Validation of local model networks for electrically stimulated muscle

In this work we use Local Model Networks (LMN) as a non-linear modelling structure. In LMN, simple locally valid models are smoothly interpolated, forming a non-linear model. If the local models have a linear structure, tools are available to analyse the LMN in terms of changes of properties (such as steady state gain, location of poles) with changing operating conditions. The existence of such tools facilitates the validation of the model by providing means to check whether the model properties correspond to known properties of the real system. We use Local Model Networks to identify empirical models for isometric muscle contraction. Real data from rabbit muscle are employed. Both fast and slow muscle are investigated. Cross-validation tests show that Local Model Network models identified for these muscles produce good representations of the system behaviour. By applying the analysis tools to the LMNs identified for the fast and slow muscle it is shown that the different properties of these two kinds of muscle are represented by the empirical models.