File size: 10,610 Bytes
938949f | 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 | """ML-based energy generation predictor for the Yeruham Vineyard PV system.
Trained on ThingsBoard production data + on-site weather (Air1 sensor).
Two models:
- **XGBoost** (primary): uses GSR, temperature, solar elevation, hour,
clearness index, wind speed. Backtested MAPE ≈ 4.6 %.
- **Linear fallback**: uses only GSR + temperature (when fewer features
are available, e.g. IMS-only forecasts). MAPE ≈ 6.9 %.
Usage::
from src.energy_predictor import EnergyPredictor
ep = EnergyPredictor() # loads saved model
daily = ep.predict_day("2026-03-15",
forecast_ghi=[...], # 24 hourly W/m²
forecast_temp=[...]) # 24 hourly °C
"""
from __future__ import annotations
import pickle
from datetime import date, datetime, timedelta, timezone
from pathlib import Path
from typing import Any, Dict, List, Optional, Sequence
import numpy as np
import pandas as pd
_MODEL_PATH = Path(__file__).resolve().parent.parent / "Data" / "energy_predictor_model.pkl"
# Site constants
_LATITUDE_DEG = 30.85
_LAT_RAD = np.radians(_LATITUDE_DEG)
_SYSTEM_CAPACITY_KW = 48.0
def _solar_sin_elevation(day_of_year: int, hour_utc: int) -> float:
"""Approximate sin(solar elevation) for Sde Boker."""
dec = np.radians(23.45 * np.sin(np.radians(360 / 365 * (day_of_year - 81))))
ha = np.radians(15 * (hour_utc + 2 - 12)) # UTC+2 ≈ local solar
return float(max(0.0, np.sin(_LAT_RAD) * np.sin(dec)
+ np.cos(_LAT_RAD) * np.cos(dec) * np.cos(ha)))
class EnergyPredictor:
"""Predict PV energy generation from weather features."""
def __init__(self, model_path: Optional[Path] = None):
path = model_path or _MODEL_PATH
if not path.exists():
raise FileNotFoundError(
f"Energy model not found at {path}. "
"Run the training notebook / script first."
)
with open(path, "rb") as f:
bundle = pickle.load(f)
self._xgb = bundle["xgb_model"]
self._xgb_features = bundle["xgb_features"]
self._lr = bundle["lr_fallback"]
self._lr_features = bundle["lr_features"]
self._meta = {
k: v for k, v in bundle.items()
if k not in ("xgb_model", "lr_fallback")
}
# ------------------------------------------------------------------
# Public API
# ------------------------------------------------------------------
def predict_hourly(
self,
target_date: str | date,
forecast_ghi: Sequence[float],
forecast_temp: Sequence[float],
forecast_wind: Optional[Sequence[float]] = None,
) -> pd.DataFrame:
"""Predict hourly energy production (kWh) for *target_date*.
Parameters
----------
target_date : str or date
ISO date, e.g. ``"2026-03-15"``.
forecast_ghi : sequence of 24 floats
Hourly Global Solar Radiation (W/m²), index 0 = 00:00 UTC.
forecast_temp : sequence of 24 floats
Hourly air temperature (°C).
forecast_wind : sequence of 24 floats, optional
Hourly wind speed (m/s). Falls back to 1.5 m/s if omitted.
Returns
-------
DataFrame with columns ``hour_utc``, ``production_kwh``,
``method`` ('xgb' or 'lr_fallback').
"""
if isinstance(target_date, str):
target_date = date.fromisoformat(target_date)
doy = target_date.timetuple().tm_yday
ghi = np.asarray(forecast_ghi, dtype=float)
temp = np.asarray(forecast_temp, dtype=float)
wind = (np.asarray(forecast_wind, dtype=float)
if forecast_wind is not None
else np.full(24, 1.5))
rows = []
for h in range(24):
sin_el = _solar_sin_elevation(doy, h)
clearness = (ghi[h] / (sin_el * 1000)
if sin_el > 0.05 else 0.0)
clearness = min(clearness, 1.5)
rows.append({
"GSR": ghi[h],
"airTemperature": temp[h],
"sin_elevation": sin_el,
"hour": h,
"clearness": clearness,
"windSpeed": wind[h],
})
df = pd.DataFrame(rows)
# Prefer XGBoost; fall back to LR if it fails
try:
preds = self._xgb.predict(df[self._xgb_features])
method = "xgb"
except Exception:
preds = self._lr.predict(df[self._lr_features])
method = "lr_fallback"
preds = np.clip(preds, 0, None)
# Zero out nighttime hours (sun below horizon)
for i in range(24):
if df.loc[i, "sin_elevation"] < 0.02 and ghi[i] < 10:
preds[i] = 0.0
return pd.DataFrame({
"hour_utc": range(24),
"production_kwh": np.round(preds, 2),
"method": method,
})
def predict_day(
self,
target_date: str | date,
forecast_ghi: Sequence[float],
forecast_temp: Sequence[float],
forecast_wind: Optional[Sequence[float]] = None,
) -> Dict[str, Any]:
"""Predict daily energy production with hourly profile.
