Deep Learning-based Optimization Research Articles

Intelligent Control Framework for Improving Energy System Stability Through Deep Learning-Based Modal Optimization Scheme

Ensuring the stability of power systems is essential to promote energy sustainability. The integrated operation of these systems is critical in sustaining modern societies and economies, responding to the increasing demand for electricity and curbing environmental consequences. This study focuses on the optimization of energy system stability through the coordination of power system stabilizers (PSSs) and power oscillation dampers (PODs) in a single-machine infinite bus energy grid configuration that has flexible AC alternating current transmission system (FACTS) devices. Intelligent control strategies using PSS and POD techniques are suggested to increase power system stability and generate supplementary control signals for both the generator excitation system and FACTS device switching control. An intelligent optimal modal control framework equipped with deep learning methods is introduced to control the generator excitation system and thyristor-controlled series capacitor (TCSC). By optimally choosing the weighting matrix Q and implementing close-loop pole shifting, an optimal modal control approach is formulated. To harness its adaptive potential in fine-tuning controller parameters, an auxiliary deep learning-based optimization algorithm with actor–critic architecture is implemented. This comprehensive technique provides a promising path to effectively reduce electromechanical oscillations, thereby enhancing voltage regulation and transient stability in power systems.

CPSGD: A Novel Optimization Algorithm and Its Application in Side-Channel Analysis

In recent years, side-channel analysis based on deep learning has garnered significant attention from researchers. A pivotal reason for this lies in the fact that deep learning-based side-channel analysis requires minimal preprocessing of side-channel data. The automatic feature extraction property of deep learning methods drastically reduces the workload for researchers, enabling them to focus more on the core issues of side-channel analysis, namely, extracting sensitive information by attacking devices. However, in prior studies, most scholars have concentrated more on the model construction process, with little research focusing on the choice of optimizers.This paper explores a novel deep learning-based optimization algorithm—CPSGD (combined projection stochastic gradient descent). The algorithm comprises two variants, designed, respectively, for unprotected side-channel analysis (CPSGD1) and desynchronized side-channel analysis (CPSGD2), and their convergence has been theoretically proven. Experimental results demonstrate that, while maintaining the neural network structure unchanged, CPSGD1 exhibits the best performance on unprotected datasets compared to other publicly available optimizers, whereas CPSGD2 performs optimally on desynchronized datasets.

822: Deep learning-based optimization of field geometry for total marrow irradiation delivered with VMAT

822: Deep learning-based optimization of field geometry for total marrow irradiation delivered with VMAT

Deep learning-based optimization of field geometry for total marrow irradiation delivered with volumetric modulated arc therapy.

Total marrow (lymphoid) irradiation (TMI/TMLI) is a radiotherapy treatment used to selectively target the bone marrow and lymph nodes in conditioning regimens for allogeneic hematopoietic stem cell transplantation. A complex field geometry is needed to cover the large planning target volume (PTV) of TMI/TMLI with volumetric modulated arc therapy (VMAT). Five isocenters and ten overlapping fields are needed for the upper body, while, for patients with large anatomical conformation, two specific isocenters are placed on the arms. The creation of a field geometry is clinically challenging and is performed by a medical physicist (MP) specialized in TMI/TMLI. To develop convolutional neural networks (CNNs) for automatically generating the field geometry of TMI/TMLI. The dataset comprised 117 patients treated with TMI/TMLI between 2011 and 2023 at our Institute. The CNN input image consisted of three channels, obtained by projecting along the sagittal plane: (1) average CT pixel intensity within the PTV; (2) PTV mask; (3) brain, lungs, liver, bowel, and bladder masks. This "averaged" frontal view combined the information analyzed by the MP when setting the field geometry in the treatment planning system (TPS). Two CNNs were trained to predict the isocenters coordinates and jaws apertures for patients with (CNN-1) and without (CNN-2) isocenters on the arms. Local optimization methods were used to refine the models output based on the anatomy of the patient. Model evaluation was performed on a test set of 15 patients in two ways: (1) by computing the root mean squared error (RMSE) between the CNN output and ground truth; (2) with a qualitative assessment of manual and generated field geometries-scale: 1=not adequate, 4=adequate-carried out in blind mode by three MPs with different expertise in TMI/TMLI. The Wilcoxon signed-rank test was used to evaluate the independence of the given scores between manual and generated configurations (p<0.05 significant). The average and standard deviation values of RMSE for CNN-1 and CNN-2 before/after local optimization were 15±2/13±3mm and 16±2/18±4mm, respectively. The CNNs were integrated into a planning automation software for TMI/TMLI such that the MPs could analyze in detail the proposed field geometries directly in the TPS. The selection of the CNN model to create the field geometry was based on the PTV width to approximate the decision process of an experienced MP and provide a single option of field configuration. We found no significant differences between the manual and generated field geometries for any MP, with median values of 4versus 4 (p=0.92), 3versus 3 (p=0.78), 4versus 3 (p=0.48), respectively. Starting from October 2023, the generated field geometry has been introduced in our clinical practice for prospective patients. The generated field geometries were clinically acceptable and adequate, even for an MP with high level of expertise in TMI/TMLI. Incorporating the knowledge of the MPs into the development cycle was crucial for optimizing the models, especially in this scenario with limited data.

