ANN based load forecasting: a parallel structure

Air-conditioning system plays an important role in modern architectures, whereas load forecasting provides a more efficient and accurate way to control air-conditioning systems. In this article, a parallel structure called 2-ANN for air-conditioner load forecasting is proposed. 2-ANN combines two independent predictors by a corrector. For forecasting, first the predictors make separate predictions, and then their predictions are compared and revised by the corrector to form a single output. 2-ANN structure is easy to analysis as well as easy to implement. In implementation, the predictors and corrector are back-propagation (BP) neural networks. The parallel structure, set up with three BP networks, is trained and tested by real-world data of air-conditioner load. In those tests, 2-ANN outperforms each one of the two predictors alone.