grid_feedback_optimizer.engine package

Submodules

grid_feedback_optimizer.engine.grad_proj_optimizer module

class grid_feedback_optimizer.engine.grad_proj_optimizer.GradientProjectionOptimizer(opt_model_data: OptimizationModelData, sensitivities: dict, alpha: float = 0.5, solver: str = 'CLARABEL', **solver_kwargs)

Bases: object

Gradient projection optimizer. Caches the CVXPY problem to allow fast updates of parameters.

solve_problem(opt_input: OptimizationInputs)

Update CVXPY parameters using structured optimization inputs, solve the cached optimization problem, and return optimized setpoints.

Parameters:

opt_input (OptimizationInputs) –

Structured input model containing:

• u_pu_measnp.ndarray

  Measured node voltages [p.u.]

• P_line_meas, Q_line_measnp.ndarray

  Measured active/reactive line power flows

• p_gen_last, q_gen_lastnp.ndarray

  Previous generator active/reactive power setpoints

• P_transformer_meas, Q_transformer_measnp.ndarray, optional

  Measured transformer active/reactive power (if applicable)

Returns:

Optimized generator setpoints of shape (n_generators, 2), where each row contains [p_opt, q_opt].

Return type:

np.ndarray

grid_feedback_optimizer.engine.powerflow module

class grid_feedback_optimizer.engine.powerflow.PowerFlowSolver(network: Network)

Bases: object

Run a power flow simulation and return bus voltages and line currents.

build_network(network: Network)

is_congested(output_data: Dict[ComponentType, ndarray] | None = None, tol_v: float = 0.0001, tol_l: float = 0.0001, tol_t: float = 0.0001)

Check if the network is congested. Returns True if any of the following occur: - Bus voltage exceeds limits (u_pu_max or u_pu_min) - Line or transformer loading exceeds 1.0 (100%)

obtain_sensitivity(delta_p: float = 1.0, delta_q: float = 1.0, loading_meas_side: str = 'from', rel_tol: float = 0.0001, rel_tol_line: float = 0.01)

Compute sensitivities of bus voltages and line/transformer power flows to small perturbations in generator power injections (p and q) around the default operating point.

Parameters:

• delta_p (float) – Active power perturbation (W).

• delta_q (float) – Reactive power perturbation (VAr).

• loading_meas_side (str) – From which side branches are monitored: “from” or “to”.

Returns:

sensitivities – { “du_dp”: array (n_bus, n_gen), “du_dq”: array (n_bus, n_gen), “dP_line_dp”: array (n_line, n_gen), “dQ_line_dp”: array (n_line, n_gen), “dP_line_dq”: array (n_line, n_gen), “dQ_line_dq”: array (n_line, n_gen), “dP_transformer_dp”: array (n_transformer, n_gen), “dQ_transformer_dp”: array (n_transformer, n_gen), “dP_transformer_dq”: array (n_transformer, n_gen), “dQ_transformer_dq”: array (n_transformer, n_gen) }

Return type:

dict

static prune_relative(A: ndarray, rel_tol: float = 0.0001)

Zero very small elements relative to the matrix scale.

run(gen_update: ndarray = None, load_update: ndarray = None)

Re-run power flow in optimization iterations. The size of the arrays should match the total numbers of gens.

grid_feedback_optimizer.engine.primal_dual_optimizer module

class grid_feedback_optimizer.engine.primal_dual_optimizer.PrimalDualOptimizer(opt_model_data: OptimizationModelData, sensitivities: dict, alpha: float = 0.5, alpha_v: float = 10.0, alpha_l: float = 10.0, alpha_t: float = 10.0, solver: str = 'CLARABEL', **solver_kwargs)

Bases: object

A primal-dual gradient projection feedback optimizer.

static calc_loading(p: ndarray | float, q: ndarray | float, s: ndarray | float) → ndarray | float

static calc_pf(p: ndarray | float, q: ndarray | float) → ndarray | float

static calc_rpf(p: ndarray | float, q: ndarray | float) → ndarray | float

solve_problem(opt_input: OptimizationInputs, grad_callback: Callable | None = None, **callback_kwargs)

Update parameters and implement primal-dual gradient projection.

Parameters:

• opt_input (OptimizationInputs) –

  Structured input model containing:

  • u_pu_measnp.ndarray

    Measured node voltages [p.u.]

  • P_line_meas, Q_line_measnp.ndarray

    Measured active/reactive line power flows

  • p_gen_last, q_gen_lastnp.ndarray

    Previous generator active/reactive power setpoints

  • P_transformer_meas, Q_transformer_measnp.ndarray, optional

    Measured transformer active/reactive power (if applicable)

• grad_callback (callable | None = None) – A function which takes grad_p and grad_q and outputs grad_p and grad_q

Returns:

Optimized generator setpoints of shape (n_generators, 2), where each row contains [p_opt, q_opt].

Return type:

np.ndarray

grid_feedback_optimizer.engine.renew_gen_projection module

class grid_feedback_optimizer.engine.renew_gen_projection.RenewGenProjection(solver: str = 'CLARABEL', **solver_kwargs)

Bases: object

Project points onto the feasible inverter operating region. Supports analytical and CVXPY-based projections.

static analytic_projection(p_max: float, s_inv: float, p: float, q: float)

Project a point (p, q) onto the feasible operating region of a PV inverter defined by active power limit p_max and apparent power limit s_inv.

Based on: Optimal Power Flow Pursuit (Appendix B) — setpoint update rule.

Parameters:

• p_max (float) – Active power limit (p_max >= 0)

• s_inv (float) – Inverter apparent power capacity (s_inv >= 0)

• p (float) – Active and reactive power values to project

• q (float) – Active and reactive power values to project

Returns:

The projected point [p_proj, q_proj] lying within the feasible PV region.

Return type:

np.ndarray

Raises:

ValueError – If projection fails or inputs are inconsistent.

projection(p_max: float, p_min: float, p: float, q: float, s_inv: float | None = None, pf_min: float | None = None, q_min: float | None = None, q_max: float | None = None)

Hybrid projection: analytic if simple, CVXPY otherwise.

grid_feedback_optimizer.engine.solve module

grid_feedback_optimizer.engine.solve.solve(network: Network, max_iter: int = 1000, tol: float = 0.001, delta_p: float = 1.0, delta_q: float = 1.0, algorithm: str = 'gp', alpha: float = 0.5, alpha_v: float = 10.0, alpha_l: float = 10.0, alpha_t: float = 10.0, record_iterates: bool = True, solver: str = 'CLARABEL', loading_meas_side: str = 'from', rel_tol: float = 0.0001, rel_tol_line: float = 0.01, **solver_kwargs)

Solve the grid optimization problem by iterating between power flow and optimization.

Module contents