The existing extensions of Petri nets supporting dynamic changes cannot ensure preservation of desired properties of the modeled systems. This paper tries to address this problem. By extending the structure of the net rewriting system (NRS) and modifying the specification of its rewriting rule, we present improved net rewriting system (INRS). INRS enjoys four important properties, i.e., locality, transitivity, reflexivity, and reversibility. A practical system subject to dynamic changes can be modeled by an INRS, and its desired properties can be preserved with the state evolvement of the INRS. The property-preserving principles of INRS and their strict proofs are provided as well. Finally, a case study is provided to demonstrate the modeling of an automated manufacturing cell with changeable structure using an INRS. The result shows that the desired properties of the cell are still preserved despite its model evolvement.