Short-term irradiance forecasting is a fundamental research problem in optimizing utility-scale photovoltaic (PV) systems. Clouds and shading conditions are significant causes of uncertainty and instability in photovoltaic (PV) power generation. To address these issues, we have developed a weather-based irradiance forecasting algorithm that can aid in planning PV array operations. Our method utilizes a 1-dimensional (1-D) temporal convolutional network (TCN). To benchmark this model, we evaluated against three alternative neural network methods on the National Solar Radiation Database (NSRDB) dataset for a time series forecasting task. We compare the prediction accuracy of our TCN against more common multilayer 1-D convolution, dense, and long-short-term memory (LSTM) neural networks. With only a single day of prior multivariate meteorological data and nine weather characteristics in a time-bound window, our TCN method demonstrates a significant 8.5% RMSE and 17% MSE performance improvement. We conducted an empirical analysis to identify weather markers that are most strongly correlated with irradiance prediction, demonstrating a strong correlation with the solar zenith angle, cloud types, and surface albedo. Monitoring these features and predicting future solar irradiance can aid in planning and optimal power generation from PV arrays.

Recent innovations in ROI camera systems have opened up the avenue for exploring energy optimization techniques like adaptive subsampling. Generally speaking, image frame capture and read-out demand high power consumption. ROI camera systems make it possible to exploit the inverse relation between energy consumption and spatiotemporal pixel readout to optimize the power efficiency of the image sensor. To this end, we develop a reinforcement learning (RL) based adaptive subsampling framework which predicts ROI trajectories and reconfigures the image sensor on-the-fly for improved power efficiency of the image sensing pipeline. In our proposed framework, a pre-trained convolutional neural network (CNN) extracts rich visual features from incoming frames and a long short-term memory (LSTM) network predicts the region of interest (ROI) and subsampling pattern for the consecutive image frame. Based on the application and the difficulty level of object motion trajectory, the user can utilize either the predicted ROI or coarse subsampling pattern to switch off the pixels for sequential frame capture, thus saving energy. We have validated our proposed method by adapting existing trackers for the adaptive subsampling framework and evaluating them as competing baselines. As a proof-of-concept, our method outperforms the baselines and achieves an average AUC score of 0.5090 on three benchmarking datasets. We also characterize the energy-accuracy tradeoff of our method vs. the baselines and show that our approach is best suited for applications that demand both high visual tracking precision and low power consumption. On the TB100 dataset, our method achieves the highest AUC score of 0.5113 out of all the competing algorithms and requires a medium-level power consumption of approximately 4 W as per a generic energy model and an energy consumption of 1.9 mJ as per a mobile system energy model. Although other baselines are shown to have better performance in terms of power consumption, they are ill-suited for applications that require considerable tracking precision, making our method the ideal candidate in terms of power-accuracy tradeoff.

There is tremendous scope for improving the energy efficiency of embedded vision systems by incorporating programmable region-of-interest (ROI) readout in the image sensor design. In this work, we study how ROI programmability can be leveraged for vision applications by anticipating where the ROI will be located in future frames and switching pixels off outside of this region. We refer to this process of ROI prediction and corresponding sensor configuration as adaptive subsampling. Our adaptive subsampling algorithms comprise an object detector and an ROI predictor (Kalman filter) which operate in conjunction to optimize the energy efficiency of the vision pipeline with the end task being object tracking. To further facilitate the implementation of our adaptive algorithms in real systems, we select a candidate algorithm and map it onto an FPGA. Leveraging Xilinx Vitis AI tools, we designed and accelerated a YOLO object detector-based adaptive subsampling algorithm. In order to further improve the algorithm post-deployment, we evaluated several competing baselines on the OTB100 and LaSOT datasets. We found that coupling the ECO tracker with the Kalman filter has a competitive AUC score of 0.4568 and 0.3471 on the OTB100 and LaSOT datasets respectively. Further, the power efficiency of this algorithm is on par with, and in a couple of instances superior to, the other baselines. The ECO-based algorithm incurs a power consumption of approximately 4 W averaged across both datasets while the YOLO-based approach requires power consumption of approximately 6 W (as per our power consumption model). In terms of accuracy-latency tradeoff, the ECO-based algorithm provides near-real-time performance (19.23 FPS) while managing to attain competitive tracking precision.

