File size: 8,245 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 | """
BaselinePredictor: hybrid FvCB + ML photosynthesis baseline for day-ahead planning.
Provides a single ``predict_day()`` method that:
1. Runs FvCB (Farquhar–Greer–Weedon) for each slot using forecast weather
2. Optionally runs a trained ML model for the same slots
3. Uses the RoutingAgent's rule-based logic to pick the better prediction per slot
4. Returns a 96-slot profile of predicted photosynthesis rate A (µmol CO₂ m⁻² s⁻¹)
This feeds into the DayAheadPlanner to estimate crop value for each slot,
replacing the current temperature-only heuristic with an actual photosynthesis
prediction that captures the Rubisco transition more accurately.
"""
from __future__ import annotations
import logging
import math
from datetime import date
from typing import List, Optional
import numpy as np
from config.settings import SEMILLON_TRANSITION_TEMP_C
logger = logging.getLogger(__name__)
class BaselinePredictor:
"""Hybrid FvCB + ML photosynthesis prediction for day-ahead planning.
Parameters
----------
fvcb_model : FarquharModel, optional
Lazy-initialised if not provided.
ml_predictor : PhotosynthesisPredictor, optional
Trained ML model. If None, FvCB-only mode is used.
routing_agent : RoutingAgent, optional
Model router for per-slot FvCB/ML selection.
If None, uses rule-based routing only (no API calls).
"""
def __init__(
self,
fvcb_model=None,
ml_predictor=None,
routing_agent=None,
):
self._fvcb = fvcb_model
self._ml = ml_predictor
self._router = routing_agent
@property
def fvcb(self):
if self._fvcb is None:
from src.models.farquhar_model import FarquharModel
self._fvcb = FarquharModel()
return self._fvcb
# ------------------------------------------------------------------
# Main API
# ------------------------------------------------------------------
def predict_day(
self,
forecast_temps: List[float],
forecast_ghi: List[float],
co2_ppm: float = 400.0,
rh_pct: float = 40.0,
) -> List[float]:
"""Predict photosynthesis rate A for each 15-min slot.
Parameters
----------
forecast_temps : list of 96 floats
Forecast air temperature (°C) per slot.
forecast_ghi : list of 96 floats
Forecast GHI (W/m²) per slot.
co2_ppm : float
Atmospheric CO₂ concentration (default 400 ppm).
rh_pct : float
Relative humidity (%) for VPD estimation (default 40%).
Returns
-------
list of 96 floats
Predicted net photosynthesis A (µmol CO₂ m⁻² s⁻¹) per slot.
0.0 for nighttime slots.
"""
assert len(forecast_temps) == 96 and len(forecast_ghi) == 96
# FvCB predictions for all 96 slots
fvcb_predictions = self._predict_fvcb(
forecast_temps, forecast_ghi, co2_ppm, rh_pct,
)
# If no ML model, return FvCB-only
if self._ml is None:
return fvcb_predictions
# ML predictions for all 96 slots
ml_predictions = self._predict_ml(forecast_temps, forecast_ghi)
# Route each slot
predictions = self._route_predictions(
forecast_temps, forecast_ghi,
fvcb_predictions, ml_predictions,
)
return predictions
# ------------------------------------------------------------------
# FvCB predictions
# ------------------------------------------------------------------
def _predict_fvcb(
self,
temps: List[float],
ghis: List[float],
co2_ppm: float,
rh_pct: float,
) -> List[float]:
"""Run FvCB for each slot. Returns 96 A values."""
predictions = []
for i in range(96):
temp = temps[i]
ghi = ghis[i]
# Nighttime or negligible light
if ghi < 50:
predictions.append(0.0)
continue
# Estimate PAR from GHI (roughly 2× conversion for photosynthetically active)
par = ghi * 2.0
# Estimate Tleaf from Tair (proxy: +2°C under sun)
tleaf = temp + 2.0
# Estimate VPD from temperature and RH
vpd = self._estimate_vpd(temp, rh_pct)
try:
result = self.fvcb.calc_photosynthesis_semillon(
PAR=par,
Tleaf=tleaf,
CO2=co2_ppm,
VPD=vpd,
Tair=temp,
)
# Returns (A, limiting_state, shading_helps)
A = result[0] if isinstance(result, tuple) else result
predictions.append(max(0.0, float(A)))
except Exception as exc:
logger.debug("FvCB failed at slot %d: %s", i, exc)
predictions.append(0.0)
return predictions
@staticmethod
def _estimate_vpd(tair_c: float, rh_pct: float) -> float:
"""Estimate VPD (kPa) from air temperature and relative humidity."""
# Tetens formula for saturated vapor pressure
es = 0.6108 * math.exp(17.27 * tair_c / (tair_c + 237.3))
ea = es * rh_pct / 100.0
return max(0.0, es - ea)
# ------------------------------------------------------------------
# ML predictions
# ------------------------------------------------------------------
def _predict_ml(
self,
temps: List[float],
ghis: List[float],
) -> List[float]:
"""Run ML model for each slot. Returns 96 A values."""
if self._ml is None:
return [0.0] * 96
try:
import pandas as pd
# Build feature DataFrame matching ML model's expected features
hours = [i * 0.25 for i in range(96)]
df = pd.DataFrame({
"air_temperature_c": temps,
"ghi_w_m2": ghis,
"hour": [int(h) for h in hours],
"minute": [int((h % 1) * 60) for h in hours],
})
# Try prediction with the best model
best_model = None
best_mae = float("inf")
for name, result in self._ml.results.items():
if result.get("mae", float("inf")) < best_mae:
best_mae = result["mae"]
best_model = name
if best_model and best_model in self._ml.models:
model = self._ml.models[best_model]
# Use whatever features the model was trained on
feature_cols = [c for c in df.columns if c in getattr(model, "feature_names_in_", df.columns)]
if feature_cols:
preds = model.predict(df[feature_cols])
return [max(0.0, float(p)) for p in preds]
except Exception as exc:
logger.warning("ML prediction failed: %s", exc)
return [0.0] * 96
# ------------------------------------------------------------------
# Routing
# ------------------------------------------------------------------
def _route_predictions(
self,
temps: List[float],
ghis: List[float],
fvcb_preds: List[float],
ml_preds: List[float],
) -> List[float]:
"""Pick FvCB or ML per slot using routing logic."""
from src.chatbot.routing_agent import RoutingAgent
predictions = []
for i in range(96):
telemetry = {
"temp_c": temps[i],
"ghi_w_m2": ghis[i],
"hour": i // 4,
}
# Use rule-based routing only (no API calls for batch prediction)
choice = RoutingAgent._rule_based_route(telemetry)
if choice is None:
# Transition zone: weight FvCB 60% / ML 40% as compromise
a = 0.6 * fvcb_preds[i] + 0.4 * ml_preds[i]
elif choice == "ml":
a = ml_preds[i]
else:
a = fvcb_preds[i]
predictions.append(a)
return predictions
|