論文

There is a great demand for high-quality indoor thermal environments while achieving comfort and saving the energy for heating, ventilation, and air conditioning (HVAC) devices offer a tradeoff. The room shapes, partitioning, and the location of diffusers cause uneven comfort and degrade thermal satisfaction. Therefore, an HVAC control method that provides uniform thermal comfort while saving energy is required. In such cases, providing uniform comfort in a room can be a low-cost and energy-efficient solution. We propose a deep reinforcement learning (DRL)-based HVAC control method to uniformize comfort in a room. The proposed method considers the use of sensors, which are already deployed in existing HVAC devices and is adaptive to rooms with different conditions. The proposed method controls the HVAC device based on the prediction of the room's thermal response using wall heatmaps of the target room. The heatmaps are obtained by using a simple measuring system of an infrared thermopile array sensor. For evaluating the proposed method's adaptivity to rooms with different conditions, a computer flud dynamics (CFD)-based simulation was designed. The coldest period throughout a year was chosen as the evaluation period; uniformizing comfort is the most difficult when the temperature difference between outside and inside a room and the validity of the proposed method for the other periods is implied. The usefulness of the proposed method for the chosen period was shown for the selected period; the proposed method achieved an improvement of 33.9% in thermal comfort performance while on average increasing energy consumption by 6.3%, against that of comparable approaches. Moreover, the comfort performance during learning is confirmed to be similar to those of the existing method.

RHiNET-2/SW is a network switch for the RHiNET-2 parallel computing system. RHiNET-2/SW enables high-speed and long-distance data transmission between PC nodes for parallel computing. In RHiNET-2/SW, a one-chip CMOS switch-LSI and eight pairs of 800-Mbit/s × 12-channel parallel optical interconnection modules are mounted into a single compact board. This switch allows high-speed 8-Gbit/s/port parallel optical data transmission over a distance of up to 100 m, and the aggregate throughput is 64 Gbit/s/board. The CMOS-ASIC switching LSI enables high-throughput (64 Gbit/s) packet switching with a single chip. The parallel optical interconnection modules enable high-speed and low-latency data transmission over a long distance. The structure and layout of the printed circuit board is optimized for high-speed, high-density device implementation to overcome electrical problems such as signal propagation-loss and crosstalk. All of the electrical interfaces are composed of high-speed CMOS-LVDS logic (800 Mbit/s/pin). We evaluated the realibility of the optical I/O port through long-term data transmission. No errors were detected during 50 hours of continuous data transmission at a data of 800 Mbit/s × 10 bits (BER: < 2.44×10-14). This test result shows that RHiNET-2/SW can provide high-throughput, long-transmission-length, and highly reliable data transmission in a practical parallel computing system.

Large many-core parallel applications become sensitive to communication latencies, suggesting the need for low-latency networks in high-performance computing systems. Switch delay dominates network latencies. To reduce the network latencies, we exploit routing cache on a switch. Routing decision based on off-chip CAM (Content Addressable Memory)-based table lookup imposes a significant delay, however, using on-chip small routing cache can bypass it when it hits. Our simulation results show that the only 256-entry routing cache hits 98% on over 4k-host systems with the matrix-transpose traffic, and the 1024-entry routing cache improves not only up to 16% of packet latency but also up to 18% of network throughput.

RHiNET-2/SW is a network switch that enables high-performance optical network-based parallel computing system in a distributed environment. The switch used in such a computing system must provide high-speed, low-latency packet switching with high reliability. Our switch allows high-speed 8-Gb/s/port optical data transmission over a distance of up to 100 m, and the aggregate throughput is 64 Gb/s. In RHiNET-2/SW, eight pairs of 800 -Mb/s × 12-channel optical interconnection modules and a one-chip CMOS ASIC switch LSI (a 784-pin BGA package) are mounted on a single compact board. To enable high-performance parallel computing, this switch must provide high-speed, highly reliable node-to-node data transmission. To evaluate the reliability of the switch, we measured the bit error rate (BER) and skew between the data channels. The BER of the signal transmission through one I/O port was better than 10-11 at a data rate of 800 Mb/s × 10 b with a large timing-budget margin (870 ps) and skew of less than 140 ps. This shows that RHiNET-2/SW can provide high-throughput, highly reliable optical data transmission between the nodes of a network-based parallel computing system.

© 2000 IEEE. The Real World Computing Partnership (RWCP) has developed a local area system network (LASN) called RHiNET-1 (RWCP High-performance NETwork, version 1) using 1.33-Gbps optical interconnections for high-performance computing using personal computers distributed in an office or laboratory environment. The network interface, RHiNET-1/NI, uses a complex programmable logic device (CPLD) based protocol controller to provide an easy evaluation platform for various protocols. It fits in a 32-bit/33-MHz PCI bus. The switch, RHiNET-1/SW, consists of a single-chip CMOS switch and external SRAM. It provides low-latency, reliable communication with a flexible topology design. We are currently evaluating protocols on RHiNET-1. RHiNET-1 will enable a new form of high-performance computing environment. We are also developing the second implementation, RHiNET-2. RHiNET-2/NI will support a 64-bit/66-MHz PCI bus. RHiNET-2/SW is an 8-Gbps/port 8x8 single-chip ASIC switch. The aggregate bandwidth of RHiNET-2/SW is 64 Gbps.

