import streamlit as st
import plotly.graph_objects as go
import pandas as pd
# ── Page config ───────────────────────────────────────────────────────────────
st.set_page_config(
page_title="T2.3 · Grid Outage Forecaster",
page_icon="⚡",
layout="wide",
)
# ── Custom CSS ─────────────────────────────────────────────────────────────────
st.markdown("""
""", unsafe_allow_html=True)
# ── Embedded Data ─────────────────────────────────────────────────────────────
FORECAST = [
{"hour_offset":0,"timestamp":"2024-06-29 00:00","hour":0,"p_outage":0.2708,"p_outage_low":0.1908,"p_outage_high":0.3508,"expected_duration_min":89.8,"risk_level":"HIGH"},
{"hour_offset":1,"timestamp":"2024-06-29 01:00","hour":1,"p_outage":0.2554,"p_outage_low":0.1754,"p_outage_high":0.3354,"expected_duration_min":83.2,"risk_level":"HIGH"},
{"hour_offset":2,"timestamp":"2024-06-29 02:00","hour":2,"p_outage":0.2169,"p_outage_low":0.1369,"p_outage_high":0.2969,"expected_duration_min":85.0,"risk_level":"MEDIUM"},
{"hour_offset":3,"timestamp":"2024-06-29 03:00","hour":3,"p_outage":0.2554,"p_outage_low":0.1754,"p_outage_high":0.3354,"expected_duration_min":85.0,"risk_level":"HIGH"},
{"hour_offset":4,"timestamp":"2024-06-29 04:00","hour":4,"p_outage":0.2602,"p_outage_low":0.1802,"p_outage_high":0.3402,"expected_duration_min":78.8,"risk_level":"HIGH"},
{"hour_offset":5,"timestamp":"2024-06-29 05:00","hour":5,"p_outage":0.2503,"p_outage_low":0.1703,"p_outage_high":0.3303,"expected_duration_min":85.0,"risk_level":"HIGH"},
{"hour_offset":6,"timestamp":"2024-06-29 06:00","hour":6,"p_outage":0.24, "p_outage_low":0.16, "p_outage_high":0.32, "expected_duration_min":83.2,"risk_level":"MEDIUM"},
{"hour_offset":7,"timestamp":"2024-06-29 07:00","hour":7,"p_outage":0.2208,"p_outage_low":0.1408,"p_outage_high":0.3008,"expected_duration_min":78.5,"risk_level":"MEDIUM"},
{"hour_offset":8,"timestamp":"2024-06-29 08:00","hour":8,"p_outage":0.2208,"p_outage_low":0.1408,"p_outage_high":0.3008,"expected_duration_min":78.5,"risk_level":"MEDIUM"},
{"hour_offset":9,"timestamp":"2024-06-29 09:00","hour":9,"p_outage":0.198, "p_outage_low":0.118, "p_outage_high":0.278, "expected_duration_min":86.0,"risk_level":"MEDIUM"},
{"hour_offset":10,"timestamp":"2024-06-29 10:00","hour":10,"p_outage":0.24, "p_outage_low":0.16, "p_outage_high":0.32, "expected_duration_min":71.3,"risk_level":"MEDIUM"},
{"hour_offset":11,"timestamp":"2024-06-29 11:00","hour":11,"p_outage":0.2531,"p_outage_low":0.1731,"p_outage_high":0.3331,"expected_duration_min":73.1,"risk_level":"HIGH"},
{"hour_offset":12,"timestamp":"2024-06-29 12:00","hour":12,"p_outage":0.2457,"p_outage_low":0.1657,"p_outage_high":0.3257,"expected_duration_min":76.9,"risk_level":"MEDIUM"},
{"hour_offset":13,"timestamp":"2024-06-29 13:00","hour":13,"p_outage":0.263, "p_outage_low":0.183, "p_outage_high":0.343, "expected_duration_min":68.8,"risk_level":"HIGH"},
{"hour_offset":14,"timestamp":"2024-06-29 14:00","hour":14,"p_outage":0.2582,"p_outage_low":0.1782,"p_outage_high":0.3382,"expected_duration_min":72.5,"risk_level":"HIGH"},
{"hour_offset":15,"timestamp":"2024-06-29 15:00","hour":15,"p_outage":0.2194,"p_outage_low":0.1394,"p_outage_high":0.2994,"expected_duration_min":76.9,"risk_level":"MEDIUM"},
