Spaces:
Runtime error
Runtime error
File size: 26,796 Bytes
f4c1c59 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 | """
PowerZoo SmartGrid: Interactive CMDP Environment Demo
HuggingFace Spaces application with Gradio + Plotly.
5 Tabs: Overview | Voltage Heatmap | Lagrangian Trajectory | Component Status | Training Dashboard
SmartGrid is a modular PV integration environment using CMDP (Constrained MDP)
with Lagrangian relaxation. 360-step annual episodes (1 step = 1 day),
homogeneous agents controlling capacitors, regulators, batteries, and PV.
"""
import numpy as np
import pandas as pd
import plotly.graph_objects as go
from plotly.subplots import make_subplots
import gradio as gr
# === Monkey-patch: fix Gradio additionalProperties schema error with Plotly ===
_original_plot_init = gr.Plot.__init__
def _patched_plot_init(self, *args, **kwargs):
_original_plot_init(self, *args, **kwargs)
if hasattr(self, "schema") and isinstance(self.schema, dict):
self.schema.pop("additionalProperties", None)
gr.Plot.__init__ = _patched_plot_init
# === Color Palette ===
COLORS = {
"primary": "#10B981", # emerald
"secondary": "#059669", # emerald dark
"accent": "#06B6D4", # cyan
"warning": "#F59E0B", # amber
"danger": "#EF4444", # red
"bg_card": "rgba(16, 185, 129, 0.05)",
"grid": "rgba(255, 255, 255, 0.08)",
"text": "#E2E8F0",
"text_dim": "#94A3B8",
}
PLOTLY_LAYOUT_DEFAULTS = dict(
template="plotly_dark",
paper_bgcolor="rgba(0,0,0,0)",
plot_bgcolor="rgba(0,0,0,0)",
font=dict(family="Inter, system-ui, sans-serif", color=COLORS["text"]),
margin=dict(l=60, r=30, t=50, b=50),
xaxis=dict(gridcolor=COLORS["grid"], zerolinecolor=COLORS["grid"]),
yaxis=dict(gridcolor=COLORS["grid"], zerolinecolor=COLORS["grid"]),
)
# Deterministic RNG for reproducible demo data
RNG = np.random.default_rng(seed=42)
# ============================================================
# Demo Data Generators
# ============================================================
BUS_NAMES_13 = [
"650", "632", "633", "634", "645", "646",
"671", "680", "684", "611", "652", "692", "675",
]
def generate_voltage_heatmap_data() -> np.ndarray:
"""Generate 360x13 voltage matrix with seasonal PV patterns.
Summer months show higher voltage from PV injection;
winter shows lower voltage from increased load.
Returns:
np.ndarray: shape (360, 13), voltage in per-unit.
"""
days = np.arange(360)
n_buses = len(BUS_NAMES_13)
# Base voltage profile: sinusoidal annual pattern
# Summer peak (day ~180) pushes voltage up from PV generation
seasonal = 0.025 * np.sin(2 * np.pi * (days - 90) / 360)
# Per-bus baseline offset (some buses naturally higher/lower)
bus_offset = RNG.uniform(-0.015, 0.015, size=n_buses)
# Build matrix
voltage = np.ones((360, n_buses))
for b in range(n_buses):
voltage[:, b] += seasonal + bus_offset[b]
# Add daily noise
noise = RNG.normal(0, 0.008, size=(360, n_buses))
voltage += noise
# Inject realistic violations:
# Summer PV overvoltage on downstream buses (indices 6-12)
for b in range(6, n_buses):
summer_mask = (days >= 120) & (days <= 240)
voltage[summer_mask, b] += RNG.uniform(0.02, 0.045, size=summer_mask.sum())
# Winter undervoltage on load-heavy buses (indices 3, 10, 12)
for b in [3, 10, 12]:
winter_mask = (days <= 60) | (days >= 300)
voltage[winter_mask, b] -= RNG.uniform(0.01, 0.035, size=winter_mask.sum())
return np.clip(voltage, 0.90, 1.12)
def generate_lagrangian_data(n_episodes: int = 500) -> dict[str, np.ndarray]:
"""Generate CMDP convergence curves for Lagrangian multiplier and constraint violation.
