دمج الذكاء الاصطناعي مع الويب وأنظمة التحكم

"""Load the irrigation model trained in session 16 and serve predictions (session 17)."""
from datetime import datetime, timedelta

import config
import db
import weather

_model = {"bundle": None, "mtime": None}


def pct_to_vwc(pct):
    """Capacitive sensor % -> volumetric water content using the 2-point calibration in config."""
    return config.VWC_AT_0_PCT + (config.VWC_AT_100_PCT - config.VWC_AT_0_PCT) * (pct / 100.0)


def _load():
    if not config.MODEL_PATH.exists():
        return None
    mtime = config.MODEL_PATH.stat().st_mtime
    if _model["mtime"] != mtime:                 # hot-reload when you retrain the model
        import joblib
        try:
            _model["bundle"] = joblib.load(config.MODEL_PATH)
        except Exception as exc:                 # e.g. trained with another scikit-learn version -> retrain (s16)
            print("model load failed, using the rule fallback:", exc)
            _model["bundle"] = None
        _model["mtime"] = mtime
    return _model["bundle"]


def predict(node_id):
    latest = db.rows("SELECT * FROM readings WHERE node_id=? ORDER BY ts DESC LIMIT 1", (node_id,))
    if not latest or latest[0]["soil_moisture_pct"] is None:
        return {"ok": False, "error": "no readings for this node yet"}
    r = latest[0]
    since = (datetime.fromisoformat(r["ts"]) - timedelta(hours=24)).isoformat(timespec="seconds")
    old = db.rows("SELECT soil_moisture_pct FROM readings WHERE node_id=? AND ts<=? AND soil_moisture_pct IS NOT NULL "
                  "ORDER BY ts DESC LIMIT 1", (node_id, since))
    w = weather.current()
    vwc = pct_to_vwc(r["soil_moisture_pct"])
    features = {
        "soil_vwc": vwc,
        "vwc_drop_24h": (pct_to_vwc(old[0]["soil_moisture_pct"]) - vwc) if old else 0.0,
        "air_temp": r["air_temp_c"] if r["air_temp_c"] is not None else 25.0,
        "air_humidity": r["air_humidity_pct"] if r["air_humidity_pct"] is not None else 60.0,
        "soil_temp": w["soil_temp"], "et0_24h": w["et0_24h"], "rain_24h": w["rain_24h"],
        "radiation": w["radiation"], "wind": w["wind"], "hour": datetime.fromisoformat(r["ts"]).hour,
    }
    bundle = _load()
    if bundle is None:      # graceful fallback: agronomic rule
        prob = 1.0 if vwc < 0.30 else 0.0
        source = "rule (model file not found - run ai/s16_irrigation_classifier.py)"
    else:
        import pandas as pd
        X = pd.DataFrame([[features[f] for f in bundle["features"]]], columns=bundle["features"])
        prob = float(bundle["model"].predict_proba(X)[0][1])
        source = "ml"
    rec = "irrigate" if prob >= config.IRRIGATION_THRESHOLD else "wait"
    if w.get("rain_next_24h", 0) >= 5:
        rec, note = "wait", "rain expected in the next 24 h"
    else:
        note = ""
    # depth to refill the 40 cm root zone to field capacity, capped at one practical irrigation (40 mm)
    litres_m2 = round(min(40.0, max(0.0, (config.VWC_AT_100_PCT * 0.8 - vwc)) * 1000 * 0.40), 1) if rec == "irrigate" else 0
    return {"ok": True, "node_id": node_id, "probability": round(prob, 3), "threshold": config.IRRIGATION_THRESHOLD,
            "recommendation": rec,
            "note": note, "source": source, "vwc": round(vwc, 3), "water_mm": litres_m2,
            "weather": w, "features": {k: round(v, 3) if isinstance(v, float) else v for k, v in features.items()}}

