epyt_control.controllers

epyt_control.controllers.pid

This module contains an implementation of a PID controller and a learner (i.e. tuning method) based on an evolutionary algorithm.

class epyt_control.controllers.pid.PidController(proportional_gain: float, integral_gain: float, derivative_gain: float, target_value: float, action_lower_bound: float | None = None, action_upper_bound: float | None = None)[source]

Bases: object

Implementation of a Proportional-Integral-Derivative (PID) controller.

Parameters:

  • proportional_gain (float) – Proportional gain coefficient.

  • integral_gain (float) – Integral gain coefficient.

  • derivative_gain (float) – Derivative gain coefficient.

  • target_value (float) – Target value (observed system state) which the controller is supposed to reach.

  • action_lower_bound (float, optional) –

    Lower bound of the computed action. Smaller control outputs will be clipped.

    The default is None.

  • action_upper_bound (float, optional) –

    Upper bound of the computed action. Control outputs exceeding this upper bound will be clipped.

    The default is None.

property action_lower_bound: float

Lower bound of the computed action. Smaller control outputs will be clipped.

Returns:

Lower bound of the computed action.

Return type:

float

property action_upper_bound: float

Upper bound of the computed action. Control outputs exceeding this upper bound will be clipped.

Returns:

Upper bound of the computed action.

Return type:

float

property derivative_gain: float

Returns the derivative gain coefficient.

Returns:

Derivative gain coefficient.

Return type:

float

property integral_gain: float

Returns the integral gain coefficient.

Returns:

Integral gain coefficient.

Return type:

float

property proportional_gain: float

Returns the proportional gain coefficient.

Returns:

Proportional gain coefficient.

Return type:

float

step(cur_value: float) → float[source]

Computes the current/next control action/signal based on the given observation (i.e. system state).

Parameters:

cur_value (float) – Current observation (i.e. system state).

Returns:

Computed action – i.e. control signal.

Return type:

float

property target_value: float

Returns the target value (observed system state) which the controller is supposed to reach.

Returns:

Target value (system state).

Return type:

float

epyt_control.controllers.lqr

This module contains implementations of different Linear Quadratic Regulator (LQR) variants.

epyt_control.controllers.lqr.linear_quadratic_regulator(current_state: numpy.ndarray, state_cost_mat: numpy.ndarray, action_cost_mat: numpy.ndarray, state_transition_mat: numpy.ndarray, action_transition_mat: numpy.ndarray, time_horizon: int, final_state_cost_mat: numpy.ndarray | None = None) → list[numpy.ndarray][source]

Computes the Linear Quadratic Regulator (LQR) control solution of a given inite-horizon & discrete-time LQR problem.

Parameters:

  • current_state (numpy.ndarray) – Current system state.

  • state_cost_mat – Cost matrix of states – i.e. a s.p.s.d. matrix specifying the cost of a given state.

  • action_cost_mat – Cost matrix of actions – i.e. a s.p.d. matrix specifying the cost of a action state.

  • state_transition_mat – State transition matrix – i.e. mapping a given state to the next state (without any action).

  • action_transition_mat – Action transition matrix – i.e. mapping specifying the state change/influence of taking an action.

  • time_horizon (int) – Time horizon.

  • final_state_cost_mat – Cost matrix of the final state – i.e. a s.p.s.d. matrix specifying the cost of the final state. If None, ‘state_cost_mat’ will be used for the final state cost.

Returns:

List of actions for reaching the specified target space.

Return type:

list[numpy.ndarray]

epyt_control.controllers.lqr.time_varying_lqr(current_state: numpy.ndarray, state_cost: Callable[[int], numpy.ndarray], action_cost: Callable[[int], numpy.ndarray], state_transition: Callable[[int], numpy.ndarray], action_transition: Callable[[int], numpy.ndarray], time_horizon: int, final_state_cost_mat: numpy.ndarray | None = None) → list[numpy.ndarray][source]

Computes the Linear Quadratic Regulator (LQR) control solution of a given inite-horizon & discrete-time LQR problem with time varying parameters.

Parameters:

  • current_state – Current system state.

  • state_cost – Time varying cost matrix of states – i.e. mapping time to a s.p.s.d. matrix specifying the cost of a given state.

  • action_cost – Time varying cost matrix of actions – i.e. mapping time to a s.p.d. matrix specifying the cost of a action state.

  • state_transition – Time varying state transition matrix – i.e. mapping time to a matrix for mapping a given state to the next state (without any action).

  • action_transition – Time varying action transition matrix – i.e. mapping time to a matri for specifying the state change/influence of taking an action.

  • time_horizon (int) – Time horizon.

  • final_state_cost_mat – Cost matrix of the final state – i.e. a s.p.s.d. matrix specifying the cost of the final state. If None, ‘state_cost’ will be used for getting final state cost.

Returns:

List of actions for reaching the specified target space.

Return type:

list[numpy.ndarray]