Lambda starts high (~10) and converges to ~2.
Violation starts at ~15% and drops below 2%.
Returns:
dict with keys: episodes, lambda_values, violation_rate
"""
episodes = np.arange(n_episodes)
# Lambda convergence: exponential decay + noise
lambda_base = 2.0 + 8.0 * np.exp(-episodes / 80)
lambda_noise = RNG.normal(0, 0.3, size=n_episodes) * np.exp(-episodes / 200)
lambda_values = np.clip(lambda_base + lambda_noise, 0.5, 12.0)
# Constraint violation rate: sigmoid-like decrease
violation_base = 0.15 / (1 + np.exp((episodes - 100) / 40))
violation_noise = RNG.uniform(-0.005, 0.005, size=n_episodes)
violation_rate = np.clip(violation_base + violation_noise + 0.012, 0.0, 0.20)
# Final episodes settle below 2%
violation_rate[400:] = np.clip(violation_rate[400:] * 0.6, 0.005, 0.02)
return {
"episodes": episodes,
"lambda_values": lambda_values,
"violation_rate": violation_rate * 100, # percentage
}
def generate_component_data(day_of_year: int) -> dict[str, np.ndarray]:
"""Generate 24-hour component operation data for a given day.
Seasonal variation affects PV output and battery cycling.
Args:
day_of_year: 0-indexed day (0=Jan1, 90=Apr1, 181=Jul1, 272=Oct1).
Returns:
dict with keys: hours, cap_kvar, reg_tap, battery_soc, pv_kw, pv_curtail
"""
hours = np.arange(24)
# Season factor (0=winter, 1=summer peak)
season_factor = 0.5 + 0.5 * np.sin(2 * np.pi * (day_of_year - 90) / 360)
# Capacitor reactive power: switched based on load/voltage
# Higher in afternoon, lower at night
load_pattern = np.array([
0.3, 0.25, 0.2, 0.2, 0.25, 0.35, 0.5, 0.65,
0.75, 0.8, 0.85, 0.9, 0.95, 1.0, 0.95, 0.9,
0.85, 0.9, 0.95, 0.85, 0.7, 0.55, 0.45, 0.35,
])
cap_kvar = load_pattern * 300 + RNG.normal(0, 15, size=24)
cap_kvar = np.clip(cap_kvar, 0, 400)
# Regulator tap position: -16 to +16, tracks voltage deviation
reg_base = np.array([
2, 2, 3, 3, 2, 1, 0, -1,
-2, -3, -4, -5, -6, -7, -6, -5,
-4, -3, -2, -1, 0, 1, 2, 2,
])
# PV pushes tap down in summer
reg_tap = reg_base - int(season_factor * 4) + RNG.integers(-1, 2, size=24)
reg_tap = np.clip(reg_tap, -16, 16)
# Battery SOC: charges from PV midday, discharges evening peak
soc = np.zeros(24)
soc[0] = 50 + season_factor * 10 # initial SOC
for h in range(1, 24):
if 9 <= h <= 15: # charging from PV
soc[h] = soc[h - 1] + (3.0 + season_factor * 2.0) + RNG.normal(0, 0.5)
elif 17 <= h <= 21: # evening discharge
soc[h] = soc[h - 1] - (4.0 + season_factor * 1.5) + RNG.normal(0, 0.5)
else:
soc[h] = soc[h - 1] + RNG.normal(0, 0.3)
soc = np.clip(soc, 10, 95)
# PV output: bell curve centered at noon, scaled by season
pv_peak = 200 + season_factor * 300 # kW
pv_raw = pv_peak * np.exp(-0.5 * ((hours - 12) / 2.8) ** 2)
pv_raw[:6] = 0
pv_raw[19:] = 0
pv_noise = RNG.normal(0, 10, size=24) * (pv_raw > 0)
pv_kw = np.clip(pv_raw + pv_noise, 0, 600)
# PV curtailment: only in summer midday when overvoltage risk
pv_curtail = np.zeros(24)
if season_factor > 0.6:
curtail_hours = (hours >= 10) & (hours <= 15)
pv_curtail[curtail_hours] = pv_kw[curtail_hours] * RNG.uniform(0.05, 0.20, size=curtail_hours.sum())
return {
"hours": hours,
"cap_kvar": cap_kvar,
"reg_tap": reg_tap.astype(float),
"battery_soc": soc,
"pv_kw": pv_kw,
"pv_curtail": pv_curtail,
}
def generate_training_data(n_episodes: int = 500) -> dict[str, np.ndarray]:
"""Generate CMDP training curves: reward, Lagrangian objective decomposition, constraint rate.
