Organic solar cells (OSCs) have emerged as a competitive alternative to conventional silicon-based photovoltaics for their inexpensive production, versatility, and reduced energy consumption. However, it is still challenging to accurately assess their performance due to the complex interactions between molecular structure and device-level features. We provide a physics-constrained graph neural network (GNN) architecture for multi-output prediction of key OSC parameters, including power conversion efficiency (PCE), open-circuit voltage, short-circuit current density, and fill factor in this study. To ensure agreement between the anticipated PCE and its physically derived formulation, a physics-guided regularization term is added. Experimental results on a dataset of 5628 samples show that the neural-only GNN achieves strong predictive performance (R2=0.630), outperforming the baseline model random forest (R2=0.537). The proposed physics-constrained GNN maintains comparable accuracy (R2=0.626) while significantly reducing physics violation (from 0.406 to 0.104). These results show that adding physics constraints makes predictions more consistent without lowering accuracy, making it a reliable way to predict OSC performance.

In this paper, we utilized machine learning (ML) algorithms to optimize Maximum Power Point Tracking (MPPT) in photovoltaic systems. Predicting the optimal voltage is important as, at that voltage, the system gains maximum energy output and efficiency. We predicted optimal voltage from irradiance and temperature datasets by applying ML algorithm. We evaluated several ML models, finding Linear Regression to perform best with an MSE (mean square error) of 0.0024 and RMSE (root mean square error) of 0.0489. The proposed solution integrates an ML (linear regression)-driven voltage predictor, a PID controller, and a DC-DC buck-boost converter in a MATLAB simulation environment. In MATLAB Simulink, we observed that the buck-boost converter produced the same voltage as the predicted value across the load.

Precise modeling of power systems is vital to ensure stability, reliability, and secure operations. In power industrial settings, model parameters can become skewed over time due to prolonged device usage or modifications made to the control systems. Doubly-Fed Induction Generator (DFIG), one of the most prevalent generators in wind farms, is sensitive to transient occurrences. Consequently, parameter calibration of DFIG becomes a crucial focal point in power system planning and operational studies. In this paper, two baseline approaches are first developed to identify the potentially harmful parameters of the DFIG system, including the Particle Swarm Optimization (PSO) method and the state-of-the-art off-policy Reinforcement Learning (RL) method, Soft Actor-Critic (SAC). The outcomes demonstrated that the SAC method outperformed PSO, resulting in an impressive reduction of 74.67% Mean Squared Error (MSE) and a more efficient testing period. In further exploration, a novel hybrid approach called SAC-PSO is developed, with SAC being the teacher of PSO to tackle scenarios with multiple potential solutions. The results exhibited an even greater enhancement over using SAC alone, leading to a remarkable reduction of 87.84% MSE during the testing phase. The proposed method can also effectively apply to a power plant incorporating multiple wind generators.

Conventional model-based probabilistic inference methods require increasingly complex models to improve Electric Vehicle (EV) battery State of Charge (SOC) estimation. Deep learning methods gained popularity in recent years with their model free estimations. However, practical constraints such as insufficient training data, model complexity for real time implementation, and generalization on new dataset hinder performance reliability. Another major practical drawback of the data driven deep learning approach is its poor convergence from an unfamiliar initial error state as training dataset does not adequately accommodate these practical error scenarios. This paper proposes an ensemble method that uses a weighted estimate of the Central Difference Kalman Filter (CDKF) and Nonlinear Autoregressive with Exogenous Input (NARX) to accurately estimate SOC in the early stages of degradation. We employ a parallel ensemble estimation method that reduces the estimation bias, improves generalization, accuracy, robustness, and reliability of the estimator. We propose a pre-estimated voting weight to combine the ensemble algorithm and employ the CDKF covariance dependent method as the optimum approach for initializing the ensemble system to achieve a robust convergence performance. The state converges on an average of 136 time-steps when initialized halfway from the true state. The average Mean Absolute Error (MAE) performance of the ensemble method is about 0.5 % with an average training data of about 31,041 time-steps. The model was validated using conventional drive cycle data and was shown to outperform its individual ensemble members and gated Recurrent Neural Networks (RNN) such as Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM) and the Bidirectional LSTM (BiLSTM).

