Dynamic graph structure and spatio-temporal representations in wind power forecasting | Science and Technology for Energy Transition (STET)

Decarbonizing Energy Systems: Smart Grid and Renewable Technologies

Open Access

Table A1

The major symbols and their definitions.

Symbol	Definition
X ∈ ℝ^N × C	Observation data matrix of wind turbines with N features and C time points.
X_t ∈ ℝ^N × 1	Wind power data vector at timestamp t, and x_i represents the i-th feature.
ϕ (·)	Nonlinear function for DSTG model fitting.
$\hat{Y} \in R^{t^{'} \times 1}$ $\widehat{Y}\in {\mathbb{R}}^{{t}^\mathrm{\prime}\times 1}$	Predicted active power values for t' time steps.
$G_{T} = (V, E, A)$ ${G}_T=(V,\mathcal{E},{A})$	Temporal graph, where V is the set of nodes, $E$ $\mathcal{E}$ is the edge set, and A is the adjacency matrix.
V ∈ ℝ^1 × N	Set of nodes representing multivariate signals.
$F_{i} \in R^{N \times F_{emmd}}$ ${{F}}_i\in {\mathbb{R}}^{N\times {F}_{\mathrm{emmd}}}$	Feature matrix for each time series variable $x_{i}$ ${x}_i$ after EEMD.
$E_{mn}$ ${\mathcal{E}}_{{mn}}$	Edge weight representing the connection between nodes z_m and z_n.
A_mn = g(z_m, z_n)	Adaptive function to determine edge connectivity based on node features.
L_gsl	Loss function for graph structure learning incorporating distance and regularization.
$I, F, O$ $\mathcal{I},\mathcal{F},\mathcal{O}$ and $H$ $\mathcal{H}$	Gates in the GAL-cell: input, forget, output gates and hidden states.
w	Trainable weight vector for GSL module.
W	Trainable matrices for different DTG layers.
$E_{t}^{G}$ ${{E}}_t^{\mathcal{G}}$	Feature matrices for the t-th graph structure.
L_mse	Mean squared error loss between true and predicted values.
L_s	Regularization terms for time smoothness.
L_total	Total loss function for model optimization.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.