Returns:
dict with keys: episodes, rewards, primal_obj, dual_penalty,
lagrangian_obj, constraint_satisfaction
"""
episodes = np.arange(n_episodes)
# Episode reward: starts bad, improves with noise
reward_base = -50 + 45 * (1 - np.exp(-episodes / 120))
reward_noise = RNG.normal(0, 3.0, size=n_episodes) * np.exp(-episodes / 300)
rewards = reward_base + reward_noise
# Primal objective J(pi): the reward part of CMDP
primal_obj = rewards.copy()
# Lagrangian multiplier (same trajectory as in lagrangian data)
lambda_vals = 2.0 + 8.0 * np.exp(-episodes / 80)
lambda_noise = RNG.normal(0, 0.2, size=n_episodes) * np.exp(-episodes / 200)
lambda_vals = np.clip(lambda_vals + lambda_noise, 0.5, 12.0)
# Constraint cost g(pi): violation magnitude
g_pi = 0.12 / (1 + np.exp((episodes - 100) / 40))
g_noise = RNG.uniform(-0.003, 0.003, size=n_episodes)
g_pi = np.clip(g_pi + g_noise + 0.008, 0.001, 0.2)
g_pi[400:] = np.clip(g_pi[400:] * 0.5, 0.002, 0.015)
# Dual penalty: lambda * g(pi)
dual_penalty = lambda_vals * g_pi * 100 # scaled for visibility
# Lagrangian objective: J(pi) - lambda * g(pi)
lagrangian_obj = primal_obj - dual_penalty
# Constraint satisfaction rate: 1 - violation_rate
violation_rate = g_pi / 0.12 # normalized
constraint_satisfaction = np.clip((1 - violation_rate) * 100, 50, 100)
constraint_satisfaction[350:] = np.clip(
constraint_satisfaction[350:] + RNG.uniform(0, 2, size=150), 96, 100
)
return {
"episodes": episodes,
"rewards": rewards,
"primal_obj": primal_obj,
"dual_penalty": dual_penalty,
"lagrangian_obj": lagrangian_obj,
"constraint_satisfaction": constraint_satisfaction,
}
# Pre-generate all demo data at module load
VOLTAGE_DATA = generate_voltage_heatmap_data()
LAGRANGIAN_DATA = generate_lagrangian_data()
TRAINING_DATA = generate_training_data()
# ============================================================
# Plot Factory Functions
# ============================================================
def plot_voltage_heatmap(time_window: str = "Full Year") -> go.Figure:
"""Create voltage heatmap with bus names on y-axis and days on x-axis.
Args:
time_window: one of 'Full Year', 'Q1', 'Q2', 'Q3', 'Q4'.
Returns:
Plotly Figure with annotated heatmap.
