Accelerated Continuous Conditional Random Fields For Load Forecasting

摘要: Increasingly, aiming to contain their rapidly growing energy expenditures, commercial buildings are equipped respond utility’s demand and price signals. Such smart consumption, however, heavily relies on accurate short-term load forecasting, such as hourly predictions for the next $n$ $(n\ge 2)$ hours. To attain sufficient accuracy these predictions, it is important exploit relationships among estimated outputs. This paper treats multi-steps ahead regression task a sequence labeling (regression) problem, adopts continuous conditional random fields (CCRF) explicitly model interconnected In particular, we improve CCRF’s computation complexity predictive with two novel strategies. First, employ tridiagonal matrix techniques significantly speed up training inference. These tackle cubic computational cost required by inversion calculations in inference of CCRF, resulting linear operations. Second, address weak feature constraint problem multi-target edge function, thus boosting performance. The proposed able convert relationship related outputs values into set “sub-relationships”, each providing more specific constraints interplays We applied approach real-world prediction systems: one electricity another gas usage. Our experimental results show that strategy can meaningfully reduce error systems, terms mean absolute percentage root square error, when compared three benchmarking methods. Promisingly, relative reduction achieved our CCRF was 50 percent.