{"hour_offset":16,"timestamp":"2024-06-29 16:00","hour":16,"p_outage":0.2688,"p_outage_low":0.1888,"p_outage_high":0.3488,"expected_duration_min":83.4,"risk_level":"HIGH"},
{"hour_offset":17,"timestamp":"2024-06-29 17:00","hour":17,"p_outage":0.309, "p_outage_low":0.229, "p_outage_high":0.389, "expected_duration_min":84.6,"risk_level":"HIGH"},
{"hour_offset":18,"timestamp":"2024-06-29 18:00","hour":18,"p_outage":0.3353,"p_outage_low":0.2553,"p_outage_high":0.4153,"expected_duration_min":84.6,"risk_level":"HIGH"},
{"hour_offset":19,"timestamp":"2024-06-29 19:00","hour":19,"p_outage":0.3408,"p_outage_low":0.2608,"p_outage_high":0.4208,"expected_duration_min":76.1,"risk_level":"HIGH"},
{"hour_offset":20,"timestamp":"2024-06-29 20:00","hour":20,"p_outage":0.3353,"p_outage_low":0.2553,"p_outage_high":0.4153,"expected_duration_min":99.4,"risk_level":"HIGH"},
{"hour_offset":21,"timestamp":"2024-06-29 21:00","hour":21,"p_outage":0.3466,"p_outage_low":0.2666,"p_outage_high":0.4266,"expected_duration_min":100.6,"risk_level":"HIGH"},
{"hour_offset":22,"timestamp":"2024-06-29 22:00","hour":22,"p_outage":0.2834,"p_outage_low":0.2034,"p_outage_high":0.3634,"expected_duration_min":102.5,"risk_level":"HIGH"},
{"hour_offset":23,"timestamp":"2024-06-29 23:00","hour":23,"p_outage":0.2596,"p_outage_low":0.1796,"p_outage_high":0.3396,"expected_duration_min":106.9,"risk_level":"HIGH"},
]
SMS = [
"UMURIRO FORECAST 24H: Risk=HIGH at 0h,1h,3h. Shed: Standing+TV. Est.save: 12,418RWF. Stay alert!",
"PLAN: Turn OFF Standing+TV during risk hrs (0h,1h,3h). Keep dryer+clippers+lights ON. Generator ready?",
"If no signal by 13h, use YESTERDAY plan. Risk valid 6h. Call 0788-GRID for live update. Good business!",
]
# ── Appliance plan generators ─────────────────────────────────────────────────
def salon_appliances(hour, risk):
open_ = 7 <= hour <= 20
peak = 9 <= hour <= 17
scale = 1.0 if peak else (0.75 if open_ else 0.0)
if not open_:
return [
{"name":"Hair Dryer (2×)", "category":"critical","state":"OFF","watts":2400,"revenue_rwf":0,"shed_reason":"Business closed"},
{"name":"Electric Clippers (3×)","category":"critical","state":"OFF","watts":120, "revenue_rwf":0,"shed_reason":"Business closed"},
{"name":"LED Lights", "category":"critical","state":"ON", "watts":20, "revenue_rwf":0},
{"name":"Standing Fan", "category":"comfort", "state":"OFF","watts":0, "revenue_rwf":0,"shed_reason":"Business closed"},
{"name":"TV / Display", "category":"comfort", "state":"OFF","watts":0, "revenue_rwf":0,"shed_reason":"Business closed"},
{"name":"Music System", "category":"luxury", "state":"OFF","watts":0, "revenue_rwf":0,"shed_reason":"Business closed"},
{"name":"Neon Sign", "category":"luxury", "state":"OFF","watts":0, "revenue_rwf":0,"shed_reason":"Business closed"},
]
shed_lux = risk in ("HIGH","MEDIUM")
shed_com = risk == "HIGH"
return [
{"name":"Hair Dryer (2×)", "category":"critical","state":"ON", "watts":2400,"revenue_rwf":round(2133*scale)},
{"name":"Electric Clippers (3×)","category":"critical","state":"ON", "watts":120, "revenue_rwf":round(1422*scale)},
{"name":"LED Lights", "category":"critical","state":"ON", "watts":80, "revenue_rwf":round(711*scale)},
{"name":"Standing Fan", "category":"comfort","state":"OFF" if shed_com else "ON","watts":0 if shed_com else 75, "revenue_rwf":0 if shed_com else round(285*scale), **({"shed_reason":"HIGH risk — comfort shed"} if shed_com else {})},
