Publications
2021 |
J. F. Torres and D. Hadjout and A. Sebaa and F. Martínez-Álvarez and A. Troncoso Deep Learning for Time Series Forecasting: A Survey (Journal Article) Big Data, 9 (1), pp. 3-21, 2021. (Abstract | Links | BibTeX | Tags: big data, deep learning, time series) @article{TORRES21, title = {Deep Learning for Time Series Forecasting: A Survey}, author = {J. F. Torres and D. Hadjout and A. Sebaa and F. Martínez-Álvarez and A. Troncoso}, url = {https://www.liebertpub.com/doi/10.1089/big.2020.0159}, doi = {10.1089/big.2020.0159}, year = {2021}, date = {2021-02-05}, journal = {Big Data}, volume = {9}, number = {1}, pages = {3-21}, abstract = {Deep learning, one of the most remarkable techniques of machine learning, has been a major success in many fields, including image processing, speech recognition, and text understanding. It is powerful engines capable of learning arbitrary mapping functions, not require a scaled or stationary time series as input, support multivariate inputs, and support multi-step outputs. All of these features together make deep learning useful tools when dealing with more complex time series prediction problems involving large amounts of data, and multiple variables with complex relationships. This paper provides an overview of the most common Deep Learning types for time series forecasting, Explain the relationships between deep learning models and classical approaches to time series forecasting. A brief background of the particular challenges presents in time-series data and the most common deep learning techniques that are often used for time series forecasting is provided. Previous studies that applied deep learning to time series are reviewed.}, keywords = {big data, deep learning, time series}, pubstate = {published}, tppubtype = {article} } Deep learning, one of the most remarkable techniques of machine learning, has been a major success in many fields, including image processing, speech recognition, and text understanding. It is powerful engines capable of learning arbitrary mapping functions, not require a scaled or stationary time series as input, support multivariate inputs, and support multi-step outputs. All of these features together make deep learning useful tools when dealing with more complex time series prediction problems involving large amounts of data, and multiple variables with complex relationships. This paper provides an overview of the most common Deep Learning types for time series forecasting, Explain the relationships between deep learning models and classical approaches to time series forecasting. A brief background of the particular challenges presents in time-series data and the most common deep learning techniques that are often used for time series forecasting is provided. Previous studies that applied deep learning to time series are reviewed. |
F. Martínez-Álvarez and A. Troncoso and H. Quintián and E. Corchado Special Issue SOCO 2019: New trends in soft computing and its application in industrial and environmental problems (Journal Article) Neurocomputing, in press , 2021. (BibTeX | Tags: big data, deep learning) @article{MARTINEZ21, title = {Special Issue SOCO 2019: New trends in soft computing and its application in industrial and environmental problems}, author = {F. Martínez-Álvarez and A. Troncoso and H. Quintián and E. Corchado}, year = {2021}, date = {2021-01-01}, journal = {Neurocomputing}, volume = {in press}, keywords = {big data, deep learning}, pubstate = {published}, tppubtype = {article} } |
2020 |
F. Divina and J. F. Torres and M. García-Torres and F. Martínez-Álvarez and A. Troncoso Hybridizing deep learning and neuroevolution: Application to the Spanish short-term electric energy consumption forecasting (Journal Article) Applied Sciences, 10 (16), pp. 5487, 2020. (Abstract | Links | BibTeX | Tags: big data, deep learning, energy, time series) @article{DIVINA2020, title = {Hybridizing deep learning and neuroevolution: Application to the Spanish short-term electric energy consumption forecasting}, author = {F. Divina and J. F. Torres and M. García-Torres and F. Martínez-Álvarez and A. Troncoso}, url = {https://www.mdpi.com/2076-3417/10/16/5487}, doi = {https://doi.org/10.3390/app10165487}, year = {2020}, date = {2020-07-30}, journal = {Applied Sciences}, volume = {10}, number = {16}, pages = {5487}, abstract = {The electric energy production would be much more efficient if accurate estimations of the future demand were available, since these would allow allocating only the resources needed for the production of the right amount of energy required. With this motivation in mind, we propose a strategy, based on neuroevolution, that can be used to this aim. Our proposal uses a genetic algorithm in order to find a sub-optimal set of hyper-parameters for configuring a deep neural network, which can then be used for obtaining the forecasting. Such a strategy is justified by the observation that the performances achieved by deep neural networks are strongly dependent on the right setting of the hyper-parameters, and genetic algorithms have shown excellent search capabilities in huge search spaces. Moreover, we base our proposal on a distributed computing platform, which allows its use on a large time-series. In order to assess the performances of our approach, we have applied it to a large dataset, related to the electric energy consumption registered in Spain over almost 10 years. Experimental results confirm the validity of our proposal since it outperforms all other forecasting techniques to which it has been compared.