Application of neural networks to HVAC load forecasting

The chief purpose of forecasting HVAC load is to store sufficient ice at off-peak times to meet the peak air conditioning load demands of the next day, and thereby achieve the goal of energy conservation. Forecasting is considered more useful than installing more ice storage systems in Taiwan because there is currently no way to forecast air conditioning load and store enough ice to meet the next-day air conditioning load demand. Real-time load forecasting can employ time series, regression analysis, and neural network methods, but neural networks are best able to forecast load. Since neural networks can achieve very high HVAC load forecasting accuracy, this method has great potential in practical applications. The ability to forecast peak air conditioning load on the next day allows sufficient ice to be stored during off-peak times to achieve the goal of energy conservation. This study obtains parameters affecting air conditioning load from weather forecasting data. Accurate weather forecasts can adequately predict next-day weather conditions, and thus enable even more accurate air conditioning load forecasts. When weather forecasts were used as neural network input parameters in an experiment, air conditioning load forecasts achieved an accuracy of 90%.