Computer algorithm to predict anode effect events

Abstract

Alumar which is one of Alcoa units, by following a corporate vision, is pursuing ways to reduce greenhouse gas emissions. Normally, the pot voltage is used to detect the anode effect events every 1 to 10 seconds. With the continuous improvements on the computer performance to handle faster scan data, we are able to read the pot voltage every 100 milliseconds. Our attempt is to use this fast scan data to distinguish the normal and the pre-anode effect voltage period. An algorithm has been created to detect this behavior, based on the speed of the voltage increase. With simulation we observed that 60 to 70% of the anode effects are predicted by the new algorithm. The other 40 to 30% are ignored in order to reduce false detections. Only 2 to 10% of the predictions do not really result in anode effects. The accuracy is strongly associated with the noise. The average voltage of prediction is 5.19V where the normal computer detection is 8V. The predictor was tested during one month in 102 pots indicating that we can predict the anode effect events from 7 to 20 seconds prior to its occurrence. These tests resulted in a reduction of 30% of time above 8 volts and 20% reduction in anode effect/potday.