Load forecasting (LF) is an essential factor in power system management. LF helps the utility maximize the utilization of power-generating plants and schedule them both reliably and economically. In this paper, a novel and hybrid forecasting method is proposed, combining a long short-term memory network (LSTM) and neural prophet (NP) through an artificial neural network. The paper aims to predict electric load for different time horizons with improved accuracy as well as consistency. The proposed model uses historical load data, weather data, and statistical features obtained from the historical data. Multiple case studies have been conducted with two different real-time data sets on three different types of load forecasting. The hybrid model is later compared with a few established methods of load forecasting found in the literature with different performance metrics: mean average percentage error (MAPE), root mean square error (RMSE), sum of square error (SSE), and regression coefficient (R). Moreover, a guideline with various attributes is provided for different types of load forecasting considering the applications of the proposed model. The results and comparisons from our test cases showed that the proposed hybrid model improved the forecasting accuracy for three different types of load forecasting over other forecasting techniques.

Energy storage technologies are being used excessively in industrial applications and in automobiles. Battery state of charge (SOC) is an important metric to be monitored in these applications to ensure proper and safe functionality. Since SOC cannot be measured directly, this paper puts forth a novel machine learning architecture to improve on the existing methods of SOC estimation. This method consists of using combined stacked bi-directional LSTM and encoder–decoder bi-directional long short-term memory architecture. This architecture henceforth represented as SED is implemented to overcome the nonparallel functionality observed in traditional RNN algorithms. Estimations were made utilizing different open-source datasets such as urban dynamometer driving schedule (UDDS), highway fuel efficiency test (HWFET), LA92 and US06. The least Mean Absolute Error observed was 0.62% at 25 °C for the HWFET condition, which confirms the good functionality of the proposed architecture.

The first part of a two-part series of papers provides a survey on recent advances in Deep Reinforcement Learning (DRL) applications for solving partially observable Markov decision processes (POMDP) problems. Reinforcement Learning (RL) is an approach to simulate the human’s natural learning process, whose key is to let the agent learn by interacting with the stochastic environment. The fact that the agent has limited access to the information of the environment enables AI to be applied efficiently in most fields that require self-learning. Although efficient algorithms are being widely used, it seems essential to have an organized investigation—we can make good comparisons and choose the best structures or algorithms when applying DRL in various applications. In this overview, we introduce Markov Decision Processes (MDP) problems and Reinforcement Learning and applications of DRL for solving POMDP problems in games, robotics, and natural language processing. A follow-up paper will cover applications in transportation, communications and networking, and industries.

Diabetes is a disease that has no permanent cure; hence early detection is required. Data mining, machine learning (ML) algorithms, and Neural Network (NN) methods are used in diabetes prediction in our research. We used the Pima Indian Diabetes (PID) dataset for our research, collected from the UCI Machine Learning Repository. The dataset contains information about 768 patients and their corresponding nine unique attributes. We used seven ML algorithms on the dataset to predict diabetes. We found that the model with Logistic Regression (LR) and Support Vector Machine (SVM) works well on diabetes prediction. We built the NN model with a different hidden layer with various epochs and observed the NN with two hidden layers provided 88.6% accuracy.

The two-part series of papers provides a survey on recent advances in Deep Reinforcement Learning (DRL) for solving partially observable Markov decision processes (POMDP) problems. Reinforcement Learning (RL) is an approach to simulate the human’s natural learning process, whose key is to let the agent learn by interacting with the stochastic environment. The fact that the agent has limited access to the information of the environment enables AI to be applied efficiently in most fields that require self-learning. It’s essential to have an organized investigation—we can make good comparisons and choose the best structures or algorithms when applying DRL in various applications. The first part of the overview introduces Markov Decision Processes (MDP) problems and Reinforcement Learning and applications of DRL for solving POMDP problems in games, robotics, and natural language processing. In part two, we continue to introduce applications in transportation, industries, communications and networking, etc. and discuss the limitations of DRL.

