# Copyright © 2026 Setti Solutions S.R.L, www.setti.md. All rights reserved. """A room's thermal resistance R and capacity C from its history (docs/methods.md §6): the planner's first-order model (edge optimizer.py `room`), θ_t = (1 − a) θ_{t−1} + a T_out,t + (Δt/C) P_t + γ G_t + c0, a = Δt / (R C), written as a regression on 15-minute steps (Δt = 0.25 h): θ_t − θ_{t−1} = a (T_out,t − θ_{t−1}) + b P_t + γ G_t + c0, b = Δt / C, with P the heating power (kW; thermal where the site meters the heat output, else electric: the planner's COP turns that into heat), G the global irradiance (kW/m², the sun through the windows) and c0 the internal gains (people, appliances). Least squares on every step where all inputs are there; then C = Δt / b, R = Δt / (a C). The other two terms go to the planner with R and C (Finding B6-a: R and C were fitted with them; without them the planner expects the room to cool faster than it does), as heat in the same units as P: gain_w = 1000 c0 / b the internal gain (W, constant), solar_m2 = γ / b the sun's heat per irradiance (m²: W of heat per W/m² of global irradiance, an effective window area). (electric fits: W and m² of electric equivalent; the box multiplies by the COP like C.) Sanity before a fit is used (else the planner keeps its defaults, R 10 K/kW and C 3 kWh/K): a > 0 and b > 0; R within 1..100 K/kW, C within 0.3..50 kWh/K, τ = RC within 3..1000 h; at least a day of steps, the heating on and off in it, 2 K of spread in T_out − θ; and the fitted model simulates the room better than the defaults do (4-hour open-loop runs from every 4th hour: RMSE in K). The site's heating power is used for every room (the sims: one zone per home): the rooms of a home with one heat meter each get the whole home's C. Per-room heating power, where metered per control, is the next step.""" from __future__ import annotations import math import numpy as np DT_H = 0.25 DEFAULT_R, DEFAULT_C = 10.0, 3.0 R_BOUNDS, C_BOUNDS, TAU_BOUNDS = (1.0, 100.0), (0.3, 50.0), (3.0, 1000.0) MIN_STEPS = 96 SIM_STEPS = 16 SIM_EVERY = 16 def simulate(theta0: float, tout: np.ndarray, p_kw: np.ndarray, ghi_kw: np.ndarray, r: float, c: float, gamma: float = 0.0, c0: float = 0.0) -> np.ndarray: a, b = DT_H / (r * c), DT_H / c out, th = [], theta0 for k in range(len(tout)): th = (1 - a) * th + a * tout[k] + b * p_kw[k] + gamma * ghi_kw[k] + c0 out.append(th) return np.array(out) def sim_rmse(theta: np.ndarray, tout: np.ndarray, p_kw: np.ndarray, ghi_kw: np.ndarray, r: float, c: float, gamma: float = 0.0, c0: float = 0.0) -> float | None: """4-hour open-loop runs from every 4th hour where all inputs exist.""" errs = [] for s in range(0, len(theta) - SIM_STEPS - 1, SIM_EVERY): seg = slice(s + 1, s + 1 + SIM_STEPS) if np.isnan(theta[s]) or np.isnan(theta[seg]).any() or np.isnan(tout[seg]).any() or np.isnan(p_kw[seg]).any(): continue sim = simulate(theta[s], tout[seg], p_kw[seg], np.nan_to_num(ghi_kw[seg]), r, c, gamma, c0) errs.append((sim - theta[seg]) ** 2) if not errs: return None return float(math.sqrt(np.concatenate(errs).mean())) def fit(theta: np.ndarray, tout: np.ndarray, p_w: np.ndarray, ghi: np.ndarray | None = None, power: str = "thermal") -> dict: """theta, tout (°C), p_w (W), ghi (W/m²) per 15-minute slot, aligned. Returns {"status": "fitted"|"default", r_k_per_kw, c_kwh_per_k, tau_h, power, n, rmse_k (simulation, fitted), rmse_default_k, r2, note}, and when fitted gain_w and solar_m2 (the module's docstring).""" n_all = len(theta) ghi = np.zeros(n_all) if ghi is None else np.where(np.isnan(ghi), 0.0, ghi) p_kw = p_w / 1000.0 g_kw = ghi / 1000.0 prev = np.concatenate([[np.nan], theta[:-1]]) dy = theta - prev x1 = tout - prev ok = ~(np.isnan(dy) | np.isnan(x1) | np.isnan(p_kw)) out = {"status": "default", "r_k_per_kw": DEFAULT_R, "c_kwh_per_k": DEFAULT_C, "tau_h": DEFAULT_R * DEFAULT_C, "power": power, "n": int(ok.sum()), "rmse_k": None, "rmse_default_k": None, "r2": None, "note": ""} if ok.sum() < MIN_STEPS: out["note"] = f"{int(ok.sum())} usable steps (at least {MIN_STEPS})" return out on = p_kw[ok] > 0.05 if on.all() or not on.any(): out["note"] = "the heating was never " + ("off" if on.all() else "on") + " in the data" return out if np.ptp(x1[ok]) < 2.0: out["note"] = "less than 2 K between outside and inside" return out X = np.column_stack([x1[ok], p_kw[ok], g_kw[ok], np.ones(int(ok.sum()))]) coef, *_ = np.linalg.lstsq(X, dy[ok], rcond=None) a, b, gamma, c0 = (float(v) for v in coef) resid = dy[ok] - X @ coef var = float(np.var(dy[ok])) out["r2"] = round(1 - float(np.var(resid)) / var, 3) if var > 0 else None if a <= 0 or b <= 0: out["note"] = f"no physical fit (a {a:.4g}, b {b:.4g})" return out c = DT_H / b r = DT_H / (a * c) tau = r * c out["fit"] = {"a": a, "b": b, "gamma": gamma, "c0": c0} if not (R_BOUNDS[0] <= r <= R_BOUNDS[1] and C_BOUNDS[0] <= c <= C_BOUNDS[1] and TAU_BOUNDS[0] <= tau <= TAU_BOUNDS[1]): out["note"] = f"outside the sanity bounds (R {r:.3g} K/kW, C {c:.3g} kWh/K, τ {tau:.3g} h)" return out fitted = sim_rmse(theta, tout, p_kw, g_kw, r, c, gamma, c0) default = sim_rmse(theta, tout, p_kw, g_kw, DEFAULT_R, DEFAULT_C) out["rmse_default_k"] = round(default, 3) if default is not None else None out["rmse_k"] = round(fitted, 3) if fitted is not None else None if fitted is None or default is None: out["note"] = "too few complete 4-hour stretches to check the fit" return out if fitted >= default: out["note"] = f"no better than the defaults ({fitted:.2f} K vs {default:.2f} K)" return out out.update(status="fitted", r_k_per_kw=round(r, 3), c_kwh_per_k=round(c, 3), tau_h=round(tau, 2), gain_w=round(1000.0 * c0 / b, 1), solar_m2=round(gamma / b, 3), note=f"simulates the room to {fitted:.2f} K (defaults {default:.2f} K)") return out