<p>The existing adaptive neuro-fuzzy inference system (ANFIS) for load frequency control in multi-area power systems has two inputs consisting of area control error (ACE) and its integer order derivative. A recently proposed ANFIS has added another input consisting of integer order integral of the ACE. In this paper, ANFIS controllers with fractional order derivative of the ACE and fractional order integral of the ACE, referred to as fractional order ANFIS (FO-ANFIS) controllers in this paper, are used instead to improve the performance of ANFIS controllers. The FO-ANFIS training dataset is obtained from the input ACE, its fractional order derivative and its fractional order integral of a cascaded fractional order PI-fractional order PID with derivative filters (FOPI-FOPIDN) tuned by a particle swarm optimization variant called adaptive dynamic inertia weight acceleration coefficient optimization algorithm. The controllers consisting of 2-input FO-ANFIS and 3-input FO-ANFIS, are tested on a three-area power system integrated with renewable energy sources. The results obtained are compared with those of their integer order ANFIS (IO-AFIS) counterparts and the FOPI-FOPIDN from which the training data was obtained using the overshoot, undershoot, settling time, steady-state error in the frequency and tie-line power responses as well as integral time absolute error values. Their real-world applicability is validated by incorporating communication time delay and governor dead band in one of the four experimental scenarios used for the evaluation. Robustness to power system parameter uncertainty is further assessed through parameter variation of ± 25%. From the results, the 3-input FO-ANFIS controller emerges as the best performing between the ANFIS controllers followed by the 2-input FO-ANFIS controller.</p>