A novel hybrid ensemble LSTM‐FFNN forecasting model for very short‐term and short‐term PV generation forecasting

Abstract

The increasing penetration of photovoltaic (PV) systems into the electrical energy systems brings forward several technical and economic issues that mostly relate to their unpredictable nature. A promising solution to many of these is the implementation of robust PV generation forecasting models. In this paper a novel hybrid Ensemble Long Short-Term Memory-Feed Forward Neural Network (ELSTM-FFNN) model is proposed, that is able to perform both very-short and short-term forecasting. The performance of the proposed model is compared with individual LSTM models, and its forecasting accuracy is assessed in two different forecasting horizons: (a) 15-min ahead and (b) 1-h ahead. Moreover, in order to fully examine the contribution of the utilized data to the performance of the model, several scenarios have been formulated for each forecasting horizon. The results indicate that the proposed ELSTM-FFNN model can increase the forecasting accuracy in both horizons between 3-11.9% and 0.2-17.8%, respectively, considering the Mean Absolute Range Normalized Error (MARNE).This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Day‐ahead scheduling of a hybrid renewable energy system based on generation forecasting using a deep‐learning approach

“…It is a recurrent neural network and has been widely used in time series forecasting applications. LSTM's aforementioned ability derives from the structure of the unit, consisting of four layers, which is able to control the information flows from the current unit to the next ones 39 …”

“…LSTM's aforementioned ability derives from the structure of the unit, consisting of four layers, which is able to control the information flows from the current unit to the next ones. 39 An autocorrelation analysis is conducted on the PV power data, to define the optimal number of previous time steps to be used as inputs to the model, as the LSTM deals with time-series forecasting. Figure 4 presents the autocorrelation curve of six sequential days.…”

Localization and Detection of Multiple Attacks in Wireless Sensor Networks Using Artificial Neural Network

“…We suppose that the model's inputs consist of the three preceding time steps. The data from the first unit flows into the second, as seen in the unfolded version [55,56].…”

Futuristic deep learning algorithms for long-term solar power prediction

“…The GAN-CNN [36] hybrid model was created for solar PV forecasting using data from 33 geographically dispersed weather stations. The ELSTM-FFNN [37] algorithm is suggested for both very short-term(VST) and shortterm(ST) predicting with increased precision. In order to combine the various LSTM models, FFNN is used in the demonstration of the ELSTM.…”

“…Individual algorithms are unable to meet the demand for more precise PV power prediction. Thus, a variety of hybrid model combinations have been developed, including genetic algorithm-ANN [15], analog ensemble (AnEn)-ANN [30], FL-ANN [31], ANN-physical method [32], wavelets transform (WT)-artificial intelligence (AI) techniques [50], time-delay neural network (TDNN)-ARMA [33], WT-partial swarm optimization (PSO)-SVM [34], ARMAnonlinear auto regression NLAG) [35], generative adversarial networks (GAN)-convolution neural networks(CNN) [36], ensemble memory-feed forward neural network (ELSTM-FFNN) [37], LSTM-CNN as well as CNN-LSTM [38], and bidirectional GRU [39].…”