This work presents a hybrid and hierarchical deep learning model for midterm load forecasting. The combines exponential smoothing (ETS), advanced long short-term memory (LSTM), ensembling. ETS extracts dynamically the main components of each individual time series enables to learn their representation. Multilayer LSTM is equipped with dilated recurrent skip connections spatial shortcut path from lower layers allow better capture long-term seasonal relationships ensure more efficient...

Short-term load forecasting (STLF) is challenging due to complex time series (TS) which express three seasonal patterns and a nonlinear trend. This article proposes novel hybrid hierarchical deep-learning (DL) model that deals with multiple seasonality produces both point forecasts predictive intervals (PIs). It combines exponential smoothing (ES) recurrent neural network (RNN). ES extracts dynamically the main components of each individual TS enables on-the-fly deseasonalization,...

In this paper, we propose a new short-term load forecasting (STLF) model based on contextually enhanced hybrid and hierarchical architecture combining exponential smoothing (ES) recurrent neural network (RNN). The is composed of two simultaneously trained tracks: the context track main track. introduces additional information to It extracted from representative series dynamically modulated adjust individual forecasted by RNN consists multiple layers stacked with dilations equipped recently...

Short-term load forecasting (STLF) is a challenging problem due to the complex nature of time series expressing multiple seasonality and varying variance. This paper proposes an extension hybrid model combining exponential smoothing dilated recurrent neural network (ES-dRNN) with mechanism for dynamic attention. We propose new gated cell - attentive cell, which implements attention weighting input vector components. The most relevant components are assigned greater weights, subsequently...

This paper describes a family of seasonal and non-seasonal time series models that can be viewed as generalisations additive multiplicative exponential smoothing to model grow faster than linear but slower exponential. Their development is motivated by fast-growing, volatile series. In particular, our have global trend smoothly change from combined with local trend. Seasonality, when used, in models, the error always heteroscedastic through parameter sigma. We leverage state-of-the-art...

Contemporary datacenters house tens of thousands servers. The servers are closely monitored for operating conditions and utilizations by collecting their performance data (e.g., CPU utilization). In this paper, we show that existing database file-system solutions not suitable warehousing collected from a large number because the scale complexity data. We describe design implementation DataGarage, system have developed at Microsoft. DataGarage is hybrid solution combines benefits DBMSs,...

Short-term load forecasting (STLF) is challenging due to complex time series (TS) which express three seasonal patterns and a nonlinear trend. This paper proposes novel hybrid hierarchical deep learning model that deals with multiple seasonality produces both point forecasts predictive intervals (PIs). It combines exponential smoothing (ES) recurrent neural network (RNN). ES extracts dynamically the main components of each individual TS enables on-the-fly deseasonalization, particularly...

With the advent of Big Data, nowadays in many applications databases containing large quantities similar time series are available. Forecasting these domains with traditional univariate forecasting procedures leaves great potentials for producing accurate forecasts untapped. Recurrent neural networks (RNNs), and particular Long Short-Term Memory (LSTM) networks, have proven recently that they able to outperform state-of-the-art methods this context when trained across all available series....

In this paper, we propose a new short-term load forecasting (STLF) model based on contextually enhanced hybrid and hierarchical architecture combining exponential smoothing (ES) recurrent neural network (RNN). The is composed of two simultaneously trained tracks: the context track main track. introduces additional information to It extracted from representative series dynamically modulated adjust individual forecasted by RNN consists multiple layers stacked with dilations equipped recently...

Time series forecasting is an active research topic in academia as well industry. Although we see increasing amount of adoptions machine learning methods solving some those challenges, statistical remain powerful while dealing with low granularity data. This paper introduces a refined Bayesian exponential smoothing model the help probabilistic programming languages including Stan. Our refinements include additional global trend, transformation for multiplicative form, noise distribution and...

This work presents a hybrid and hierarchical deep learning model for mid-term load forecasting. The combines exponential smoothing (ETS), advanced Long Short-Term Memory (LSTM) ensembling. ETS extracts dynamically the main components of each individual time series enables to learn their representation. Multi-layer LSTM is equipped with dilated recurrent skip connections spatial shortcut path from lower layers allow better capture long-term seasonal relationships ensure more efficient...

Power systems operate under uncertainty originating from multiple factors that are impossible to account for deterministically. Distributional forecasting is used control and mitigate risks associated with this uncertainty. Recent progress in deep learning has helped significantly improve the accuracy of point forecasts, while accurate distributional still presents a significant challenge. In paper, we propose novel general approach capable predicting arbitrary quantiles. We show our can be...

Real-world time series often exhibit complex interdependencies that cannot be captured in isolation. Global models model past data from multiple related globally while producing series-specific forecasts locally are now common. However, their for each individual remain isolated, failing to account the current state of its neighbouring series. Multivariate like multivariate attention and graph neural networks can explicitly incorporate inter-series information, thus addressing shortcomings...

In Smyl et al. [Local and global trend Bayesian exponential smoothing models. International Journal of Forecasting, 2024.], a generalised model was proposed that is able to capture strong trends volatility in time series. This method achieved state-of-the-art performance many forecasting tasks, but its fitting procedure, which based on the NUTS sampler, very computationally expensive. this work, we propose several modifications original model, as well bespoke Gibbs sampler for posterior...

This paper introduces a recurrent neural network approach for predicting user lifetime value in Software as Service (SaaS) applications. The accounts three connected time dimensions. These dimensions are the cohort (the date joined), age-in-system since joined service) and calendar is an (i.e., contemporaneous information).The networks use multi-cell architecture, where each cell resembles long short-term memory network. applied to both acquisition (new users) rolling (existing user) values...