Benefiting from the powerful discriminative feature learning capability of convolutional neural networks (CNNs), deep learning techniques have achieved remarkable performance improvement for the task of salient object detection (SOD) in recent years. However, most existing deep SOD models do not fully exploit informative contextual features, which often leads to suboptimal detection performance in the presence of a cluttered background. This paper presents a context-aware attention module that detects salient objects by simultaneously constructing connections between each image pixel and its local and global contextual pixels. Specifically, each pixel and its neighbors bidirectionally exchange semantic information by computing their correlation coefficients, and this process aggregates contextual attention features both locally and globally. In addition, an attention-guided hierarchical network architecture is designed to capture fine-grained spatial details by transmitting contextual information from deeper to shallower network layers in a top-down manner. Extensive experiments on six public SOD datasets show that our proposed model demonstrates superior SOD performance against most of the current state-of-the-art models under different evaluation metrics.
14 February 2022 in IEEE Open Access Journal of Power and Energy
As the penetration of renewable energy resources increases, their intrinsic uncertainty and randomness raise significant concerns about the grid’s reliability. With the development of information technologies, residential air-conditioners serve increasing services, including secondary frequency regulation, peak load shaving, and capacity reserve. However, these resources can hardly be allocated economically and reliably. On the one hand, it is unrealistic to provide advanced and expensive control modes for all AC units due to the limited investment cost. On the other hand, the uncertainty of residents’ behavior varying with control modes affects the reliability. Few previous works effectively make a trade-off between economics and reliability, which restricts the value of residential air-conditioners. Thus, this paper proposed an economical and reliable allocation method based on a real survey dataset and the online learning multi-armed bandits framework. This method explores the inherent relationship between the allocation plan’s economics and reliability. We also propose an allocation model which enables system operators to design an economic and reliable plan for each residential area considering ancillary services requirements. A case study in a modified IEEE 9 nodes system illustrates the effectiveness of the proposed method.
In this paper, a portable assistance system is designed to help the visually impaired to detect the traffic light. The designed system is realized on the basis of the AdaBoost algorithm, which is fast and robust in object detections. In order to accelerate the AdaBoost-based approach, a flexible parallel architecture is implemented on the field-programmable gate array (FPGA) platform. The architecture is designed utilizing the parallelism of computations in the AdaBoost-based detection. The computations of the window integral image are implemented in parallel, and the confidences of the weak classifiers are calculated in parallel. The parameters of the weak classifiers are trained by the AdaBoost algorithm with multi-layer features in the MATLAB software, and then are configured on the FPGA platform via the Vivado design suite before the detection process. The parallelism optimized architecture is implemented on an Artix-7 FPGA at 200 MHZ. Experiments show that it can detect the traffic light in videos with a rate of 30 frames per second (fps).
25 September 2024 in IEEE Transactions on Power Systems
Production cost minimization (PCM) simulation is commonly employed for assessing the operational efficiency, economic viability, and reliability, providing valuable insights for power system planning and operations. However, solving a PCM problem is time-consuming, consisting of numerous binary variables for simulation horizon extending over months and years. This hinders rapid assessment of modern energy systems with diverse planning requirements. Existing methods for accelerating PCM tend to sacrifice accuracy for speed. In this paper, we propose a stable relay learning optimization ( s-RLO ) approach within the Branch and Bound (B&B) algorithm. The proposed approach offers rapid and stable performance, and ensures optimal solutions. The two-stage s-RLO involves an imitation learning ( IL ) phase for accurate policy initialization and a reinforcement learning ( RL ) phase for time-efficient fine-tuning. When implemented on the popular SCIP solver, s-RLO returns the optimal solution up to 2× faster than the default relpscost rule and 1.4× faster than IL , or exhibits a smaller gap at the predefined time limit. The proposed approach shows stable performance, reducing fluctuations by approximately 50% compared with IL . The efficacy of the proposed s-RLO approach is supported by numerical results.
Compared with the medium voltage ac (MVAC) collection system, the medium voltage dc (MVDC) one for renewable energy sources has many advantages. High-power dc/dc converters are one of the key stages of the MVDC collection system to boost the voltage generated by photovoltaic or wind turbine. A novel hybrid three-level dc/dc converter utilizing a blocking capacitor to realize zero-voltage zero-current-switching (ZVZCS) is proposed. A higher overall efficiency can be achieved by reducing conduction losses. Detailed experimental results on a scaled-down hardware prototype rated at 150 V/750 V/1 kW are demonstrated to verify the proposed converter performance.