Partial shading (PS) conditions create challenging issues for the maximum power point tracking (MPPT) algorithms in photovoltaic (PV) systems such as getting stuck in the local maximum power points (LMPP), slow tracking time and fluctuations in the generated power during tracking the global maximum power point (GMPP). To address these issues, this article brings contributions by proposing an improved PSO-based MPPT technique tailored for PS conditions. First, to highlight the weakness of the traditional PSO-MPPT technique, its stability and steady state behavior in the PS conditions is analyzed in depth. Second, the required criteria to achieve a stable response are obtained. Finally, a novel technique to estimate the convex area of the power versus voltage (P-V) curve is presented where the GMPP is located using two voltage boundaries. The performance of the proposed MPPT technique is experimentally validated. The results highlight the capabilities of the proposed technique in finding the GMPP with a rapid convergence and small fluctuations. The proposed MPPT is applicable to most PV inverters under any PS conditions.

The design of multiple human activity recognition applications in areas such as healthcare, sports and safety relies on wearable sensor technologies. However, when making decisions based on the data acquired by such sensors in practical situations, several factors related to sensor data alignment, data losses, and noise, among other experimental constraints, deteriorate data quality and model accuracy. To tackle these issues, this paper presents a data-driven iterative learning framework to classify human locomotion activities such as walk, stand, lie, and sit, extracted from the Opportunity dataset. Data acquired by twelve 3-axial acceleration sensors and seven inertial measurement units are initially de-noised using a two-stage consecutive filtering approach combining a band-pass Finite Impulse Response (FIR) and a wavelet filter. A series of statistical parameters are extracted from the kinematical features, including the principal components and singular value decomposition of roll, pitch, yaw and the norm of the axial components. The novel interactive learning procedure is then applied in order to minimize the number of samples required to classify human locomotion activities. Only those samples that are most distant from the centroids of data clusters, according to a measure presented in the paper, are selected as candidates for the training dataset. The newly built dataset is then used to train an SVM multi-class classifier. The latter will produce the lowest prediction error. The proposed learning framework ensures a high level of robustness to variations in the quality of input data, while only using a much lower number of training samples and therefore a much shorter training time, which is an important consideration given the large size of the dataset.

With the recent widespread deployment of Photovoltaic (PV) panels in the northern snow-prone areas, performance analysis of these panels is getting much more importance. Partial or full reduction in energy yield due to snow accumulation on the surface of PV panels, which is referred to as snow loss, reduces their operational efficiency. Developing intelligent algorithms to accurately predict the future snow loss of PV farms is addressed in this article for the first time. The article proposes daily snow loss prediction models using machine learning algorithms solely based on meteorological data. The algorithms include regression trees, gradient boosted trees, random forest, feed-forward and recurrent artificial neural networks, and support vector machines. The prediction models are built based on the snow loss of a PV farm located in Ontario, Canada which is calculated using a 3-stage model and hourly data records over a 4-year period. The stages of the aforementioned model consist of: stage I: yield determination, stage II: power loss calculation, and stage III: snow loss extraction. The functionality of the proposed prediction models is validated over the historical data and the optimal hyperparameters are selected for each model to achieve the best results. Among all the models, gradient boosted trees obtained the minimum prediction error and thus the best performance. The results achieved prove the effectiveness of the proposed models for the prediction of daily snow loss of PV farms.

