Comparison of Principal-Component-Analysis-Based Extreme Learning Machine Models for Boiler Output Forecasting