Accurate Photovoltaic Power Forecasting in 5G Networks: A Novel Neural Network Approach

Accurate Photovoltaic Power Forecasting in 5G Networks

Abstract

This study addresses the challenge of integrating photovoltaic (PV) power generation into 5G base stations to reduce energy consumption. A novel improved Firefly Algorithm-Back Propagation (IFA-BP) neural network model is proposed for enhanced PV power prediction accuracy. The methodology combines Circle chaos mapping for optimized population initialization with nonlinear mutational perturbation to strengthen global search capabilities. Critical input parameters are selected through grey correlation analysis to improve model efficiency. Comparative analysis with conventional BP and FA-BP models is conducted using historical operational data from 5G base station installations. Experimental results across diverse weather conditions demonstrate the model’s superior performance, achieving a Mean Absolute Percentage Error (MAPE) of 4.7943% and coefficient of determination (R2) of 0.9895 under optimal solar conditions. The system maintains robust prediction capability in challenging environments, yielding a MAPE of 12.1988% and R2 of 0.9793 during cloudy weather. These findings highlight the model’s effectiveness in stabilizing power supply systems for 5G infrastructure while optimizing renewable energy utilization.