6.1 PRODUCTION DASHBOARD EXAMPLE Figure 8 illustrates a three-hour operating window (11:00–14:00 on 12/02/2026) in which a change in ore type is detected (cluster transition). This regime change is reflected in the SAG mill behavior and provides a representative stress test for short-term forecasting under nonstationary conditions. Figure 9 shows the model’s online multihorizon forecasts of SAG throughput (TPH) at 5-, 15-, and 30-minute horizons over the same period. For each horizon, the point forecast (mixture expectation) is plotted together with uncertainty bands (P10–P90), which are obtained from samples drawn from the predicted mixture distribution. As the operating regime evolves, the uncertainty bands expand or contract accordingly, communicating forecast confidence in real time. Despite the ore-type transition, the probabilistic forecasts track the main trend of the measured TPH and provide consistent horizon-dependent behavior: shorter horizons react faster, while longer horizons are smoother and capture medium-term tendencies. Figure 8 – Ore-type change detected during a 3-hour operating window (12/02/2026, 11:00– 14:00), shown as a cluster transition with associated probability. Figure 9 – Online probabilistic SAG throughput forecasts at 5, 15, and 30-minute horizons for the same operating window, including P10–P90 uncertainty bands. 6.2 WHAT-IF ANALYSIS To assess the ability of the probabilistic twin to compare operational alternatives before execution, we implemented a what-if study using the CVAE+MDN-RNN model over a 30-minute
RkJQdWJsaXNoZXIy MTM0Mzk2