A fuzzy nearest neighbor neural network statistical model for predicting demand for natural gas and energy cost savings in public buildings

• Rooftop HVAC sensors were monitored daily at various temperatures.
• Real-time decision support system forecasts MCF demand for natural gas.
• Model predicts cost savings based on price, operating expenses, cost to gas on, etc.
• Statistics, ANN, fuzzy logic, and nearest neighbor methods are used to compare demand.

This paper addresses the problem of predicting demand for natural gas for the purpose of realizing energy cost savings. Daily monitoring of a rooftop unit wireless sensor system provided feedback for a decision support system that supplied the demand for the required number of million cubic feet of natural gas used to control heating, ventilation, and air conditioning systems. The system was modeled with artificial neural networks (ANNs). Data on the consumption of the system were collected for 111 days beginning September 21, 2012. The input/output data were used to train the ANN. The ANN approximated the data very well, showing that it can be used to predict demand for natural gas. A fuzzy nearest neighbor neural network statistical model consisting of four components was used. The predictive models were implemented by comparing regression, fuzzy logic, nearest neighbor, and neural networks. In addition, to optimize natural gas demand, we used the fuzzy regression nearest neighbor ANN model cost function to investigate the variables of price, operating expenses, cost to drill new wells, cost to turn gas on, oil price and royalties.