Load extraction from actual operation data for data-driven ultra-short-term room air-conditioning load prediction

Data-driven modeling achieves excellent performance in building load prediction by mining the actual load characteristic from operation data. When predicting room air-conditioning load, however, the measured data of cooling capacity cannot be used as the load label in the modeling as customary. The reason lies in the unstable automatic control of the air-conditioner, which makes the cooling capacity oscillate at the actual load. The mismatch of cooling capacity and actual load will affect prediction modeling. This paper first sets up a simulation experiment to quantitatively analyze this effect. The results indicate that the more the cooling capacity deviates from the load, the greater the error of the prediction model. To solve this problem, a load extraction method based on the Fourier decomposition is proposed, which involves transforming the actual cooling capacity data into the frequency domain and extracting the low-frequency components as load data, which is further used for modeling. The experiments show that the accuracy between extracted data and real load exceeds 90%. Additionally, the load extraction method improves the prediction accuracy by up to 17.55% compared to the model using cooling capacity data directly. Finally, a practical case demonstrates the effectiveness of the proposed load extraction method.