A Novel Approach for Supporting Approximation Representation of Linear Prediction Residuals by means of Simulated-Annealing-Based Optimization