Using LSTM and GRU to Predict SAG Mill Energy Consumption
MetadataShow full item record
In mine operations, comminution is the most demanding energy consumer. Within comminution, semi-autogenous grinding mills are by far the most intensive consumers. Classic techniques to forecast their future energy consumption relies on feeding ore characterization and some operational variables. However, two of their main assumptions are working in steady-state and without up/downstream process bottlenecks. This work explores the capability of Recurrent Neural Networks techniques to capture input time dependencies based only on operational information. Two of the most widely used recurrent networks are compared: Long-Short Term Memory and Gated Recurrent Unit. Results show high performance on small supports (30 minutes) to capture local variability and long-trends while larger time supports (8 hours) capture time-trends but have difficulties on realizing local variability. Finally, there is no clear advantage between choosing between one technique and the other since both show good results.