"""
# Slice by quarter
slices = {
"Full Year": (0, 360),
"Q1 (Jan-Mar)": (0, 90),
"Q2 (Apr-Jun)": (90, 180),
"Q3 (Jul-Sep)": (180, 270),
"Q4 (Oct-Dec)": (270, 360),
}
start, end = slices.get(time_window, (0, 360))
data = VOLTAGE_DATA[start:end, :].T # shape: (n_buses, n_days)
days = list(range(start, end))
# Custom colorscale: blue (low) -> green (normal) -> red (high)
colorscale = [
[0.0, "#2563EB"], # blue: severe undervoltage
[0.25, "#3B82F6"], # blue: undervoltage
[0.40, "#10B981"], # green: entering safe zone
[0.50, "#059669"], # dark green: nominal 1.0 pu
[0.60, "#10B981"], # green: leaving safe zone
[0.75, "#F59E0B"], # amber: overvoltage warning
[1.0, "#EF4444"], # red: severe overvoltage
]
fig = go.Figure(data=go.Heatmap(
z=data,
x=days,
y=BUS_NAMES_13,
colorscale=colorscale,
zmin=0.92,
zmax=1.10,
colorbar=dict(
title=dict(text="Voltage (p.u.)", side="right"),
tickvals=[0.93, 0.95, 1.00, 1.05, 1.08],
ticktext=["0.93", "0.95", "1.00", "1.05", "1.08"],
),
hovertemplate=(
"Day: %{x}<br>"
"Bus: %{y}<br>"
"Voltage: %{z:.4f} p.u."
"<extra></extra>"
),
))
# Add violation boundary lines
fig.add_hline(y=None) # hlines not applicable for heatmap y
# Instead, add shapes for voltage limit annotations
fig.add_annotation(
text="V_min=0.95 | V_max=1.05",
xref="paper", yref="paper",
x=1.0, y=1.05,
showarrow=False,
font=dict(size=11, color=COLORS["warning"]),
xanchor="right",
)
layout_overrides = {**PLOTLY_LAYOUT_DEFAULTS}
layout_overrides["yaxis"] = dict(
gridcolor=COLORS["grid"],
zerolinecolor=COLORS["grid"],
type="category",
title_text="Bus Name",
)
fig.update_layout(
**layout_overrides,
height=520,
title=f"Bus Voltage Profile - {time_window}",
xaxis_title="Day of Year",
)
return fig
def plot_lagrangian_trajectory() -> go.Figure:
"""Create dual subplot: lambda convergence + constraint violation over episodes.
Returns:
Plotly Figure with 1x2 subplots.
"""
data = LAGRANGIAN_DATA
ep = data["episodes"]
fig = make_subplots(
rows=1, cols=2,
subplot_titles=(
"Lagrangian Multiplier (lambda) Convergence",
"Voltage Constraint Violation Rate",
),
horizontal_spacing=0.12,
)
# Lambda convergence
fig.add_trace(
go.Scatter(
x=ep, y=data["lambda_values"],
mode="lines",
name="lambda",
line=dict(color=COLORS["primary"], width=1.8),
opacity=0.7,
),
row=1, col=1,
)
# Smoothed lambda (moving average)
window = 20
lambda_smooth = np.convolve(data["lambda_values"], np.ones(window) / window, mode="valid")
fig.add_trace(
go.Scatter(
x=ep[window - 1:], y=lambda_smooth,
mode="lines",
name="lambda (smoothed)",
line=dict(color=COLORS["warning"], width=2.5),
),
row=1, col=1,
)
# Target lambda reference
fig.add_hline(
y=2.0, line_dash="dot", line_color=COLORS["text_dim"],
annotation_text="converged ~2.0",
annotation_position="bottom right",
row=1, col=1,
)
# Constraint violation rate
fig.add_trace(
go.Scatter(
x=ep, y=data["violation_rate"],
mode="lines",
name="violation %",
line=dict(color=COLORS["danger"], width=1.8),
opacity=0.7,
),
row=1, col=2,
)
# Smoothed violation
viol_smooth = np.convolve(data["violation_rate"], np.ones(window) / window, mode="valid")
fig.add_trace(
go.Scatter(
x=ep[window - 1:], y=viol_smooth,
mode="lines",
name="violation % (smoothed)",
line=dict(color=COLORS["accent"], width=2.5),
),
row=1, col=2,
)
# Target violation threshold
fig.add_hline(
y=2.0, line_dash="dot", line_color=COLORS["text_dim"],
annotation_text="target < 2%",
annotation_position="bottom right",
row=1, col=2,
)
fig.update_xaxes(title_text="Training Episode", row=1, col=1)
fig.update_xaxes(title_text="Training Episode", row=1, col=2)
fig.update_yaxes(title_text="Lambda Value", row=1, col=1)
fig.update_yaxes(title_text="Violation Rate (%)", row=1, col=2)
fig.update_layout(
**PLOTLY_LAYOUT_DEFAULTS,
height=450,
showlegend=True,
legend=dict(orientation="h", yanchor="bottom", y=1.08, xanchor="center", x=0.5),
)
return fig
def plot_component_status(season: str = "Day 1 (Winter)") -> go.Figure:
"""Create grouped bar/line chart showing 24-hour component operation.
