Rear-end collisions often cause serious traffic accidents. Conventionally, in intelligent transpo... more Rear-end collisions often cause serious traffic accidents. Conventionally, in intelligent transportation systems (ITS), radar collision warning methods are highly accurate in determining the inter-vehicle distance via detecting the rear-end of a vehicle; however, in poor weather conditions such as fog, rain, or snow, the accuracy is significantly affected. In recent years, the advent of Vehicle to Vehicle (V2V) and Vehicle to Infrastructure (V2I) communication systems has introduced new methods for solving the rear-end collision problem. Nevertheless, there is still much left for improvement. For instance, weather conditions have an impact on human-related factors such as response time. To address the issue of collision detection under low visibility conditions, we propose a Visibility-based Collision Warning System (ViCoWS) design that includes four models for prediction horizon estimation, velocity prediction, headway distance prediction, and rear-end collision warning. Based on the history of velocity data, future velocity volumes are predicted. Then, the prediction horizon (number of future time slots to consider) is estimated corresponding to different weather conditions. ViCoWs can respond in real-time to weather conditions with correct collision avoidance warnings. Experiment results show that the mean absolute percentage error of our velocity prediction model is less than 11%. For non-congested traffic under heavy fog (very low visibility of 120 m), ViCoWS warns a driver by as much as 4.5 s prior to a possible future collision. If the fog is medium with a low visibility of 160 m, ViCoWs can give warnings by about 2.1 s prior to a possible future collision. In contrast, the Forward Collision Probability Index (FCPI) method gives warnings by only about 0.6 s before a future collision. For congested traffic under low visibility conditions, ViCoWS can warn a driver by about 1.9 s prior to a possible future collision. In this case, the FCPI method gives 1.2 s for the driver to react before collision.
In recent years, people are trying to make consumer electronics more powerful and have started to... more In recent years, people are trying to make consumer electronics more powerful and have started to embed chips in these products to increase intelligence. Therefore, there should be a powerful application program to control the consumer electronics. For the above reasons, a distributed real-time framework and development environment is proposed, which can be used to produce distributed real-time applications for easily controlling electronics distributed around the world. Our framework consists of three modules: Standalone System, Control Client, and Host Agent. Each standalone system can be controlled by a control client and the host agents act as bridges connecting the standalone systems. Compared to conventional object-oriented application frameworks, our environment is not only modeled in UML, but also network-based, completely written in Java and hence highly portable, remotely controllable, and able to produce customized graphical user-inter faces for applications. Applications developed using the environment show its feasibility as a useful development aid.
Landslides could cause huge damages to properties and severe loss of lives. Landslides can be det... more Landslides could cause huge damages to properties and severe loss of lives. Landslides can be detected by analyzing the environmental data collected by wireless sensor networks (WSNs). However, environmental data are usually complex and undergo rapid changes. Thus, if landslides can be predicted, people can leave the hazardous areas earlier. A good prediction mechanism is, thus, critical. Currently, a widely-used method is Artificial Neural Networks (ANNs), which give accurate predictions and exhibit high learning ability. Through training, the ANN weight coefficients can be made precise enough such that the network works in analogy to a human brain. However, when there is an imbalanced distribution of data, an ANN will not be able to learn the pattern of the minority class; that is, the class having very few data samples. As a result, the predictions could be inaccurate. To overcome this shortcoming of ANNs, this work proposes a model switching strategy that can choose between different predictors, according to environmental states. In addition, ANN-based error models have also been designed to predict future errors from prediction models and to compensate for these errors in the prediction phase. As a result, our proposed method can improve prediction performance, and the landslide prediction system can give warnings, on average, 44.2 min prior to the occurrence of a landslide.
