Evolutionary neural-networks based optimisation for short-term load forecasting

• Autorzy: Grzenda M. , Macukow B.
• Języki publikacji: EN

Abstrakty

EN

The purpose of short-term load forecasting is to optimise the power supply volume in short time horizon. There is no straightforward mapping rule between the type of time period and the resulting power consumption. Still, it is inevitable for the overall efficiency of the power system to rely on a good prediction model. Our paper illustrates a novel approach based on evolutionary programming. Feedforward networks are being evolved by the ECoMLP method in order to properly solve the optimisation problem, defined as minimisation of the prediction error. All the results have been obtained using the data from the Polish Power System. The data used for the training and tests has been chosen so as to reflect both short-time and long-time dependencies between time period category and load of the system. The primary feature of the described method is a novel self-adaptive procedure that is a part of a sophisticated design algorithm serving to select both network architecture and weight connections. Due to the application of this procedure, no time consuming tests are required to train and retrain neural prediction models. Therefore, the method makes it possible to construct and maintain prediction models for load forecasting without expert knowledge about neural networks.

Słowa kluczowe

EN

evolutionary programming; neural networks; optimisation

PL

optymalizacja; programowanie ewolucyjne; sieć neuronowa

• Wydawca: Systems Research Institute, Polish Academy of Sciences
• Czasopismo: Control and Cybernetics
• Rocznik: 2002
• Tom: Vol. 31, no 2
• Strony: 371--382
• Opis fizyczny: Bibliogr. 11 poz.,

Twórcy

autor

Grzenda M.

autor

Macukow B.

• Warsaw University of Technology, Faculty of Mathematics and Information Science, Plac Politechniki 1, 00-661 Warszawa, Poland,