Forecast will never be identical to the actual series. After all this is a model. To quote famous statistician George Box "essentially, all models are wrong, but some are useful". Unfortunately there isn't a magic way to select lag orders, however there are several strategies you can use to improve results. For a start take a look at a surveys common strategies for lag order selection here:
Other examples that use autocorrelation and paritial autocorrelation along with aic and bic:
