Modelling and identification: progress and new challenges

Modelling and identification has seen enormous progress and paradigm shifts over the last 40 years. The systems and control community has played a major role in these developments. And yet, the task of modelling and identification still represents the major cost in any advanced control project. It is still viewed as a difficult problem that must be left to the experts. So this raises the questions: what progress has been accomplished? what exactly are the bottlenecks? what are the new challenges? This talk will first outline the major steps accomplished over this 40-year period. It will then focus on the new challenges which can be encapsulated in one major goal: reduce the cost of the identification. This focus on the reduction of the modelling and identification cost is quite recent and takes on several different forms. The cost can be reduced by tuning the model to the application for which it is intended (e.g. control application), by reducing the experiment time, by reducing the model complexity, by optimizing the data collection experiment (optimal experiment design), by producing methods and algorithms that rely as little as possible on the expertise of the user.