run_topochange.py

from __future__ import annotations

from collections.abc import Callable, Sequence
import pickle
import datetime as dt
import io
import os
from typing import Any

import cvxpy as cp
import matplotlib.pyplot as plt
import numpy as np
from tqdm.auto import tqdm

from cbc.base import (
    CBCBase, CBCConst, CBCConstWithNoise, project_into_X_set)
from network_utils import (
    create_56bus, create_RX_from_net, meta_gen_X_set, read_load_data)
from robust_voltage_control import (
    VoltPlot, robust_voltage_control)
from utils import wrap_write_newlines

os.environ["OMP_NUM_THREADS"] = "1"
os.environ["OPENBLAS_NUM_THREADS"] = "1"
os.environ["MKL_NUM_THREADS"] = "1"
os.environ["VECLIB_MAXIMUM_THREADS"] = "1"
os.environ["NUMEXPR_NUM_THREADS"] = "1"

# hide top and right splines on plots
plt.rcParams['axes.spines.right'] = False
plt.rcParams['axes.spines.top'] = False


def create_sel(
    cbc_alg: str, n: int, T: int, X_init: np.ndarray, v: np.ndarray,
    gen_X_set: Callable[[cp.Variable], list[cp.Constraint]],
    X_true: np.ndarray, obs_nodes: Sequence[int] | None,
    log: tqdm | io.TextIOBase | None, config: dict[str, Any],
    eta: float, nsamples: int, alpha: float,
    Vpar: tuple[np.ndarray, np.ndarray], δ: float, seed: int
) -> tuple[CBCBase, bool]:
    """
    Args
    - cbc_alg: str, one of ['const', 'lsq', 'proj', 'steiner']
    - n: int, # of buses
    - T: int, maximum # of time steps
    - X_init: np.array, shape [n, n], initial guess for X matrix, must be
        PSD and entry-wise >= 0
    - v: np.array, shape [n], initial squared voltage magnitudes
    - gen_X_set: function, takes an optimization variable (cp.Variable) and
        returns a list of constraints (cp.Constraint) describing the
        convex, compact uncertainty set for X
    - X_true: np.array, shape [n, n], true X matrix, optional
    - obs_nodes: list of int, nodes that we can observe voltages for
    - log: object with .write() function, defaults to tqdm
    - config: dict, only used by CBCProjection* and CBCSteiner
    - eta: float, noise bound, only used by CBCProjection* and CBCSteiner
    - nsamples: int, # of observations to use for defining the convex set
    - alpha: float, weight on CBCProjection slack variable,
        set to 0 to turn off slack variable
    - Vpar: tuple (Vpar_min, Vpar_max), box description of Vpar
        - each Vpar_* is a np.array of shape [n]
    - δ: float, weight of noise term in CBC norm when learning eta,
        only used by CBCProjectionWithNoise
    - seed: int, only used by CBCProjection* and CBCSteiner

    Returns:
    - sel: CBC algorithm
    - requires_X_psd: whether X returned by CBC algorithm will be PSD
    """
    sel: CBCBase
    require_X_psd = True
    if cbc_alg == 'const':
        if δ == 0:
            sel = CBCConst(
                n=n, T=T, X_init=X_init, v=v,
                gen_X_set=gen_X_set, X_true=X_true, obs_nodes=obs_nodes, log=log)
        else:
            sel = CBCConstWithNoise(
                n=n, T=T, X_init=X_init, v=v,
                gen_X_set=gen_X_set, X_true=X_true, obs_nodes=obs_nodes, log=log)
    elif cbc_alg == 'lsq':
        assert δ == 0
        from cbc.lsq import CBCLsq
        sel = CBCLsq(
            n=n, T=T, X_init=X_init, v=v,
            gen_X_set=gen_X_set, X_true=X_true, obs_nodes=obs_nodes, log=log)
        require_X_psd = False
    elif cbc_alg == 'proj':
        from cbc.projection import CBCProjection, CBCProjectionWithNoise
        config.update(alpha=alpha, nsamples=nsamples)
        if δ == 0:
            sel = CBCProjection(
                n=n, T=T, X_init=X_init, v=v,
                gen_X_set=gen_X_set, eta=eta, nsamples=nsamples, alpha=alpha,
                Vpar=Vpar, X_true=X_true, obs_nodes=obs_nodes, log=log, seed=seed)
        else:
            sel = CBCProjectionWithNoise(
                n=n, T=T, X_init=X_init, eta_init=0, v=v,
                gen_X_set=gen_X_set, eta=eta, nsamples=nsamples, δ=δ,
                Vpar=Vpar, X_true=X_true, obs_nodes=obs_nodes, log=log, seed=seed)
        # save_dict.update(w_inds=sel.w_inds, vpar_inds=sel.vpar_inds)
    elif cbc_alg == 'steiner':
        assert δ == 0
        from cbc.steiner import CBCSteiner
        dim = n * (n+1) // 2
        config.update(nsamples=nsamples, nsamples_steiner=dim)
        sel = CBCSteiner(
            eta=eta, n=n, T=T, nsamples=nsamples, nsamples_steiner=dim,
            v=v, gen_X_set=gen_X_set, Vpar=Vpar,
            X_init=X_init, X_true=X_true, obs_nodes=obs_nodes, log=log, seed=seed)
    else:
        raise ValueError('unknown cbc_alg')
    return sel, require_X_psd