Smart Health Monitoring System of Agricultural Machines: Deep Learning-based Optimization with IoT and AI

Implementing intelligent monitoring systems for Agricultural Machinery (AM) is hindered by the intricate and costly nature of the Internet of Things (IoT) sensor technologies. The heavy reliance on cloud and fog computing, the availability of network infrastructure, and the need for expert knowledge pose challenges in rural areas that lack network connectivity. Using edge devices, such as smartphones, which possess significant computational capabilities, is a potential solution that has not yet been fully realized in the commercial sphere. Furthermore, the increasing demand from users for economically viable and user-friendly technology serves as a driving force for transitioning away from expensive and intricate sensors towards more cost-effective alternatives. In the IoT era, there is anticipated to be a widespread network connection between a vast array of AM and service centers. Using smartphone applications has increased the potential for edge computation on smartphones to significantly aid in network traffic control. The development of an Artificial Intelligence (AI) - -based data analytic method poses a significant challenge due to the need to optimize for the limited computational capabilities of smartphones. However, the users’ demand for affordable technology renders it resistant to easy penetration. This paper uses IoT and AI to propose a Smart Health Monitoring System for Agricultural Machines with Deep Learning-based Optimization (SHMAM-DLO). This paper aims to propose a Fusion Genetic Algorithm (FGA) methodology and Artificial Neural Network (ANN) for optimization during monitoring the health of AM. The proposed approach enables cost-effective utilization of smartphone end devices by leveraging their built-in microphones instead of relying on expensive IoT sensors.

Deep learning-based optimization method for detecting data anomalies in power usage detection devices

Abstract In this paper, the self-attention layer of a graph convolutional neural network is first constructed to output the important information in the network structure. The migration learning network model is established, and the sample data are preprocessed and trained sequentially. The final processing results are used as the initial data for abnormal power consumption detection. Introduce Bayes’ theorem to optimize the hyperparameters of the model. The optimized model is applied in the abnormal power consumption detection system to identify abnormal power consumption events and provide specific processing solutions. Through the detection of the system, it was found that the voltage of the test user dropped from a 100V cliff to about 20V in late November, which was determined by the system to be a power consumption abnormality, and, therefore, an operation and maintenance order was issued. The site survey revealed that the data was in line with the system detection. Calculating the power consumption information of another user, the phase voltage of this user stays around 85-100V, far below 150V, so the undercounting of power is verified for the user, and the amount of power that should be recovered is 201.22kW.