Facial expression recognition (FER) is a promising but challenging area of Computer Vision (CV). Many researchers have devoted significant resources to exploring FER in recent years, but an impediment remains: classifiers perform well on fine resolution images but have difficulty recognizing in-the-wild human emotional states. In order to solve the aforementioned issue, we introduced three novel designs and implemented them in neural networks. More specifically, we utilized an asymmetric pyramidal network (APNet) and employed multi-scale kernels instead of identical size kernels. In addition, square kernels were replaced by a sequence of square, horizontal, and vertical convolutions. This structure can increase the description ability of convolutional neural networks (CNN) and transfer multi-scale features between different layers. Additionally, when training CNN, we adopted stochastic gradient descent with gradient centralization (SGDGC) where it centralizes gradients to have zero mean and makes the training process more efficient and stable. To verify the effectiveness of APNet with SGDGC, we used three of the most popular in-the-wild emotion datasets, FER-2013, CK+, and JAFFE, for our experiments. The results of our experiment and comparisons with state-of-the-art designs from others demonstrate that our method outperforms all the single model methods and has comparable performance with model fusion methods.

With the increasing amount of available medical data, computing power and network speed, modern medical imaging is facing an unprecedented amount of data to analyze and interpret. Phenomena such as Big Data-omics stemming from several diagnostic procedures and novel multi-parametric imaging modalities tend to produce almost unmanageable quantities of data. The paper addresses the aforementioned context by assuming that a novel paradigm in massive data processing and automation becomes necessary in order to improve diagnostics and facilitate personalized and precision medicine for each patient. Traditional machine learning concepts have demonstrated many shortcomings when it comes to correctly diagnose fatal diseases. At the same time static graph networks are unable to capture the fluctuations in brain processing and monitor disease evolution. Therefore, artificial intelligence and deep learning are increasingly applied in oncologic medical imaging because they excel at providing quantitative assessments of biomedical imaging characteristics. On the other hand, novel concepts borrowed from modern control have paved the path for a dynamic graph theory that can predict neurodegenerative disease evolution and replace longitudinal studies. We chose two important topics, brain data processing and oncologic imaging to show the relevance of these concepts. We believe that these novel paradigms will impact multiple facets of radiology but are convinced that it is unlikely that they will replace radiologists any time in the near future since there are still many challenges in the clinical implementation.

We are reviewing various notions of Artificial Intelligence (AI) based on its unsuccessful history (as the success of AI has too many reports). Then we suggest changes based on notions of possibility from fuzzy math, ergodic process for examining possibility space rather than a stochastic time space, speed and memory in neural networks, which follow the supervised least mean square (LMS) deep learning or unsupervised gradient descent learning based on minimizing cost function of Helmholtz free energy, etc. The first Gen AI is Marvin Minsky's rule-based AI which has taken into account implicitly the first attribute Poincare-Neumann Ergodicity that has foregone the small Rosenblatt neural nets. The second Gen AI is Google learnable rule-based system added either to LMS Deep Learning or Maxwell–Boltzmann irreversible thermodynamics Unsupervised Deep Learning (unlabeled inputs) that has beaten Go champion Lee Se-Dol. The third Gen AI shall take into account open set possibilities used in Fuzzy Logic (not yet in digital automation computer scientists [ACS]). In order to compute “where we are,” in turns of 5W (when, where, what, why, and how) needed for route planning, geolocation, and situation awareness, we generated the data from the dynamic equations of motion of Newton, Kubo-Langevin's Einstein fluctuation-dissipation, multiple satellite triangulation GPS, advanced EO, IR, RF sensors suite, as well as Lyaponov Control theory. We wish to tabulate the peak occurrence frequency of those dynamic data near the peak of a triangle shape function in rounded brackets into Zadeh-Freeman fuzzy membership functions (FMF).

In this paper, we present our work on high-efficiency multi-junction polymer and hybrid solar cells. The transfer matrix method is used for optical modeling of an organic solar cell, which was inspired by the McGehee Group in Stanford University. The software simulation calculates the optimal thicknesses of the active layers to provide the best short circuit current (JSC) value. First, we show three designs of multi-junction polymer solar cells, which can absorb sunlight beyond the 1000 nm wavelengths. Then we present a novel high-efficiency hybrid (organic and inorganic) solar cell, which can absorb the sunlight with a wavelength beyond 2500 nm. Approximately 12% efficiency was obtained for the multi-junction polymer solar cell and 20% efficiency was obtained from every two-, three- and four-junction hybrid solar cell under 1 sun AM1.5 illumination.