def run(ε: float, q_max: float, cbc_alg: str, eta: float,
        norm_bound: float, norm_bound_init: float | None = None,
        noise: float = 0, modify: str | None = None, δ: float = 0.,
        obs_nodes: Sequence[int] | None = None,
        ctrl_nodes: Sequence[int] | None = None,
        known_bus_topo: int = 0, known_line_params: int = 0,
        nsamples: int = 100, seed: int = 123,
        is_interactive: bool = False, savedir: str = '',
        pbar: tqdm | None = None,
        tag: str = '') -> str:
    """
    Args
    - ε: float, robustness
    - q_max: float, maximum reactive power injection
    - cbc_alg: str, one of ['const', 'lsq', 'proj', 'steiner']
    - eta: float, maximum ‖w‖∞
    - norm_bound: float, size of uncertainty set
    - norm_bound_init: float or None, norm of uncertainty set from which
        X_init is sampled
    - noise: float, network impedances modified by fraction Uniform(±noise)
    - modify: str, how to modify network, one of [None, 'perm', 'linear', 'rand']
    - δ: float, weight of noise term in CBC norm when learning eta
    - obs_nodes: list of int, nodes that we can observe voltages for,
        set to None if we observe all voltages
    - ctrl_nodes: list of int, nodes that we can control voltages for,
        set to None if we control all voltages
    - known_bus_topo: int in [0, n], n = # of buses (excluding substation),
        when topology is known for buses/lines in {0, ..., known_bus_topo-1}
    - known_line_params: int in [0, known_bus_topo], when line parameters
        (little x_{ij}) are known ∀ i,j in {0, ..., known_line_params-1}
    - nsamples: int, # of samples to use for computing consistent set,
        only used when cbc_alg is 'proj' or 'steiner'
    - seed: int, random seed
    - is_interactive: bool, whether to output to screen, or log to disk
    - savedir: str, path to folder for saving outputs ('' for current dir)
    - pbar: tqdm instance
    - tag: str, arbitrary tag to add to filename ('' for no tag)

    Returns: str, filename (without extension)
    """
    assert δ >= 0, 'δ must be >= 0'
    assert 'topochange' in tag
    if known_line_params > 0:
        assert f'knownlines{known_line_params}' in tag
    elif known_bus_topo > 0:
        assert f'knowntopo{known_bus_topo}' in tag

    if savedir != '':
        os.makedirs(savedir, exist_ok=True)
    tz = dt.timezone(dt.timedelta(hours=-8))  # PST
    start_time = dt.datetime.now(tz)

    config: dict[str, Any] = dict(
        cbc_alg=cbc_alg, q_max=q_max, ε=ε, eta=eta, δ=δ,
        obs_nodes=obs_nodes, ctrl_nodes=ctrl_nodes, seed=seed,
        known_bus_topo=known_bus_topo, known_line_params=known_line_params)
    filename = os.path.join(savedir, f'CBC{cbc_alg}')

    if δ > 0:
        filename += f'_δ{δ}_η{eta}'

    # read in data
    if noise > 0 or modify is not None:
        config.update(norm_bound=norm_bound, norm_bound_init=norm_bound_init)
        if noise > 0:
            config.update(noise=noise)
            filename += f'_noise{noise}'
        if modify is not None:
            config.update(modify=modify)
            filename += f'_{modify}'
        if norm_bound_init is not None:
            filename += f'_norminit{norm_bound_init}'
        filename += f'_norm{norm_bound}_seed{seed}'

    net = create_56bus()
    R, X = create_RX_from_net(net, noise=0)  # true R and X
    p, qe = read_load_data()  # in MW and MVar
    T, n = p.shape
    change_t = int(T/2)  # = 7210

    # change edges (33, 40) -> (1, 40), and (46, 48) -> (10, 48)
    # - here, buses are indexed with substation = 0, then other nodes are 1,...,55
    net_mod = create_56bus()
    net_mod.line.loc[39, 'from_bus'] = 1
    net_mod.line.loc[47, 'from_bus'] = 10
    R_mod, X_mod = create_RX_from_net(net_mod)

