evo-thermal-sysid

System identification for first-order thermal HVAC systems.

Functions

• run_calibration_test(sim, heater_power=50.0, duration=60.0) - Run open-loop step test, returns calibration_log dict
• first_order_step_response(t, K, tau, T_amb, U_step) - First-order model: T_amb + KU_step(1-exp(-t/tau))
• identify_system_params(calibration_log) - Fit K and tau from calibration data using curve_fit
• filter_temperature_data(temp_data, window_length=11, polyorder=2) - Savitzky-Golay filter
• calculate_fit_metrics(T_actual, T_predicted) - Returns (r_squared, rmse)
• save_calibration_results(calibration_log, estimated_params) - Save JSON files to /root/

Usage

import sys
sys.path.insert(0, '/app/environment/skills/evo-thermal-sysid/scripts')
from sysid_utils import run_calibration_test, identify_system_params, save_calibration_results

sys.path.insert(0, '/root')
from hvac_simulator import HVACSimulator

sim = HVACSimulator()
calib_log = run_calibration_test(sim, heater_power=50.0, duration=60.0)
params = identify_system_params(calib_log)
save_calibration_results(calib_log, params)

Key Domain Knowledge

• First-order thermal model: dT/dt = (1/tau) * (K*u + T_amb - T)
• Step response: T(t) = T_amb + KU_step(1 - exp(-t/tau))
• K ~ 0.12 C/% power, tau ~ 40s for typical HVAC
• Use Savitzky-Golay filter (preserves exponential shape) not moving average
• Use curve_fit with bounds ([0.001, 1.0], [1.0, 200.0]) and maxfev=10000
• Calibration needs >= 30s duration, >= 20 data points
• R-squared > 0.95 indicates good first-order fit