AI Technologies Against Imbalances: Modern System For Forecasting Electricity Output From RES

Since the introduction of imbalance responsibility for renewable energy producers, interest in electricity forecasting systems has been increasing every year. The main requirement is forecast accuracy to minimize imbalances. What are the existing forecasting methods? How is forecast accuracy achieved? What is the role of a competent and reliable provider of renewable energy forecasting services? Hryhorii Hutsol, Head of the Electricity Volumes Forecasting in KNESS Group, dwells upon these issues and more.

How is the forecast accuracy achieved?
A number of factors should be taken into account when forecasting the electricity output of solar power plants: the plant’s shape and location, the plant’s length from north to south and from west to east, the panels’ angle of inclination, the dust content of the panels, the range, type and condition of the plant’s equipment. However, the most important factor that affects the generation forecast accuracy is the meteorological parameters forecast, in particular solar insolation on the surface of the PV module, which, in turn, is crucially influenced by the level of cloudiness since the value of the insolation level on an inclined plane can be calculated with little or no error. Analytics shows that 95% of all deviations in electricity volumes between forecast and actual values are caused by an error in the meteorological parameters forecast.

What are the forecasting methods?

Total Sky Imagery Method — allows for quite accurate electricity generation forecasting, but only for half an hour in advance. The method involves taking sequential photos of clouds, analyzing them (thickness, type, speed, movement direction), and calculating generation based on the data obtained, taking into account the irradiation power. However, the forecast accuracy for more than 2 hours is reduced because it is difficult to predict the cloud movement and changes in cloud geometry. In addition, the technical equipment used in this approach is rather expensive.

Method for analyzing satellite images — allows to obtain data for cloudiness analysis over large areas and calculate a 6-hour forecast. Forecasts for the day solar intervals are obtained by processing real-time satellite images from several geostationary satellites. This method requires the use of powerful computing technology.

Numerical Weather Prediction, or NWP, involves the use of specialized software that generates a weather forecast using mathematical and physical algorithms. This requires a large array of accurate initial data collected during meteorological research. Thus, it is possible to forecast the generation of a solar power plant more than 2 weeks in advance.

KNESS Group started developing software for solar power plants’ generation forecasting back in 2016, when the first drafts of the mechanisms of the new electricity market model appeared, according to which it was envisaged to establish responsibility for settling imbalances for producers under the feed-in tariff. Accordingly, the methods described above were also analyzed and tested by our team to create an optimal solution for forecasting the volume of electricity supplied by RES.

Which forecasting method is the most optimal, in our opinion?

KNESS forecasts electricity generation using its own developed PV.Forecast system, which integrates a set of mathematical and statistical models combined with artificial intelligence. This system uses various inputs, such as historical power generation data, a set of meteorological data, as well as other factors that may affect power generation, including power outages, power plant maintenance, etc.

The advantage of this system is that the neural network is trained with the data of each specific energy facility, taking into account its characteristics, and produces forecasting results which are close to accurate for different time horizons.

The use of artificial intelligence makes it possible to increase the accuracy of PV generation forecasts with an increase in the complex lifetime. The average annual error of the forecasting system for a new station is about 18-20% for forecasting the next day’s generation. Refinement of the forecast schedule reduces the error to 9-12%.

What are the key stages of forecasting using a neural network in PV.Forecast?

Data collection: collecting various data (Meteo Soft) related to electricity production and factors that may affect it, such as weather, equipment characteristics, power plant capacity, etc.

Data preparation: data processing, normalization, and validation for further use by the neural network.

Neural network building: creating a neural network using machine learning algorithms such as recurrent neural networks (RNN) and deep neural networks (DNN).

Model training: training a neural network using historical data to determine the connection between input data and electricity productio.

Forecasting: using a trained model to predict future electricity production based on input data (Forecast Soft).

Benefits of the PV.Forecast system by KNESS

High accuracy: neural networks can detect complex, non-obvious correlations between inputs and power generation and take into account many different factors that affect generation.

Automation: the system operates automatically after training, which reduces human intervention in the forecasting process.

Forecasting horizon: the system allows forecasting trading schedules from 2 hours in advance to 31 days in advance, which allows the RES producer to plan and optimize the operation of the PV plant in time.

Adaptability: neural networks can quickly adapt to changing conditions, such as weather, making them effective for adjusting forecasts in real-time.

Taking into account many factors: the system can take into account many different factors that affect electricity production, which provides more accurate forecasts.

All of these advantages make the PV.Forecast power generation forecasting system using a neural network an important tool for optimizing electricity production and supply.

What is the role of a competent and reliable RES generation forecasting service provider?

As of today, more than 160 generation facilities with a total capacity of more than 1030 MW are users of the generation forecasting service by KNESS Energy, which is a part of KNESS Group. Our company offers a full range of services related to forecasting, which, among other things, includes the automated submission of forecast schedules through the Distribution Soft software.

Our offer:

For a producer that is a member of the balancing group of SE “Guaranteed Buyer”:

• Calculation of the forecast for up to 7 days in advance with its upload to the personal account on the platform of the SE “Guaranteed Buyer”.
• Hourly intraday update of the forecast schedule.
• Prompt consideration of planned and emergency orders in the forecast schedules.
• Access to a personal account in PV. SCADA with analytics of forecast data and a monthly report on the operation of the PV.Forecast forecasting system.
• Calculation of the imbalances settling cost.

In the case of the electricity sale generated by PV power plants on the market of bilateral agreements (when PV power plants are not part of the balancing group of SE “Guaranteed Buyer”):

• Calculation of the forecast for 2 days (D-2) preceding the trading day (D).
• Uploading a trading chart to the market management system.
• Submitting the trading schedule to the customer’s and buyer’s e-mail addresses.
• Calculating a long-term forecast for the next month.

Benefits of the Distribution Soft system by KNESS Energy

Automation: Distribution Soft system by KNESS Energy automatically updates trading schedules, ensuring continuity in electricity supply. RES producers do not need to spend time and resources on manual updates of schedules.

Hourly refinement: intraday hourly adjustment makes it possible to adjust schedules in real-time with the maximum possible adjustment horizon in accordance with the current legislation to take into account changes in energy production.

Autonomy: due to the uninterrupted operation of the system, the RES producer always receives accurate and up-to-date forecasts, minimizing the risk of large financial losses.

Client support: KNESS Energy provides highly qualified support to its clients, and our experts are ready to address all inquiries.

Why do renewable energy producers choose us?

Experience: our company has deep knowledge and many years of experience in the field of electricity generation forecasting.

Accuracy and reliability: our forecasts are based on modern technologies using neural networks and statistical methods, which make it possible to achieve high accuracy.

Efficiency: our solutions are aimed at efficient energy management, helping business owners reduce costs and increase profitability.

Individual approach: providing an individual approach to each client, developing solutions that meet specific requirements.

Comprehensive services: KNESS Energy offers a full range of forecasting services.

Innovation: we are always aware of the latest technological and industrial trends. Our company is constantly updating its approaches and tools in order to always provide the best solutions.

Learn more on the website.