Smart mobile devices demand higher performance of embedded systems. With the recent advances in V... more Smart mobile devices demand higher performance of embedded systems. With the recent advances in VLSI technology, multicore processors (e.g. ARM Coretex A-15 or Qualcomm SnapDragon, both are 4 cores at 2.5 GHz available in 2012) are getting popular in today's embedded system; multicore systems can be considered as the established trend in modern architectures. While these systems are more powerful than ever, application specific architectures, energy consumption, and their applications remain challenge. Silicon resources are becoming increasingly abundant, heterogeneous processing elements can be combined together with many cores on a chip by processor designers. Such architectures, while providing important improvements in system performance, also pose new research issues that need to be resolved. This ''Special Issue on Embedded Multicore Systems: Architecture, Performance and Application'' presents a collection of highquality papers from the embedded system research community. The 6 accepted papers were selected from a total of 16 submissions. Each accepted paper has gone through at least one revision to enhance the paper quality. The authors have done an excellent job of presenting their research materials. We are sure this issue will be very informative for all the readers who are engaged in modern embedded system research. The numerical linear algebra operations can be benefit from the hardware accelerators. A study of recent tiled decomposition algorithms for multicore architectures is presented in our first paper, entitled ''Scalable Matrix Decompositions with Multiple Cores on FPGAs''. The tiled algorithms for scalable QR and LU matrix decomposers are studied. The performance of the proposed approach implemented on commercially available FPGAs is also presented. The Network-on-Chip (NoC) has emerged as a scheme for onchip communication of multicore system. Processors, memory elements and other modules are placed throughout the network and connected with routers. Examining the routing algorithms, the second paper, ''Non-minimal, Turn-Model Based NoC Routing'', presents three non-minimal NoC routing algorithms. The issues of deadlock, livelock and backward compatibility are examined. The performance and cost of the proposed routing algorithms are analyzed and presented in this paper. The paper, entitled ''VBON: Towards Efficient On-Chip Networks via Hierarchical Virtual Bus'', proposes a novel architecture called VBON that incorporates shared buses into NoCs in a hierarchical way such that both unicast and multicast communication can be achieved at low cost. The point-to-point links of the NoC are used as bus transaction links. The extensions to NoC router is also minimal. Implementations and experiments show that VBON can achieve low latency while sustaining high throughput for unicast, as well as, multicast types of communications.
Page 1. RECDNFIGURABLE 5Y5TEM DE5IBN AND VERIFICATION R$> ■err-m-1 r -1 PaD-An... more Page 1. RECDNFIGURABLE 5Y5TEM DE5IBN AND VERIFICATION R$> ■err-m-1 r -1 PaD-Ann Hsiung i[fiM,f ^n MarcD D. BantambragicL Chun-Hsian Huang (d* CRC Press Taylor & Francis Croup Page 2. RECONFIGURABLE SYSTEM DESIGN AND VERIFICATION Page 3. ...
Unified Modeling Language (UML), an industry de-facto standard, has been used to analyze dynamica... more Unified Modeling Language (UML), an industry de-facto standard, has been used to analyze dynamically partially reconfigurable systems (DPRS) that can reconfigure their hardware functionalities on-demand at runtime. To make model-driven architecture (MDA) more realistic and applicable to the DPRS design in an industrial setting, a model-based verification and estimation (MOVE) framework is proposed in this work. By taking advantage of the inherent features of DPRS and considering real-time system requirements, a semiautomatic model translator converts the UML models of DPRS into timed automata models with transition urgency semantics for model checking. Furthermore, a UML-based hardware/software co-design platform (UCoP) is proposed to support the direct interaction between the UML models and the real hardware architecture. The two-phase verification process, including exhaustive functional verification and physical-aware performance estimation, is completely model-based, thus reducing system verification efforts. We used a dynamically partially reconfigurable network security system (DPRNSS) as a case study. The related experiments have demonstrated that the model checker in MOVE can alleviate the impact of the state-space-explosion problem. Compared to the synthesis-based estimation method having inaccuracies ranging from 43.4% to 18.4%, UCoP can provide accurate and efficient platform-specific verification and estimation through actual time measurements.
In model checking a system design against a set of properties, coverage estimation is frequently ... more In model checking a system design against a set of properties, coverage estimation is frequently used to measure the amount of system behavior being checked by the properties. A popular coverage estimation method is to mutate the system model and check if the mutation can be detected by the given properties. For each mutation and each property, a full model check is required by some state-of-the-art coverage estimation methods. With such repeated model checking, mutation-based coverage estimation becomes significantly time-consuming. To alleviate this problem, a partial model checking (PMC) technique is proposed to recheck only those system states that were affected by a mutation, thus unnecessary rechecking of a large portion of the system states is avoided and time is saved. The PMC method has been integrated into the State Graph Manipulators model checker. Applying the proposed method to several examples showed that PMC has a saving of 50% to 70% in the coverage estimation time, and a reduction of 90% in mode visits.
