Wind energy is one of the most promising resources for the mitigation of greenhouse gas emissions that contribute to anthropogenic global warming. However, the large proliferation of wind power generators is causing several critical issues in power systems due to their variable power generated profiles. For this reason, a large number of learning techniques, e.g. integrating Vector Auto-Regressive and Neural Network-based models, were proposed in the literature for mitigating wind power uncertainty issues. Unfortunately, these methodologies show several limitations, e.g. the huge number of parameters and/or the heavy computational cost, which hinder their deployment in modern power system operation, where prompt and reliable wide-area wind power generation forecasts are requested for supporting time-critical decision making on several time horizons. To try addressing this issue, this paper proposes the Dynamic Adaptive Feature-based Temporal Ensemble (DAFT-E) forecasting approach, which relies on an extensive feature engineering, a fast feature selection step and an ensemble of computationally inexpensive models to reduce the computational complexity of the forecasting task, while still preserving predictive accuracy. The experimental results, which benchmark DAFT-E against multivariate (VAR and deep learning) alternatives on two real case studies, show that the proposed approach outperforms state-of-the-art and representation learning models according to several forecasting accuracy metrics.
31 August 2018 in International Journal of Data Science and Analytics
The limitation of transmission lines thermal capacity plays a crucial role in the safety and reliability of power systems. Dynamic thermal line rating approaches aim to estimate the transmission line’s temperature and assess its compliance with the limitations above. Existing physics-based standards estimate the temperature based on environment and line conditions measured by several sensors. This manuscript shows that estimation accuracy can be improved by adopting a data-driven Digital Twin approach. The proposed method exploits machine learning by learning the input–output relation between the physical sensors data and the actual conductor temperature, serving as a digital equivalent to physics-based standards. An experimental assessment on real data, comparing the proposed approach with the IEEE 738 standard, shows a reduction of 60% of the Root Mean Squared Error and a decrease in the maximum estimation error from above 10 °C to below 7 °C. These preliminary results suggest that the Digital Twin provides more accurate and robust estimations, serving as a complement, or a potential alternative, to traditional methods.
18 October 2020 in Technology and Economics of Smart Grids and Sustainable Energy
The massive penetration of renewable power generation in modern power grids is an effective way to reduce the impact of energy production on global warming. Unfortunately, the wind power generation may affect the regular operation of electrical systems, due to the stochastic and intermittent nature of the wind. For this reason, reducing the uncertainty about the wind evolution, e.g. by using short-term wind power forecasting methodologies, is a priority for system operators and wind producers to implement low-carbon power grids. Unfortunately, though the complexity of this task implies the comparison of several alternative forecasting methodologies and dimensionality reduction techniques, a general and robust procedure of model assessment still lacks in literature. In this paper the authors propose a robust methodology, based on extensive statistical analysis and resampling routines, to supply the most effective wind power forecasting method by testing a vast ensemble of methodologies over multiple time-scales and on a real case study. Experimental results on real data collected in an Italian wind farm show the potential of ensemble approaches integrating both statistical and machine learning methods.
The availability of massive amounts of temporal data opens new perspectives of knowledge extraction and automated decision making for companies and practitioners. However, learning forecasting models from data requires a knowledgeable data science or machine learning (ML) background and expertise, which is not always available to end-users. This gap fosters a growing demand for frameworks automating the ML pipeline and ensuring broader access to the general public. Automatic machine learning (AutoML) provides solutions to build and validate machine learning pipelines minimizing the user intervention. Most of those pipelines have been validated in static supervised learning settings, while an extensive validation in time series prediction is still missing. This issue is particularly important in the forecasting community, where the relevance of machine learning approaches is still under debate. This paper assesses four existing AutoML frameworks (AutoGluon, H2O, TPOT, Auto-sklearn) on a number of forecasting challenges (univariate and multivariate, single-step and multi-step ahead) by benchmarking them against simple and conventional forecasting strategies (e.g., naive and exponential smoothing). The obtained results highlight that AutoML approaches are not yet mature enough to address generic forecasting tasks once compared with faster yet more basic statistical forecasters. In particular, the tested AutoML configurations, on average, do not significantly outperform a Naive estimator. Those results, yet preliminary, should not be interpreted as a rejection of AutoML solutions in forecasting but as an encouragement to a more rigorous validation of their limits and perspectives.
1 January 2024 in Database: The Journal of Biological Databases and Curation
While biomedical relation extraction (bioRE) datasets have been instrumental in the development of methods to support biocuration of single variants from texts, no datasets are currently available for the extraction of digenic or even oligogenic variant relations, despite the reports in literature that epistatic effects between combinations of variants in different loci (or genes) are important to understand disease etiologies. This work presents the creation of a unique dataset of oligogenic variant combinations, geared to train tools to help in the curation of scientific literature. To overcome the hurdles associated with the number of unlabelled instances and the cost of expertise, active learning (AL) was used to optimize the annotation, thus getting assistance in finding the most informative subset of samples to label. By pre-annotating 85 full-text articles containing the relevant relations from the Oligogenic Diseases Database (OLIDA) with PubTator, text fragments featuring potential digenic variant combinations, i.e. gene–variant–gene–variant, were extracted. The resulting fragments of texts were annotated with ALAMBIC, an AL-based annotation platform. The resulting dataset, called DUVEL, is used to fine-tune four state-of-the-art biomedical language models: BiomedBERT, BiomedBERT-large, BioLinkBERT and BioM-BERT. More than 500 000 text fragments were considered for annotation, finally resulting in a dataset with 8442 fragments, 794 of them being positive instances, covering 95% of the original annotated articles. When applied to gene–variant pair detection, BiomedBERT-large achieves the highest F1 score (0.84) after fine-tuning, demonstrating significant improvement compared to the non-fine-tuned model, underlining the relevance of the DUVEL dataset. This study shows how AL may play an important role in the creation of bioRE dataset relevant for biomedical curation applications. DUVEL provides a unique biomedical corpus focusing on 4-ary relations between two genes and two variants. It is made freely available for research on GitHub and Hugging Face. Database URL: https://huggingface.co/datasets/cnachteg/duvel or https://doi.org/10.57967/hf/1571