Interval Prediction of Air Conditioning Load Based on Quantile Regression BiTCN-BiGRU-Attention Model*

Aiming at the characteristics of load instability and large volatility in the process of air conditioning operation, to better improve the prediction accuracy of air conditioning load and quantization the uncertainty of prediction results, this paper proposed an improved neural network model composed of Bidirectional Temporal Convolutional Network based on Attention mechanism (BiTCN) and Bidirectional Gated Recurrent Unit (BiGRU). Quantile Regression (QR) was applied to realize interval prediction of air conditioning load. The model was trained and tested adopted measured data from an office and a machine room in both summer and winter. The results showed that the average MAPE and R2 of the prediction model were 2.3% and 0.980, respectively, while the average PICP (Prediction Interval Coverage Probability) and PIMWP (Prediction Interval based Mean Width Percentage) were 0.965 and 0.213, respectively. It was verified that the proposed method had a good prediction effect on both point prediction and interval prediction of air conditioning load.