    # ==== FIXED PARAMETERS ====
    v_min, v_max = (11.4**2, 12.6**2)  # +/-5%, units kV^2
    v_nom = 12**2  # nominal squared voltage magnitude, units kV^2
    v_sub = v_nom  # fixed squared voltage magnitude at substation, units kV^2

    vpars = qe @ X + p @ R + v_sub  # shape [T, n]
    vpars_mod = qe @ X_mod + p @ R_mod + v_sub  # shape [T, n]

    vpars_combined = np.concatenate([vpars[:change_t], vpars_mod[change_t:]])
    Vpar_min = np.min(vpars_combined, axis=0)  # shape [n]
    Vpar_max = np.max(vpars_combined, axis=0)  # shape [n]
    Vpar = (Vpar_min, Vpar_max)

    Pv = 0.1
    Pu = 10

    # weights on slack variables: alpha for CBC, β for robust oracle
    alpha = 1000  # only used when not learning eta, set to 0 to turn off slack variable
    β = 100

    config.update(v_min=v_min, v_max=v_max, v_nom=v_nom, Pv=Pv, Pu=Pu, β=β)
    # ==== end of FIXED PARAMETERS ====

    filename += tag
    filename += start_time.strftime('_%Y%m%d_%H%M%S')
    if is_interactive:
        log = tqdm
    else:
        log = wrap_write_newlines(open(f'{filename}.log', 'w'))
        print(f'filename: {filename}')
    log.write(f'filename: {filename}')

    start = 0  # starting time step

    # randomly initialize a network matrix
    _, X_init = create_RX_from_net(net, noise=noise, modify=modify,
                                   check_pd=True, seed=seed)
    save_dict = dict(X_init=X_init)
    if norm_bound_init is not None:
        assert norm_bound_init < norm_bound
        var_X = cp.Variable(X.shape, PSD=True)
        init_X_set = meta_gen_X_set(
            norm_bound=norm_bound_init, X_true=X, net=net,
            known_bus_topo=known_bus_topo, known_line_params=known_line_params
        )(var_X)
        project_into_X_set(X_init=X_init, var_X=var_X, log=log,
                           X_set=init_X_set, X_true=X)
        X_init = var_X.value

    gen_X_set = meta_gen_X_set(
        norm_bound=norm_bound, X_true=X, net=net,
        known_bus_topo=known_bus_topo, known_line_params=known_line_params)

    volt_plot = VoltPlot(
        v_lims=(np.sqrt(v_min), np.sqrt(v_max)),
        q_lims=(-q_max, q_max))

    pkl_name = None
    # pkl_name = 'CBCproj_δ20_η10_noise1.0_perm_norm1.0_seed8_knownlines14_topochange_20240110_120849.pkl'
    # pkl_name = 'CBCproj_δ20_η10_noise1.0_perm_norm1.0_seed9_knownlines14_topochange_20240110_125001.pkl'
    # pkl_name = 'CBCproj_δ20_η10_noise1.0_perm_norm1.0_seed8_topochange_20240110_120831.pkl'
    # pkl_name = 'CBCproj_δ20_η10_noise1.0_perm_norm1.0_seed9_topochange_20240110_140752.pkl'
    if pkl_name is None:
        sel, require_X_psd = create_sel(
            cbc_alg=cbc_alg, n=n, T=T-start, X_init=X_init, v=vpars[start],
            gen_X_set=gen_X_set, X_true=X, obs_nodes=obs_nodes, log=log,
            config=config, eta=eta, nsamples=nsamples, alpha=alpha, Vpar=Vpar,
            δ=δ, seed=seed)

        vs, qcs, dists, params, consistent_arrs = robust_voltage_control(
            vpars=vpars_combined[start:T],
            v_lims=(v_min, v_max), q_lims=(-q_max, q_max), v_nom=v_nom,
            X=X, require_X_psd=require_X_psd, Pv=Pv * np.eye(n), Pu=Pu * np.eye(n),
            eta=eta, ε=ε, β=β, sel=sel, δ=δ,
            ctrl_nodes=ctrl_nodes, pbar=pbar, log=log,
            volt_plot=volt_plot if is_interactive else None,
            change_net=(change_t, X_mod))