Wireless Sensor Nnetworks (WSNs) are becoming increasingly widespread owing to the introduction of the Internet of Things (IoT) and the networks' comparatively low cost and usability. WSNs are usually battery-powered, owing to their low power consumption, which however limits their computational performance. The lifetime of a WSN strongly depends on its communication-related power consumption. Time Slotted Channel Hopping (TSCH) solves the problem of power consumption by featuring both time-division and frequency- division multiplexing in the system's Media Access Control (MAC) layer. In this paper, Pruning-based Coloring Scheduling (PRCOS) is proposed, which is an algorithm for TSCH scheduling that accounts for the power consumption to extend the system's lifetime. Using mathematical analysis, we show that the proposed algorithm decreases the transfer time and energy consumption by factors of 0.43 and 0.57, respectively, compared with existing algorithms.

This paper introduces the time synchronization approaches to the Institute of Electrical and Electronics Engineers (IEEE) P1451.0 standard-based sensor networks for Internet of Things (IoT) applications. A time synchronization architecture of IEEE P1451.0 standard-based sensor networks is described including two-level time synchronization systems in IEEE P1451.0 and P1451.1.X standards-based wide-area network (WAN) and IEEE P1451.0 and P1451.5.X standards-based local area networks (LANs). However, this paper mainly focuses on the time synchronization approach of IEEE P1451.0 and P1451.1.6 standards-based WANs and provides two implementations of time synchronization of IEEE P1451.0 and P1451.1.6 using wireline and wireless networks with their preliminary results to verify that the time synchronization approach of IEEE P1451.1.6 functions properly. In addition, the time synchronization transducer electronic data sheets (TEDS) of P1451.1.6 is described.

The recent emergence of smartphones, cloud computing, and the Internet of Things has brought about the explosion of data creation. By collating and merging these enormous data with other information, services that use information become more sophisticated and advanced. However, at the same time, the consideration of privacy violations caused by such merging is indispensable. Various anonymization methods have been proposed to preserve privacy. The conventional perturbation-based anonymization method of location data adds comparatively larger noise, and the larger noise makes it difficult to utilize the data effectively for secondary use. In this research, to solve these problems, we first clarified the definition of privacy preservation and then propose TMfc-anonymity according to the definition.

Video traffic on the Internet has been increasing rapidly and accounts for a large percentage of the total traffic. To process the increasing number of videos, edge computing is preferable for load balancing and bandwidth reduction. However, edge areas have less computational resources than cloud areas, and high-performance GPUs for processing videos at high speed are not always present. Therefore, a memory-saving and high-Throughput video analysis method is necessary for analyzing videos in edge areas. In this paper, a video object detection method using single-frame detection and motion vector tracking is proposed. This method is classified as a pixel and compressed domain analysis method and is realized by compensating motion using the motion vectors that already exist in the compressed domain. This method is divided into two processes: CNN-based object detection and motion vector-based object detection. In addition, a network-Transparent platform for video reconstruction in edge areas is constructed. The network-Transparent service can be installed without modifying the existing end-device network settings, network configuration, and routing. The platform enables video object detection services to be added on without modification of these settings.

The secondary use of data in consideration of individual privacy has been gaining attention. Data anonymization needs to be applied to enable both secondary use and privacy protection. In distributing anonymized data, there is a concern that data users might re-distribute the data to a third party without permission. To tackle this issue, a watermarking technique that correlates data users (IDs) with anonymized data has been studied. In existing studies, a non-blind detection method has been employed to obtain the IDs; however, the problem is the use of original data in the detection, which deteriorates the applicability. In this paper, a watermarking technique in combination with a blind detection method is proposed. We evaluate and compare the proposed watermarking method with an existing non-blind method in terms of information loss, tolerance against attacks, and the processing time of watermarking and detection.

The development of smart communities has provided additional sources of data for innovative new services. In residential areas, Home Energy Management Systems (HEMS) provide a data stream from the home, including room temperature and the status of heating, ventilation, and air conditioning (HVAC) appliances. The data is of interests to both residents and utilities for energy saving purposes. The Green New Deal is a scheme that subsides building improvement to increase energy efficiency by insulating buildings and introducing energy-saving electric appliances. However, the scheme is often criticized due to high installation costs. Therefore, a simple and low cost method to assess building thermal performance is important to select those homes, that can benefit the most from the scheme. In this study, a novel method for assessing building thermal performance based on HEMS data is proposed. The thermal equilibrium of a house is modeled to state the system identification problem to estimate the unknown parameters, such as heat transfer coefficient and heat capacity of a house. The unknown parameters are estimated by using Kalman Filter. In this way, the thermal performance of building can be assessed remotely while residents live their daily lives. Estimated parameters are then validated by considering those elements affected by the behavior of residents. The proposed method was found to estimate the heat transfer coefficient well and thus aid in effectively targeting Green New Deal projects.