}, keywords = {big data, deep learning, energy, time series}, pubstate = {published}, tppubtype = {article} } The electric energy production would be much more efficient if accurate estimations of the future demand were available, since these would allow allocating only the resources needed for the production of the right amount of energy required. With this motivation in mind, we propose a strategy, based on neuroevolution, that can be used to this aim. Our proposal uses a genetic algorithm in order to find a sub-optimal set of hyper-parameters for configuring a deep neural network, which can then be used for obtaining the forecasting. Such a strategy is justified by the observation that the performances achieved by deep neural networks are strongly dependent on the right setting of the hyper-parameters, and genetic algorithms have shown excellent search capabilities in huge search spaces. Moreover, we base our proposal on a distributed computing platform, which allows its use on a large time-series. In order to assess the performances of our approach, we have applied it to a large dataset, related to the electric energy consumption registered in Spain over almost 10 years. Experimental results confirm the validity of our proposal since it outperforms all other forecasting techniques to which it has been compared. |
F. Martínez-Álvarez and G. Asencio-Cortés and J. F. Torres and D. Gutiérrez-Avilés and L. Melgar-García and R. Pérez-Chacón and C. Rubio-Escudero and A. Troncoso and J. C. Riquelme Coronavirus Optimization Algorithm: A bioinspired metaheuristic based on the COVID-19 propagation model (Journal Article) Big Data, 8 (4), pp. 308-322, 2020. (Abstract | Links | BibTeX | Tags: big data, deep learning, energy, time series) @article{MARTINEZ-ALVAREZ20, title = {Coronavirus Optimization Algorithm: A bioinspired metaheuristic based on the COVID-19 propagation model}, author = {F. Martínez-Álvarez and G. Asencio-Cortés and J. F. Torres and D. Gutiérrez-Avilés and L. Melgar-García and R. Pérez-Chacón and C. Rubio-Escudero and A. Troncoso and J. C. Riquelme}, url = {https://www.liebertpub.com/doi/full/10.1089/big.2020.0051}, doi = {10.1089/big.2020.0051}, year = {2020}, date = {2020-07-22}, journal = {Big Data}, volume = {8}, number = {4}, pages = {308-322}, abstract = {This work proposes a novel bioinspired metaheuristic, simulating how the coronavirus spreads and infects healthy people. From a primary infected individual (patient zero), the coronavirus rapidly infects new victims, creating large populations of infected people who will either die or spread infection. Relevant terms such as reinfection probability, super-spreading rate, social distancing measures or traveling rate are introduced into the model in order to simulate the coronavirus activity as accurately as possible. The infected population initially grows exponentially over time, but taking into consideration social isolation measures, the mortality rate and number of recoveries, the infected population gradually decreases. The Coronavirus Optimization Algorithm has two major advantages when compared to other similar strategies. Firstly, the input parameters are already set according to the disease statistics, preventing researchers from initializing them with arbitrary values. Secondly, the approach has the ability to end after several iterations, without setting this value either. Furthermore, a parallel multi-virus version is proposed, where several coronavirus strains evolve over time and explore wider search space areas in less iterations. Finally, the metaheuristic has been combined with deep learning models, in order to find optimal hyperparameters during the training phase. As application case, the problem of electricity load time series forecasting has been addressed, showing quite remarkable performance.}, keywords = {big data, deep learning, energy, time series}, pubstate = {published}, tppubtype = {article} } This work proposes a novel bioinspired metaheuristic, simulating how the coronavirus spreads and infects healthy people. From a primary infected individual (patient zero), the coronavirus rapidly infects new victims, creating large populations of infected people who will either die or spread infection. Relevant terms such as reinfection probability, super-spreading rate, social distancing measures or traveling rate are introduced into the model in order to simulate the coronavirus activity as accurately as possible. The infected population initially grows exponentially over time, but taking into consideration social isolation measures, the mortality rate and number of recoveries, the infected population gradually decreases. The Coronavirus Optimization Algorithm has two major advantages when compared to other similar strategies. Firstly, the input parameters are already set according to the disease statistics, preventing researchers from initializing them with arbitrary values. Secondly, the approach has the ability to end after several iterations, without setting this value either. Furthermore, a parallel multi-virus version is proposed, where several coronavirus strains evolve over time and explore wider search space areas in less iterations. Finally, the metaheuristic has been combined with deep learning models, in order to find optimal hyperparameters during the training phase. As application case, the problem of electricity load time series forecasting has been addressed, showing quite remarkable performance. |
