Litcius/Paper detail

An Efficient Method for Wind Power Generation Forecasting by LSTM in Consideration of Overfitting Prevention

Soichiro Ookura, Hiroyuki Mori

2020IFAC-PapersOnLine25 citationsDOIOpen Access PDF

Abstract

This paper proposes an efficient method for wind power generation forecasting by Long Short Term Memory (LSTM) of Deep Neural Network (DNN). It is one of recurrent neural networks that make use of past output of the network, but replaces hidden layers of the conventional networks with the LSTM Block with memory and three gates of input, output and forget. Artificial and Deep Neural Networks are inclined to overfit leaning data in learning process. This paper proposes a modified LSTM that considers to prevent LSTM from overfitting with two strategies. One is Dropout to exclude some nodes randomly and change network topology while the other is Weight Decay that evaluates smaller weights between neurons. The effectiveness of the proposed method is demonstrated for real data of wind power generation.

Topics & Concepts

OverfittingDropout (neural networks)Computer scienceArtificial neural networkArtificial intelligenceBlock (permutation group theory)Wind powerMachine learningProcess (computing)Power (physics)Deep learningEngineeringMathematicsGeometryElectrical engineeringQuantum mechanicsPhysicsOperating systemEnergy Load and Power ForecastingImage and Signal Denoising MethodsNeural Networks and Applications