The proposed surrogate loss function for the ASP can be represented by 1010101010101010
\begin{aligned}
& \min_{\bm{S}(.) \in \mathcal{S}_{DNN^+},\{\bm{S}: \mathcal{X} \rightarrow \Re^{+}\}} \frac{1}{2T(c^{\mathcal{W}} + c^{\mathcal{I}})}\sum_{t=1}^{T}f\left(\hat{p}^t ,p^t\right) = \\
& \frac{1}{2T(c^{\mathcal{W}} + c^{\mathcal{I}})}\left(c^{\mathcal{W}}\left \lVert \sum_{i=2}^{n} \left(\bm{W}_{i-1} + \bm{p}_{i-1}-\bm{\hat{p}}_{i-1}\right)_{+}\right \rVert_{1} \right)+ \\
& \qquad \qquad \qquad \left( c^{\mathcal{I}} \left \lVert \sum_{i=2}^{n} \left (-\bm{W}_{i-1}+\bm{\hat{p}}_i-\bm{p}_i\right)_{+}\right \rVert_{1}\right)
\end{aligned}
where q = \frac{c^{\mathcal{W}}}{c^{\mathcal{W}} + c^{\mathcal{I}}}, \hat{p}_{i} = \bm{S}(\bm{x}_{i},\bm{\omega},\bm{b}), and W_{i} is calculated from a modified version of Lindley’s recursion lindley1952theory. Note that we use (.)_{+} indicating \max(. , 0).