Energy Systems: Technology and Management

Abstract Nowadays, global optimization algorithms are widely used for solving engineering problems. These algorithms are complex and characterized by various properties, for instance, they can be non-linear, non-separable, symmetric, asymmetrical, smooth with narrow ridges, unimodal, and multimodal, and can involve non-differentiable functions and high dimensionality. These properties create challenges for existing optimization algorithms, and the global optimum finding, is one of the long-standing goals in this area of study. In this paper, an efficient hybrid metaheuristic search algorithm based on swarm intelligent has been developed. In order to evaluate the performance of the presented algorithm, it has been implemented on 13 benchmark functions and two real engineering problems in the fields of power engineering. The proposed functions are among the constrained and unconstrained optimization problems. Nevertheless, the proposed method has optimized the desired problems well and has performed better than GA, PSO and CBSA methods, which are among the well-known optimization methods.

Abstract Utilization of electric vehicles (EVs) is rapidly increasing in power systems. Due to the impact of EVs on the system performance, from the distribution network's point of view, it is vital to optimally site and size electric vehicle charging stations (EVCSs). This paper establishes an optimization framework for optimal allocation of EVCSs in distribution network. For this goal, development cost of EVCSs and network losses are included in the objective function and the planning problem is solved considering load flow calculations and the constraints. To effectively solve the problem, crow search algorithm (CSA) and differential CSA (CSAdif) are applied and the results are compared. Furthermore, impact of the load uncertainty is investigated on the planning problem. Over the IEEE-33 bus distribution network with photovoltaics (PVs) and wind turbines (WTs) penetration, it can be drawn that it is beneficial to install EVCSs near renewable resources to simply charge batteries. On the other hand, by consideration of the load uncertainty, it is preferred to install more connectors in the stations. Furthermore, compared to CSA, CSAdif finds more promising results in terms of accuracy.

Abstract This paper proposes an Enhanced Whale Optimizer (EWO), a variant of the recently introduced Whale Optimizer (WO), motivated by the cooperative and foraging behaviors of groups of Whales. Although the original WO achieves a good exploration–exploitation balance, it often gets trapped in premature convergence and reduces population diversity at later search stages. To overcome these limitations, two adaptive strategies are combined: Opposition-Based Learning (OBL) and the Golden-Sine Strategy (GSS). The OBL mechanism ensures diversity of the initial population and global exploration, and assures a more comprehensive exploration of the search domain. The GSS, conversely, adaptively balances exploration and exploitation, resulting in improved convergence and enhanced local optimization capabilities. Performance of the proposed EWO is also confirmed with the CEC-2019 benchmark suite and several difficult real-world engineering problems, such as economic load dispatch, microgrid energy management, and Traveling Salesman Problem (TSP). Comparative analyses with well-known algorithms such as Particle Swarm Optimization (PSO), Dwarf Mongoose Optimization Algorithm (DMO), and White Shark Optimizer (WO) confirm EWO's better accuracy, convergence rate, and stability. Statistical tests and boxplot visualization also confirm its stability and consistency. It is evidenced that EWO has smaller mean errors, lower standard deviations, and higher reliability across multiple problem areas, citing its strength as a strong and general-purpose optimization algorithm suitable for both benchmark and engineering problems.

Abstract The growing demand for reduced energy consumption in industrial and commercial buildings has positioned energy optimization as one of the central challenges in the building industry. This study examines the design and implementation of intelligent systems aimed at optimizing energy use in industrial and commercial facilities through the integration of the Internet of Things (IoT) and machine learning techniques. The primary objective is to employ advanced technologies to decrease overall energy consumption, enhance operational efficiency, and maintain favorable environmental conditions within these buildings. The paper first introduces smart building systems and the role of IoT in energy management, followed by an explanation of how machine learning algorithms can be utilized for energy prediction and optimization. It then outlines the design of IoT-based architectures for monitoring and controlling energy consumption, incorporating a network of sensors and smart devices. The findings indicate that leveraging these technologies can result in a 25–30 percent reduction in energy usage. Furthermore, the proposed systems demonstrate effective control over temperature, humidity, and lighting, ultimately lowering energy costs and improving occupant comfort.

Abstract Increasing competition among electricity market players (mainly generation companies [GenCos]) emphasizes the significance of well-maintained operational planning, risk management, and decision-making. Electricity price forecasting plays a critical role in this respect. However, forecasting electricity prices is a complex and challenging task because price signals are nonlinear, non-stationary, and time-varying. Price fluctuations in power markets are both periodic and random, with price spikes being examples of events that severely affect forecasting performance. Although many forecasting methods have been proposed, few forecasting methods focus specifically on price spike information. There are two major challenges in price spike forecasting: predicting when a spike will occur and how large it will be. A typical assumption is that neural network (NN) information about expected future data points is equally valuable, or even more valuable. Prediction is mainly treated using an online NN approach. This paper proposes a novel approach for forecasting both the occurrence and value of electricity price spikes. The developed method is based on an improved neural network-based algorithm that calculates the upper and lower bounds of the price signal. Compared to traditional techniques, the proposed approach predicts both the price spike class and intensity in a single step. The method was evaluated using real data from several electricity markets. The obtained results indicate the accuracy and efficiency of the proposed method. The proposed method achieves up to 98% spike occurrence prediction accuracy (PJM 2006, $150/MWh threshold) and 96.9% (PJM 2008), correctly identifying most price spikes in real market data.

Abstract Nowadays, renewable energy resources like wind energy generation are the most important and critical issues for fossil fuel replacement and environmental rescue. Although recent wind power generation has developed rapidly, the planning of these generations is still difficult because the accurate forecasting of wind speed is challenging due to its volatile and uncertain nature, particularly in coastal areas. Therefore, this study proposes a real-time wind speed forecasting model based on a hybrid convolutional neural network architecture, where multiple meteorological variables are filtered through a cross-screening strategy for multivariate time series forecasting. To screen the parameters for training, an evaluation was conducted based on multi-feature wind speed datasets. The performances of the proposed and baseline models in terms of real-time wind speed forecasting accuracy and volatility were evaluated based on a comparative study of error metrics. To benchmark the results, the proposed model was compared with other state-of-the-art paradigms based on error metrics. The results indicate that the proposed CNN-LSTM model outperformed all baseline models in terms of training efficiency and delivered values with higher accuracy than the other models. Based on different test datasets, the CNN-LSTM model with the same hyperparameter configurations predicted more accurately than other baseline models under the collected datasets. For 10-meter datasets the RMSE index obtained as 0.25, 0.42 and 0.25. Also, for 50-meter datasets the RMSE index obtained as 0.26, 0.51 and 0.25. These results prove the proposed model accuracy.