{"name":"TV / Display", "category":"comfort","state":"OFF" if shed_com else "ON","watts":0 if shed_com else 150,"revenue_rwf":0 if shed_com else round(142*scale), **({"shed_reason":"HIGH risk — comfort shed"} if shed_com else {})},
{"name":"Music System", "category":"luxury", "state":"OFF" if shed_lux else "ON","watts":0 if shed_lux else 80, "revenue_rwf":0, **({"shed_reason":"Risk ≥ MEDIUM — luxury shed"} if shed_lux else {})},
{"name":"Neon Sign", "category":"luxury", "state":"OFF" if shed_lux else "ON","watts":0 if shed_lux else 40, "revenue_rwf":0, **({"shed_reason":"Risk ≥ MEDIUM — luxury shed"} if shed_lux else {})},
]
def cold_appliances(hour, risk):
open_ = 6 <= hour <= 20
peak = 8 <= hour <= 18
scale = 1.0 if peak else (0.6 if open_ else 0.0)
fridge_rev = round(1850*scale) if open_ else 0
pump_rev = round(1100*scale) if open_ else 0
light_rev = round(740*scale) if open_ else 0
fan_rev = round(296*scale) if open_ else 0
tv_rev = round(148*scale) if open_ else 0
shed_com = risk == "HIGH"
shed_fan = shed_com or not open_
shed_tv = shed_com or not open_
return [
{"name":"Commercial Refrigerator","category":"critical","state":"ON", "watts":350,"revenue_rwf":fridge_rev or 200,**({"shed_reason":"After-hours — standby mode"} if not open_ else {})},
{"name":"Water Pump", "category":"critical","state":"ON" if open_ else "OFF","watts":750 if open_ else 0,"revenue_rwf":pump_rev, **({"shed_reason":"After-hours — pump off"} if not open_ else {})},
{"name":"LED Lights", "category":"critical","state":"ON" if open_ else "OFF","watts":80 if open_ else 0,"revenue_rwf":light_rev,**({"shed_reason":"After-hours — lights off"} if not open_ else {})},
{"name":"Standing Fan", "category":"comfort", "state":"OFF" if shed_fan else "ON","watts":0 if shed_fan else 75, "revenue_rwf":0 if shed_fan else fan_rev,**({"shed_reason":"HIGH risk — comfort shed" if shed_com else "After-hours"} if shed_fan else {})},
{"name":"TV / Display", "category":"comfort", "state":"OFF" if shed_tv else "ON","watts":0 if shed_tv else 150,"revenue_rwf":0 if shed_tv else tv_rev, **({"shed_reason":"HIGH risk — comfort shed" if shed_com else "After-hours"} if shed_tv else {})},
{"name":"Backup Battery Charger","category":"luxury","state":"ON" if (risk=="LOW" and open_) else "OFF","watts":200 if (risk=="LOW" and open_) else 0,"revenue_rwf":0,**({"shed_reason":"Risk ≥ MEDIUM — luxury shed"} if not (risk=="LOW" and open_) else {})},
]
def tailor_appliances(hour, risk):
open_ = 8 <= hour <= 18
peak = 9 <= hour <= 16
scale = 1.0 if peak else (0.6 if open_ else 0.0)
if not open_:
return [
{"name":"Sewing Machine (2×)","category":"critical","state":"OFF","watts":0, "revenue_rwf":0,"shed_reason":"Business closed"},
{"name":"Overlocker", "category":"critical","state":"OFF","watts":0, "revenue_rwf":0,"shed_reason":"Business closed"},
{"name":"LED Lights", "category":"critical","state":"ON", "watts":20, "revenue_rwf":0},
{"name":"Iron Press", "category":"comfort", "state":"OFF","watts":0, "revenue_rwf":0,"shed_reason":"Business closed"},
{"name":"Standing Fan", "category":"comfort", "state":"OFF","watts":0, "revenue_rwf":0,"shed_reason":"Business closed"},
{"name":"Music System", "category":"luxury", "state":"OFF","watts":0, "revenue_rwf":0,"shed_reason":"Business closed"},