Ion Channel sensors have several applications including DNA sequencing, biothreat detection, and medical applications. Ion-channel sensors mimic the selective transport mechanism of cell membranes and can detect a wide range of analytes at the molecule level. Analytes are sensed through changes in signal patterns. Papers in the literature have described different methods for ion channel signal analysis. In this paper, we describe a series of new graphical tools for ion channel signal analysis which can be used for research and education. The paper focuses on the utility of this tools in biosensor classes. Teaching signal processing and machine learning for ion channel sensors is challenging because of the multidisciplinary content and student backgrounds which include physics, chemistry, biology and engineering. The paper describes graphical ion channel analysis tools developed for an on-line simulation environment called J-DSP. The tools are integrated and assessed in a graduate bio-sensor course through computer laboratory exercises.

Domain generalization (DG) methods aim to develop models that generalize to settings where the test distribution is different from the training data. In this paper, we focus on the challenging problem of multi-source zero shot DG (MDG), where labeled training data from multiple source domains is available but with no access to data from the target domain. A wide range of solutions have been proposed for this problem, including the state-of-the-art multi-domain ensembling approaches. Despite these advances, the na\"ive ERM solution of pooling all source data together and training a single classifier is surprisingly effective on standard benchmarks. In this paper, we hypothesize that, it is important to elucidate the link between pre-specified domain labels and MDG performance, in order to explain this behavior. More specifically, we consider two popular classes of MDG algorithms -- distributional robust optimization (DRO) and multi-domain ensembles, in order to demonstrate how inferring custom domain groups can lead to consistent improvements over the original domain labels that come with the dataset. To this end, we propose (i) Group-DRO++, which incorporates an explicit clustering step to identify custom domains in an existing DRO technique; and (ii) DReaME, which produces effective multi-domain ensembles through implicit domain re-labeling with a novel meta-optimization algorithm. Using empirical studies on multiple standard benchmarks, we show that our variants consistently outperform ERM by significant margins (1.5% - 9%), and produce state-of-the-art MDG performance. Our code can be found at https://github.com/kowshikthopalli/DREAME

With rising concerns over climate change, there is an increasing need for renewable energy sources. Photovoltaic(PV) systems are one of the most environmentally friendly ways of producing energy. However, the fluctuations in power outputs from utility scale PV arrays makes it difficult to incorporate them into electric grids. The power output is directly related to the irradiance and the irradiance is related to the surface albedo, which is the fraction of sunlight reflected by a surface. If we can predict the surface albedo, we can predict the power output. Using random forest regression, we can make predictions of the power output based on various features. In response to this prediction, the topology of the system may be reconfigured.

—Domain generalization (DG) methods aim to develop models that generalize to settings where the test distribution is different from the training data. In this paper, we focus on the challenging problem of multi-source zero shot DG, where labeled training data from multiple source domains is available but with no access to data from the target domain. Though this problem has become an important topic of research, surprisingly, the simple solution of pooling all source data together and training a single classifier is highly competitive on standard benchmarks. More importantly, even sophisticated approaches that explicitly optimize for invariance across different domains do not necessarily provide non-trivial gains over ERM. In this paper, for the first time, we study the important link between pre-specified domain labels and the generalization performance. Using a motivating case-study and a new variant of a distributional robust optimization algorithm, GroupDRO++, we first demonstrate how inferring custom domain groups can lead to consistent improvements over the original domain labels that come with the dataset. Subsequently, we introduce a general approach for multi-domain generalization, MulDEns, that uses an ERM-based deep ensembling backbone and performs implicit domain re-labeling through a meta-optimization algorithm. Using empirical studies on multiple standard benchmarks, we show that MulDEns does not require tailoring the augmentation strategy or the training process specific to a dataset, consistently outperforms ERM by significant margins, and produces state-of-the-art generalization performance, even when compared to existing methods that exploit the domain labels.

