Algorithm 2.

Overview over the deep learning algorithm adapted for control inputs

1:	Load observed data
2:	Initialize deep learning models, optimization model, learning rate strategy
3:	Detect and label intervals		▷ Section 6.3.2
4:	for n_epochs do
5:	$\text{mean\_loss}\leftarrow 0$
6:	$\text{initial\_conditions}\leftarrow [\overrightarrow{x}(t=0)]$
7:	for $i\in \left\{1,\dots ,\text{n\_intervals}\}\right.$ do
8:	$\mathrm{i}\mathrm{n}\mathrm{p}\leftarrow \left[T_i^{\prime },\mathrm{g}\mathrm{l}\mathrm{u}\left(T_i^{\prime -}\right)\right.$, $\text{initial\_conditions}\left.[i-1]\right]$
9:	est ← model.predict(inp)
10:	initial_conditions ← initial_condition + est
11	end for
12
13	while $\mathrm{t}$, $\overrightarrow{x}=\mathrm{l}\mathrm{o}\mathrm{a}\mathrm{d}\_\mathrm{b}\mathrm{a}\mathrm{t}\mathrm{c}\mathrm{h}()$ do
14:	$\mathrm{i}\mathrm{n}\mathrm{p}\leftarrow [t,\mathrm{glu}\left(t\right),\text{initial\_conditions}[i\left]\right]$		▷ Where $t\in \left[T_i^{\prime +},T_{i+1}^{\prime -}\right]$
15:	$\hat{\overrightarrow{x}}=\mathrm{m}\mathrm{o}\mathrm{d}\mathrm{e}\mathrm{l}.\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{c}\mathrm{t}(\mathrm{t})$
16:	Compute Losses
17:	Compute ${\nabla }_{\overrightarrow{\theta }}ℒ$
18:	if clip_gradients then
19:	$\text{clip\_gradients}({\nabla }_{\overrightarrow{\theta }}ℒ)$
20:	end if
21:	optimization_step()
22:	$\text{mean\_loss}\leftarrow \text{mean\_loss}+ℒ$
23:	end while
24:	register_lr(mean_loss)	▷ Reduces LR if necessary
25:	end for
26:	Save results