Robust streamflow forecasting: a Student’s t-mixture vector autoregressive model

Favereau, Marcel; Lorca, Álvaro; Negrete-Pincetic, Matías; Vicuña, Sebastián

doi:10.1007/s00477-022-02241-y

Cited by 4 publications

(3 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Results are averaged over all test executions (10% of sequences). 5 The results highlight that at least one variant of the method (GAP-LSTM-Weighted or GAP-LSTM-Default) achieves the best forecasting accuracy in terms of average RMSE, for all the analyzed datasets with the exception of Beijing Air Quality, where our method provides the second-best results. Particularly, GAP-LSTM-Weighted performs better on the Lightsource and PEMS-SF Weather datasets, possibly due to the presence of highly correlated clusters of nodes operating similarly, as well as the high similarity and regularity of solar energy prediction and traffic curves in these datasets.…”

Section: Resultsmentioning

confidence: 76%

“…Popular models include autoregressive integrated moving average (ARIMA) [1], [2] and Prophet [3]. Other works combine autoregressive approaches and evolutionary algorithms to better explore model variants in the search space and select better performing models [4], or combine them with fuzzy or other types of statistical modeling [5], [6], [7]. Among their benefits, it is worthwhile mentioning their ease of use and their determinism, given their statistical nature.…”

Section: A Autoregressive Modelsmentioning

confidence: 99%

“…While autoregressive models capture temporal autocorrelation, they are typically unable to analyze multivariate data and effectively exploit spatial patterns [1], [2], [3], [4], [5], [6], [7]. Vector autoregression (VAR)-based approaches are able to model spatial dependencies, albeit in a very basic form, by means of coefficients learned solely for a target feature of interest [8], [9].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

GAP-LSTM: Graph-Based Autocorrelation Preserving Networks for Geo-Distributed Forecasting

Altieri,

Corizzo,

Ceci

2024

IEEE Trans. Neural Netw. Learning Syst.

View full text Add to dashboard Cite

Forecasting methods are important decision support tools in geo-distributed sensor networks. However, challenges such as the multivariate nature of data, the existence of multiple nodes, and the presence of spatio-temporal autocorrelation increase the complexity of the task. Existing forecasting methods are unable to address these challenges in a combined manner, resulting in a suboptimal model accuracy. In this article, we propose GAP-LSTM, a novel geo-distributed forecasting method that leverages the synergic interaction of graph convolution, attention-based long short-term memory (LSTM), 2-D-convolution, and latent memory states to effectively exploit spatio-temporal autocorrelation in multivariate data generated by multiple nodes, resulting in improved modeling capabilities. Our extensive evaluation, involving real-world datasets on traffic, energy, and pollution domains, showcases the ability of our method to outperform state-of-the-art forecasting methods. An ablation study confirms that all method components provide a positive contribution to the accuracy of the extracted forecasts. The method also provides an interpretable visualization that complements forecasts with additional insights for domain experts.

show abstract

Section: Resultsmentioning

confidence: 76%