TPC#

Transient Predictive Control (TPC) is a DDPC method with the following key properties:

  • Small online optimization problem,

  • Predictors are causal, and

  • not biased when closed-loop training data is used.

TPC is presented in the following research papers:

Parameters and Variables#

Symbol

Definition

Dimensions

\(H_p,\ H_u\)

Multistep predictor

\(n\_y_f \times n\_z_p\); \(n\_y_f \times n\_u_f\)

\(\Lambda_{uu},\ \Lambda_{uz},\ \Lambda_{uy}\)

Regularization matrices

\(n\_u_f \times n\_u_f\); \(n\_u_f \times n\_z_p\); \(n\_u_f \times n\_y_f\)

Optimization Formulation#

\[\begin{split}\min_{u_f,y_f} &\quad \|y_f - y_r\|_Q^2 + \|u_f - u_r\|_R^2 + r(u_f, y_r, z_p) \\ \text{s.t.} &\quad y_f = H_u u_f + H_p z_p \\ &\quad u_f \in \mathcal{U}, \quad y_f \in \mathcal{Y}\end{split}\]

where:

\[r(u_f, y_r, z_p) := u_f^T \Lambda_{uu}u_f+2u^T_f \Lambda_{uz}z_p+2u_f^T \Lambda_{uy}y_r\]

Closed-Form Solution Derivation#

Set \(y_f\) into the cost equation leads to:

\[\|H_u u_f + H_p z_p - y_r\|_Q^2 + \|u_f - u_r\|_R^2 + u_f^T \Lambda_{uu}u_f+2u^T_f \Lambda_{uz}z_p+2u_f^T \Lambda_{uy}y_r\]

Now taking the derivative in respect to \(u_f\) and set it to zero.

\[(H_u ^T Q H_u + R + \Lambda_{uu})u_f^\star = (-H_u ^TQH_p -\Lambda_{uz})z_p + (H_u^TQ - \Lambda_{uy})y_r + R u_r\]

Defining the help matrices:

\[\begin{split}F_1 &= (H_u ^T Q H_u + R + \Lambda_{uu})^{-1} \\ F_2 &= -H_u ^TQH_p -\Lambda_{uz} \\ F_3 &= H_u^TQ - \Lambda_{uy}\end{split}\]

Leads to the following closed-form gain matrices:

\[\begin{split}K_{z_p} &= F_1 F_2 \\ K_{y_r} &= F_1 F_3 \\ K_{u_r} &= F_1 R\end{split}\]

With the optimal \(u_f^\star\) calculated as:

\[u_f^* = K_{z_p} z_p + K_{y_r} y_r + K_{u_r} u_r\]