Artificial Bee Colony Based Learning of Local Linear Neuro-Fuzzy Models

One of the powerful methods in identification and prediction tasks is local linear neuro-fuzzy (LLNF) modeling, and it has been proven to be robust and accurate. Estimating the right parameters of an LLNF model is the main problem in establishing the proper one. Usually Model Tree (LOLIMOT) algorithm is used for learning of the LLNF models, but it is a less accurate technique. LLNF modeling approach is based on divide and conquer strategy. It divides the problem space into different partitions and solves the sub-problems. In order to do so, it needs to establish three kinds of parameters, which two of them (Center and Sigma) are fuzzy and employed to create the partitions, and the other one establishes connections between different parts of the model and inputs. The main task, here, is to compute the two internal fuzzy parameters. Once they have been calculated, the other parameter can be computed simply. Artificial Bee Colony (ABC) is a population-based method, which is usually used for optimization problems. It works on multidimensional real-valued functions, which are not necessarily continuous or differentiable. Hence, it's a good option for estimating the internal parameters of LLNF model. In this paper, we will introduce a new learning algorithm of LLNF models based on ABC method. At the end, the comparison results with LOLIMOT has shown that our method is better in accuracy.