Args:
season: one of the seasonal day selections.
Returns:
Plotly Figure with 2x2 subplots for each component type.
"""
day_map = {
"Day 1 (Winter)": 1,
"Day 91 (Spring)": 91,
"Day 182 (Summer)": 182,
"Day 273 (Autumn)": 273,
}
day = day_map.get(season, 1)
data = generate_component_data(day)
hours = data["hours"]
fig = make_subplots(
rows=2, cols=2,
subplot_titles=(
"Capacitor Reactive Power",
"Regulator Tap Position",
"Battery State of Charge",
"PV Output & Curtailment",
),
vertical_spacing=0.15,
horizontal_spacing=0.10,
)
# Capacitor kVar
fig.add_trace(
go.Bar(
x=hours, y=data["cap_kvar"],
name="Cap kVar",
marker_color=COLORS["primary"],
opacity=0.85,
),
row=1, col=1,
)
# Regulator tap
fig.add_trace(
go.Scatter(
x=hours, y=data["reg_tap"],
mode="lines+markers",
name="Reg Tap",
line=dict(color=COLORS["accent"], width=2),
marker=dict(size=6),
),
row=1, col=2,
)
fig.add_hline(y=0, line_dash="dot", line_color=COLORS["text_dim"], row=1, col=2)
# Battery SOC
fig.add_trace(
go.Scatter(
x=hours, y=data["battery_soc"],
mode="lines+markers",
name="SOC %",
line=dict(color=COLORS["warning"], width=2.5),
marker=dict(size=5),
fill="tozeroy",
fillcolor="rgba(245, 158, 11, 0.1)",
),
row=2, col=1,
)
# SOC bounds
fig.add_hline(y=20, line_dash="dot", line_color=COLORS["danger"], row=2, col=1)
fig.add_hline(y=90, line_dash="dot", line_color=COLORS["danger"], row=2, col=1)
# PV output + curtailment stacked
fig.add_trace(
go.Bar(
x=hours, y=data["pv_kw"] - data["pv_curtail"],
name="PV Delivered (kW)",
marker_color=COLORS["secondary"],
),
row=2, col=2,
)
fig.add_trace(
go.Bar(
x=hours, y=data["pv_curtail"],
name="PV Curtailed (kW)",
marker_color=COLORS["danger"],
opacity=0.7,
),
row=2, col=2,
)
# Axis labels
for row, col in [(1, 1), (1, 2), (2, 1), (2, 2)]:
fig.update_xaxes(title_text="Hour", row=row, col=col)
fig.update_yaxes(title_text="kVar", row=1, col=1)
fig.update_yaxes(title_text="Tap Position", row=1, col=2)
fig.update_yaxes(title_text="SOC (%)", row=2, col=1)
fig.update_yaxes(title_text="Power (kW)", row=2, col=2)
fig.update_layout(
**PLOTLY_LAYOUT_DEFAULTS,
height=600,
barmode="stack",
title=f"Component Operation - {season} (Day {day})",
showlegend=True,
legend=dict(orientation="h", yanchor="bottom", y=1.06, xanchor="center", x=0.5),
)
return fig
def plot_training_rewards() -> go.Figure:
"""Plot episode reward curve over CMDP training.
