Editing Fuzzy control system (section)

==Fuzzy qualitative simulation==
Before an Artificial Intelligence system is able to plan the action sequence, some kind of [[Mathematical model|model]] is needed. For video games, the model is equal to the game rules. From the programming perspective, the game rules are implemented as a [[Physics engine]] which accepts an action from a player and calculates, if the action is valid. After the action was executed, the game is in follow up state. If the aim isn't only to play [[mathematical game]]s but determine the actions for real world applications, the most obvious bottleneck is, that no game rules are available. The first step is to model the domain. [[System identification]] can be realized with precise mathematical equations or with [[Fuzzy rule]]s.<ref>{{cite thesis |degree=PhD |title=Fuzzy qualitative simulation and diagnosis of continuous dynamic systems |author=Shen, Qiang |date=September 1991 |publisher=University of Edinburgh |hdl=1842/7307 |hdl-access=free}}</ref>

Using Fuzzy logic and [[Adaptive neuro fuzzy inference system|ANFIS]] systems (Adaptive network based fuzzy inference system) for creating the [[Internal model (motor control)#Forward models|forward model]] for a domain has many disadvantages.<ref>{{cite conference |title=Generating fuzzy models from deep knowledge: robustness and interpretability issues |author=Guglielmann, Raffaella |author2=Ironi, Liliana |conference=European Conference on Symbolic and Quantitative Approaches to Reasoning and Uncertainty |pages=600–612 |year=2005 |publisher=Springer |doi=10.1007/11518655_51}}</ref> A [[qualitative simulation]] isn't able to determine the correct follow up state, but the system will only guess what will happen if the action was taken. The Fuzzy qualitative simulation can't predict the exact numerical values, but it's using imprecise [[natural language]] to speculate about the future. It takes the current situation plus the actions from the past and generates the expected follow up state of the game.

The output of the ANFIS system isn't providing correct information, but only a [[Fuzzy set]] notation, for example [0,0.2,0.4,0]. After converting the set notation back into numerical values the accuracy get worse. This makes Fuzzy qualitative simulation a bad choice for practical applications.<ref>{{cite journal |title=Fuzzy qualitative trigonometry |author=Liu, Honghai |author2=Coghill, George M |author3=Barnes, Dave P |journal=International Journal of Approximate Reasoning |volume=51 |number=1 |pages=71–88 |year=2009 |publisher=Elsevier |doi=10.1016/j.ijar.2009.07.003|s2cid=47212 |url=http://pure.aber.ac.uk/ws/files/124705/Published_Fuzzy_Qualitative_Trigonometry_IJAR.pdf |archive-url=https://web.archive.org/web/20200506021452/http://pure.aber.ac.uk/ws/files/124705/Published_Fuzzy_Qualitative_Trigonometry_IJAR.pdf |archive-date=2020-05-06 |url-status=live |doi-access=free }}</ref>