We propose a technique to reduce compulsory misses of packet processing cache (PPC), which largely affects both throughput and energy of core routers. Rather than prefetching data, our technique called response prediction cache (RPC) speculatively stores predicted data into PPC without additional access to the low-throughput and power-consuming memory (i.e., TCAM). RPC predicts the data related to a response flow at the arrival of the corresponding request flow, based on the request-response model of internet communications. RPC can improve the cache miss rate, throughput, and energy-efficiency of PPC systems by 15.3%, 17.9%, and 17.8%, respectively.

As various data services are provided to realize Society 5.0, the usage of personal data is estimated to increase along with the explosive increase in data traffic. It is important that the protection of privacy keeps pace with increases in the exchange of data containing personal information. Starting from the enforcement of the General Data Protection Regulation (GDPR), stricter privacy protection regulations are expanding to more countries. These restrictions require that personal data should be hidden or anonymized before they are propagated over the network. The secondary usage of data is assumed in smart communities, and systems that can protect privacy are required for secure network infrastructures. In this study, we propose an edge-based computing platform that manages the privacy of users on the network of a smart community. For the platform, we prepared three models: The Basic, Preceding Packet, and Piggyback models. These are considered OpenFlow models, and the network efficiency for each was evaluated.

A smart factory or Industry 4.0 is creating an epoch for manufacturing and its production lines. It reduces the total cost by monitoring and predicting the expected faults of factory lines and products. One of the essential challenges is to develop a technology to detect and predict abnormalities at an early stage without human resources. For this reason, the automation of anomaly detection is now attracting attention. Many statistical and machine-learning methods have been studied for anomaly detection. In this study, we focus on a refrigeration system for large storage, where the failures of the system will cause enormous losses. Moreover, this type of system was independently designed according to the environment, location, and storage items. Under this condition, it is difficult to train discriminative models for anomaly detection using training data that include failure data. In addition, it is indispensable to provide a basis for determining whether the system is abnormal to achieve future treatments. Therefore, deep generative models are used to achieve unsupervised abnormality detection. Because the sensing system's cost for detecting system failures should be reduced, the proposed system uses low-cost microphone arrays to monitor sounds and source locations. The system also provides a rationale by visualizing and mentioning irregular sounds. Furthermore, this study compared various deep generative models in terms of accuracy and showed that the Efficient GAN-based method achieved the highest accuracy.

Energy optimization in the built environment is receiving more attention in the last decade. This leads to remarkable technological advancements in energy monitoring, data communication, energy storage, control, and data analysis applications. Energy monitoring can be aggregate (Non-Intrusive load monitoring -NILM) or disaggregate (Intrusive load monitoring -ILM) based on the location in which the monitoring takes place. Sensors communicate via wired or wireless means, and several protocols are developed to support communications in smart homes. This real-time data can be sent to remote controllers for further analysis, training and prediction applications using Internet of Things (IoT) that provides a cost-effective solution for energy management systems. This manuscript presents an overview of current innovation in smart homes for the advancement of building technology.

The growing demand for computing resources leads to a serious problem of excessive energy consumption in data centers. In recent studies, energy consumption of both computing and cooling equipment is drawing attention. For improving the energy efficiency of cooling equipment such as computer room air conditioners (CRACs), it is neccesary to predict temperatures in data centers and to optimize thermal management in data centers. In this study, we propose a temperature prediction method for servers in a data center using a neural network. We used the prediction result for distributing task targeting temperature-based load balancing. First, we conducted an experiment in a real data center to evaluate the prediction accuracy of the proposed method. We then simulated task distribution based on the predicted temperatures and compared the maximum CPU temperature with a non-predictive approach. The results indicated that the proposed method can reduce future CPU temperatures successfully compared to the non-predictive approach, though in exchange for high computational cost.

This study aims to execute error correction service of sensor data at fog and edge, which are the computing environments for distributing the concentrated processes in a cloud by using hierarchical computing resources. Service integrity of error correction was evaluated as an essential application of fog and edge computing environment, which can use the locality of fog and edge. Errors or outliers in data may cause a serious problem in system controlling and visualization application for behavior change when by using the malformed data or visualization of the data. The proposed methods can solve the data congestion problems at a cloud in the future IoT era.

Human Activity Recognition (HAR) is a well-studied scientific area that has gained much traction with the rise of Internet of Things (IoT). Despite the interest in HAR for a wide spectrum of domains (technological, medical, etc.) only a few works exist, which study the variability in IoT data. To correctly perceive this variability, it is essential to dynamically model the evolving context of daily-life activities. Additionally, it is required to reduce the calculation cost of HAR, which is crucial for security and real-time applications. For the purpose of dynamically modeling, three context-aware approaches are formalized along with a context-free baseline. This study demonstrates improvements in terms of both of accuracy and calculation cost by considering variability in IoT data; our experimental study on real datasets reduced calculation cost by 20% while increasing accuracy by 20%.