Improved day-ahead predictions of load and renewable generation by optimally exploiting multi-scale dependencies

Capturing Spatio-Temporal Dependencies in the Probabilistic Forecasting of Distribution Locational Marginal Prices

“…the input gate i t which memorizes the new information revealed over time (2), the forget gate f t which has the ability to discard irrelevant information from the past (3), and the output gate o t that extracts the relevant information from the memory content c t to compute the LSTM state h t (4). Since the neural network is composed of multiple LSTM neurons, the information can be either propagated or eliminated among different units such that the tool is potentially able to model any complex nonlinear signals, resulting in performance enhancement [32]. The standard LSTM is implemented by the following composite function H LST M :…”

Non‐linear hybrid approach for the scheduling of merchant underground pumped hydro energy storage

“…The scenarios are here predicted with the two-stage procedure presented in [26]. Firstly, multivariate probabilistic forecasts (under the form of densities) are generated for each time step of the scheduling horizon using recurrent neural networks [27]. Secondly, a copula-based sampling strategy (from the forecasted densities) is implemented to obtain time trajectories that embody both the temporal information of individual variables (e.g.…”

“…Complex geometries thereby result in highly nonlinear functions. Owing to friction and turbulence within the penstock, the net head is always lower than the gross head (27). This penstock head loss (26) is usually modelled as a quadratic function of the water flow, whose coefficient depends on the pipe characteristics [34] h h, τ…”

“…9 for an actual Belgian cavity (located in Maizeret) are approximated by third-degree polynomials. In addition, it is necessary to accurately model both head variations (which depend on the geometry of reservoirs) using (24)- (27), and UPC (18) and (19). Finally, since GAs cannot handle simultaneously integer variables and equality constraints, the latter need to be transformed into two inequality constraints.…”

Chance-Constrained Scheduling of Underground Pumped Hydro Energy Storage in Presence of Model Uncertainties

“…An important advantage of the proposed procedure is that modeling errors and endogenous model uncertainties can be considered independently from the exogenous sources of uncertainty (that do not influence the UPHES state, such as market conditions), which are here modeled with scenarios to fully exploit the knowledge of the associated probability distributions (that can be efficiently forecasted [37]- [38]). Results from a case study on a hypothetical UPHES plant on an actual candidate site demonstrate that the proposed chanceconstrained method allows determining the risk attitude that maximizes profits, thereby outperforming their risk-neutral, deterministic equivalents.…”