Minimizing charging cost of plug-in electric vehicle based on ensemble machine learning techniques and optimal control / Ali Najm Abdulnabi Alkawaz

Ali Najm , Abdulnabi Alkawaz (2024) Minimizing charging cost of plug-in electric vehicle based on ensemble machine learning techniques and optimal control / Ali Najm Abdulnabi Alkawaz. PhD thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (213Kb)
	PDF (Thesis PhD) Download (1924Kb)

Abstract

Plug-in Electric Vehicles (PEVs) offer an environmentally friendly alternative to conventional internal combustion engine vehicles, promising reduced greenhouse gas emissions and conservation of oil reserves. However, their increasing integration into the electric grid raises concerns about stability and reliability owing to heightened charging demands. Random or uncoordinated charging within the distribution network can exacerbate these challenges, leading to increased charging costs for PEV owners and potential strain on the grid. In the present study, a smart and coordinated charging approach with centralized control is proposed, aimed at minimizing charging costs and supporting grid stability. This is achieved using the optimal control (OC) technique, considering fixed fluctuations in electricity prices and various driving patterns. Empirical results indicate that, compared to a random charging plan, smart and coordinated charging strategies can reduce charging costs by up to 73% and 21%, respectively. A new hybrid machine learning (ML) model was proposed for electricity price forecasting (EPF) to further optimize charging costs. It integrates the linear automatic relevance determination (ARD) model, addressing trend and seasonality, with the ensemble bagging extra tree regression (ETR) model, capturing interactions. Validated with a Nord Pool market dataset, this approach surpassed other hybrid models, achieving reductions in testing mean absolute error and root mean square error values by 32.1% and 21.5%, respectively. Random PEV charging activities are well-recognized for their potential impacts on both vehicle owners and electric networks, resulting in elevated charging costs and degraded distribution system performance such as power loss, and voltage deviation. To address these challenges, scheduling of PEV charging activities is essential. Coordinated and smart charging strategies incorporated with EPF using ML and OC methods have been devised. These strategies not only lead to cost savings for PEV owners but also benefit electric utilities. To guide PEV charging decisions based on price forecasts, three ML classifiers were employed: neural network, naïve Bayes, and an ensemble approach. Empirical results show that the ensemble ML classifier outperforms its counterparts in various charging strategies. Remarkably, the proposed ensemble smart charging strategy recorded a PEV charging cost of £15, which was significantly lower than the £280 incurred using the random charging strategy. Moreover, compared to the base random charging case, the proposed ensemble-based smart and coordinated approaches reduced the charging cost by approximately 94% and 40%, respectively. To illustrate the impacts of the proposed coordinated and smart charging techniques compared to random charging at various levels of PEV penetration levels (16%, 28%, and 41%), both modified and standard IEEE 69-bus radial distribution systems were employed, smart and coordinated PEV charging showcased better performance in terms of power consumption, voltage drop, and system loss than those of random charging. Overall, this study indicates that an ensemble-based coordinated and smart PEV charging strategy is a promising approach for efficiently managing electricity usage. As PEV adoption increases, smart charging technologies are likely to become more widespread and help drive the transition to a more sustainable energy future.

Actions (For repository staff only : Login required)