PbS quantum dots (QDs) are a promising nanostructured material for solar cells. However, limited works have been done to explore the active layer thickness, layer deposition techniques, stability improvement, and cost reduction for PbS QD solar cells. We address those issues of device fabrication herein and suggest their possible solutions. In our work, to get the maximum current density from a PbS QD solar cell, we estimated the optimized active layer thickness using Matlab simulation. After that, we fabricated a high-performance and low-cost QD photovoltaic (PV) device with the simulated optimized active layer thickness. We implemented this low-cost device using a 10 mg/mL PbS concentration. Here, spin coating and drop-cast layer deposition methods were used and compared. We found that the device prepared by the spin coating method was more efficient than that by the drop cast method. The spin-coated PbS QD solar cell provided 6.5% power conversion efficiency (PCE) for the AM1.5 light spectrum. Besides this, we observed that Cr (chromium) interfaced with the Ag (Cr–Ag) electrode can provide a highly air-stable electrode.

Neural activity in the brain releases electrical impulses that can be captured using electrodes and analyzed. This signal, same as other types of real-world signals, contains noise and artifacts and needs preprocessing for better visualization and analysis. In this paper, MATLAB and specifically the EEGLAB add-on are used to perform the pre-processing and clustering on the set of data extracted from an extensive experiment.

Computer aided applications in Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) are increasingly gaining attention as important tools to asses the risk of breast cancer. Chest wall detection and whole breast segmentation require effective solutions to increase the potential benefits of computer aided tools for tumor detection. Here we propose a 3D extension of Gabor filtering for detection of wall-like regions in medical imaging, and prove its effectiveness in chest-wall detection.

In this study, a multi-stage optimization procedure is proposed to develop deep neural network models which results in a powerful deep learning pipeline called intelligent deep learning (iDeepLe). The proposed pipeline is then evaluated by a challenging real-world problem, the modeling of the spectral acceleration experienced by a particle during earthquakes. This approach has three main stages to optimize the deep model topology, the hyper-parameters, and its performance, respectively. This pipeline optimizes the deep model via adaptive learning rate optimization algorithms for both accuracy and complexity in multiple stages, while simultaneously solving the unknown parameters of the regression model. Among the seven adaptive learning rate optimization algorithms, Nadam optimization algorithm has shown the best performance results in the current study. The proposed approach is shown to be a suitable tool to generate solid models for this complex real-world system. The results also show that the parallel pipeline of iDeepLe has the capacity to handle big data problems as well.

In this paper, we present our work on Maximum Power Point Tracking (MPPT) using neural network. The MATLAB/Simulink is used to establish a model of photovoltaic array. The Simulink model is tested with different temperature and irradiation and resultant I-V and P-V characteristics proved the validation of Simulink model of PV array. We collected a set of data from the Simulink model of PV array after simulated under a range of irradiation and temperature. The data collected from the system is used to train the neural network. When we tested the neural network with different irradiance and temperature, we see that the neural network can accurately predict the maximum power point of a photovoltaic array. In this paper, the backpropagation training algorithm is used to train the neural network. Comparisons of MPPT with P & O algorithm and without MPPT tracker are also shown in this paper. It is demonstrated that the neural network based MPPT tracking require less time and provide more accurate results than the P&O algorithm based MPPT.

In this paper, we present our work on maximum power point tracking (MPPT) using neural networks. The MATLAB Simulink simulation platform is used to establish a model of photovoltaic array. The Simulink model is tested with different temperature and irradiation parameters, and the resultant I-V and P-V characteristics proved the validation of the Simulink model of PV array. In this paper, the backpropagation training algorithm is used to train a multilayer neural network using on data points collected from Simulink model of PV array simulated under a range of irradiations and temperatures. The collected data from the system is then tested by the neural network. We show that the neural network can accurately predict the maximum power point of a photovoltaic array. Comparisons between MPPT using the perturb and observe (P & O) algorithm and without MPPT are also shown in this paper. It is demonstrated that the neural network-based MPPT require less time and provide more accurate results than the P&O algorithm-based MPPT.