I am looking for a smooth (continuous differentiable) approximation of the following five-phased linear function: $$ P(U, R, P_r) = \begin{cases} R(U_{\max} - U) + R(U_{up} - U_{\max}) + P_r; & U \geq U_{\max}\\ R(U_{up} - U) + P_r; & U_{up} < U < U_{\max}\\ P_r; & U_{low} \leq U \leq U_{up}\\ R(U_{low} - U) + P_r; & U_{\min} < U < U_{low}\\ R(U_{\min} - U) + R(U_{low} - U_{\min}) + P_r; & U \leq U_{\min}\\ \end{cases}, $$ Here $U_{\max} \geq U_{up} \geq U_{low} \geq U_{\min} $ are the breakpoints and bounds of the variable $U$. $P_r$ and $R$ are a given setpoint value and slope of the line respectively. Also, $P$ has bounds as $P_{\min} \leq P \leq P_{\max}$.

I am trying to define these constraints as nested max functions and then apply the smooth approximation. Such that: $$F(x) = \max\{0,x\}, \quad for \quad x \in \mathbb{R}$$

By picking any positive number $\epsilon > 0$:

$$F^\epsilon(x) = \epsilon \ln \left(1+ \exp(x/\epsilon)\right), \quad for \quad x \in \mathbb{R}$$

with bound verification:

$$F^\epsilon(x) - \epsilon \ln2 \leq F(x) \leq F^\epsilon(x)$$

  • $\begingroup$ If I further simplify the function as: $$ P(U, R, P_r) = \begin{cases} R(U_{up} - U) + P_r; & U_{up} < U \leq U_{max}\\ P_r; & U_{low} \leq U \leq U_{up}\\ R(U_{low} - U) + P_r; & U_{min} \leq U < U_{low}\\ \end{cases}, $$ $\endgroup$ Feb 2 at 0:35


You must log in to answer this question.