Federated learning is a multiple device collaboration setup designed to solve machine learning pr... more Federated learning is a multiple device collaboration setup designed to solve machine learning problems under framework for aggregation and knowledge transfer in distributed local data. This distributed model ensures the privacy of data at each local node. Owing to its relevance, there has been extensive research activities and outcomes in federated learning with expanded applicability to different areas by the research community. As such, there is a vast research archive made available by the community with research work and articles related to the various aspects of federated learning such as applications, challenges, privacy, functionalities, and design. With respect to the function and design of federated learning, client selection, aggregation, knowledge transfer, management of distributed data (Non-IID), Incentive of data and communication cost are of paramount importance. Any effective design of federated learning requires these aspects to be well considered.There are numerous survey articles found among the available literature that focus on its application and challenges, opportunities, data privacy and protection, as well as on federated learning on internet of things, federated learning on edge computing, etc.
2018 International Conference on Smart Green Technology in Electrical and Information Systems (ICSGTEIS), 2018
In metropolitan areas, common traffic issues include traffic congestion, traffic accidents, air p... more In metropolitan areas, common traffic issues include traffic congestion, traffic accidents, air pollution, and energy consumption occur. To resolve this issues, Intelligent Transportation Systems (ITS) have been evolved by many researchers. One of the important sub-systems in the development of ITS is a Traffic Management System (TMS) which attempts to reduce a traffic congestion. In fact, TMS itself relies on the estimation of traffic flow, therefore providing such an accurate traffic flow prediction is needed. For this reason, we aim to provide an accurate traffic flow prediction to facilitate this system. In this works, a Supervised Deep Learning Based Traffic Flow Prediction (SDLTFP) was proposed which is a type of fully-connected deep neural network (FC-DNN). Timely prediction is also a major issue in guaranteeing reliable traffic flow prediction. However, training a deep network could be time-consuming, and overfitting is might be happening, especially when feeding small data into the deep architecture. The network is learned perfectly during the training, but in testing with the new data, it could fail to generalize the model. We adopt the Batch Normalization (BN) and Dropout techniques to help the network training. SGD and momentum are carried out to update the weight. We then take advantage of open data as historical traffic data which are then used to predict future traffic flow with the proposed method and model above. Experiments show that the Mean Absolute Percentage Error (MAPE) for our traffic flow prediction is within 5 % using sample data and between 15% to 20% using out of the sample data. Training a deep network faster with BN and Dropout reduces the overfitting.
2019 IEEE International Conference on Consumer Electronics - Taiwan (ICCE-TW), 2019
Irregularities in the use of electrical energy could result in failure of power grids and blackou... more Irregularities in the use of electrical energy could result in failure of power grids and blackouts. Anomaly detection is required not only for ensuring grid safety, but also to prevent illegal hacking. Long-term data are recorded for such anomaly detection. However, due to the nonlinear characteristics of the time series data, correlation regression becomes difficult when using non-deep learning techniques. To address this issue, we use Long-Short Term Memory (LSTM) recurrency techniques. We experiment with two datasets from two different smart meters, which are sampled once every 30 seconds for one month. A total of 120,000 data samples were used for training and 40,000 data samples for testing. From the experiment results, the testing accuracy, True-Positive Rate, and False Positive Rate were 0.92, 0.81, and 0.50, respectively. Further, to demonstrate that the LSTM model can actually be designed at the network edge, we implemented the model and the trained weights on a Raspberry Pi platform. The inference time for each sample was 935 μs, which is short enough for realizing edge-based anomaly detection.
2019 IEEE International Conference on Consumer Electronics - Taiwan (ICCE-TW), 2019
Future intersection crossings for autonomous vehicles will not be controlled by traffic signals, ... more Future intersection crossings for autonomous vehicles will not be controlled by traffic signals, rather a controller will be used for communication among vehicles that need to cross an intersection. In this work, we propose an innovative management system called Deep Reinforcement Learning-based Autonomous Intersection Management (DRLAIM) system, which is the first system to use deep reinforcement learning. We train the system to learn a good intersection control policy by interacting with traffic environment through reinforcement learning. The brake-safe control model is used to ensure the safety of each autonomous vehicle while crossing. Experiment results show that after training using reinforcement learning, the throughput of intersection control model increased by 83%. In comparison with the Fast First Service (FFS) policy, the average waiting time of DRLAIM reduced by about 1.2% to 11.4%.