F. Martínez-Álvarez and A. Troncoso and H. Quintián and E. Corchado Special issue: HAIS16-IGPL (Journal Article) Logic Journal of the IGPL, 28 (1), pp. 1-3, 2020. (Abstract | Links | BibTeX | Tags: big data, deep learning, pattern recognition) @article{IGPL20b, title = {Special issue: HAIS16-IGPL}, author = {F. Martínez-Álvarez and A. Troncoso and H. Quintián and E. Corchado}, url = {https://doi.org/10.1093/jigpal/jzz066}, doi = {10.1093/jigpal/jzz066}, year = {2020}, date = {2020-02-01}, journal = {Logic Journal of the IGPL}, volume = {28}, number = {1}, pages = {1-3}, abstract = {Following, Fournier-Viger et al. propose to integrate the concept of correlation in high-utility itemset mining to find profitable itemsets that are highly correlated, using the all-confidence and bond measures. An efficient algorithm named FCHM (fast correlated high-utility itemset miner) is proposed to efficiently discover correlated high-utility itemsets. Two versions of the algorithm are proposed, named FCHMall-confidence and FCHMbond based on the all-confidence and bond measures, respectively. An experimental evaluation was done using four real-life benchmark data sets from the high-utility itemset mining literature: mushroom, retail, kosarak and foodmart. Results show that FCHM is efficient and can prune a huge amount of weakly correlated high-utility itemsets.}, keywords = {big data, deep learning, pattern recognition}, pubstate = {published}, tppubtype = {article} } Following, Fournier-Viger et al. propose to integrate the concept of correlation in high-utility itemset mining to find profitable itemsets that are highly correlated, using the all-confidence and bond measures. An efficient algorithm named FCHM (fast correlated high-utility itemset miner) is proposed to efficiently discover correlated high-utility itemsets. Two versions of the algorithm are proposed, named FCHMall-confidence and FCHMbond based on the all-confidence and bond measures, respectively. An experimental evaluation was done using four real-life benchmark data sets from the high-utility itemset mining literature: mushroom, retail, kosarak and foodmart. Results show that FCHM is efficient and can prune a huge amount of weakly correlated high-utility itemsets. |
2019 |
J. F. Torres and D. Gutiérrez-Avilés and A. Troncoso and F. Martínez-Álvarez Random Hyper-Parameter Search-Based Deep Neural Network for Power Consumption Forecasting (Conference) IWANN 15th International Work-Conference on Artificial Neural Networks, 11506 , Lecture Notes in Computer Science 2019. (Links | BibTeX | Tags: deep learning, energy, time series) @conference{TORRES19-2, title = {Random Hyper-Parameter Search-Based Deep Neural Network for Power Consumption Forecasting}, author = {J. F. Torres and D. Gutiérrez-Avilés and A. Troncoso and F. Martínez-Álvarez}, url = {https://link.springer.com/chapter/10.1007/978-3-030-20521-8_22}, doi = {https://doi.org/10.1007/978-3-030-20521-8_22}, year = {2019}, date = {2019-05-16}, booktitle = {IWANN 15th International Work-Conference on Artificial Neural Networks}, volume = {11506}, pages = {259-269}, series = {Lecture Notes in Computer Science}, keywords = {deep learning, energy, time series}, pubstate = {published}, tppubtype = {conference} } |
J. F. Torres and A. Troncoso and I. Koprinska and Z. Wang and F. Martínez-Álvarez Big data solar power forecasting based on deep learning and multiple data sources (Journal Article) Expert Systems, 36 , pp. id12394, 2019. (Links | BibTeX | Tags: deep learning, energy, time series) @article{TORRES19-1, title = {Big data solar power forecasting based on deep learning and multiple data sources}, author = {J. F. Torres and A. Troncoso and I. Koprinska and Z. Wang and F. Martínez-Álvarez}, url = {https://onlinelibrary.wiley.com/doi/abs/10.1111/exsy.12394}, doi = {https://doi.org/10.1111/exsy.12394}, year = {2019}, date = {2019-03-01}, journal = {Expert Systems}, volume = {36}, pages = {id12394}, keywords = {deep learning, energy, time series}, pubstate = {published}, tppubtype = {article} } |
2018 |
