Anomaly detection is a key factor in the processing of large amounts of sensor data from Wireless Sensor Networks (WSN). Efficient anomaly detection algorithms can be devised performing online node-local computations and reducing communication overhead, thus improving the use of the limited hardware resources. This work introduces a fixed-point embedded implementation of Online Sequential Extreme Learning Machine (OS-ELM), an online learning algorithm for Single Layer Feed forward Neural Networks (SLFN). To overcome the stability issues introduced by the fixed precision, we apply correction mechanisms previously proposed for Recursive Least Squares (RLS). The proposed implementation is tested extensively with generated and real-world datasets, and compared with RLS, Linear Least Squares Estimation, and a rule-based method as benchmarks. The methods are evaluated on the prediction accuracy and on the detection of anomalies. The experimental results demonstrate that fixed-point OS-ELM can be successfully implemented on resource-limited embedded systems, with guarantees of numerical stability. Furthermore, the detection accuracy of fixed-point OS-ELM shows better generalization properties in comparison with, for instance, fixed-point RLS. © 2013 IEEE.
Energy conservation is crucial in wireless ad hoc sensor network design to increase network lifetime. Since communication consumes a major part of the energy used by a sensor node, efficient communication is important. Topology control aims at achieving more efficient communication by dropping links and reducing interference among simultaneous transmissions by adjusting the nodes’ transmission power. Since dropping links make a network more susceptible to node failure, a fundamental problem in wireless sensor networks is to find a communication graph with minimum interference and minimum power assignment aiming at an induced topology that can satisfy fault-tolerant properties. In this paper, we examine and propose linear integer programming formulations and a hybrid meta-heuristic GRASP/VNS (Greedy Randomized Adaptive Search Procedure/Variable Neighborhood Search) to determine the transmission power of each node while maintaining a fault-tolerant network and simultaneously minimize the interference and the total power consumption. Optimal biconnected topologies for moderately sized networks with minimum interference and minimum power are obtained using a commercial solver. We report computational simulations comparing the integer programming formulations and the GRASP/VNS, and evaluate the effectiveness of three meta-heuristics in terms of the tradeoffs between computation time and solution quality. We show that the proposed meta-heuristics are able to find good solutions for sensor networks with up to 400 nodes and that the GRASP/VNS was able to systematically find the best lower bounds and optimal solutions.
Existing research on the recognition of Activities of Daily Living (ADL) from simple sensor networks assumes that only a single person is present in the home. In real life there will be situations where the inhabitant receives visits from family members or professional health care givers. In such cases activity recognition is unreliable. In this paper, we investigate the problem of detecting multiple persons in an environment equipped with a sensor network consisting of binary sensors. We conduct a real-life experiment for detection of visits in the oce of the supervisor where the oce is equipped with a video camera to record the ground truth. We collected data during two months and used two models, a Naive Bayes Classier and a Hidden Markov Model for a visitor detection. An evaluation of these two models shows that we achieve an accuracy of 83% with the NBC and an accuracy of 92% with a HMM, respectively.
MULTIFILE
The bi-directional communication link with the physical system is one of the main distinguishing features of the Digital Twin paradigm. This continuous flow of data and information, along its entire life cycle, is what makes a Digital Twin a dynamic and evolving entity and not merely a high-fidelity copy. There is an increasing realisation of the importance of a well functioning digital twin in critical infrastructures, such as water networks. Configuration of water network assets, such as valves, pumps, boosters and reservoirs, must be carefully managed and the water flows rerouted, often manually, which is a slow and costly process. The state of the art water management systems assume a relatively static physical model that requires manual corrections. Any change in the network conditions or topology due to degraded control mechanisms, ongoing maintenance, or changes in the external context situation, such as a heat wave, makes the existing model diverge from the reality. Our project proposes a unique approach to real-time monitoring of the water network that can handle automated changes of the model, based on the measured discrepancy of the model with the obtained IoT sensor data. We aim at an evolutionary approach that can apply detected changes to the model and update it in real-time without the need for any additional model validation and calibration. The state of the art deep learning algorithms will be applied to create a machine-learning data-driven simulation of the water network system. Moreover, unlike most research that is focused on detection of network problems and sensor faults, we will investigate the possibility of making a step further and continue using the degraded network and malfunctioning sensors until the maintenance and repairs can take place, which can take a long time. We will create a formal model and analyse the effect on data readings of different malfunctions, to construct a mitigating mechanism that is tailor-made for each malfunction type and allows to continue using the data, albeit in a limited capacity.