IECON 2016 - 42nd Annual Conference of the IEEE Industrial Electronics Society, 2016
The traditional centralized power system is gradually being replaced by smart grids. However, an ... more The traditional centralized power system is gradually being replaced by smart grids. However, an important design issue is how to perform accurate demand-response such that the power distribution management is effective. This includes two sub-problems, namely the accurate prediction of future electricity demand-response situations and the optimization of power distribution. In this work, we propose a novel Model Predictive Optimization (MPO) method for the advanced distribution management system in smart grids. Future electricity situations (surplus/deficit) are predicted using a customized Autoregressive Integrated Moving Average (ARIMA) model. Pairing between buyers and sellers of electricity are performed based on not only the current situation, but also considering future situations. As a result, trading pairs with overall near-optimal cost are found through concurrent and multiple instances of Particle Swarm Optimization (PSO), along with conflict resolution. Experimental results on 30 micro-grids show the error rate of the ARIMA prediction model to be less than 10%. The proposed MPO method saves totally 19.38% overall trading cost, if predictions are made for 4 future time slots.
IGARSS 2020 - 2020 IEEE International Geoscience and Remote Sensing Symposium, 2020
Landslides could cause huge threats to lives and cause property damages. In the landslide predict... more Landslides could cause huge threats to lives and cause property damages. In the landslide prediction system, environmental information can be collected through sensors to detect the possibility of landslide occurrences. However, the data collected may be lost due to sensor failures, external interferences or other environmental factors, which may affect the accuracy of landslide predictions. In order to solve the problem of missing data, we propose a data reconstruction method based on rainfall intensity and soil moisture, which reconstructs missing data based on temporal relationships. It is based on the data trend in the past period of time. A Long Short-Term Memory (LSTM) deep neural network is trained to predict the data value in missing time slots. We use the predicted data to compensate for the missing data so as the elevate the accuracy not only of data, but also landslide predictions. Our method is compared with other reconstruction methods. The proposed LSTM model exhibit a smaller RMSE than the Linear Extrapolation (LE) method. Even if 90% of random data is lost, the RMSE results for the data reconstruction by LE and LSTM are, respectively, 0.033 and 0.036 for rainfall data and 0.029 and 0.032 for soil moisture data.
Rear-end collisions often cause serious traffic accidents. Conventionally, in intelligent transpo... more Rear-end collisions often cause serious traffic accidents. Conventionally, in intelligent transportation systems (ITS), radar collision warning methods are highly accurate in determining the inter-vehicle distance via detecting the rear-end of a vehicle; however, in poor weather conditions such as fog, rain, or snow, the accuracy is significantly affected. In recent years, the advent of Vehicle to Vehicle (V2V) and Vehicle to Infrastructure (V2I) communication systems has introduced new methods for solving the rear-end collision problem. Nevertheless, there is still much left for improvement. For instance, weather conditions have an impact on human-related factors such as response time. To address the issue of collision detection under low visibility conditions, we propose a Visibility-based Collision Warning System (ViCoWS) design that includes four models for prediction horizon estimation, velocity prediction, headway distance prediction, and rear-end collision warning. Based on the history of velocity data, future velocity volumes are predicted. Then, the prediction horizon (number of future time slots to consider) is estimated corresponding to different weather conditions. ViCoWs can respond in real-time to weather conditions with correct collision avoidance warnings. Experiment results show that the mean absolute percentage error of our velocity prediction model is less than 11%. For non-congested traffic under heavy fog (very low visibility of 120 m), ViCoWS warns a driver by as much as 4.5 s prior to a possible future collision. If the fog is medium with a low visibility of 160 m, ViCoWs can give warnings by about 2.1 s prior to a possible future collision. In contrast, the Forward Collision Probability Index (FCPI) method gives warnings by only about 0.6 s before a future collision. For congested traffic under low visibility conditions, ViCoWS can warn a driver by about 1.9 s prior to a possible future collision. In this case, the FCPI method gives 1.2 s for the driver to react before collision.
In recent years, people are trying to make consumer electronics more powerful and have started to... more In recent years, people are trying to make consumer electronics more powerful and have started to embed chips in these products to increase intelligence. Therefore, there should be a powerful application program to control the consumer electronics. For the above reasons, a distributed real-time framework and development environment is proposed, which can be used to produce distributed real-time applications for easily controlling electronics distributed around the world. Our framework consists of three modules: Standalone System, Control Client, and Host Agent. Each standalone system can be controlled by a control client and the host agents act as bridges connecting the standalone systems. Compared to conventional object-oriented application frameworks, our environment is not only modeled in UML, but also network-based, completely written in Java and hence highly portable, remotely controllable, and able to produce customized graphical user-inter faces for applications. Applications developed using the environment show its feasibility as a useful development aid.