Taming Connectedness in Machine-Learning-Based Topology Optimization with Connectivity Graphs

Despite the remarkable advancements in machine learning (ML) techniques for topology optimization, the predicted solutions often overlook the necessary structural connectivity required to meet the load-carrying demands of the resulting designs. Consequently, these predicted solutions exhibit subpar structural performance because disconnected components are unable to bear loads effectively and significantly compromise the manufacturability of the designs.In this paper, we propose an approach to enhance the topological accuracy of ML-based topology optimization methods by employing a predicted dual connectivity graph. We show that the tangency graph of the Maximal Disjoint Ball Decomposition (MDBD), which accurately captures the topology of the optimal design, can be used in conjunction with a point transformer network to improve the connectivity of the design predicted by Generative Adversarial Networks and Convolutional Neural Networks. Our experiments show that the proposed method can significantly improve the connectivity of the final predicted structures. Specifically, in our experiments the error in the number of disconnected components was reduced by a factor of 4 or more without any loss of accuracy. We demonstrate the flexibility of our approach by presenting examples including various boundary conditions (both seen and unseen), domain resolutions, and initial design domains. Importantly, our method can seamlessly integrate with other existing deep learning-based optimization algorithms, utilize training datasets with models using any valid geometric representations, and naturally extend to three-dimensional applications.

Crack segmentation of imbalanced data: The role of loss functions

Loss functions, which govern a deep learning-based optimization process, have been widely investigated to handle the class imbalanced data issue in crack segmentation. However, their performance varies according to models and datasets, making it challenging to choose the most appropriate ones. To address this issue, the paper conducts a large-scale performance comparison of twelve commonly used loss functions on four benchmark datasets. Various aspects are considered using a statistical test-based ranking scheme, which integrates accuracy, sensitivity to threshold change, and varying degrees of imbalance severity for a comprehensive comparison. The results show that most loss functions achieve relatively similar accuracy on mildly imbalanced datasets, while weighted binary cross-entropy loss, Focal loss, Dice-based loss, and compound loss functions significantly outperform others as imbalance severity increases. In general, Focal Tversky loss function exhibits excellent performance in handling the imbalanced data issue.

Fast topology optimization of phononic crystal‐based metastructures for vibration isolation by deep learning

AbstractA novel deep learning‐based optimization (DLBO) methodology is proposed for rapidly optimizing phononic crystal‐based metastructure topologies. DLBO eliminates the need for pre‐optimized data by leveraging the learned relation from metastructure features to bandgaps. It enables optimization based on qualitative/quantitative descriptions and forms a regular generalization domain to avoid misjudgments. DLBO achieves similar or better results to genetic algorithm (GA) and only requires 0.01% of the time GA costs. Metastructures with different periodic constants and filling fractions are also optimized, offering insights for balancing space, material, and vibration isolation. Based on a newly defined objective function, an economical metastructure is customized for subway‐induced vibrations; and its performance on vibration isolation is verified through a 3D finite element model. Additionally, the datasets and codes in this study are shared.

Proactive Resource Request for Disaster Response: A Deep Learning-Based Optimization Model

In the realm of disaster response operations, effective resource management is crucial. This research introduces an innovative approach that proactively determines the optimal quantities of resources that should be requested by local agencies. This determination is based on both current and anticipated demands, thereby ensuring a more efficient and effective response to disasters. The approach first utilizes a method that combines deep learning and temporal point process for predicting irregularly spaced future demands, and then, it formulates the resource allocation problem faced with randomly arrived demands as a stochastic optimization model. The superiority of this approach over existing resource allocation methods is demonstrated using both real-world data and simulated scenarios. The findings highlight the need for a shift from reactive to proactive strategies. Moreover, the research emphasizes the potential of advanced techniques, such as deep learning and stochastic optimization, in disaster management. These techniques can provide valuable tools for policy makers and practitioners in the field, enabling them to make more informed and effective decisions. Policies that encourage the adoption of such optimized resource allocation strategies could lead to more effective disaster response operations.

