmission planning.
Operators can procure weather forecasts from various vendors and meteorological research institutions, or they can develop their own methodology. According to the US-based National Renewable Energy Laboratory, forecasting can be carried out using physical methods or statistical methods. Physical methods use weather data and numerical weather prediction (NWP) models to estimate energy production. Statistical methods use NWP data along with historical and real-time generation data to arrive at precise estimates.
A July 2019 technical guide titled “Using Forecasting Systems to Reduce Cost and Improve Dispatch of Variable Renewable Energy” by the Energy Sector Management Assistance Program (ESMAP) of the World Bank describes various methods of forecasting variable renewable energy. The key techniques described in this guide can be grouped under physical methods, statistical methods and hybrid methods.
Forecasting models based on physical methods
Weather data like air temperature, pressure and surface roughness is used to make meteorological predictions that complement local conditions. Such models include NWP, remote sensing and local sensing. NWP models use weather data from radiosondes or weather satellites and mathematical representations of atmosphere for predictions. Remote sensing models use data on satellite-based weather measurements, which are acquired from various sources that can help provide good estimates without using many local sensors.
Local sensing models use high resolution spatial and temporal data that is captured from points located at or close to the site to assess actual field conditions.
Forecasting models based on statistical methods
Statistical methods are based on patterns and rely on gathering large amounts of historical data to assess and predict energy output. In statistical methods, a reference model or baseline is often required for comparing and predicting output. This forecasting reference is used in many complex statistical methods to accurately predict output. Persistence models are simple statistical models that assume that if the conditions remain unchanged between two time periods, the forecast will also remain the same. Time series-based modelling and statistical methods use data collected over a period of time to assess and predict future output. More advanced artificial intelligence (AI) methods use neural networks instead of regression models to predict output using historical data, weather conditions and the relationship between the two. There has been large-scale uptake of AI tools for both solar and wind power projects to enable more precise forecasting and scheduling. These advanced digital techniques are faster and more precise in assessing the large volumes of data available. Now, many operators are even using digital twins to further improve forecast accuracy. Forecasting models based on hybrid methods: As the name suggests, hybrid models use both physical and statistical methods to forecast output. As they combine the characteristics of both, they are more effective and accurate.
Key challenges and the way forward
With more renewable energy being fed into the grid, it is becoming quite critical to incorporate advanced forecasting and scheduling techniques to ensure grid stability. However, there are still many issues that need to be addressed before proper forecasting tools can be deployed. First, there remains some level of uncertainty in accurate weather forecasting that needs to be taken care of as solar and wind energy generation depend significantly on the weather. Second, refined regulatory frameworks must be designed that take grid balancing and ancillary services into consideration. Third, grid operators and personnel need to be trained regularly regarding the latest developments in this space and more efforts need to be directed towards incorporating advanced AI tools for forecasting. Finally, with large volumes of distributed renewable energy assets and with electric vehicles being connected to the grid, forecasting and course correction in real time with advanced tools and techniques will play a vital role in stable grid operations