matlab - Using Linear Prediction Over Time Series to Determine Next K Points -

I have a time series of data spots of sunspots and would like to infer based on a subset of these points Compare the points in the series and then the correctness.

I'm just presenting for linear prediction using Multback and so I have decided that I will go to the path of using the following code segment within a loop so that each point given data There is a prediction outside the training set by the end of:

 % x is the number of data, training set X starts from the beginning of some subset% of 'unknown'% end of the training set From Point to expand predict when Uru does (ie training set and the difference in the length of the data vectors)% x_pred x start p = length (Prshikshn_set); Coeffs = lpc (training_set, p); For I = 1: unknown next value = -coffs (2: end) * x_preped (end-unknown -1 + i: -1: end-unknown -1 + i-p + 1); '; X_ Pred (end-unknown + i) = next value; End error = ideal (x - x_pred)  

I have three questions:

1) Is it fair that what I have described? I ask because my mistake is too big (> 100) when it estimates the last 20 digits of the dataset, which contains hundreds of digits.

2) Do I correctly interpret LPC's second argument? That is, it means 'order' or the number of points that you want to use in the next point prediction?

3) If this is a more efficient, single line function in matlab, then call me to replace looping and calculate all the necessary predictions for me, as a set of my training. Have some subsets of aggregate data?

I tried to search for the LPC matlab tutorial, but it did not seem to be predicting as I mentioned that my requirements are needed, I am also using it as a reference. Therefore, after much discussion and experimentation, I have considered the above mentioned approach as right and do not see it. To perform the above mentioned functions, have a single matte function. Experiencing large errors is justified because I am using a linear prediction algorithm (i.e. sunscot prediction) for a problem which is the underlying non-linear behavior.

Hope this will help someone else to work in something like this.