Designing directional adhesive pillars using deep learning-based optimization, 3D printing, and testing

Nature-inspired fibrillar adhesives have versatile applications, such as wall-climbing robots and grippers, which require residue-free and repeatable adhesion. To meet the requirements of excellent adhesion strength combined with controllability of adhesion and detachment, directional adhesive pillars with anisotropic adhesion properties have been extensively investigated. However, existing designs that simply mimic nature suffer from relatively weak adhesion and insufficient directionality, because most of them were designed without considering a sufficiently large design space. In this study, we rigorously defined adhesive directionality and systematically investigated the optimal pillar shape to obtain an excellent combination of adhesion strength and directionality using deep-learning-based optimization. A data-driven model based on artificial neural networks was trained using finite element analysis data obtained from 199,466 adhesive pillar shapes, which can predict the directionality and adhesion strength of given pillar shapes. Optimization was performed using the trained neural network to obtain the optimal pillar shape under the geometric constraints necessary to secure the reliability of the pillar. Finally, we suggest adhesive pillar shapes having severe directionality and high adhesive strength at the same time. For validation of our optimization model and optimized result, the proposed pillar shapes were fabricated using a 3D polyjet printer, and their adhesion strength and directionality were tested. We noticed that optimized pillar shapes we obtained have superior directionality and adhesion strength in real.

Enhancing Friction Stir Welding in Fishing Boat Construction through Deep Learning-Based Optimization

In the present study, the authors have attempted to present a novel approach for the prediction, analysis, and optimization of the Friction Stir Welding (FSW) process based on the Deep Neural Network (DNN) model. To obtain the DNN structure with high accuracy, the most focus has been on the number of hidden layers and the activation functions. The DNN was developed by a small database containing results of tensile and hardness tests of welded 7075-T6 aluminum alloy. This material and the production method were selected based on the application in the construction of fishing boat flooring, because on the one hand, it faces the corrosion caused by proximity to sea water and on the other hand, due to direct contact with human food, i.e., fish etc., antibacterial issues should be considered. All the major parameters of the FSW process, including axial force, rotational speed, and traverse speed as well as tool diameter and tool hardness, were considered to investigate their correspondence effects on the tensile strength and hardness of welded zone. The most important achievement of this research showed that by using SAE for pre-training of neural networks, higher accuracy can be obtained in predicting responses. Finally, the optimal values for various welding parameters were reported as rotational speed: 1600 rpm, traverse speed: 65 mm/min, axial force: 8 KN, shoulder and pin diameters: 15.5 and 5.75 mm, and tool hardness: 50 HRC.

Optimizing Machine Learning Algorithms for Landslide Susceptibility Mapping along the Karakoram Highway, Gilgit Baltistan, Pakistan: A Comparative Study of Baseline, Bayesian, and Metaheuristic Hyperparameter Optimization Techniques.

Algorithms for machine learning have found extensive use in numerous fields and applications. One important aspect of effectively utilizing these algorithms is tuning the hyperparameters to match the specific task at hand. The selection and configuration of hyperparameters directly impact the performance of machine learning models. Achieving optimal hyperparameter settings often requires a deep understanding of the underlying models and the appropriate optimization techniques. While there are many automatic optimization techniques available, each with its own advantages and disadvantages, this article focuses on hyperparameter optimization for well-known machine learning models. It explores cutting-edge optimization methods such as metaheuristic algorithms, deep learning-based optimization, Bayesian optimization, and quantum optimization, and our paper focused mainly on metaheuristic and Bayesian optimization techniques and provides guidance on applying them to different machine learning algorithms. The article also presents real-world applications of hyperparameter optimization by conducting tests on spatial data collections for landslide susceptibility mapping. Based on the experiment's results, both Bayesian optimization and metaheuristic algorithms showed promising performance compared to baseline algorithms. For instance, the metaheuristic algorithm boosted the random forest model's overall accuracy by 5% and 3%, respectively, from baseline optimization methods GS and RS, and by 4% and 2% from baseline optimization methods GA and PSO. Additionally, for models like KNN and SVM, Bayesian methods with Gaussian processes had good results. When compared to the baseline algorithms RS and GS, the accuracy of the KNN model was enhanced by BO-TPE by 1% and 11%, respectively, and by BO-GP by 2% and 12%, respectively. For SVM, BO-TPE outperformed GS and RS by 6% in terms of performance, while BO-GP improved results by 5%. The paper thoroughly discusses the reasons behind the efficiency of these algorithms. By successfully identifying appropriate hyperparameter configurations, this research paper aims to assist researchers, spatial data analysts, and industrial users in developing machine learning models more effectively. The findings and insights provided in this paper can contribute to enhancing the performance and applicability of machine learning algorithms in various domains.

