Analysis and forecasting of electricity prices using an improved time series ensemble approach: an application to the Peruvian electricity market

AIMS Mathematics, Journal Year: 2024, Volume and Issue: 9(8), P. 21952 - 21971

Published: Jan. 1, 2024

<p>In today's electricity markets, accurate price forecasting provides valuable insights for decision-making among participants, ensuring reliable operation of the power system. However, complex characteristics time series hinder accessibility to forecasting. This study addressed this challenge by introducing a novel approach predicting prices in Peruvian market. involved preprocessing monthly addressing missing values, stabilizing variance, normalizing data, achieving stationarity, and seasonality issues. After this, six standard base models were employed model series, followed applying three ensemble forecast filtered series. Comparisons conducted between predicted observed using mean error accuracy measures, graphical evaluation, an equal statistical test. The results showed that proposed was efficient tool Moreover, outperformed earlier studies. Finally, while numerous global studies have been from various perspectives, no analysis has undertaken learning market.</p>

Language: Английский

---

Back to Basics: The Power of the Multilayer Perceptron in Financial Time Series Forecasting

Mathematics, Journal Year: 2024, Volume and Issue: 12(12), P. 1920 - 1920

Published: June 20, 2024

The economic time series prediction literature has seen an increase in research leveraging artificial neural networks (ANNs), particularly the multilayer perceptron (MLP) and, more recently, transformer networks. These ANN models have shown superior accuracy compared to traditional techniques such as autoregressive integrated moving average (ARIMA) models. most recent of this type network, recurrent or Transformers models, are composed complex architectures that require sufficient processing capacity address problems, while MLP is based on densely connected layers and supervised learning. A deep understanding limitations necessary appropriately choose ideal model for each tasks. In article, we show how a simple architecture allows better adjustment than other including shorter time. This premise use will not always allow results.

Language: Английский

---

Walking Back the Data Quantity Assumption to Improve Time Series Prediction in Deep Learning

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(23), P. 11081 - 11081

Published: Nov. 28, 2024

Deep learning techniques have significantly advanced time series prediction by effectively modeling temporal dependencies, particularly for datasets with numerous observations. Although larger are generally associated improved accuracy, the results of this study demonstrate that assumption does not always hold. By progressively increasing amount training data in a controlled experimental setup, best predictive metrics were achieved intermediate iterations, variations up to 66% RMSE and 44% MAPE across different models datasets. The findings challenge notion more necessarily leads better generalization, showing additional observations can sometimes result diminishing returns or even degradation metrics. These emphasize importance strategically balancing dataset size model optimization achieve robust efficient performance. Such insights offer valuable guidance forecasting, especially contexts where computational efficiency accuracy must be optimized.

Language: Английский

---

Advances in time series forecasting: innovative methods and applications

AIMS Mathematics, Journal Year: 2024, Volume and Issue: 9(9), P. 24163 - 24165

Published: Jan. 1, 2024

Language: Английский

---

Enhancing sewage flow prediction using an integrated improved SSA-CNN-Transformer-BiLSTM model

AIMS Mathematics, Journal Year: 2024, Volume and Issue: 9(10), P. 26916 - 26950

Published: Jan. 1, 2024

<p>Accurate prediction of sewage flow is crucial for optimizing treatment processes, cutting down energy consumption, and reducing pollution incidents. Current models, including traditional statistical models machine learning have limited performance when handling nonlinear high-noise data. Although deep excel in time series prediction, they still face challenges such as computational complexity, overfitting, poor practical applications. Accordingly, this study proposed a combined model based on an improved sparrow search algorithm (SSA), convolutional neural network (CNN), transformer, bidirectional long short-term memory (BiLSTM) prediction. Specifically, the CNN part was responsible extracting local features from series, Transformer captured global dependencies using attention mechanism, BiLSTM performed temporal processing features. The SSA optimized model's hyperparameters to improve accuracy generalization capability. validated dataset actual plant. Experimental results showed that introduced mechanism significantly enhanced ability handle data, effectively hyperparameter selection, improving training efficiency. After introducing SSA, CNN, modules, $ {R^{\text{2}}} increased by 0.18744, RMSE (root mean square error) decreased 114.93, MAE (mean absolute 86.67. difference between predicted peak/trough monitored within 3.6% appearance 2.5 minutes away time. By employing multi-model fusion approach, achieved efficient accurate highlighting potential application prospects field treatment.</p>

Language: Английский

---

Enhancing sewage flow prediction using an integrated improved SSA-CNN-Transformer-BiLSTM model

AIMS Mathematics, Journal Year: 2024, Volume and Issue: 9(10), P. 26916 - 26950

Published: Jan. 1, 2024

<p>Accurate prediction of sewage flow is crucial for optimizing treatment processes, cutting down energy consumption, and reducing pollution incidents. Current models, including traditional statistical models machine learning have limited performance when handling nonlinear high-noise data. Although deep excel in time series prediction, they still face challenges such as computational complexity, overfitting, poor practical applications. Accordingly, this study proposed a combined model based on an improved sparrow search algorithm (SSA), convolutional neural network (CNN), transformer, bidirectional long short-term memory (BiLSTM) prediction. Specifically, the CNN part was responsible extracting local features from series, Transformer captured global dependencies using attention mechanism, BiLSTM performed temporal processing features. The SSA optimized model's hyperparameters to improve accuracy generalization capability. validated dataset actual plant. Experimental results showed that introduced mechanism significantly enhanced ability handle data, effectively hyperparameter selection, improving training efficiency. After introducing SSA, CNN, modules, $ {R^{\text{2}}} increased by 0.18744, RMSE (root mean square error) decreased 114.93, MAE (mean absolute 86.67. difference between predicted peak/trough monitored within 3.6% appearance 2.5 minutes away time. By employing multi-model fusion approach, achieved efficient accurate highlighting potential application prospects field treatment.</p>

Language: Английский

---