An Autoencoder Wavelet Based Deep Neural Network with Attention Mechanism for Multistep Prediction of Plant Growth

EasyChair Preprint 3490

Abstract

Multi-step prediction is considered of major significance for time series analysis in many real-life problems. Existing methods mainly focus on one-step-ahead forecasting, since multiple step forecasting generally fails due to accumulation of prediction errors. This paper presents a novel approach for predicting plant growth in agriculture, focusing on prediction of plant Stem Diameter Variations (SDV). The proposed approach consists of three main steps. At first, wavelet decomposition is applied to the original data, as to facilitate model fitting to them. Then an encoder-decoder framework is developed using Long Short-Term Memory (LSTMs) and used for appropriate feature extraction from the data. Finally, a recurrent neural network including LSTMs and an attention mechanism is proposed for modelling long-term dependencies in the time series data. Experimental results are presented which illustrate the good performance of the proposed approach and that it significantly outperforms the existing models, in terms of error criteria such as RMSE, MAE and MAPE.

Keyphrases: Attention Mechanism, Deep Neural Networks, LSTMs, multistep prediction, plant growth prediction, time series analysis, wavelet analysis

@booklet{EasyChair:3490,
  author    = {Bashar Alhnaity and Stefanos Kollias and Georgios Leontidis and Shouyong Jiang and Bert Schamp and Simon Pearson},
  title     = {An Autoencoder Wavelet Based Deep Neural Network with Attention Mechanism for Multistep Prediction of Plant Growth},
  howpublished = {EasyChair Preprint 3490},
  year      = {EasyChair, 2020}}