Image sensors with programmable region-of-interest (ROI) readout are a new sensing technology important for energyefficient embedded computer vision. In particular, ROIs can subsample the number of pixels being readout while performing single object tracking in a video. In this paper, we develop adaptive sampling algorithms which perform joint object tracking and predictive video subsampling. We utilize an object detection consisting of either mean shift tracking or a neural network, coupled with a Kalman filter for prediction. We show that our algorithms achieve mean average precision of 0.70 or higher on a dataset of 20 videos in software. Further, we implement hardware acceleration of mean shift tracking with Kalman filter adaptive subsampling on an FPGA. Hardware results show a 23 × improvement in clock cycles and latency as compared to baseline methods and achieves 38FPS real-time performance. This research points to a new domain of hardware-software co-design for adaptive video subsampling in embedded computer vision.

Compressive sensing cameras hold the promise of cost-effective hardware, lower data rates, and improved video quality, particularly outside the visible spectrum. However, these improvements involve significant computational cost, as sensor output must be reconstructed in order to form an image viewable by a human. This paper describes a prototype automated detection and tracking system using a compressive sensing camera that does not rely on computationally costly image reconstructions. It operates on raw sensor data for an approximately ten-fold improvement in computation time over a comparable reconstruct-then-track algorithm. The detector is successful at a sensing rate of 0.3, comparable to that required for high-quality image reconstructions. If initialized with the location of a target, the tracker holds the target at a sensing rate of 0.005, below the boundary where reconstruction breaks down. These results show not only that direct tracking from compressive cameras is possible, but also give support to the pursuit of direct inference from compressive sensors of all types.

The production of PV energy is affected by external conditions such as partial shading, varying temperatures, and soiling of PV panels. Among these, partial shading causes a significant reduction in power. On the other hand, the energy production can also be affected if a PV array system has faulty modules [13] as discussed in Chapter 3. Partial shading although temporary causes voltage and current mismatch losses and can cause a significant reduction in the power supplied to the grid thereby limiting the performance of the PV array [92, 93]. The previous chapters (Chapter 3 and 4) dealt with fault detection and shading prediction. In addition to these tasks, another desirable feature in solar monitoring includes, reconfiguring the panel wiring to form different topologies or circuit connections to allow the PV array systems to produce increased power output under shading conditions. Reconfiguring the connections between the panels in a PV array is a powerful strategy to mitigate the impact of partial shading. Conventionally, utility-scale and roof-top PV arrays are connected in fixed topologies for example in Series-Parallel (SP) [13] where a fixed number of PV panels are connected in series and parallel to form an array. These fixed PV array system topologies were determined taking into consideration the weather condition for the entire year. This design does not provide the maximum power levels for a given day or season. A fixed topology is unable to bypass failed or under-performing modules, reducing the overall array output. To improve the array output power, reconfigurable systems that can change their topology need to be developed.

Effective patient care mandates rapid, yet accurate, diagnosis. With the abundance of non-invasive diagnostic measurements and electronic health records (EHR), manual interpretation for differential diagnosis has become time-consuming and challenging. This has led to wide-spread adoption of AI-powered tools, in pursuit of improving accuracy and efficiency of this process. While the unique challenges presented by each modality and clinical task demand customized tools, the cumbersome process of making problem-specific choices has triggered the critical need for a generic solution to enable rapid development of models in practice. In this spirit, we develop DDxNet, a deep architecture for time-varying clinical data, which we demonstrate to be well-suited for diagnostic tasks involving different modalities (ECG/EEG/EHR), required level of characterization (abnormality detection/phenotyping) and data fidelity (single-lead ECG/22-channel EEG). Using multiple benchmark problems, we show that DDxNet produces high-fidelity predictive models, and sometimes even provides significant performance gains over problem-specific solutions.