Returns:
Plotly Figure with raw + smoothed reward curves.
"""
data = TRAINING_DATA
ep = data["episodes"]
window = 20
fig = go.Figure()
# Raw rewards
fig.add_trace(go.Scatter(
x=ep, y=data["rewards"],
mode="lines",
name="Episode Reward (raw)",
line=dict(color=COLORS["primary"], width=1),
opacity=0.4,
))
# Smoothed rewards
smooth = np.convolve(data["rewards"], np.ones(window) / window, mode="valid")
fig.add_trace(go.Scatter(
x=ep[window - 1:], y=smooth,
mode="lines",
name="Episode Reward (smoothed)",
line=dict(color=COLORS["primary"], width=2.5),
))
fig.update_layout(
**PLOTLY_LAYOUT_DEFAULTS,
height=400,
title="CMDP Episode Reward (SmartGrid 34-Bus PV)",
xaxis_title="Training Episode",
yaxis_title="Episode Reward",
legend=dict(orientation="h", yanchor="bottom", y=1.02, xanchor="right", x=1),
)
return fig
def plot_lagrangian_decomposition() -> go.Figure:
"""Plot Lagrangian objective decomposition: J(pi) - lambda*g(pi).
Shows primal objective, dual penalty, and combined Lagrangian objective.
Returns:
Plotly Figure with three overlaid curves.
"""
data = TRAINING_DATA
ep = data["episodes"]
window = 25
fig = go.Figure()
# Helper: smooth curve
def _smooth(arr: np.ndarray) -> tuple[np.ndarray, np.ndarray]:
s = np.convolve(arr, np.ones(window) / window, mode="valid")
return ep[window - 1:], s
# Primal objective J(pi)
x, y = _smooth(data["primal_obj"])
fig.add_trace(go.Scatter(
x=x, y=y,
mode="lines",
name="J(pi) Primal Objective",
line=dict(color=COLORS["accent"], width=2.2),
))
# Dual penalty lambda*g(pi)
x, y = _smooth(data["dual_penalty"])
fig.add_trace(go.Scatter(
x=x, y=y,
mode="lines",
name="lambda * g(pi) Dual Penalty",
line=dict(color=COLORS["danger"], width=2.2, dash="dash"),
))
# Lagrangian objective L = J(pi) - lambda*g(pi)
x, y = _smooth(data["lagrangian_obj"])
fig.add_trace(go.Scatter(
x=x, y=y,
mode="lines",
name="L(pi, lambda) Lagrangian",
line=dict(color=COLORS["warning"], width=2.8),
))
fig.update_layout(
**PLOTLY_LAYOUT_DEFAULTS,
height=400,
title="Lagrangian Objective Decomposition: L = J(pi) - lambda * g(pi)",
xaxis_title="Training Episode",
yaxis_title="Objective Value",
legend=dict(orientation="h", yanchor="bottom", y=1.02, xanchor="center", x=0.5),
)
return fig
def plot_constraint_satisfaction() -> go.Figure:
"""Plot constraint satisfaction rate over training.
Returns:
Plotly Figure with satisfaction rate and 95% target line.
