Automated generation of process simulation scenarios from declarative changes

Abstract

This thesis addresses the aforementioned limitation by introducing a chatbot-based solution that allows the creation of simulation scenarios with changes specified at both the simulation parameter level and the control flow level. The implementation of changes at the simulation parameter level is achieved through the development of a chatbot using the Rasa Framework, which provides a user-friendly web interface for interacting with the chatbot. The proposed chatbot underwent manual validation using simulation models for three event logs representing different processes. It successfully generated scenarios aligned with the desired modifications. Demonstrated at the 2022 ICPM conference as a Demo paper, the chatbot received positive feedback and generated significant interest from the audience. The live demonstration allowed attendees to interact with the chatbot, validating its effectiveness and user experience. While formal validation using datasets and performance metrics was not conducted, the positive reception at the conference serves as an initial validation of its potential value in real-world applications. On the other hand, to handle changes at the control flow level, the thesis proposes an innovative approach that enables the specification of such changes declaratively, without the need for manual modifications to the underlying procedural model. This approach utilizes a generative deep learning model to generate traces that resemble the specified process change. Based on these generated traces, a stochastic process model is derived and utilized as the foundation for constructing a modified simulation model for further analysis and evaluation. An experimental evaluation of the proposed approach demonstrates that the generated simulation models achieve a level of accuracy comparable to manually created models that directly modify the original process model. This evaluation highlights the effectiveness and reliability of the proposed solution in generating modified simulation models for hypothetical analysis. Overall, this research presents a comprehensive and innovative approach to address the shortcomings in generating simulation scenarios with changes at both the simulation parameter and control flow levels. The chatbot-based solution and the utilization of generative deep learning models contribute to the automation and ease of generating accurate and reliable simulation models, enabling effective hypothetical analysis for process improvement and optimization.