Using mixed RP/SP models with non linear effects in forecasting