Abstract
The Speaker Recursive algorithms with forgetting factor are well-known algorithms for on-line identification. The forgetting factor in these algorithms facilitates the gradual discounting for the past data. However, it is found that recursive algorithms are best suited when the underlying process is time varying. This approach is not very useful when the underlying process is a non-linear process. To this end, Just-In-Time (JIT), also known as locally weighted regression, modeling is of help. This approach finds the most relevant data to the current query from the historical data base. Using this data, a local model is built corresponding to the query. This Just-In-Time approach is approximation of a non-linearity using local linear models. In this talk, we propose a unified approach which merges traditional recursive approach in JIT framework. This unified approach is capable of addressing time varying as well as non-linearity issues in modeling. The usefulness of this approach is illustrated on an industrial case study.
[an error occurred while processing this directive]