        # save data
        with open(f'{filename}.pkl', 'wb') as f:
            pickle.dump(file=f, obj=dict(
                vs=vs, qcs=qcs, dists=dists, params=params, consistent_arrs=consistent_arrs,
                config=config, **save_dict))

    else:
        with open(os.path.join(savedir, pkl_name), 'rb') as f:
            pkl = pickle.load(f)
            vs = pkl['vs']
            qcs = pkl['qcs']
            dists = pkl['dists']
            params = pkl['params']
            consistent_arrs = pkl['consistent_arrs']

    assert np.all(vs[change_t + 1] == 0) and np.all(qcs[change_t + 1] == 0)

    start = change_t
    sel2, require_X_psd = create_sel(
        cbc_alg=cbc_alg, n=n, T=T-start, X_init=X_init, v=vs[change_t],
        gen_X_set=gen_X_set, X_true=X_mod, obs_nodes=obs_nodes, log=log,
        config=config, eta=eta, nsamples=nsamples, alpha=alpha, Vpar=Vpar,
        δ=δ, seed=seed)
    sel2.q[0] = qcs[change_t]

    vs2, qcs2, dists2, params2, consistent_arrs2 = robust_voltage_control(
        vpars=vpars_combined[start:T],
        v_lims=(v_min, v_max), q_lims=(-q_max, q_max), v_nom=v_nom,
        X=X_mod, require_X_psd=require_X_psd, Pv=Pv * np.eye(n), Pu=Pu * np.eye(n),
        eta=eta, ε=ε, β=β, sel=sel2, δ=δ,
        ctrl_nodes=ctrl_nodes, pbar=pbar, log=log,
        volt_plot=volt_plot if is_interactive else None)

    elapsed = (dt.datetime.now(tz) - start_time).total_seconds()

    # overwrite with additional data
    with open(f'{filename}.pkl', 'wb') as f:
        pickle.dump(file=f, obj=dict(
            vs=vs, qcs=qcs, dists=dists, params=params, consistent_arrs=consistent_arrs,
            vs2=vs2, qcs2=qcs2, dists2=dists2, params2=params2, consistent_arrs2=consistent_arrs2,
            config=config, elapsed=elapsed, **save_dict))

    # plot and save figure
    volt_plot.update(qcs=qcs,
                     vs=np.sqrt(vs),
                     vpars=np.sqrt(vpars),
                     dists=(dists['t'], dists['X_true']))
    volt_plot.fig.savefig(f'{filename}.svg', pad_inches=0, bbox_inches='tight')
    volt_plot.fig.savefig(f'{filename}.pdf', pad_inches=0, bbox_inches='tight')

    if not is_interactive:
        log.close()
    return filename


if __name__ == '__main__':
    savedir = 'out'
    # all_nodes = np.arange(55)
    # exclude = np.array([8, 18, 21, 30, 39, 45, 54]) - 1
    # obs_nodes = np.setdiff1d(all_nodes, exclude).tolist()
    obs_nodes = None
    for seed in [8]:  # 8, 9, 10, 11]:  # for norm_bound=1.0, noise=1.0
    # for seed in [55, 56, 57, 58]:  # for norm_bound=0.5, noise=0.5
        run(
            ε=0.1,
            q_max=0.24,
            cbc_alg='proj',  # 'proj',
            eta=10,
            norm_bound=1.0,
            norm_bound_init=None,
            noise=1.0,
            modify='perm',
            δ=20,
            obs_nodes=obs_nodes,
            ctrl_nodes=obs_nodes,
            known_line_params=0,
            known_bus_topo=14,
            seed=seed,
            pbar=tqdm(),
            is_interactive=False,
            savedir=savedir,
            tag='_knowntopo14_topochange')  # choose from ['', '_partialobs', '_partialctrl', '_knowntopoX', '_knownlinesX']

    # fixed X*, known eta
    # run(
    #     ε=0.1,
    #     q_max=0.24,
    #     cbc_alg='const',
    #     eta=8.65,
    #     norm_bound=0.,
    #     norm_bound_init=None,
    #     noise=0,
    #     modify=None,
    #     δ=0,
    #     obs_nodes=None,
    #     ctrl_nodes=None,
    #     known_line_params=0,
    #     known_bus_topo=0,
    #     seed=None,
    #     pbar=tqdm(),
    #     is_interactive=False,
    #     savedir=savedir,
    #     tag='')

    # fixed X*, unknown eta
    # run(
    #     ε=0.1,
    #     q_max=0.24,
    #     cbc_alg='const',
    #     eta=10,
    #     norm_bound=0.,
    #     norm_bound_init=None,
    #     noise=0,
    #     modify=None,
    #     δ=20,
    #     obs_nodes=None,
    #     ctrl_nodes=None,
    #     known_line_params=0,
    #     known_bus_topo=0,
    #     seed=None,
    #     pbar=tqdm(),
    #     is_interactive=False,
    #     savedir=savedir,
    #     tag='')