The impact of electricity price forecasts on production plans in the district heating system: A comparative study between different forecasting methods
Examensarbete för masterexamen
This thesis covers the investigation of electricity price forecasts and the effect they have on production plans and production costs in the district heating system. Four different forecasts were tested on four different models representing district heating systems. The forecasts were investigated in terms of error between the forecast and the actual spot price and the DH-models were optimized in GAMS to find the minimum cost. The first forecast is the naive benchmark which copies previous events. The second forecast is a rolling average performed on the naive benchmark. These two simpler forecasts are compared with a linear regression forecast and a random forest forecast which both are constructed using machine learning-algorithms. The first model of the district heating system contains a combined heat and power plant coupled with a heat pump. In the second model heat storage was added since that is a fundamental equipment in many heat producing plants today. In the two last models an electric boiler and a bio-oil boiler were added respectively. The different units were added to make the system more reality like and for the purpose of investigating different features and how they were affected by the forecasts. The simple models with few units are less dependent on the forecasts since they have less options of how to produce heat. Thus a limit in the electricity price decides if the electricity producer or the electricity consumer should produce the heat. The linear regression outperformed the other forecasts in terms of error and cost for all models of the district heating plant. Also, a rolling average on a benchmark forecast can make the outcome in form of production cost better. When thermal energy storage (TES) was implemented the production costs were decreased and the difference in cost between the forecasts were larger. The hypothetical future system, using danish electricity prices with more fluctuations which can be expected in electricity systems with more renewable electricity generation, was harder to predict.
Forecast , reference , electricity price , production cost , production plan , naive , rolling average , regression , random forest