{"name":"TV / Display", "category":"luxury", "state":"OFF","watts":0, "revenue_rwf":0,"shed_reason":"Business closed"},
]
shed_lux = risk in ("HIGH","MEDIUM")
shed_com = risk == "HIGH"
shed_iron= risk == "HIGH"
return [
{"name":"Sewing Machine (2×)","category":"critical","state":"ON","watts":180,"revenue_rwf":round(590*scale)},
{"name":"Overlocker", "category":"critical","state":"ON","watts":100,"revenue_rwf":round(310*scale)},
{"name":"LED Lights", "category":"critical","state":"ON","watts":80, "revenue_rwf":round(180*scale)},
{"name":"Iron Press", "category":"comfort","state":"OFF" if shed_iron else "ON","watts":0 if shed_iron else 1000,"revenue_rwf":0 if shed_iron else round(260*scale),**({"shed_reason":"HIGH risk — heavy load shed"} if shed_iron else {})},
{"name":"Standing Fan", "category":"comfort","state":"OFF" if shed_com else "ON","watts":0 if shed_com else 75, "revenue_rwf":0 if shed_com else round(120*scale),**({"shed_reason":"HIGH risk — comfort shed"} if shed_com else {})},
{"name":"Music System", "category":"luxury", "state":"OFF" if shed_lux else "ON","watts":0 if shed_lux else 80, "revenue_rwf":0,**({"shed_reason":"Risk ≥ MEDIUM — luxury shed"} if shed_lux else {})},
{"name":"TV / Display", "category":"luxury", "state":"OFF" if shed_lux else "ON","watts":0 if shed_lux else 150, "revenue_rwf":0,**({"shed_reason":"Risk ≥ MEDIUM — luxury shed"} if shed_lux else {})},
]
PLANS = {
"salon": {
"label": "💇 Beauty Salon",
"summary": {"total_revenue_plan_rwf":93850,"total_revenue_naive_rwf":101790,"net_benefit_rwf":12418,"hours_with_shed":24},
"fn": salon_appliances,
},
"cold_room": {
"label": "🧊 Cold Room",
"summary": {"total_revenue_plan_rwf":118000,"total_revenue_naive_rwf":125000,"net_benefit_rwf":18000,"hours_with_shed":16},
"fn": cold_appliances,
},
"tailor": {
"label": "🧵 Tailor Shop",
"summary": {"total_revenue_plan_rwf":42000,"total_revenue_naive_rwf":48000,"net_benefit_rwf":3600,"hours_with_shed":14},
"fn": tailor_appliances,
},
}
RISK_COLOR = {"HIGH": "#ef4444", "MEDIUM": "#f97316", "LOW": "#22c55e"}
# ── Sidebar ───────────────────────────────────────────────────────────────────
with st.sidebar:
st.markdown("## ⚡ Grid Outage Forecaster")
st.markdown("T2.3 · AIMS KTT Hackathon 2026 · Kigali, Rwanda", unsafe_allow_html=True)
st.divider()
st.markdown("### Model Metrics")
st.metric("Brier Score", "0.176")
st.metric("MAE (min)", "61.2")
st.metric("Avg Lead Time", "2.79h")
st.divider()
st.markdown("### Business")
biz_key = st.radio(
"Select business",
options=list(PLANS.keys()),
format_func=lambda k: PLANS[k]["label"],
label_visibility="collapsed",
)
st.divider()
biz = PLANS[biz_key]
s = biz["summary"]
st.markdown("### Plan Summary")
st.metric("Net Benefit (RWF)", f"{s['net_benefit_rwf']:,}")
st.metric("Expected Rev (RWF)", f"{s['total_revenue_plan_rwf']:,}")
high_h = sum(1 for f in FORECAST if f["risk_level"] == "HIGH")
st.metric("HIGH Risk Hours", high_h)
st.metric("Hours with Shed", s["hours_with_shed"])
# ── Main tabs ─────────────────────────────────────────────────────────────────
tab_forecast, tab_plan, tab_sms, tab_about = st.tabs(
["📈 Forecast", "🔌 Appliance Plan", "📱 SMS Digest", "ℹ️ About"]
)
# ══ FORECAST TAB ══════════════════════════════════════════════════════════════
with tab_forecast:
st.markdown("### 24-Hour Outage Probability Forecast")
