Enhanced Salp Search Algorithm for Optimization Extreme Learning Machine and Application to Dew Point Temperature Prediction

Abstract

Extreme learning machine (ELM) is popular as a method of training single hidden layer feedforward neural networks. However, the ELMs optimized by the traditional gradient descent algorithms cannot fundamentally solve the influence of the random selection of the input weights and biases. Therefore, this paper proposes a method of extreme learning machine optimized by an enhanced salp search algorithm (NSSA-ELM). Salp search algorithm (SSA) is a metaheuristic algorithm, to improve the performance of SSA exploration and avoid getting stuck in local optima, the neighborhood centroid opposite‑based learning is used to optimize SSA. This method maintains the diversity of the population, which is conducive to avoid local optimization and accelerate convergence. This paper performs classification tests on NSSA and other metaheuristic-optimized ELMs on ten datasets, and regression tests on 5 datasets. Finally, the prediction ability of dew point temperature is evaluated. The meteorological data of five climatically representative cities in China from 2016 to 2022 were collected to predict the dew point temperature. The experimental results show that the NSSA-ELM is the best model, and its generalization performance and accuracy are better than other models.