The bi-directional communication link with the physical system is one of the main distinguishing features of the Digital Twin paradigm. This continuous flow of data and information, along its entire life cycle, is what makes a Digital Twin a dynamic and evolving entity and not merely a high-fidelity copy. There is an increasing realisation of the importance of a well functioning digital twin in critical infrastructures, such as water networks. Configuration of water network assets, such as valves, pumps, boosters and reservoirs, must be carefully managed and the water flows rerouted, often manually, which is a slow and costly process. The state of the art water management systems assume a relatively static physical model that requires manual corrections. Any change in the network conditions or topology due to degraded control mechanisms, ongoing maintenance, or changes in the external context situation, such as a heat wave, makes the existing model diverge from the reality. Our project proposes a unique approach to real-time monitoring of the water network that can handle automated changes of the model, based on the measured discrepancy of the model with the obtained IoT sensor data. We aim at an evolutionary approach that can apply detected changes to the model and update it in real-time without the need for any additional model validation and calibration. The state of the art deep learning algorithms will be applied to create a machine-learning data-driven simulation of the water network system. Moreover, unlike most research that is focused on detection of network problems and sensor faults, we will investigate the possibility of making a step further and continue using the degraded network and malfunctioning sensors until the maintenance and repairs can take place, which can take a long time. We will create a formal model and analyse the effect on data readings of different malfunctions, to construct a mitigating mechanism that is tailor-made for each malfunction type and allows to continue using the data, albeit in a limited capacity.
The integration of renewable energy resources, controllable devices and energy storage into electricity distribution grids requires Decentralized Energy Management to ensure a stable distribution process. This demands the full integration of information and communication technology into the control of distribution grids. Supervisory Control and Data Acquisition (SCADA) is used to communicate measurements and commands between individual components and the control server. In the future this control is especially needed at medium voltage and probably also at the low voltage. This leads to an increased connectivity and thereby makes the system more vulnerable to cyber-attacks. According to the research agenda NCSRA III, the energy domain is becoming a prime target for cyber-attacks, e.g., abusing control protocol vulnerabilities. Detection of such attacks in SCADA networks is challenging when only relying on existing network Intrusion Detection Systems (IDSs). Although these systems were designed specifically for SCADA, they do not necessarily detect malicious control commands sent in legitimate format. However, analyzing each command in the context of the physical system has the potential to reveal certain inconsistencies. We propose to use dedicated intrusion detection mechanisms, which are fundamentally different from existing techniques used in the Internet. Up to now distribution grids are monitored and controlled centrally, whereby measurements are taken at field stations and send to the control room, which then issues commands back to actuators. In future smart grids, communication with and remote control of field stations is required. Attackers, who gain access to the corresponding communication links to substations can intercept and even exchange commands, which would not be detected by central security mechanisms. We argue that centralized SCADA systems should be enhanced by a distributed intrusion-detection approach to meet the new security challenges. Recently, as a first step a process-aware monitoring approach has been proposed as an additional layer that can be applied directly at Remote Terminal Units (RTUs). However, this allows purely local consistency checks. Instead, we propose a distributed and integrated approach for process-aware monitoring, which includes knowledge about the grid topology and measurements from neighboring RTUs to detect malicious incoming commands. The proposed approach requires a near real-time model of the relevant physical process, direct and secure communication between adjacent RTUs, and synchronized sensor measurements in trustable real-time, labeled with accurate global time-stamps. We investigate, to which extend the grid topology can be integrated into the IDS, while maintaining near real-time performance. Based on topology information and efficient solving of power flow equation we aim to detect e.g. non-consistent voltage drops or the occurrence of over/under-voltage and -current. By this, centrally requested switching commands and transformer tap change commands can be checked on consistency and safety based on the current state of the physical system. The developed concepts are not only relevant to increase the security of the distribution grids but are also crucial to deal with future developments like e.g. the safe integration of microgrids in the distribution networks or the operation of decentralized heat or biogas networks.