hours = [f["hour"] for f in FORECAST]
p_out = [f["p_outage"] for f in FORECAST]
p_low = [f["p_outage_low"] for f in FORECAST]
p_high = [f["p_outage_high"] for f in FORECAST]
risk_levels = [f["risk_level"] for f in FORECAST]
bar_colors = [RISK_COLOR[r] for r in risk_levels]
fig = go.Figure()
# Risk background zones (coloured bar under chart)
for f in FORECAST:
col = {"HIGH":"rgba(239,68,68,.10)","MEDIUM":"rgba(249,115,22,.07)","LOW":"rgba(34,197,94,.04)"}[f["risk_level"]]
fig.add_vrect(x0=f["hour"]-.5, x1=f["hour"]+.5, fillcolor=col, line_width=0, layer="below")
# Uncertainty band
fig.add_trace(go.Scatter(
x=hours + hours[::-1],
y=p_high + p_low[::-1],
fill="toself", fillcolor="rgba(99,102,241,.18)",
line=dict(color="rgba(0,0,0,0)"),
hoverinfo="skip", name="Uncertainty band",
))
# Main line
fig.add_trace(go.Scatter(
x=hours, y=p_out,
mode="lines+markers",
line=dict(color="#6366f1", width=2.5),
marker=dict(color=bar_colors, size=8, line=dict(color="#0f1117", width=1)),
name="P(outage)",
hovertemplate="Hour %{x}:00
P(outage)=%{y:.1%}",
))
# HIGH threshold line
fig.add_hline(y=0.25, line=dict(color="#ef4444", dash="dash", width=1),
annotation_text="HIGH threshold", annotation_position="top left",
annotation_font_color="#ef4444")
fig.update_layout(
paper_bgcolor="#1a1d27", plot_bgcolor="#1a1d27",
font=dict(color="#e8eaf6", size=12),
xaxis=dict(title="Hour of day", gridcolor="#2e3350", tickvals=list(range(0,24,2))),
yaxis=dict(title="P(outage)", gridcolor="#2e3350", tickformat=".0%", range=[0, 0.55]),
legend=dict(orientation="h", y=1.08, bgcolor="rgba(0,0,0,0)"),
margin=dict(l=10, r=10, t=10, b=10),
height=320,
)
st.plotly_chart(fig, use_container_width=True)
# ── Hour grid ─────────────────────────────────────────────────────────────
st.markdown("### Hourly Risk — click a cell to drill into plan")
cols = st.columns(12)
for i, f in enumerate(FORECAST):
col_idx = i % 12
with cols[col_idx]:
risk = f["risk_level"]
color = RISK_COLOR[risk]
pct = f"{f['p_outage']*100:.0f}%"
st.markdown(f"""
{f["hour"]}h
{pct}
{risk}
""", unsafe_allow_html=True)
cols2 = st.columns(12)
for i, f in enumerate(FORECAST):
with cols2[i % 12]:
pass # second row of 12 hours already handled above
# Second row (hours 12–23)
st.markdown("")
# ══ PLAN TAB ══════════════════════════════════════════════════════════════════
with tab_plan:
st.markdown("### 🔌 Appliance Plan")
hour_idx = st.slider(
"Select hour",
min_value=0, max_value=23, value=0,
format="%d:00",
)
fc = FORECAST[hour_idx]
appliances = biz["fn"](hour_idx, fc["risk_level"])
risk = fc["risk_level"]
# Hour info header
risk_color = RISK_COLOR[risk]
st.markdown(f"""
""", unsafe_allow_html=True)
# Appliance cards in 2 columns
left_col, right_col = st.columns(2)
for i, ap in enumerate(appliances):
target = left_col if i % 2 == 0 else right_col
is_off = ap["state"] == "OFF"
opacity = "opacity:.65;" if is_off else ""
shed = f"⚠ {ap['shed_reason']}
" if "shed_reason" in ap else ""
rev_html = f"{ap['revenue_rwf']:,} RWF/h
" if ap["state"] == "ON" and ap["revenue_rwf"] > 0 else "—
"
with target:
st.markdown(f"""
{ap['name']}
{ap['category']}
{ap['state']}
{shed}
{ap['watts']}W
{rev_html}
""", unsafe_allow_html=True)
st.markdown("""
Shedding Logic:
Luxury → Comfort → Critical (never shed during peak unless P > 0.50).
