Influence of neural network training parameters on short-term wind forecasting

Abstract

This paper investigates factors which can affect the accuracy of short-term wind speed prediction when done over long periods spanning different seasons. Two types of Neural Networks are used to forecast power generated via specific horizontal axis wind turbines.Meteorological data used is for a specific Western Australian location.Results reveal that seasonal variations affect the prediction accuracy of the wind resource, but the magnitude of this influence strongly depends on the details of the Neural Network deployed. Factors investigated include the span of the time series needed to initially train the networks, the temporal resolution of this data, the length of training pattern within the overall span which are used to implement the predictions and whether the inclusion of solar irradiance data can appreciably affect wind speed prediction accuracy. There appears to be a relatively complex relationship between these factors and the accuracy of wind speed prediction via Neural Networks. Predicting wind speed based on Neural Networks trained using wind speed and solar irradiance data also increases the prediction accuracy of wind power generated, as can the type of network selected.

Predictive power management strategies for stand-alone hydrogen systems: Lab-scale validation

“…The load profile used to test the experimental setup is used to train RBF-NNs to perform 30, 40, 50, 60 and 120 s ahead PH. The procedure that is used to train the RBF-NNs has been explained by the author in an earlier work [20]. More details about the structure, activation function and advantages of RBF-NN can also be found in the mentioned work.…”

“…NNs are widely used for prediction applications such as wind energy [20], electric loads [58] and solar irradiance [59]. A previous study by the authors of this paper has proven that Radial Basis Function NNs (RBF-NNs) are good forecasting tools for nonlinear time series dynamics [20].…”

“…Because measured profiles of renewable energy resources such as wind may not always be available, especially for remote locations, a PMS that is able to predict (in real-time) the availability of renewables and load demand is highly desirable but challenging to obtain. Therefore in most instances, recourse to several (off-line, not real-time) methods is used to forecast renewable resources using techniques such as Neural Networks (NNs) [20] or the ASHRAE clear-sky model [21].…”

Multi-objective optimisation of renewable hybrid energy systems with desalination

“…As such, they are an excellent candidate for stand-alone power generation at reduced or negligible operational emissions [1]. However, renewable sources are highly stochastic and experience seasonal fluctuations [2]. Thus energy storage devices such as batteries and hydrogen are often used in stand-alone (hybrid) energy systems [3e6].…”

A regime-switching recurrent neural network model applied to wind time series

“…The neural network approaches build nonlinear models that feature elaborate relationships between the selected variables. Some neural networks developed for processing wind series include: feedforward multilayer perceptrons (MLP) [5,10,15,25,43] and recurrent neural networks (RNN) [3,7,26,33,44]. These MLP networks take lagged inputs passed with tapped-delay lines.…”