Advances in Non-Invasive Biomedical Signal Sensing and Processing with Machine LearningThis book presents the modern technological advancements and revolutions in the biomedical sector. Progress in the contemporary sensing, Internet of Things (IoT) and machine learning algorithms and architectures have introduced new approaches in the mobile healthcare. A continuous observation of patients with critical health situation is required. It allows monitoring of their health status during daily life activities such as during sports, walking and sleeping. It is realizable by intelligently hybridizing the modern IoT framework, wireless biomedical implants and cloud computing. Such solutions are currently under development and in testing phases by healthcare and governmental institutions, research laboratories and biomedical companies. The biomedical signals such as electrocardiogram (ECG), electroencephalogram (EEG), Electromyography (EMG), phonocardiogram (PCG), Chronic Obstructive Pulmonary (COP), Electrooculography (EoG), photoplethysmography (PPG), and image modalities such as positron emission tomography (PET), magnetic resonance imaging (MRI) and computerized tomography (CT) are non-invasively acquired, measured, and processed via the biomedical sensors and gadgets. These signals and images represent the activities and conditions of human cardiovascular, neural, vision and cerebral systems. Multi-channel sensing of these signals and images with an appropriate granularity is required for an effective monitoring and diagnosis. It renders a big volume of data and its analysis is not feasible manually. Therefore, automated healthcare systems are in the process of evolution. These systems are mainly based on biomedical signal and image acquisition and sensing, preconditioning, features extraction and classification stages. The contemporary biomedical signal sensing, preconditioning, features extraction and intelligent machine and deep learning-based classification algorithms are described.
Effective Load Pattern Classification by Processing the Smart Meter Data Based on Event-Driven Processing and Machine LearningThe technological advancements have evolved the deployment of smart meters. A fine-grained metering data collection and analysis is necessary to bring benefits to multiple smart grid stakeholders. The classical sensing mechanism is time-invariant. Therefore, it results in the collection, transmission, and processing of a large amount of unnecessary data. This work employs the event-driven sensing mechanism to achieve real-time data compression. Afterward, the novel adaptive rate techniques are employed for the data conditioning, segmentation, and extraction of features. The pertinent features regarding the appliances’ consumption patterns are afterward used for their identification. It is realized by employing the mature Support Vector Machine and k-Nearest Neighbor classifiers. Results confirm a 3.4 times compression gain and the computational effectiveness of the suggested solution while securing 95.4% classification precision. It shows the benefits of integrating the proposed method in the realization of current energy efficiency services like enumerated consumption billing, effective load identification, and dynamic load management.
Effective Power Quality Disturbances Identification Based on Event-Driven Processing and Machine LearningPower quality (PQ) disturbances cause rigorous issues in smart grids and industries. The identification of PQ disturbances and effective prevention of such events are essential. In this framework, vital aspects are a precise understanding and a real-time treatment of the PQ disturbances. A novel tactic is described in this chapter. It is founded on the basis of event-driven processing, analysis and machine learning for successful and efficient detection of PQ disturbances. The definition is based on an sophisticated combination of event-driven signal amplification and segmentation and local extraction of features and categorization to achieve an effective and appropriate precision method.Two rigorous classifiers namely k-Nearest Neighbor (KNN) and Naïve Bias are used for the automatic identification. For a case study, the framework functionality is checked, and findings are reported. Compared to conventional equivalents, the first order of magnitude reduction in the cumulative sample count is accomplished by the invented method. It leads to substantial benefit in the reduction of computational complexity and effectiveness in terms of power consumption and delays in processing. The designed framework also attains an appropriate precision of categorization for the case of four-class PQ disturbances.