Algorithm 2  Fuzzy Deep learning

Input: All time-invariant vector ${\overline{y}}_I^{\left(i\right)}$ and time-varying set ${\overline{Y}}_V^{\left(i\right)}$, with $i=1,2,\dots ,M$. Output: $N_c$ DNN models, ${\mathrm{\Theta }}_i$, with $i=1,2,\dots ,N_c$, which forecasts future                 load consumption.   Step 1: Initialize the parameters, $N_c$, m, $\overline{A}$, and $\left|\right|·{\left|\right|}_{\overline{A}}$, and the threshold, $ϵ$.   Step 2: Generate the $N_c$ cluster centres, $\overline{V}=\{{\overline{v}}_1,{\overline{v}}_2,\dots ,{\overline{v}}_{N_c}\}$, using             the FCM in Algorithm 1.   Step 3: Determine the membership of ${\overline{y}}_I^{\left(i\right)}$ to the $k^{th}$ cluster, ${\widehat{u}}_k\left(y_I^{\left(i\right)}\right)$ using (15).   Step 4: Determine the index vector, ${\widehat{p}}_j=\{p(j,1),p(j,2),\dots ,p(j,O_j)\}$ using (16),             with $j=1,2,\dots ,M$, which indexed the time-varying sets in the $j^{th}$ cluster.   Step 5: Use all time-varying sets ${\overline{Y}}_V^{p(j,k)}$ with $k=1,2,\dots ,O_j$ to develop the             DNN model, ${\mathrm{\Theta }}_j$, using deep learning.   Step 6: Return The DNN models ${\mathrm{\Theta }}_j$ with $j=1,2,\dots ,N_c$.