Study and analysis of SARIMA and LSTM in forecasting time series data

Dubey, Ashutosh Kumar; Kumar, Abhishek; García‐Díaz, Vicente; Sharma, Arpit; Kanhaiya, Kishan

doi:10.1016/j.seta.2021.101474

Cited by 104 publications

(44 citation statements)

References 50 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Additionally, the selection of this model is based on the reported advantages of its implementation to support DL-based models: straightforward implementation without the need of extensive data preprocessing and decomposition methods, small memory requirement, invariance to the diagonal re-scalation of gradients, suitability for models that involve large datasets, and/or large number of parameters [48,49]. Finally, the mean square error (MSE) is the selected loss function to be minimized, based on the successful results reported of this loss function in conjunction with the Adam optimizer [49][50][51].…”

Section: : End Formentioning

confidence: 99%

Short-Term Forecasting of Wind Energy: A Comparison of Deep Learning Frameworks

2021

View full text Add to dashboard Cite

Wind energy has been recognized as the most promising and economical renewable energy source, attracting increasing attention in recent years. However, considering the variability and uncertainty of wind energy, accurate forecasting is crucial to propel high levels of wind energy penetration within electricity markets. In this paper, a comparative framework is proposed where a suite of long short-term memory (LSTM) recurrent neural networks (RNN) models, inclusive of standard, bidirectional, stacked, convolutional, and autoencoder architectures, are implemented to address the existing gaps and limitations of reported wind power forecasting methodologies. These integrated networks are implemented through an iterative process of varying hyperparameters to better assess their effect, and the overall performance of each architecture, when tackling one-hour to three-hours ahead wind power forecasting. The corresponding validation is carried out through hourly wind power data from the Spanish electricity market, collected between 2014 and 2020. The proposed comparative error analysis shows that, overall, the models tend to showcase low error variability and better performance when the networks are able to learn in weekly sequences. The model with the best performance in forecasting one-hour ahead wind power is the stacked LSTM, implemented with weekly learning input sequences, with an average MAPE improvement of roughly 6, 7, and 49%, when compared to standard, bidirectional, and convolutional LSTM models, respectively. In the case of two to three-hours ahead forecasting, the model with the best overall performance is the bidirectional LSTM implemented with weekly learning input sequences, showcasing an average improved MAPE performance from 2 to 23% when compared to the other LSTM architectures implemented.

show abstract

Section: : End Formentioning

confidence: 99%

Short-Term Forecasting of Wind Energy: A Comparison of Deep Learning Frameworks

2021

View full text Add to dashboard Cite

show abstract

“…Dhakal et al [3] used LSTM to forecast accurate particulate matter (PM2.5), while Mussumeci and Coelho [4] applied LSTM to predict dengue in 790 cities in Brazil and reported that the model is better than other machine learning approaches, such as LASSO and Random Forest. Further, Dubey et al [5] compared LSTM to ARIMA (autoregressive integrated moving average) and SARIMA (seasonal ARIMA) in the forecasting case to the daily energyconsumption time series data. Other implementations of LSTM for forecasting were conducted in other fields and datasets, such as the energy sector [6], agriculture [7], weather [8], [9], public health [10], wind power generation [11], power grid [12], data center energy consumption [13], street lighting system based weather [14], and air pollution [15].…”

Section: Introductionmentioning

confidence: 99%

Robust LSTM With Tuned-PSO and Bifold-Attention Mechanism for Analyzing Multivariate Time-Series

et al. 2022

View full text Add to dashboard Cite

The need for accurate time-series results is badly demanding. LSTM has been applied for forecasting time series, which is generated when variables are observed at discrete and equal time intervals. Nevertheless, the problem of determining hyperparameters with a relatively high random rate will reduce the accuracy of the prediction results. This paper aims to promote LSTM with tuned-PSO and Bifold-Attention mechanism. PSO optimizes LSTM hyperparameters, and Bifold-attention mechanism selects the optimal input for LSTM. An accurate, adaptive, and robust time-series forecasting model is the main contribution, compared with ARIMA, MLP, LSTM, PSO-LSTM, A-LSTM, and PSO-A-LSTM. The model comparison is based on the accuracy of each model in forecasting Beijing PM2.5, Beijing Multi-Site, Air Quality, Appliances Energy, Wind Speed, and Traffic Flow. Proposed PSO Bifold-Attention LSTM has lower MAPE and RMSE than baselines. In other words, the model outperformed all LSTM base models in this study. The proposed model's accuracy is adaptable in daily, weekly, and monthly multivariate time-series datasets. This ground-breaking innovation is valuable for time-series analysis research, particularly the implementation of deep learning for time-series forecasting.

show abstract

“…The growth of technology and digitization has significantly transformed numerous industries utilizing machine learning for their prediction process. Many data-driven or algorithm-based solutions, from naive to complex, are applied in various scientific fields [12,6,4,5,7,11,13]. These developed algorithms lack a unified and dedicated framework for forecasting sequential data irrespective of any domain.…”

Section: Introductionmentioning

confidence: 99%