Improved Extreme Learning Machine Power Load Forecasting Based on Firefly Optimization Algorithms

Abstract

Power load forecasting is a crucial safeguard for the reliable, efficient, and safe operation of the power system, which is connected to the smooth operation of society at all levels. In practical applications, extreme learning machines have advantages like quick learning rates and minimal training error, but they have poor stability and generalization skills. The learning of the sample lacks relevance because the weight matrix between the hidden layer and the output layer in the limit learning is chosen at random. Because of its simple algorithm flow and strong global optimization capabilities, the firefly algorithm helps to simplify the calculation process. To address the drawbacks of the extreme learning machine and combine its benefits, this paper integrates the firefly algorithm into limit learning and makes use of its potent optimization capabilities to determine the connection weight between the extreme learning machine’s hidden layer and output layer when the training error is minimal.