"""
data = TRAINING_DATA
ep = data["episodes"]
window = 20
fig = go.Figure()
# Raw
fig.add_trace(go.Scatter(
x=ep, y=data["constraint_satisfaction"],
mode="lines",
name="Constraint Satisfaction (raw)",
line=dict(color=COLORS["secondary"], width=1),
opacity=0.35,
))
# Smoothed
smooth = np.convolve(data["constraint_satisfaction"], np.ones(window) / window, mode="valid")
fig.add_trace(go.Scatter(
x=ep[window - 1:], y=smooth,
mode="lines",
name="Constraint Satisfaction (smoothed)",
line=dict(color=COLORS["secondary"], width=2.5),
))
# 95% target
fig.add_hline(
y=95, line_dash="dot", line_color=COLORS["warning"],
annotation_text="95% target",
annotation_position="bottom right",
)
# 98% excellent
fig.add_hline(
y=98, line_dash="dot", line_color=COLORS["primary"],
annotation_text="98% excellent",
annotation_position="top right",
)
layout_overrides = {**PLOTLY_LAYOUT_DEFAULTS}
layout_overrides["yaxis"] = dict(
range=[50, 102],
gridcolor=COLORS["grid"],
zerolinecolor=COLORS["grid"],
title_text="Satisfaction (%)",
)
fig.update_layout(
**layout_overrides,
height=350,
title="Voltage Constraint Satisfaction Rate",
xaxis_title="Training Episode",
legend=dict(orientation="h", yanchor="bottom", y=1.02, xanchor="right", x=1),
)
return fig
# ============================================================
# Build Gradio App
# ============================================================
def build_app() -> gr.Blocks:
"""Construct the Gradio Blocks application with 5 tabs."""
with gr.Blocks(
title="PowerZoo SmartGrid: CMDP Environment Demo",
theme=gr.themes.Soft(primary_hue="emerald"),
) as app:
# Header
gr.Markdown(
"""
# PowerZoo SmartGrid: Modular PV Integration with CMDP
**Constrained MDP** | **Lagrangian Relaxation** | **360-Step Annual Episodes** | **Homogeneous Agents**
"""
)
with gr.Tabs():
# --------------------------------------------------------
# Tab 1: Overview
# --------------------------------------------------------
with gr.Tab("Overview"):
gr.Markdown(
"""
## SmartGrid Environment
SmartGrid is a **modular PV integration environment** built on the Constrained Markov
Decision Process (CMDP) framework with Lagrangian relaxation. Unlike standard MDP
formulations that fold voltage constraints into the reward, SmartGrid separates the
**economic objective** (power loss minimization, control smoothness) from the
**safety constraint** (voltage regulation within 0.95-1.05 p.u.).
### Key Specifications
| Property | Value |
|----------|-------|
| **Agent Type** | Homogeneous (shared policy) |
| **Episode Length** | 360 steps (1 step = 1 day, annual cycle) |
| **Framework** | CMDP with Lagrangian multiplier |
| **Controlled Devices** | Capacitors, Regulators, Batteries, PV Systems |
| **Voltage Limits** | 0.95 - 1.05 p.u. (ANSI C84.1) |
| **Reward** | Power loss + control cost + PV utilization |
| **Constraint** | Voltage violation rate (squared hinge loss) |
### What Makes SmartGrid Unique
- **CMDP Formulation**: The Lagrangian multiplier `lambda` automatically balances
economic performance against voltage safety. No manual penalty tuning required.
- **OOP Circuit Modeling**: Component-based architecture (Capacitor, Regulator,
Battery, PV) with SmartGrid's own `Circuit` class wrapping OpenDSS.
- **Annual Episodes**: 360-step episodes capture seasonal load/PV variation,
enabling agents to learn long-horizon strategies.
- **Curriculum Learning**: Three-phase training (exploration -> optimization ->
refinement) with progressive constraint tightening.
### Supported IEEE Test Systems
| System | Buses | Agents | Complexity |
|--------|-------|--------|------------|
| **13-Bus** | 13 | 2-4 | Rapid prototyping |
| **34-Bus_PV** | 34 | 6-9 | PV integration studies |
| **123-Bus** | 123 | 12-20 | Medium-scale validation |
| **8500-Node** | 8500 | 50+ | Large-scale stress test |
PV variants available: Conservative, Optimized, Aggressive penetration levels.
