Improving cooling load prediction reliability for HVAC system using Monte-Carlo simulation to deal with uncertainties in input variables

Abstract

Accurate and reliable cooling load prediction is the basis for model predictive control (MPC) of HVAC systems. Cooling load prediction process is constrained by many input variables. These input variables contain different kinds of uncertainties that should not be ignored in pursuing a reliable prediction result. However, previous load prediction methods usually improve the prediction accuracy by optimizing model structure but fail to consider the influence of input variables uncertainties. To reduce the influence of uncertainty of input variables forecast data scientifically, this paper proposes an improved cooling load prediction reliability method in which the input variables are calibrated offline with Monte-Carlo simulations and stochastic treatment before inputting them into the prediction model. A machine learning algorithm (support vector machine, SVM) is used as the prediction model. Uncertainties in weather parameters, indoor heat gains, and historical cooling load were taken into consideration. Uncertainty analysis results showed that prediction accuracy with calibrated input variables was much better (R2 = 0.9627) than those predicted with forecast input data directly (R2 = 0.9488), which was closer to load prediction obtained by using measured input data (R2 = 0.9632). The coefficient of variation (CV) value using the calibrated input data was less than 15%. The reliable cooling load prediction results are beneficial for more informed, reliable decisions on MPC application, thereby promoting energy conservation potential of HVAC systems.