In this paper, a characteristic load decomposition (CLD)-based day-ahead load forecasting scheme is proposed for a mixed-use complex. The aggregated load of the complex is composed of the mixtures of different electricity usage patterns, and short-term load forecasting can be implemented by summing disaggregated sub-load predictions. However, tracing all usage patterns of sub-loads for prediction may be infeasible because of limited resources for measurement and analysis. To prevent this infeasibility, the proposed scheme focuses on effective decomposition using the sub-loads of typical characteristic load profiles and their representative pilot signals. Separate forecasts are obtained for the decomposed characteristic sub-loads using a hybrid scheme, which combines day-type conditioned linear prediction with long short-term memory regressions. Complex campus load data are considered to evaluate the proposed CLD-based hybrid forecasting. The evaluation results show that the proposed scheme outperforms conventional hybrid or similar-day-based forecasting approaches. Even when sub-load measurements are available only for a limited period, the CLD scheme can be applied for the extended training data through virtual disaggregations.
In this paper, we propose a deep reinforcement learning (DRL) based predictive control scheme for reducing the energy consumption and energy cost of pumping systems in wastewater treatment plants (WWTP), in which the pumps are operated in a binary mode, using on/off signals. As global energy consumption increases, the efficient operation of energy-intensive facilities has also become important. A WWTP in Busan, Republic of Korea is used as the target of this study. This WWTP is a large energy-consuming facility, and the pumping station accounts for a significant portion of the energy consumption of the WWTP. The framework of the proposed scheme consists of a deep neural network (DNN) model for forecasting wastewater inflow and a DRL agent for controlling the on/off signals of the pumping system, where proximal policy optimization (PPO) and deep Q-neural network (DQN) are employed as the DRL agents. To implement smart control with DRL, a reward function is designed to consider the energy consumption amount and electricity price information. In particular, new features and penalty factors for pump switching, which are essential for preventing pump wear, are also considered. The performance of our designed DRL agents is compared with those of WWTP experts and conventional approaches such as scheduling method and model predictive control (MPC), in which integer linear programming (ILP) optimization is employed. Results show that the designed agents outperform the other approaches in terms of compliance with operating rules and reducing energy costs.
This study proposes a novel authentication scheme that provides enhanced security for Internet of Things (IoT) applications by integrating a radio frequency (RF) physical unclonable function (PUF) with a device PUF. While traditional PUF-based authentication methods are lightweight, they are vulnerable to replay attacks because challenge-response pairs (CRPs) are exposed to adversaries over wireless channels. To prevent CRP exposure, we propose a joint scheme that integrates the physical layer features of wireless channels with those of the device PUF. Our authentication process consists of two stages: an enrollment stage and an authentication stage. During the enrollment stage, the physical features of the static random-access memory (SRAM) in the IoT devices are shared with the server, which generates a hashing model using the amplitudes of channel state information (CSI) as the RF-PUF and transfers it to the IoT device. In the authentication stage, the server and IoT devices exchange pilot signals to estimate the shared CSI. Both parties then generate the challenge information using the hashing model, which depends on the CSI. The challenge information is mapped to the response of the device PUF as the cryptographic key. If both parties are legitimate, the proposed model is updated through CSI amplitudes. We evaluated the proposed authentication protocol using a testbed based on Raspberry Pi and acquired a CSI dataset and SRAM-PUF using universal software radio peripheral and Arduino, respectively. Numerical results demonstrate that our method effectively defends against diverse attacks, even in critical CRP exposure scenarios.
Generally, photovoltaic (PV) fault detection approaches can be divided into two groups: end-to-end and threshold methods. The end-to-end method typically uses a deep neural network (DNN) to learn fault patterns from labeled datasets, which directly detect whether faults occur or not. The threshold method first estimates power generation and uses thresholds to detect atypical deviations of measured values from estimated ones. The former method heavily relies on fault-labeled data and, therefore, requires the collection of abnormal event records, which is usually difficult, due to the sparseness of these events. The latter method typically uses statistical approaches, such as 3-sigma, to find thresholds, and it can be practically utilized without fault labels. However, setting a threshold with a proper confidence interval is still challenging, as PV power generation is sensitive to variations in environmental conditions, such as irradiance, ambient temperature, wind speed and humidity. In this paper, we propose a novel deep reinforcement learning (DRL)-based label-free fault detection scheme in which thresholds are dynamically assigned with suitable confidence intervals under varying environmental conditions. Various weather properties were used as input features (i.e., states) to a DRL agent, and proper thresholds were estimated in real time from the actions of the DRL agent. To this end, a reward function was designed for learning proper thresholds without fault labels under different weather conditions. To evaluate the performance of the proposed scheme, the PV dataset of the National Institute of Standards and Technology (NIST) was used, as it includes paired records of local weather and PV generations. The DRL-based scheme was compared with static and conventional dynamic threshold methods, based on statistical approaches. The results revealed that the proposed scheme outperformed the existing methods, providing a 5.67% higher -score in the NIST dataset.
In this study, flexible vehicle-to-grid (V2G) coordination schemes are proposed for office buildings equipped with electric vehicle (EV) charging stations. EVs can be connected to electric grid during the rest hours and have a potential to provide V2G or vehicle-to-building (V2B) services such as electric load distributions and demand responses. Especially for smart buildings, the charging stations can be operated efficiently by integrating distributed energy resources (DER) such as Photovoltaic (PV) and battery energy storage systems (BESS). In this scenario, the charging coordination problem can be simplified as the integer linear programming (ILP), by assuming constant-rate charging with known schedules of visiting EVs at the station. Numerical evaluations are conducted with the building's expected daily electricity load, PV generation, and electricity price datasets. Public service buildings are investigated for flexible V2G supports, based on the actual vehicles' in and out patterns of the local office. The proposed coordination scheme shows potential gain of 14.3% of energy cost reduction compared to the first-come first-serve approach, under medium sized smart building scenario integrated with the PV and BESS.
With the development of the internet of things (IoT), the power grid has become intelligent using massive IoT sensors, such as smart meters. Generally, installed smart meters can collect large amounts of data to improve grid visibility and situational awareness. However, the limited storage and communication capacities can restrain their infrastructure in the IoT environment. To alleviate these problems, efficient and various compression techniques are required. Deep learning-based compression techniques such as auto-encoders (AEs) have recently been deployed for this purpose. However, the compression performance of the existing models can be limited when the spectral properties of high-frequency sampled power data are widely varying over time. This paper proposes an AE compression model, based on a frequency selection method, which improves the reconstruction quality while maintaining the compression ratio (CR). For efficient data compression, the proposed method selectively applies customized compression models, depending on the spectral properties of the corresponding time windows. The framework of the proposed method involves two primary steps: (i) division of the power data into a series of time windows with specified spectral properties (high-frequency, medium-frequency, and low-frequency dominance) and (ii) separate training and selective application of the AE models, which prepares them for the power data compression that best suits the characteristics of each frequency. In simulations on the Dutch residential energy dataset, the frequency-selective AE model shows significantly higher reconstruction performance than the existing model with the same CR. In addition, the proposed model reduces the computational complexity involved in the analysis of the learning process.