Model structures and fitting criteria for system identification with neural networks Marco Forgione 1 and Dario Piga 1 1 IDSIA Dalle Molle Institute for Artificial Intelligence SUPSI-USI, Manno, Switzerland October 29, 2021 To cite this work, please use the following bibtex entry: @inproceedings{forgione2020model, title={Model structures and fitting criteria for system identification with neural networks}, author={Forgione, Marco and Piga, Dario}, booktitle={Proc. of the 14th IEEE International Conference Application of Information and Communication Technologies, Tashkent, Uzbekistan}, year={2020} } Using the plain bibtex style, the bibliographic entry should look like: M. Forgione, D. Piga. Model structures and fitting criteria for system iden- tification with neural networks. In Proc. of the 14th IEEE International Con- ference Application of Information and Communication Technologies, Tashkent, Uzbekistan, 2020. Abstract This paper focuses on the identification of dynamical systems with tailor-made model structures, where neural networks are used to approx- imate uncertain components and domain knowledge is retained, if avail- able. These model structures are fitted to measured data using different criteria including a computationally efficient approach minimizing a regu- larized multi-step ahead simulation error. The neural network parameters are estimated along with the initial conditions used to simulate the out- put signal in small-size subsequences. A regularization term is included in 1 arXiv:1911.13034v2 [cs.LG] 28 Oct 2021