Landslides could cause huge damages to properties and severe loss of lives. Landslides can be det... more Landslides could cause huge damages to properties and severe loss of lives. Landslides can be detected by analyzing the environmental data collected by wireless sensor networks (WSNs). However, environmental data are usually complex and undergo rapid changes. Thus, if landslides can be predicted, people can leave the hazardous areas earlier. A good prediction mechanism is, thus, critical. Currently, a widely-used method is Artificial Neural Networks (ANNs), which give accurate predictions and exhibit high learning ability. Through training, the ANN weight coefficients can be made precise enough such that the network works in analogy to a human brain. However, when there is an imbalanced distribution of data, an ANN will not be able to learn the pattern of the minority class; that is, the class having very few data samples. As a result, the predictions could be inaccurate. To overcome this shortcoming of ANNs, this work proposes a model switching strategy that can choose between different predictors, according to environmental states. In addition, ANN-based error models have also been designed to predict future errors from prediction models and to compensate for these errors in the prediction phase. As a result, our proposed method can improve prediction performance, and the landslide prediction system can give warnings, on average, 44.2 min prior to the occurrence of a landslide.
Smart mobile devices demand higher performance of embedded systems. With the recent advances in V... more Smart mobile devices demand higher performance of embedded systems. With the recent advances in VLSI technology, multicore processors (e.g. ARM Coretex A-15 or Qualcomm SnapDragon, both are 4 cores at 2.5 GHz available in 2012) are getting popular in today's embedded system; multicore systems can be considered as the established trend in modern architectures. While these systems are more powerful than ever, application specific architectures, energy consumption, and their applications remain challenge. Silicon resources are becoming increasingly abundant, heterogeneous processing elements can be combined together with many cores on a chip by processor designers. Such architectures, while providing important improvements in system performance, also pose new research issues that need to be resolved. This ''Special Issue on Embedded Multicore Systems: Architecture, Performance and Application'' presents a collection of highquality papers from the embedded system research community. The 6 accepted papers were selected from a total of 16 submissions. Each accepted paper has gone through at least one revision to enhance the paper quality. The authors have done an excellent job of presenting their research materials. We are sure this issue will be very informative for all the readers who are engaged in modern embedded system research. The numerical linear algebra operations can be benefit from the hardware accelerators. A study of recent tiled decomposition algorithms for multicore architectures is presented in our first paper, entitled ''Scalable Matrix Decompositions with Multiple Cores on FPGAs''. The tiled algorithms for scalable QR and LU matrix decomposers are studied. The performance of the proposed approach implemented on commercially available FPGAs is also presented. The Network-on-Chip (NoC) has emerged as a scheme for onchip communication of multicore system. Processors, memory elements and other modules are placed throughout the network and connected with routers. Examining the routing algorithms, the second paper, ''Non-minimal, Turn-Model Based NoC Routing'', presents three non-minimal NoC routing algorithms. The issues of deadlock, livelock and backward compatibility are examined. The performance and cost of the proposed routing algorithms are analyzed and presented in this paper. The paper, entitled ''VBON: Towards Efficient On-Chip Networks via Hierarchical Virtual Bus'', proposes a novel architecture called VBON that incorporates shared buses into NoCs in a hierarchical way such that both unicast and multicast communication can be achieved at low cost. The point-to-point links of the NoC are used as bus transaction links. The extensions to NoC router is also minimal. Implementations and experiments show that VBON can achieve low latency while sustaining high throughput for unicast, as well as, multicast types of communications.
Page 1. RECDNFIGURABLE 5Y5TEM DE5IBN AND VERIFICATION R$> ■err-m-1 r -1 PaD-An... more Page 1. RECDNFIGURABLE 5Y5TEM DE5IBN AND VERIFICATION R$> ■err-m-1 r -1 PaD-Ann Hsiung i[fiM,f ^n MarcD D. BantambragicL Chun-Hsian Huang (d* CRC Press Taylor & Francis Croup Page 2. RECONFIGURABLE SYSTEM DESIGN AND VERIFICATION Page 3. ...
