Natural Gas Load Forecasting using Fuzzy Sigmoid Kernel Support Vector Machines with Genetic Algorithms

Natural gas load forecasting is very crucial for effective accurate function of pipeline networks. To ensure a balanced network operation and provide confident gas supply with minimum cost, precise forecasting is necessary. In this study, the Support Vector Machines (SVM) is employed to propose a new load forecasting method. To improve the prediction capability, facilitate the hardware implementation, and provide a straightforward interpretability, the sigmoid kernel in SVM is combined with the fuzzy logic approach. Moreover, the Genetic Algorithms (GA) approach is employed to calculate the optimal values of the SVM parameters. A data set containing two years load values is applied to the SVM based prediction model to predict the corresponding values for the next 31days. According to the short-term load predicted values for the city natural gas, the proposed SVM with fuzzy sigmoid kernel gives superior prediction capability compared with the neural network. Moreover, the average CPU execution time is considerably decreased in favor of hardware implementation for load forecasting.