"""Live weather for the farm from the free Open-Meteo forecast API (no key). Cached 30 minutes."""
import json
import ssl
import time
import urllib.request
from datetime import datetime

import config

_cache = {"t": 0, "data": None}
URL = ("https://api.open-meteo.com/v1/forecast?latitude={lat}&longitude={lon}"
       "&hourly=et0_fao_evapotranspiration,precipitation,soil_temperature_0cm,shortwave_radiation,wind_speed_10m"
       "&past_days=1&forecast_days=2&timezone=Africa%2FCairo")


def _ssl_context():
    """Use certifi's up-to-date CA bundle when available (old Windows Python installs ship expired CAs)."""
    try:
        import certifi
        return ssl.create_default_context(cafile=certifi.where())
    except ImportError:
        return ssl.create_default_context()


def _fetch():
    with urllib.request.urlopen(URL.format(lat=config.FARM_LAT, lon=config.FARM_LON), timeout=8,
                                context=_ssl_context()) as r:
        return json.load(r)["hourly"]


def current():
    """Return the weather features the irrigation model needs (with safe fallbacks offline)."""
    if time.time() - _cache["t"] > 1800 or _cache["data"] is None:
        try:
            _cache["data"] = _fetch()
            _cache["t"] = time.time()
        except Exception:
            _cache["data"] = None
    h = _cache["data"]
    if not h:
        return {"online": False, "et0_24h": 4.0, "rain_24h": 0.0, "soil_temp": 25.0,
                "radiation": 400.0, "wind": 10.0, "et0_next_24h": 4.0}
    now = datetime.now().strftime("%Y-%m-%dT%H:00")
    i = h["time"].index(now) if now in h["time"] else len(h["time"]) // 2
    past = slice(max(0, i - 23), i + 1)
    nxt = slice(i + 1, i + 25)
    return {
        "online": True,
        "et0_24h": round(sum(v or 0 for v in h["et0_fao_evapotranspiration"][past]), 2),
        "rain_24h": round(sum(v or 0 for v in h["precipitation"][past]), 2),
        "soil_temp": h["soil_temperature_0cm"][i],
        "radiation": h["shortwave_radiation"][i],
        "wind": h["wind_speed_10m"][i],
        "et0_next_24h": round(sum(v or 0 for v in h["et0_fao_evapotranspiration"][nxt]), 2),
        "rain_next_24h": round(sum(v or 0 for v in h["precipitation"][nxt]), 2),
    }

الطقس يُخزن مؤقتاً 30 دقيقة (Cache) — لا داعي لطلب Open-Meteo مع كل تحديث للوحة. وعند انقطاع الإنترنت يعود بقيم احتياطية معقولة بدلاً من إسقاط الـ API. واستخدمنا certifi لأن نسخ Python القديمة على ويندوز تحمل شهادات SSL منتهية.

async function loadPredictions() {
    const nodes = [...document.querySelectorAll('#control select[name=node_id] option')].map(o => o.value);
    const out = await Promise.all(nodes.map(n => api(`/api/v1/predict/${n}`)));   // all nodes in parallel
    $('#predictions').innerHTML = out.map((p, i) => !p || !p.ok
        ? `<div class="pred"><b>${nodes[i]}</b><span class="muted">لا توجد بيانات كافية</span></div>`
        : `<div class="pred ${p.recommendation}">
              <b>${p.node_id}</b>
              <span class="rec">${p.recommendation === 'irrigate' ? '💧 يحتاج ري خلال 24 ساعة' : '✅ لا يحتاج ري الآن'}</span>
              <span class="prob"><i style="--v:${Math.round(p.probability * 100)}%"></i>${Math.round(p.probability * 100)}%</span>
              <small>${p.water_mm ? `كمية مقترحة ≈ ${p.water_mm} مم` : ''} ${p.note || ''} · ET0 غداً ${p.weather.et0_next_24h} مم</small>
           </div>`).join('');
}
setInterval(loadPredictions, 300000);   // every 5 minutes is enough: soil changes slowly

# db.py - new table
CREATE TABLE IF NOT EXISTS predictions (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    ts TEXT, node_id TEXT, probability REAL, recommendation TEXT, threshold REAL,
    model_mtime REAL,              -- which model version produced it
    outcome INTEGER                -- filled 24 h later: 1 = irrigation was needed
);

# predictor.py - at the end of predict()
db.execute("INSERT INTO predictions (ts, node_id, probability, recommendation, threshold, model_mtime) "
           "VALUES (?,?,?,?,?,?)", (r["ts"], node_id, prob, rec, config.IRRIGATION_THRESHOLD, _model["mtime"]))

# monitor.py - nightly: label yesterday's predictions and compute the live F1
#   outcome = 1 if an irrigation command was executed (commands.status='done') within 24 h after ts,
#             or soil_moisture_pct dropped below SOIL_DRY_PCT within 24 h