Class KoopmanMPCSolver
Ths class KoopmanMPCSolver is used to solve the optimal control problem using Koopman operator and model predictive control (MPC).
API Documentation¶
__init__(dynamics, koopman, dictionary, ref_traj, time_horizon, observable_pos, lambda_param)
¶
Initializes an instance of the KoopmanMPCSolver class.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
dynamics
|
KoopmanDynamics
|
The Koopman dynamics system. |
required |
koopman
|
ParamKoopman
|
The parametric Koopman operator. |
required |
dictionary
|
TrainableDictionary or Dictionary
|
A function that maps the state to the dictionary of observables. |
required |
ref_traj
|
Tensor
|
A tensor containing the reference trajectory, with shape (trajectory_length, state_dimension). |
required |
time_horizon
|
int
|
The time horizon for model predictive control. |
required |
observable_pos
|
int
|
The position(s) of target observables in dictionary. |
required |
lambda_param
|
float
|
A regularization parameter used to balance the control error and tracking error. |
required |
solve(state0, control_min, control_max, method='powell', disp=False)
¶
Solve the optimal control problem for a given initial state and control bounds. Args: state0 (torch.Tensor): The initial state of the system. control_min (float or list of floats): Minimum allowable value(s) for the control input(s). control_max (float or list of floats): Maximum allowable value(s) for the control input(s). method (str, optional): Optimization method to use. Defaults to 'powell'. disp (bool, optional): Whether to print convergence messages. Defaults to False.
Returns:
| Type | Description |
|---|---|
Tensor
|
The optimal control sequence over the given time horizon. |