Table 2.

Summary of parameters and variables in the proposed architecture.

Symbol	Description
X	Input sequence (time series of interest)
$x_t$	Input at time step t
h	Sequence of hidden states from LSTM encoder
$h_t$	Hidden state at time step t
W	Weight matrices
b	Bias vectors
$i_t$	Input gate vector at time step t
$f_t$	Forget gate vector at time step t
$o_t$	Output gate vector at time step t
$g_t$	Candidate gate vector at time step t
$c_t$	Candidate memory cell at time step t
c	Feature map of the input sequence from CNN
${\alpha }_{t,i}$	Attention weights for encoder hidden state $h_i$ and CNN output $c_i$
$u_t$	Context vector at time step t
$Q_t$	Query matrix at time step t
$K_t$	Key matrix at time step t
$V_t$	Value matrix at time step t
${\beta }_{t,i}$	Multi-head attention weights for key-value pairs
$v_t$	Multi-head context vector at time step t
${\eta }_1$	Intermediate output of the GRN
${\eta }_2$	Intermediate output of the GRN
${\widehat{y}}_t$	Quantile regression output at time step t