Prediction horizon error analysis in thermal consumption models for control applications

Accurate consumption prediction models are crucial for optimizing building control applications, enhancing energy efficiency, reducing costs, and improving occupant comfort. However, prediction errors can significantly impact performance; overestimations lead to excessive energy consumption, higher operational costs, and increased carbon emissions, while underestimations result in inadequate heating, cooling, or lighting, negatively affecting comfort and productivity. This paper extends previous research by analysing the behaviour of prediction errors in six models of varying complexity. Using real consumption data from a large retail building in Madrid, the models predict energy demand across different time horizons, ranging from 1 hour to 24 hours. Results indicate that autoregressive models outperform others in short-term predictions but lose accuracy as the forecast horizon increases. Additionally, incorporating indexed parameters effectively mitigates error dispersion, improving model reliability over extended prediction periods.