Based on generalized intuitionistic fuzzy Kripke structures (GIFKSs), we propose a novel state transition self-learning framework to address the problems of state modeling and parameter adaptability in dynamic systems. In this work, we assume that the states of GIFKS are unknown, and the only available knowledge is the sensor variables that exhibit ambiguous relationships with these states. To bridge the gap between these sensor variables and the states, we connect the sensor data to the atomic propositions representing the states using a set of intuitionistic fuzzy functions. The first learning model for GIFKS, called the GIFKS with Intuitionistic Fuzzifier Sets (GIFKS - IFSs), combines a standard GIFKS model with a collection of intuitionistic fuzzy functions that link the system’s states to various sensor variables. These intuitionistic fuzzy functions allow for more flexible handling of uncertainty by considering both membership and non-membership degrees in the fuzzy relations between the sensor data and system states. To optimize the parameters of the intuitionistic fuzzy function that describes the state transition mechanism and the elements of the atomic propositional transition matrix, we derive learning algorithms based on stochastic gradient descent. The proposed self-learning model demonstrates superior performance in both theoretical and experimental settings, offering strong parameter adaptability and algorithmic convergence.