Case study: Energy quantity forecast in the gas industry

A service provider in the energy industry has asked COSMO CONSULT to develop a model for forecasting short-term energy needs to automate planning processes and achieve savings in energy purchasing. The company thus improved its forecast accuracy by about five percent and saved about half a million euros.

• Forecasting gas quantities with different features
• Considering specific behaviors of the gas forecasting time series
• Incorporating and modeling seasonal volatility
• Achieving forecasting quality up to 94%

In the energy industry, short-term energy forecasts are standard practices used to ensure that the purchasing of quantities is as cost-effective as possible. A service provider for smaller public utilities and providers calculated the forecast based on a simple mathematical formula. The disadvantage: The forecast formula could not respond to short-term trends or shifts in the price levels. The forecast was therefore error-prone especially during certain seasons.

COSMO CONSULT was commissioned to improve that basic energy quantity forecast in order to automate planning processes and achieve savings in energy purchasing.

As a first step, the project team eliminated the outliers and then constructed specific features using nonlinear optimization techniques. In addition to trend and seasonal components, the seasonal variance and calendar effects were taken into account. The weather data, on the other hand, were discarded because they would have not systematically improved the results. To determine the trend and seasonal component, classical forecasting methods such as ARIMA and exponential smoothing were used. Machine learning methods such as gradient boosting were also used to model the stochastic component. The corresponding R code was contained in a DLL file and was easily integrated into the customer's system. The developed algorithm can automatically remove outliers and perform forecasting with the optimized model, so the results can be displayed in the desired form.

The service provider achieved a drastic improvement in forecast accuracy of about 5% with simultaneous savings of approximately € 500,000. In addition, planning processes were improved and energy purchasing became more efficient.