Comparison of various artificial neuron models for very short-term load forecasting
Loading...
Date
2022-08
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
G.B. Pant University of Agriculture and Technology, Pantnagar, District Udham Singh Nagar, Uttarakhand. PIN - 263145
Abstract
Load forecasting has always been a crucial component of the operational and managerial aspect of efficient power system planning. Since there are several factors on which load forecasting depends, it becomes necessary to find out the level of impact these factors put on it. There are several techniques and models which can be used to forecast load on the basis of requirement such as- regression based model, fuzzy time series based model, support vector machine based model, and artificial neuron network based models etc. Artificial neural network (ANN) based models are considered as one of the popular methods for different levels of forecasting, hence used in the study. Data preparation is performed by transforming the historical electric load of Uttarakhand state adopting Max-Min normalization. The prepared data is partitioned into the categories of training and testing data for further application of the conventional and different multiplicative neuron models. The conventional ANN model having eight input nodes and one output neuron was evaluated with different combinations of activation functions. This model achieved the mean-square-error (MSE) of 0.0007. Among various multiplicative neuron models applied for VSTLF, the overall performance of QIFNM is found to be the best. The QIFNM having a single neuron achieved the MSE of 0.0020. As per, Akaike Information Criterion (AIC), all multiplicative neuron models performed better than the MLP based conventional ANN and QIFNM came out to be the best model among all. The performance analysis of these models revealed that a single multiplicative neuron can be used for VSTLF with better performance as per AIC than that of a network of several neurons of the conventional model.