Cooling load prediction through recurrent neural networks

In this paper, we focus on recurrent neural networks and investigate their applicability to some identification or prediction problems. After reviewing the well-known learning algorithms for recurrent neural networks, called back propagation through time (BPTT) and real-time recurrent learning (RTRL), we investigate their performance when applied to relatively small-scale problems and evaluate the computational complexity of them. Following these investigations, we apply the recurrent neural networks to large-scale cooling load prediction problems in a district heating and cooling system. However, the computational complexity is enormous and learning within practical time seems to be very difficult. For decreasing such difficulties, we propose a model that preserves output values observed within an appropriate period. Through a lot of numerical simulations, it is shown that the proposed model has an ability to learn long cycle time series within relatively short time. >