A HYBRID APPROACH FOR EV CHARGING LOAD FORECASTING USING SIGNAL NOISE REDUCTION AND TIME SERIES ANALYSIS

Abstract

Vehicular Ad-Hoc Networks (VANETs) are a key component of intelligent transportation systems, enabling vehicles to communicate for safer and efficient traffic management. In India, rapid urbanization and increasing vehicle density have led to congestion and suboptimal route selection. According to recent reports, Indian cities witness over 30% higher average travel times due to traffic congestion. VANETs can optimize vehicle routing, improve network communication, and enhance road safety by analyzing traffic density, vehicle speed, and network conditions. The objective of this research is to predict optimal vehicle routes and traffic load patterns using machine learning. It aims to automate decision-making in vehicular networks for enhanced routing efficiency and network reliability. Traditionally, traffic and route management rely on manual monitoring using traffic signals, CCTV cameras, and periodic reports from traffic personnel. Vehicles follow pre-defined maps or human instructions to select routes, and charging stations depend on manual scheduling for load management. Manual systems are time-consuming, error-prone, and unable to adapt to real-time traffic fluctuations. They cannot predict congestion, optimize vehicle spacing, or balance charging loads dynamically. Human intervention limits scalability and often results in inefficient route selection and increased travel time. This research aims to overcome the limitations of manual systems by leveraging machine learning for real-time route and load optimization. Improvements include predictive accuracy, automated decisionmaking, adaptive traffic management, and efficient load distribution in vehicular networks, enabling scalable and reliable traffic solutions. The proposed system uses machine learning models for classification and regression. GP Classifier predicts whether a route or charging condition is optimal using Gaussian Process-based probabilistic modeling. KNN Classifier classifies load or route patterns based on similarities to historical data. FusionMind Classifier, a hybrid MLP+Random Forest model, captures complex patterns for accurate classification. For regression, SGD, KNN, and FusionMind Regressors forecast numerical values like vehicle spacing or charging load. These models enable predictive, adaptive, and automated traffic and charging management, reducing congestion and improving system reliability.