Predicting the indoor thermal data for heating season based on short-term measurements to calibrate the simulation set-points

Abstract

This study aims to predict the hourly thermal performance data during the heating season based on short-term measured data by comparing the results with the simulated data for a big scale residential building. The measurement period was done for about 31.5 days during the heating season. Alternatively, the heating season was evaluated with the simulated heating consumption data as the period from November 15th to March 31st (137 days) in order to predict the data for inside temperature and relative humidity. Relatedly, Artificial Neural Network (ANN) was designed for the prediction model. Outside dry-bulb temperature, outside dew-point temperature, wind speed, wind direction, atmospheric pressure, solar azimuth, and heating consumption were set as independent variables (inputs) along with temperature and humidity as targeted variables of the model. Four FeedForword – BackPropagation of ANNs were used as a network for each measurement point inside of the building where each ANN includes three layers, defined as input layers, hidden layer, and output layer. A thermal dataset for the heating season was composed by validating the result of the prediction with measured data, which saved about 77% of the heating season measurement's time. Then, by comparing composed dataset with the simulated thermal data, it was obtained that the heating system is performing well and close to the expectations. The approached prediction work provides the possibility to apply real-time calibration with the monitoring system which was already implemented in the building. Therefore, the produced dataset would ensure the quality of the real-time measured data for any unexpected condition during the building operation. as well as the accuracy of simulated data for any unforeseen inputs. Results show that the ANN model might be used effectively to provide useful predictions for the indoor thermal data not only to cover the missing measured data, but also to support the calibration process, the monitoring system and validating the simulation set points.