A Novel Hybrid Differential Evolution-PSNN for Fuzzy Time Series Forecasting

Abstract

Over the years, fuzzy time series (FTS) has been used more popularly for forecasting the real-time series data. Generally, in FTS method, the membership value has not been considered for forecasting purpose, which leads as a drawback in the forecasting process. Recently, some researchers have overcome this problem by introducing the artificial neural network (ANN) concept to find the fuzzy relation. Again, the ANN has multimodal training problems, so the gradient descent algorithms will get stock in local minima. Therefore, the current study proposed a new hybridizing approach, which has used the differential evolution algorithm (DE) with PSHONN for forecasting the time series data. The hybrid DE [1] evolutionary algorithm has only two parameters such as parent $\left( {x_{i} } \right)$ and child $\left( {u_{i} } \right)$; therefore, it is more effective and faster than other evolutionary algorithms. Twelve different time series data sets are applied in our proposed model and the obtained result is compared with CRO [2, 3] and Jaya [4, 5]. In all cases, we found that the effectiveness and performance of the Jaya-PSNN were very poor and DE-PSNN model outperformed on most of the experimental results.