A highly sensitive multi-channel SPR-PCF based biosensor with deep learning prediction approach

A very sensitive multi-channel refractive index-based biosensor with a detection range of 1.26 to 1.36 has been proposed. Titanium oxide (TiO2) was utilized as the dielectric, and gold as the plasmonic material in the sensor. The suggested sensor achieved a maximum sensitivity value of 48,000 nm/RIU and 7220 RIU-1 using the wavelength and amplitude interrogation techniques with a pitch layer variation of 5.55–5.95 μm and a metal layer variation of 30–50 nm. The suggested sensor can detect many compounds with amazing sensitivity, such as glucose, sucrose, ethanol, methanol, human intestinal mucosa, sevoflurane, and bio-chemicals. Additionally, a deep learning model that forecasts six optical properties has been proposed. The hyperparameters of the deep neural network model are tuned extensively to maximize accuracy. Our model had a Mean Squared Error of 0.006 and was significantly faster than conventional techniques. The results show that deep learning can predict optical features in photonic sensors, which can identify specific analytes or biomolecules in a sample without costly and time-consuming simulations. Photonic sensor researchers and engineers may utilize the suggested model on vast datasets and adjust it to new settings. This study may open the door to the use of deep learning-based optimization techniques to enhancebiosensors.

Fast optimization of multichip modules using deep learning coupled with Bayesian method

In this study, we develop an approach based on deep learning and the Bayesian method for fast optimization of the thermal placement of the multichip modules (MCMs). Specifically, a deep learning-based reduced-order model (ROM) is proposed to replace the CFD model to accelerate the optimization. The ROM applies the convolutional neural network (CNN) as the coding layers, the deconvolutional CNN as the decoding layers, and the signed distance function (SDF) as the input of the ROM network. It enables fast and accurate prediction of the steady temperature field and the junction temperature of each chip with various chip numbers, chip powers, and chip placements. According to the results, the average error of the temperature field is only 0.81%, and the time cost for each temperature field prediction is less than 0.01 s, which is three orders faster than the numerical simulations. Profiting from the excellent performance of the ROM, the Bayesian optimization speeds up two orders of magnitude, and the coefficient of determination, R2, reaches 0.9991, which shows the huge potential of the deep learning-based optimization method to achieve the fast and accurate thermal placement design of the MCMs.

Deep Learning-Based Optimization of Visual–Auditory Sensory Substitution

Visual–auditory sensory substitution systems can aid blind people in traveling to various places and recognizing their own environments without help from others. Although several such systems have been developed, they are either not widely used or are limited to laboratory-scale research. Among various factors that hinder the widespread use of these systems, one of the most important issues to consider is the optimization of the algorithms for sensory substitution. This study is the first attempt at exploring the possibility of using deep learning for the objective quantification of sensory substitution. To this end, we used generative adversarial networks to investigate the possibility of optimizing the vOICe algorithm, a representative visual–auditory sensory substitution method, by controlling the parameters of the method for converting an image to sound. Furthermore, we explored the effect of the parameters on the conversion scheme for the vOICe system and performed frequency-range and frequency-mapping-function experiments. The process of sensory substitution in humans was modeled to use generative models to assess the extent of visual perception from the substituted sensory signals. We verified the human-based experimental results against the modeling results. The results suggested that deep learning could be used for evaluating the efficiency of algorithms for visual–auditory sensory substitutions without labor-intensive human behavioral experiments. The introduction of deep learning for optimizing the visual–auditory conversion method is expected to facilitate studies on various aspects of sensory substitution, such as generalization and estimation of algorithm efficiency.