In Chapter 2, we elaborate on several faults (soiling, module degradation, shading) that cause variation in the power output of a PV array. However, since the power output is based on sun’s irradiance there is high uncertainty and intermittency due to variable weather conditions which makes it difficult to incorporate PV arrays into existing power grids. In this chapter, we consider cloud movement as one of the category of PV panel shading. Cloud cover is one of the major reasons that hampers the widespread penetration of PV in power grids. There are numerous solar power forecasting models available where cloud velocity serves as one of the major parameters in the model [62–64]. There is extensive literature on getting cloud motion vectors using satellite imagery [65–69], irradiance sensors [70, 71], and sky-cameras [72]. In this chapter we elaborate on a novel cloud movement prediction method which views the cloud videos as dynamic textures. Sequences of moving scenes such as clouds, forest fires, boiling water, etc. with certain statistical stationarity properties in time have been modeled as spatio-temporal textures or dynamic textures in literature [73, 74].

The efficiency of solar energy farms requires detailed analytics and information on each panel regarding voltage, current, temperature, and irradiance. Monitoring utility-scale solar arrays was shown

An increase in grid-connected photovoltaic arrays creates a need for efficient and reliable fault detection. In this paper, machine learning strategies for fault detection are presented. An Artificial Neural Network was studied with the goal of detecting three photovoltaic module conditions. In addition, an unsupervised approach was successfully implemented using the -means clustering algorithm, successfully detecting arc and ground faults. To distinguish and localize additional faults such as shading and soiling, a supervised approach is adopted using a Radial Basis Function Network. A solar array dataset with voltage, current, temperature, and irradiance was examined. This dataset had labeled data with normal conditions and faults due to soiling and shading. A radial basis network was trained to classify faults, resulting in an error rate below 2% on synthetic data with realistic levels of noise.

Ground-based sky imaging has won popularity due to its higher temporal and spatial resolution when compared with satellite or air-borne sky imaging systems. Cloud identification and segmentation is the first step in several areas, such as climate research and lately photovoltaic power generation forecast. Cloud-sky segmentation involves several variables including sun position and type and altitude of clouds. We proposed a training free cloud/sky segmentation based on a threshold that adapts to the cloud formation conditions. Experimental results show that the proposed method reaches higher detection accuracy against state-of-the-art algorithms; additionally, qualitative results over hemispherical high dynamic range (HDR) sky images are provided. The proposed cloud segmentation method can be applied to shading prediction for photovoltaic (PV) systems.

Compressed sensing (CS) allows signals and images to be reliably inferred from undersampled measurements. Exploiting CS allows the creation of new types of high-performance sensors including infrared

Modern audio source separation techniques rely on optimizing sequence model architectures such as, 1D-CNNs, on mixture recordings to generalize well to unseen mixtures. Specifically, recent focus is on time-domain based architectures such as Wave-U-Net which exploit temporal context by extracting multi-scale features. However, the optimality of the feature extraction process in these architectures has not been well investigated. In this paper, we examine and recommend critical architectural changes that forge an optimal multi-scale feature extraction process. To this end, we replace regular $1-$D convolutions with adaptive dilated convolutions that have innate capability of capturing increased context by using large temporal receptive fields. We also investigate the impact of dense connections on the extraction process that encourage feature reuse and better gradient flow. The dense connections between the downsampling and upsampling paths of a U-Net architecture capture multi-resolution information leading to improved temporal modelling. We evaluate the proposed approaches on the MUSDB test dataset. In addition to providing an improved performance over the state-of-the-art, we also provide insights on the impact of different architectural choices on complex data-driven solutions for source separation.