Robots are expected to recognize the properties of objects in order to handle them safely and efficiently in a variety of applications, such as health and elder care, manufacturing, or high-risk environments. This paper explores the issue of surface characterization by monitoring the signals acquired by a novel bio-inspired tactile probe in contact with ridged surfaces. The tactile module comprises a nine Degree of Freedom Microelectromechanical Magnetic, Angular Rate, and Gravity system (9-DOF MEMS MARG) and a deep MEMS pressure sensor embedded in a compliant structure that mimics the function and the organization of mechanoreceptors in human skin as well as the hardness of the human skin. When the modules tip slides over a surface, the MARG unit vibrates and the deep pressure sensor captures the overall normal force exerted. The module is evaluated in two experiments. The first experiment compares the frequency content of the data collected in two setups: one when the module is mounted over a linear motion carriage that slides four grating patterns at constant velocities; the second when the module is carried by a robotic finger in contact with the same grating patterns while performing a sliding motion, similar to the exploratory motion employed by humans to detect object roughness. As expected, in the linear setup, the magnitude spectrum of the sensors’ output shows that the module can detect the applied stimuli with frequencies ranging from 3.66 Hz to 11.54 Hz with an overall maximum error of ±0.1 Hz. The second experiment shows how localized features extracted from the data collected by the robotic finger setup over seven synthetic shapes can be used to classify them. The classification method consists on applying multiscale principal components analysis prior to the classification with a multilayer neural network. Achieved accuracies from 85.1% to 98.9% for the various sensor types demonstrate the usefulness of traditional MEMS as tactile sensors embedded into flexible substrates.

Transfer of learning or leveraging a pre-trained network and fine-tuning it to perform new tasks has been successfully applied in a variety of machine intelligence fields, including computer vision, natural language processing and audio/speech recognition. Drawing inspiration from neuroscience research that suggests that both visual and tactile stimuli rouse similar neural networks in the human brain, in this work, we explore the idea of transferring learning from vision to touch in the context of 3D object recognition. In particular, deep convolutional neural networks (CNN) pre-trained on visual images are adapted and evaluated for the classification of tactile data sets. To do so, we ran experiments with five different pre-trained CNN architectures and on five different datasets acquired with different technologies of tactile sensors including BathTip, Gelsight, force-sensing resistor (FSR) array, a high-resolution virtual FSR sensor, and tactile sensors on the Barrett robotic hand. The results obtained confirm the transferability of learning from vision to touch to interpret 3D models. Due to its higher resolution, tactile data from optical tactile sensors was demonstrated to achieve higher classification rates based on visual features compared to other technologies relying on pressure measurements. Further analysis of the weight updates in the convolutional layer is performed to measure the similarity between visual and tactile features for each technology of tactile sensing. Comparing the weight updates in different convolutional layers suggests that by updating a few convolutional layers of a pre-trained CNN on visual data, it can be efficiently used to classify tactile data. Accordingly, we propose a hybrid architecture performing both visual and tactile 3D object recognition with a MobileNetV2 backbone. MobileNetV2 is chosen due to its smaller size and thus its capability to be implemented on mobile devices, such that the network can classify both visual and tactile data. An accuracy of 100% for visual and 77.63% for tactile data are achieved by the proposed architecture.

Power prediction for photovoltaic (PV) installations in northern snow-prone areas remains a challenging problem. The behavior of a partially/fully snow-covered PV panel can be complex depending on the snow-related factors, disregarding, which causes large prediction errors in snow conditions. To address this issue, a systematic analysis of the effects of 16 meteorological parameters on the hourly power prediction for systems in a snow-prone area is conducted in this article. According to the best of our knowledge, such comprehensive analysis on various snow-related conditions datasets, i.e., full, snow-free condition, snow condition, and snow-cover condition datasets, for several PV systems is performed for the first time in the literature. The three latter datasets are extracted from the full dataset using a proposed three-step procedure. Moreover, different computational intelligence techniques are implemented to develop hourly prediction models for each dataset of each system. A detailed comparison is then performed between the performance of the proposed models, the Marion model, a modified version of the Marion model, the classic PV model, and a computational intelligence model combined with the Marion model's snow cover detection method. The hourly values of the electrical and meteorologicalparameters for 17 PV systems across Canada, with an aggregated time period of more than 55 years, have been extracted to perform the study. As the results show, categorizing data using the proposed three-step procedure and developing specific computational intelligence models for each condition can significantly improve the prediction accuracy especially when a full/partial snow cover is probable on the panels.