You will need to implement a predictor model in Simulink. You can "convert" an estimated idss model into a predictor model. Suppose sys is the estimated idss model. Then you can create a one-step ahead predictor model as follows:
[a,b,c,d] = ssdata(sys);
sysp = idss([a-k*c],[k b-k*d],c,[zeros(ny), d],'ts',sys.Ts, 'NoiseVariance', sys.NoiseVariance);
where ny = number of outputs and nu = number of inputs of original model sys. sysp is a model with nu+ny inputs and ny outputs. The input to be used for simulation of sysp must be [y, u], where y is the measured output data (ny columns) and u is the measured input data (nu columns) that you want to use for prediction.
For an arbitrary horizon, construction of "sysp" would not be so straight-forward. Look up the PREDICT command which returns sysp as its 3rd output argument. See:
