The Tripura power engineer and the grid friction
⚡ Ashutosh Dey, 34, works as an electrical engineer at OTPC Palatana thermal power station—726 MW, operating at 65% capacity due to gas constraints and Bangladesh grid-demand fluctuations. Load-rejection events (where the turbine trips offline to protect itself) happen two to three times a month. On Tuesday, April 16, 2026, at 2:18 AM, Bangladesh demand spiked. Ashutosh held the plant at 440 MW instead of ramping to 560 MW as ordered, preventing a load-rejection. The voltage reached 0.86 per unit—just above the SCADA limit of 0.85. He acted early, based on a voltage trend he had learned to predict. But he deviated from dispatch orders to do it.
🚨 The problem
Bangladesh grid demand is unpredictable—scheduled 200 MW at 1 AM, requested +180 MW at 2:10 AM. OTPC is told to ramp from 380 to 560 MW in forty minutes. But forty-minute ramps at 560 MW (the plant's absolute maximum) are at the edge of safe turbine control. If Bangladesh's grid destabilizes while OTPC is mid-ramp, voltage collapses, the plant trips offline. ₹5–7 lakhs/hour revenue lost. A two-hour investigation cycle. Any engineer in the room would know Ashutosh had held at 440 MW instead of following dispatch orders—which was, technically, outside his authority.
🚀 How GabFORGE helped
In May 2026, Ashutosh's friend Rahul sent him a message: "Some thermal plants in Odisha are using an AI product that reads shift logs and predicts grid instability." Ashutosh uploaded April's shift logs—frequency waveforms, voltage trends, MW output, dispatch orders. The agent analyzed them and returned: Your April had 15 potential load-rejection events. SCADA detected 3. The other 12 occurred 5–30 minutes before the limit was reached. One event on April 16 at 02:15 was very close—voltage reached 0.86. Would this have predicted the April 16 event before you took action? Yes. The agent would have given the prediction at 02:08—seven minutes before you acted.
🔍 Identified near-miss events — flagged 12 events SCADA missed entirely. 💬 Provided early warning — voltage trend prediction gave 7–8 minutes advance notice. 📞 Enabled proactive decision — instead of reacting when voltage was collapsing, Ashutosh could call TERC at 02:08 and negotiate a dispatch reduction.
Ashutosh presented the analysis to the operations director. Pandit said: Install it on the control-room tablets. Do not advertise this to Delhi. The agent is now standard on every shift, flagging load-rejection predictions before the voltage collapses. ✅
🇮🇳 Why this matters
An AI reading shift data is not replacing the engineer. It is extending the engineer's senses backward in time—showing what the grid is about to do, not what it has already done.
The long version contains [details from original article].