Multi-source　mechanical　signals　at　different　locations　of　the　ball　mill　system　have　different　contributions　for　constructing　mill　load　parameter　forecasting　(MLPF)　model.　It　is　necessary　to　select　suitable　multi-source　mechanical　signals　and　their　frequency　spectral　feature　subsets.　Aim　at　these　problems,　a　new　MLPF　approach　based　on　LASSO　algorithm　and　selective　ensemble　(SEN)　modeling　strategy　by　using　multi-source　single-scale　mechanical　frequency　spectrum　is　proposed.　At　first,　Fast　Fourier　transformer　(FFT)　is　used　to　obtain　single-scale　frequency　spectrum　of　these　multi-source　(channel)　mechanical　signals.　Then,　SEN-LASSO　modeling　approach　based　on　candidate　regularization　parameter　set　is　used　to　obtain　single-scale　frequency　spectral　feature　subsets　and　to　build　single-channel　SEN　MLPF　model　jointly.　Finally,　SEN　strategy　is　employed　again　to　construct　the　final　MLPF　model　by　selecting　and　combining　these　single-channel　SEN　MLPF　models.　Simulation　results　based　on　a　laboratory-scale　ball　mill　validate　effectiveness　of　this　approach　(C)　2018,　IFAC　(International　Federation　of　Automatic　Control)　Hosting　by　Elsevier　Ltd.　All　rights　reserved.