Add UTR221_ML_previsao_falha_com.py – 6h pressure prediction during communication failure

[Copilot]

Operators need to predict UTR221 pressure behavior for up to 6 hours when communication with the UTR is lost. Previously no script existed for this scenario; the only prediction scripts required live or historical timestamped data from the sensor.

New script: ML APP/UTR221_ML_previsao_falha_com.py

• Model training: same RandomForest pipeline as UTR221_ML_validacao.py (fit on urt220.xlsx / urt221.xlsx, ~99.62% accuracy)
• Terminal input: collects all required lag features interactively at runtime
  • Failure timestamp (flexible format: DD/MM/YYYY HH:MM or ISO)
  • B1/B2 pump status from UTR220 (CMB_220_S2A_EST, CMB_220_S2B_EST) — lag 1
  • Level from UTR220 (LIT_220_RA2_000) — lag 1
  • Pressure from UTR221 (PIT_221_S01_000) — lag 1, lag 2, lag 60, and 1-hour rolling mean
• Recursive prediction: 360 steps (6 h) from the failure timestamp, with pump on/off control logic based on setpoints (1.55 / 1.68 bar)
• Outputs (named by failure timestamp):
  • previsao_falha_DDMM_HHMM.png — pressure curve + dual-axis pump state + setpoint lines
  • previsao_falha_DDMM_HHMM.xlsx — minute-by-minute predictions

Horário da falha de comunicação (DD/MM/AAAA HH:MM): 11/04/2026 14:35
Status B1 (CMB_220_S2A_EST) [0=desligado / 1=ligado]: 1
Status B2 (CMB_220_S2B_EST) [0=desligado / 1=ligado]: 0
Último nível UTR220 [m]: 1.72
Pressão mais recente       (lag 1, há  1 min) [bar]: 1.65
Pressão anterior            (lag 2, há  2 min) [bar]: 1.64
Pressão há 60 min           (lag 60, há 1 hora) [bar]: 1.60
Média de pressão última hora (60 min)           [bar]: 1.63

Naming fix: feature previously called nivel_221_lag1 (misleading — it carries pressure, not level) renamed to pressao_221_lag1 throughout.