Unified Modeling Language (UML), an industry de-facto standard, has been used to analyze dynamica... more Unified Modeling Language (UML), an industry de-facto standard, has been used to analyze dynamically partially reconfigurable systems (DPRS) that can reconfigure their hardware functionalities on-demand at runtime. To make model-driven architecture (MDA) more realistic and applicable to the DPRS design in an industrial setting, a model-based verification and estimation (MOVE) framework is proposed in this work. By taking advantage of the inherent features of DPRS and considering real-time system requirements, a semiautomatic model translator converts the UML models of DPRS into timed automata models with transition urgency semantics for model checking. Furthermore, a UML-based hardware/software co-design platform (UCoP) is proposed to support the direct interaction between the UML models and the real hardware architecture. The two-phase verification process, including exhaustive functional verification and physical-aware performance estimation, is completely model-based, thus reducing system verification efforts. We used a dynamically partially reconfigurable network security system (DPRNSS) as a case study. The related experiments have demonstrated that the model checker in MOVE can alleviate the impact of the state-space-explosion problem. Compared to the synthesis-based estimation method having inaccuracies ranging from 43.4% to 18.4%, UCoP can provide accurate and efficient platform-specific verification and estimation through actual time measurements.
In model checking a system design against a set of properties, coverage estimation is frequently ... more In model checking a system design against a set of properties, coverage estimation is frequently used to measure the amount of system behavior being checked by the properties. A popular coverage estimation method is to mutate the system model and check if the mutation can be detected by the given properties. For each mutation and each property, a full model check is required by some state-of-the-art coverage estimation methods. With such repeated model checking, mutation-based coverage estimation becomes significantly time-consuming. To alleviate this problem, a partial model checking (PMC) technique is proposed to recheck only those system states that were affected by a mutation, thus unnecessary rechecking of a large portion of the system states is avoided and time is saved. The PMC method has been integrated into the State Graph Manipulators model checker. Applying the proposed method to several examples showed that PMC has a saving of 50% to 70% in the coverage estimation time, and a reduction of 90% in mode visits.
Federated learning is a multiple device collaboration setup designed to solve machine learning pr... more Federated learning is a multiple device collaboration setup designed to solve machine learning problems under framework for aggregation and knowledge transfer in distributed local data. This distributed model ensures the privacy of data at each local node. Owing to its relevance, there has been extensive research activities and outcomes in federated learning with expanded applicability to different areas by the research community. As such, there is a vast research archive made available by the community with research work and articles related to the various aspects of federated learning such as applications, challenges, privacy, functionalities, and design. With respect to the function and design of federated learning, client selection, aggregation, knowledge transfer, management of distributed data (Non-IID), Incentive of data and communication cost are of paramount importance. Any effective design of federated learning requires these aspects to be well considered.There are numerous survey articles found among the available literature that focus on its application and challenges, opportunities, data privacy and protection, as well as on federated learning on internet of things, federated learning on edge computing, etc.
2018 International Conference on Smart Green Technology in Electrical and Information Systems (ICSGTEIS), 2018
In metropolitan areas, common traffic issues include traffic congestion, traffic accidents, air p... more In metropolitan areas, common traffic issues include traffic congestion, traffic accidents, air pollution, and energy consumption occur. To resolve this issues, Intelligent Transportation Systems (ITS) have been evolved by many researchers. One of the important sub-systems in the development of ITS is a Traffic Management System (TMS) which attempts to reduce a traffic congestion. In fact, TMS itself relies on the estimation of traffic flow, therefore providing such an accurate traffic flow prediction is needed. For this reason, we aim to provide an accurate traffic flow prediction to facilitate this system. In this works, a Supervised Deep Learning Based Traffic Flow Prediction (SDLTFP) was proposed which is a type of fully-connected deep neural network (FC-DNN). Timely prediction is also a major issue in guaranteeing reliable traffic flow prediction. However, training a deep network could be time-consuming, and overfitting is might be happening, especially when feeding small data into the deep architecture. The network is learned perfectly during the training, but in testing with the new data, it could fail to generalize the model. We adopt the Batch Normalization (BN) and Dropout techniques to help the network training. SGD and momentum are carried out to update the weight. We then take advantage of open data as historical traffic data which are then used to predict future traffic flow with the proposed method and model above. Experiments show that the Mean Absolute Percentage Error (MAPE) for our traffic flow prediction is within 5 % using sample data and between 15% to 20% using out of the sample data. Training a deep network faster with BN and Dropout reduces the overfitting.