J. F. Torres and A. Galicia and A. Troncoso and F. Martínez-Álvarez A scalable approach based on deep learning for big data time series forecasting (Journal Article) Integrated Computer-Aided Engineering, 25 (4), pp. 335-348, 2018. (Abstract | Links | BibTeX | Tags: deep learning, energy, time series) @article{ICAE2018, title = {A scalable approach based on deep learning for big data time series forecasting}, author = {J. F. Torres and A. Galicia and A. Troncoso and F. Martínez-Álvarez}, url = {https://content.iospress.com/articles/integrated-computer-aided-engineering/ica580}, doi = {10.3233/ICA-180580}, year = {2018}, date = {2018-01-01}, journal = {Integrated Computer-Aided Engineering}, volume = {25}, number = {4}, pages = {335-348}, abstract = {This paper presents a method based on deep learning to deal with big data times series forecasting. The deep feed forward neural network provided by the H2O big data analysis framework has been used along with the Apache Spark platform for distributed computing. Since H2O does not allow the conduction of multi-step regression, a general-purpose methodology that can be used for prediction horizons with arbitrary length is proposed here, being the prediction horizon, h, the number of future values to be predicted. The solution consists in splitting the problem into h forecasting subproblems, being h the number of samples to be simultaneously predicted. Thus, the best prediction model for each subproblem can be obtained, making easier its parallelization and adaptation to the big data context. Moreover, a grid search is carried out to obtain the optimal hyperparameters of the deep learning-based approach. Results from a real-world dataset composed of electricity consumption in Spain, with a ten-minute frequency sampling rate, from 2007 to 2016 are reported. In particular, the accuracy and runtimes versus computing resources and size of the dataset are analyzed. Finally, the performance and the scalability of the proposed method is compared to other recently published techniques, showing to be a suitable method to process big data time series.}, keywords = {deep learning, energy, time series}, pubstate = {published}, tppubtype = {article} } This paper presents a method based on deep learning to deal with big data times series forecasting. The deep feed forward neural network provided by the H2O big data analysis framework has been used along with the Apache Spark platform for distributed computing. Since H2O does not allow the conduction of multi-step regression, a general-purpose methodology that can be used for prediction horizons with arbitrary length is proposed here, being the prediction horizon, h, the number of future values to be predicted. The solution consists in splitting the problem into h forecasting subproblems, being h the number of samples to be simultaneously predicted. Thus, the best prediction model for each subproblem can be obtained, making easier its parallelization and adaptation to the big data context. Moreover, a grid search is carried out to obtain the optimal hyperparameters of the deep learning-based approach. Results from a real-world dataset composed of electricity consumption in Spain, with a ten-minute frequency sampling rate, from 2007 to 2016 are reported. In particular, the accuracy and runtimes versus computing resources and size of the dataset are analyzed. Finally, the performance and the scalability of the proposed method is compared to other recently published techniques, showing to be a suitable method to process big data time series. |
J. F. Torres and A. Troncoso and I. Koprinska and Z. Wang and F. Martínez-Álvarez Deep learning for big data time series forecasting applied to solar power (Conference) SOCO 13th International Conference on Soft Computing Models in Industrial and Environmental Applications, Advances in Intelligent Systems and Computing 2018. (Links | BibTeX | Tags: deep learning, energy, time series) @conference{SOCO2018, title = {Deep learning for big data time series forecasting applied to solar power}, author = {J. F. Torres and A. Troncoso and I. Koprinska and Z. Wang and F. Martínez-Álvarez}, url = {https://link.springer.com/chapter/10.1007/978-3-319-94120-2_12}, year = {2018}, date = {2018-01-01}, booktitle = {SOCO 13th International Conference on Soft Computing Models in Industrial and Environmental Applications}, series = {Advances in Intelligent Systems and Computing}, keywords = {deep learning, energy, time series}, pubstate = {published}, tppubtype = {conference} } |
2017 |
J. F. Torres and A. Troncoso and F. Martínez-Álvarez Deep Learning - Based Approach for Time Series Forecasting with Application to Electricity Load (Conference) IWINAC International Work-Conference on the Interplay Between Natural and Artificial Computation, Lecture Notes in computer Science 2017. (Links | BibTeX | Tags: deep learning, energy, time series) @conference{IWINAC2017, title = {Deep Learning - Based Approach for Time Series Forecasting with Application to Electricity Load}, author = {J. F. Torres and A. Troncoso and F. Martínez-Álvarez}, url = {https://link.springer.com/chapter/10.1007/978-3-319-59773-7_21}, year = {2017}, date = {2017-01-01}, booktitle = {IWINAC International Work-Conference on the Interplay Between Natural and Artificial Computation}, series = {Lecture Notes in computer Science}, keywords = {deep learning, energy, time series}, pubstate = {published}, tppubtype = {conference} } |