Dynamics　of　cold　tandem　rolling　processes　are　difficult　to　be　simulated　with　mechanism　models　due　to　their　high　non-linearities,　multivariable　and　strong　couplings,　time-varying　distributed　parameters　and　so　on.　A　novel　hybrid　intelligent　dynamic　modelling　approach　is　proposed　based　on　the　combination　of　a　linearised　state　space　model　derived　from　various　mechanism　equations,　a　case-based　reasoning　algorithm　for　multi-state　space　models　selection,　a　genetic　algorithm　for　optimisation　of　case　attributes,　an　adaptive　fractal　filtering　algorithm　for　the　identification　of　state　space　model　parameters,　a　neural　network-based　simulation　error　compensation　model　for　the　strip　exit　velocity.　With　actual　data　from　a　2030　mm　five-stand　cold　tandem　rolling　system　of　a　steel　plant,　simulation　experiments　verify　the　effectiveness　of　the　proposed　approach.　Furthermore,　a　rolling　simulation　system　is　developed　based　on　the　proposed　model,　and　the　virtual　tandem　rolling　experiments　with　the　simulation　system　also　validate　that　the　proposed　model　can　accurately　simulate　the　dynamics　variation　from　different　types　of　disturbances　of　cold　tandem　rolling　processes.