Returns dict matching the EnergyService.predict() schema.
"""
hourly = self.predict_hourly(
target_date, forecast_ghi, forecast_temp, forecast_wind,
)
total_kwh = hourly["production_kwh"].sum()
peak_idx = hourly["production_kwh"].idxmax()
peak_hour = int(hourly.loc[peak_idx, "hour_utc"])
peak_kwh = float(hourly.loc[peak_idx, "production_kwh"])
# Convert UTC hours → local display (Israel = UTC+2/+3)
hourly_profile = [
{"hour": int(row["hour_utc"]), "energy_kwh": round(row["production_kwh"], 2)}
for _, row in hourly.iterrows()
]
return {
"date": str(target_date),
"daily_kwh": round(float(total_kwh), 1),
"peak_hour": peak_hour,
"peak_hour_kwh": round(peak_kwh, 2),
"hourly_profile": hourly_profile,
"source": f"ML energy predictor ({hourly.iloc[0]['method']})",
"model_mape_pct": self._meta.get("test_mape_pct"),
}
def predict_day_from_weather_df(
self,
target_date: str | date,
weather_df: pd.DataFrame,
) -> Dict[str, Any]:
"""Predict from a DataFrame that has hourly GSR/airTemperature columns.
Accepts either ThingsBoard Air1 format or IMS format (ghi_w_m2).
"""
if weather_df.empty:
return {"date": str(target_date), "daily_kwh": None,
"error": "Empty weather DataFrame"}
# Resolve column names
ghi_col = next((c for c in weather_df.columns
if c in ("GSR", "ghi_w_m2", "GHI")), None)
temp_col = next((c for c in weather_df.columns
if c in ("airTemperature", "air_temperature_c", "temperature")), None)
wind_col = next((c for c in weather_df.columns
if c in ("windSpeed", "wind_speed_ms", "wind_speed")), None)
if ghi_col is None or temp_col is None:
return {"date": str(target_date), "daily_kwh": None,
"error": f"Missing columns. Need GSR + temp, got {list(weather_df.columns)}"}
# Resample to 24 hourly values
df = weather_df.copy()
if hasattr(df.index, "hour"):
hourly = df.resample("h").mean()
else:
hourly = df
# Ensure 24 hours (pad missing with 0 for GHI, forward-fill temp)
ghi_24 = np.zeros(24)
temp_24 = np.full(24, 15.0) # sensible default
wind_24 = np.full(24, 1.5)
for _, row in hourly.iterrows():
h = row.name.hour if hasattr(row.name, "hour") else 0
if 0 <= h < 24:
ghi_24[h] = row[ghi_col] if pd.notna(row.get(ghi_col)) else 0
if pd.notna(row.get(temp_col)):
temp_24[h] = row[temp_col]
if wind_col and pd.notna(row.get(wind_col)):
wind_24[h] = row[wind_col]
return self.predict_day(target_date, ghi_24, temp_24, wind_24)
def backtest(
self,
tb_client: Any,
days_back: int = 7,
) -> pd.DataFrame:
"""Compare ML predictions vs actual production for recent days.
Returns DataFrame with columns: date, actual_kwh, predicted_kwh,
error_kwh, error_pct.
"""
from datetime import timezone as tz
import pytz
utc = pytz.UTC
rows = []
now = datetime.now(tz=utc)
for d in range(days_back, 0, -1):
day = now - timedelta(days=d)
day_start = utc.localize(datetime(day.year, day.month, day.day))
day_end = day_start + timedelta(days=1)
# Actual production
try:
actual_df = tb_client.get_asset_timeseries(
"Plant", ["production"],
start=day_start, end=day_end,
limit=500, interval_ms=3_600_000, agg="SUM",
)
actual_kwh = actual_df["production"].sum() / 1000 if not actual_df.empty else 0
except Exception:
actual_kwh = 0
if actual_kwh < 10:
continue
# Weather for that day → prediction
try:
wx = tb_client.get_timeseries(
"Air1", ["GSR", "airTemperature", "windSpeed"],
start=day_start, end=day_end,
limit=100, interval_ms=3_600_000, agg="AVG",
)
if wx.empty:
continue
pred = self.predict_day_from_weather_df(
day_start.strftime("%Y-%m-%d"), wx,
)
pred_kwh = pred.get("daily_kwh") or 0
except Exception:
continue
rows.append({
"date": day_start.strftime("%Y-%m-%d"),
"actual_kwh": round(actual_kwh, 1),
"predicted_kwh": round(pred_kwh, 1),
"error_kwh": round(pred_kwh - actual_kwh, 1),
"error_pct": round((pred_kwh - actual_kwh) / actual_kwh * 100, 1),
})
return pd.DataFrame(rows)
@property
def metadata(self) -> Dict[str, Any]:
"""Model training metadata."""
return dict(self._meta)
|