Improved hybrid sparrow search algorithm for an extreme learning machine neural network for short‐term photovoltaic power prediction in 5G energy‐routing base stations

Abstract

Given the advancements in solar power generation and fifth-generation (5G) technologies, it is crucial to reduce energy consumption based on accurate predictions of the photovoltaic power requirements of 5G base stations (BSs) connected to renewable energy sources. To ensure sufficient prediction performance even under complex weather conditions and promote two-way flow of energy and information in the power supply system of the 5G BS, this work first proposes a reference scenario for a 5G BS with an energy router. Then, the comprehensive grey relational analysis is used to determine the optimal photovoltaic input variables. Finally, the improved logistic distribution, Laplacian distribution with inverse incomplete gamma-function weight factor, and nonlinear mutation perturbation strategy are introduced to improve the ability of the sparrow search algorithm to avoid locally optimal solutions for an extreme learning machine of the short-term photovoltaic power prediction model. The proposed model is applied to actual measured data for validation, and the results show that this model has the lowest mean percentage error as well as highest prediction accuracy under different weather conditions. The coefficient of determination for the proposed prediction model is greater than 99%; therefore, it is expected to improve the photovoltaic prediction performances of 5G energy-routing BSs as well as promote sustainable development and sharing of energy and information.