### CMDP Objective
The agent optimizes the Lagrangian:
**L(pi, lambda) = J(pi) - lambda * g(pi)**
where `J(pi)` is the primal reward (economic), `g(pi)` is the constraint cost
(voltage violations), and `lambda` is the dual variable updated via gradient ascent.
"""
)
# --------------------------------------------------------
# Tab 2: Voltage Heatmap
# --------------------------------------------------------
with gr.Tab("Voltage Heatmap"):
gr.Markdown(
"""
## Bus Voltage Heatmap (IEEE 13-Bus Demo)
Visualize voltage profiles across all buses over the year.
**Blue** = undervoltage (<0.95), **Green** = normal (0.95-1.05), **Red** = overvoltage (>1.05).
Summer PV injection causes overvoltage on downstream buses; winter loads pull voltage down.
"""
)
time_dropdown = gr.Dropdown(
choices=["Full Year", "Q1 (Jan-Mar)", "Q2 (Apr-Jun)", "Q3 (Jul-Sep)", "Q4 (Oct-Dec)"],
value="Full Year",
label="Time Window",
)
heatmap_plot = gr.Plot(value=plot_voltage_heatmap("Full Year"))
time_dropdown.change(
fn=plot_voltage_heatmap,
inputs=time_dropdown,
outputs=heatmap_plot,
)
# --------------------------------------------------------
# Tab 3: Lagrangian Trajectory
# --------------------------------------------------------
with gr.Tab("Lagrangian Trajectory"):
gr.Markdown(
"""
## CMDP Lagrangian Convergence
The Lagrangian multiplier `lambda` starts high (~10) to enforce strict voltage constraints,
then converges to ~2 as the policy learns to satisfy constraints naturally.
The constraint violation rate drops from ~15% to below 2%.
This dual convergence is the signature behavior of CMDP training with Lagrangian relaxation.
"""
)
lagrangian_plot = gr.Plot(value=plot_lagrangian_trajectory())
# --------------------------------------------------------
# Tab 4: Component Status
# --------------------------------------------------------
with gr.Tab("Component Status"):
gr.Markdown(
"""
## 24-Hour Component Operation
View how capacitors, regulators, batteries, and PV systems operate across a full day.
Select different seasons to see how operation patterns change with load and PV availability.
"""
)
season_radio = gr.Radio(
choices=["Day 1 (Winter)", "Day 91 (Spring)", "Day 182 (Summer)", "Day 273 (Autumn)"],
value="Day 182 (Summer)",
label="Select Day of Year",
)
component_plot = gr.Plot(value=plot_component_status("Day 182 (Summer)"))
season_radio.change(
fn=plot_component_status,
inputs=season_radio,
outputs=component_plot,
)
# --------------------------------------------------------
# Tab 5: Training Dashboard
# --------------------------------------------------------
with gr.Tab("Training Dashboard"):
gr.Markdown(
"""
## CMDP Training Dashboard (SmartGrid 34-Bus PV)
Training curves showing both primal (reward) and dual (constraint) convergence.
The Lagrangian objective decomposes into J(pi) and the penalty term lambda * g(pi).
"""
)
gr.Markdown("### Episode Reward")
reward_plot = gr.Plot(value=plot_training_rewards())
gr.Markdown("### Lagrangian Objective Decomposition")
decomp_plot = gr.Plot(value=plot_lagrangian_decomposition())
gr.Markdown("### Constraint Satisfaction Rate")
constraint_plot = gr.Plot(value=plot_constraint_satisfaction())
# Footer
gr.Markdown(
"""
---
**PowerZoo** · MIT License · [XJTU-RL](https://github.com/XJTU-RL) · IEEE TSG 2025
"""
)
return app
# ============================================================
# Launch
# ============================================================
if __name__ == "__main__":
app = build_app()
app.launch(server_name="0.0.0.0", server_port=7860, share=False)
|