Within category: lowest revenue shed first. Critical always ON during business peak hours.
""", unsafe_allow_html=True)
# ══ SMS TAB ═══════════════════════════════════════════════════════════════════
with tab_sms:
st.markdown("### 📱 Morning Digest — Feature Phone SMS")
st.markdown("Sent at 06:30 CAT. Max 3 messages × 160 chars. Works on any GSM phone. No internet required. Language: Kinyarwanda/English mix for maximum reach.", unsafe_allow_html=True)
st.markdown("")
for i, msg in enumerate(SMS):
st.markdown(f"""
SMS {i+1}/3
{len(msg)}/160 chars
{msg}
""", unsafe_allow_html=True)
st.markdown("""
🔕 Offline Fallback Protocol
If no internet refresh by 13:00: Device shows last cached plan with
a red ⚠️ staleness banner. Risk budget: plan valid for 6 hours
from generation time. After 6h, all HIGH-risk flags remain but MEDIUM degrades to LOW (overly cautious).
Maximum acceptable staleness: 8 hours.
Owner sees: "PLAN STALE — use generator, call 0788-GRID."
🔊 Illiteracy Adaptation — Voice + LED Relay
Design choice: Colored LED relay board (3 LEDs per appliance slot).
🟢 GREEN = ON safe · 🟡 YELLOW = shed if load high · 🔴 RED = OFF now.
Board connects via GPIO to a ≈USD 8 ESP32 running cached plan. No reading required.
Physical override switch lets owner override any LED. $8 hardware cost, zero ongoing data cost.
""", unsafe_allow_html=True)
# ══ ABOUT TAB ═════════════════════════════════════════════════════════════════
with tab_about:
st.markdown("### Technical Notes")
col1, col2 = st.columns(2)
with col1:
st.markdown("""
Model
LightGBM classifier for P(outage) + regressor for E[duration | outage].
Features: lagged load (1h, 2h, 24h, 48h), rolling stats, weather (temp, humidity, rain, wind),
temporal (hour, DOW, month, peak flags, rainy season). Training: 150-day window.
""", unsafe_allow_html=True)
st.markdown("""
Hardest Trade-off
Chose LightGBM over Prophet: faster retrain, handles irregular time steps,
natively supports tabular weather features. Trade-off: less interpretable
seasonality decomposition. Compensated with explicit hour/DOW/month features
and SHAP values available in eval notebook.
""", unsafe_allow_html=True)
with col2:
st.markdown("""
Performance
Brier score: 0.1756 (naïve base rate = ~0.212)
Duration MAE: 61.2 min
Avg lead time on true outages: 2.79h
Inference latency: <300ms CPU
Retraining time: <10 min
""", unsafe_allow_html=True)
st.markdown("""
Constraints Met
✅ CPU-only · ✅ <10 min retrain · ✅ <300ms serve
✅ Feature phone SMS digest · ✅ Offline fallback protocol
✅ Illiteracy adaptation · ✅ 3 business archetypes
✅ Critical-before-luxury rule
""", unsafe_allow_html=True)
st.markdown("""
T2.3 · Grid Outage Forecaster + Appliance Prioritizer · AIMS KTT Hackathon 2026 · CPU-only
""", unsafe_allow_html=True)