2019 IEEE International Conference on Consumer Electronics - Taiwan (ICCE-TW), 2019
Irregularities in the use of electrical energy could result in failure of power grids and blackou... more Irregularities in the use of electrical energy could result in failure of power grids and blackouts. Anomaly detection is required not only for ensuring grid safety, but also to prevent illegal hacking. Long-term data are recorded for such anomaly detection. However, due to the nonlinear characteristics of the time series data, correlation regression becomes difficult when using non-deep learning techniques. To address this issue, we use Long-Short Term Memory (LSTM) recurrency techniques. We experiment with two datasets from two different smart meters, which are sampled once every 30 seconds for one month. A total of 120,000 data samples were used for training and 40,000 data samples for testing. From the experiment results, the testing accuracy, True-Positive Rate, and False Positive Rate were 0.92, 0.81, and 0.50, respectively. Further, to demonstrate that the LSTM model can actually be designed at the network edge, we implemented the model and the trained weights on a Raspberry Pi platform. The inference time for each sample was 935 μs, which is short enough for realizing edge-based anomaly detection.
2019 IEEE International Conference on Consumer Electronics - Taiwan (ICCE-TW), 2019
Future intersection crossings for autonomous vehicles will not be controlled by traffic signals, ... more Future intersection crossings for autonomous vehicles will not be controlled by traffic signals, rather a controller will be used for communication among vehicles that need to cross an intersection. In this work, we propose an innovative management system called Deep Reinforcement Learning-based Autonomous Intersection Management (DRLAIM) system, which is the first system to use deep reinforcement learning. We train the system to learn a good intersection control policy by interacting with traffic environment through reinforcement learning. The brake-safe control model is used to ensure the safety of each autonomous vehicle while crossing. Experiment results show that after training using reinforcement learning, the throughput of intersection control model increased by 83%. In comparison with the Fast First Service (FFS) policy, the average waiting time of DRLAIM reduced by about 1.2% to 11.4%.
IECON 2016 - 42nd Annual Conference of the IEEE Industrial Electronics Society, 2016
The traditional centralized power system is gradually being replaced by smart grids. However, an ... more The traditional centralized power system is gradually being replaced by smart grids. However, an important design issue is how to perform accurate demand-response such that the power distribution management is effective. This includes two sub-problems, namely the accurate prediction of future electricity demand-response situations and the optimization of power distribution. In this work, we propose a novel Model Predictive Optimization (MPO) method for the advanced distribution management system in smart grids. Future electricity situations (surplus/deficit) are predicted using a customized Autoregressive Integrated Moving Average (ARIMA) model. Pairing between buyers and sellers of electricity are performed based on not only the current situation, but also considering future situations. As a result, trading pairs with overall near-optimal cost are found through concurrent and multiple instances of Particle Swarm Optimization (PSO), along with conflict resolution. Experimental results on 30 micro-grids show the error rate of the ARIMA prediction model to be less than 10%. The proposed MPO method saves totally 19.38% overall trading cost, if predictions are made for 4 future time slots.
IGARSS 2020 - 2020 IEEE International Geoscience and Remote Sensing Symposium, 2020
Landslides could cause huge threats to lives and cause property damages. In the landslide predict... more Landslides could cause huge threats to lives and cause property damages. In the landslide prediction system, environmental information can be collected through sensors to detect the possibility of landslide occurrences. However, the data collected may be lost due to sensor failures, external interferences or other environmental factors, which may affect the accuracy of landslide predictions. In order to solve the problem of missing data, we propose a data reconstruction method based on rainfall intensity and soil moisture, which reconstructs missing data based on temporal relationships. It is based on the data trend in the past period of time. A Long Short-Term Memory (LSTM) deep neural network is trained to predict the data value in missing time slots. We use the predicted data to compensate for the missing data so as the elevate the accuracy not only of data, but also landslide predictions. Our method is compared with other reconstruction methods. The proposed LSTM model exhibit a smaller RMSE than the Linear Extrapolation (LE) method. Even if 90% of random data is lost, the RMSE results for the data reconstruction by LE and LSTM are, respectively, 0.033 and 0.036 for rainfall data and 0.029 and 0.032 for soil moisture data.
Uploads
Papers by Pao-Ann Hsiung