Deep learning-based optimization of a hydrogen and oxygen production system for patients in hospital with alkaline electrolyzer

The hybrid renewable system's potential to create standard type E capsules for Covid-19 patients was explored in this study. In addition to delivering the requisite energy to the building, standard oxygen capsules were produced using the electrolysis of water using nanomaterial-supported electrolysis in the hydrogen storage system. In addition to the simulation, multi-objective optimization was done using a deep learning neural network and a genetic algorithm to maximize the number of oxygen capsules generated in a year and the system price, and the system's front beam was acquired. The system can produce 19530 units of type E oxygen capsules in a year, and the price of the electrolyzer and fuel cell is 120296 Euros at the best point of the front beam, considering both the objective variable of price and the number of produced oxygen capsules. In this scenario, the electrolyzer and fuel cell have rated powers of 61.9 kW and 15.3 kW, respectively. After determining the optimal point, researchers investigatedthe connection between meteorological data and other system characteristics including the amount of hydrogen in the tank, the number of oxygen capsules generated each hour, fuel cell power, and the electrolyzer. Lastly, the system's capacity to lower the amount of power required for the office building from the municipal network was investigated, indicating the system's excellent capability in this respect.

A deep learning-based optimization framework of two-dimensional hydrofoils for tidal turbine rotor design

Convolutional Neural Network (CNN) is a commonly used deep learning algorithm due to its excellent capability in identification of structural features and parameter predictions in many domains. In addition, it has incomparable advantages of high analysis efficiency and generalization performance. However, it has been questioned in the research community on whether CNN method can be applied to effectively predict hydrofoil performance for hydraulic machinery design. To this end, this paper demonstrates a novel optimization platform using CNN for hydrofoil performance prediction, which can effectively and accurately obtain the optimized hydrofoils results in aid of the structural design of tidal turbine. The prediction model uses signed distance function (SDF) to graphically represent the shape of the hydrofoil which is subsequently imported into CNN as the network input. Three different hydrofoil performance properties including the lift coefficient, drag coefficient and pressure coefficient of surface are used as output to train the neural network. In order to guarantee the accuracy of the forecasting model, Computational Fluid Dynamics (CFD) method characterized by high precision is applied to generate the dataset for neural network training. The results show that it can accurately predict the hydrodynamic parameters at a lower angle of attack with extremely short period of time. On top of the established hydrofoil performance prediction model, the Pareto curve of the optimized hydrofoils is obtained and applied to the design of 3D horizontal axis tidal turbine (HATT) blades. It proves that the optimization platform is effective and versatile in a manner that achieves both accurate and rapid prediction/optimization of the hydrofoil, which greatly facilitates to apply it for the tidal turbine rotor design.

Deep learning-based optimization using a limited number of experiments for roll-to-roll UV embossing

Deep learning-based optimization using a limited number of experiments for roll-to-roll UV embossing

Model-Driven Learning for Generic MIMO Downlink Beamforming With Uplink Channel Information

Accurate downlink channel information is crucial to the beamforming design, but it is difficult to obtain in practice. This paper investigates a deep learning-based optimization approach of the downlink beamforming to maximize the system sum rate, when only the uplink channel information is available. Our main contribution is to propose a model-driven learning technique that exploits the structure of the optimal downlink beamforming to design an effective hybrid learning strategy with the aim to maximize the sum rate performance. This is achieved by jointly considering the learning performance of the downlink channel, the power and the sum rate in the training stage. The proposed approach applies to generic cases in which the uplink channel information is available, but its relation to the downlink channel is unknown and does not require an explicit downlink channel estimation. We further extend the developed technique to massive multiple-input multiple-output scenarios and achieve a distributed learning strategy for multicell systems without an inter-cell signalling overhead. Simulation results verify that our proposed method provides the performance close to the state of the art numerical algorithms with perfect downlink channel information and significantly outperforms existing data-driven methods in terms of the sum rate.