No abstract

Feedforward neural network models are created for prediction of daily heating energy consumption of a NTNU university campus Gløshaugen using actual measured data for training and testing. Improvement of prediction accuracy is proposed by using neural network ensemble. Previously trained feed-forward neural networks are first separated into clusters, using k-means algorithm, and then the best network of each cluster is chosen as member of an ensemble. Two conventional averaging methods for obtaining ensemble output are applied; simple and weighted. In order to achieve better prediction results, multistage ensemble is investigated. As second level, adaptive neuro-fuzzy inference system with various clustering and membership functions are used to aggregate the selected ensemble members. Feedforward neural network in second stage is also analyzed. It is shown that using ensemble of neural networks can predict heating energy consumption with better accuracy than the best trained single neural network, while the best results are achieved with multistage ensemble.

For the prediction of heating energy consumption of university campus, neural network ensemble is proposed. Actual measured data are used for training and testing the models. Improvement of the predicton accuracy using k-means clustering for creating subsets used to train individual radial basis function neural networks is examined. Number of clusters is varying from 2 to 5. The outputs of ensemble members are aggregated using simple, weighted and median based averaging. It is shown that ensembles achieve better prediction results than the individual network.

No abstract

The paper presents a new approach to the design of the classical PD controller for the plant in the closed loop control system. Proposed controller has two unknown adjustable parameters which are designed by the algebraic method in the two dimensional parameter space, by using the newly discovered characteristic polynomial of the row nondegenerate full transfer function matrix. The system relative stability, in regard to the chosen damping coefficient is achieved. The optimality criterion is the minimal value of the performance index and sum of squared error is taken as an objective function. The output error used in the performance index is influenced by all actions on the system at the same time: by nonzero initial conditions and by nonzero inputs. In the classical approach the output error is influenced only by nonzero inputs. Experimental results confirm that by taking into account nonzero initial conditions, an optimal solution is obtained, which is not the case with the classical method where optimization is performed at zero initial conditions.