Model-based Optimization Research Articles

Accelerated Lifetime Model-Based Design Optimization Strategy With Efficiency, Reliability, and Cost Trade-Off for High-Power Modular AFE Rectifiers

Accelerated Lifetime Model-Based Design Optimization Strategy With Efficiency, Reliability, and Cost Trade-Off for High-Power Modular AFE Rectifiers

Data Quality-Aware Mixed-Precision Quantization via Hybrid Reinforcement Learning.

Mixed-precision quantization mostly predetermines the model bit-width settings before actual training due to the non-differential bit-width sampling process, obtaining suboptimal performance. Worse still, the conventional static quality-consistent training setting, i.e., all data is assumed to be of the same quality across training and inference, overlooks data quality changes in real-world applications which may lead to poor robustness of the quantized models. In this article, we propose a novel data quality-aware mixed-precision quantization framework, dubbed DQMQ, to dynamically adapt quantization bit-widths to different data qualities. The adaption is based on a bit-width decision policy that can be learned jointly with the quantization training. Concretely, DQMQ is modeled as a hybrid reinforcement learning (RL) task that combines model-based policy optimization with supervised quantization training. By relaxing the discrete bit-width sampling to a continuous probability distribution that is encoded with few learnable parameters, DQMQ is differentiable and can be directly optimized end-to-end with a hybrid optimization target considering both task performance and quantization benefits. Trained on mixed-quality image datasets, DQMQ can implicitly select the most proper bit-width for each layer when facing uneven input qualities. Extensive experiments on various benchmark datasets and networks demonstrate the superiority of DQMQ against existing fixed/mixed-precision quantization methods.

Residual Deep Reinforcement Learning with Model-based Optimization for Inverter-based Volt-Var Control

Residual Deep Reinforcement Learning with Model-based Optimization for Inverter-based Volt-Var Control

Optimization Assisted by Neural Network-Based Machine Learning in Electromagnetic Applications

We introduce an optimization assisted by a neural network (ONN) predictor to the electromagnetic community. ONN belongs to the class of the surrogate model-based optimization approaches, and approximates the objective function using a non-linear approximator – neural network. We provide a comprehensive description of the ONN algorithm and the details of its mathematical formulations. We apply ONN to optimize three popular benchmark functions and compare its performance with some commonly used optimization algorithms, namely particle swarm optimization, genetic algorithm, and Bayesian optimization. For the first time, we demonstrate ONN’s applicability and effectiveness in antenna design problems by optimizing a six-element Yagi-Uda antenna and by solving a challenging ten-dimensional dual-band slotted patch antenna constrained optimization problem. To achieve this, ONN is linked with a full-wave electromagnetic simulation solver through an application user interface. The optimized slotted patch is fabricated and measured to demonstrate how ONN can be part of the full antenna design process. Our empirical results indicate that ONN requires less objective function evaluations to reach the same qualitative point and reaches better optimal points for the same number of iterations for the studied benchmark functions and antenna optimization problems compared to the aforementioned baseline optimization algorithms.

Optimization-Driven DRL-Based Joint Beamformer Design for IRS-Aided ITSN Against Smart Jamming Attacks

This paper investigates an intelligent reflecting surfaces (IRS) aided anti-jamming communication strategy in the integrated terrestrial-satellite network (ITSN), where the IRS is exploited to mitigate jamming interference and enhance the integrated system communication performance. In such a network, the terrestrial network and satellite network are co-existing with a spectrum-sharing scheme in the presence of a multi-antenna jammer. We aim at maximizing the weighted sum rate (WSR) of all users by jointly optimizing the terrestrial beamformers and IRS phase shifts while considering the signal-to-interference-plus-noise ratio (SINR) requirements of legitimate users. Different from the non-convex optimization techniques utilized in the IRS-related problem, a novel optimization-driven deep reinforcement learning (DRL) algorithm is proposed, which leverages both the robustness of model-free learning approaches and the efficiency of model-based optimization methods. In the optimization module of the proposed algorithm, we analyze the smart jammer under the unknown jamming model and derive a lower bound of the anti-jamming uncertainty, such that the IRS-aided anti-jamming problem can be solved by alteration method with second-order cone programming (SOCP) algorithm and semidefinite relaxation (SDR) technique. Simulation results demonstrate that the IRS can enhance the anti-jamming performance efficiently, and the proposed optimization-driven DRL algorithm can improve both the learning rate and the system performance compared with existing solutions.

Adaptive Model-Based Reinforcement Learning for Fast-Charging Optimization of Lithium-Ion Batteries

Fast charging problem of lithium-ion batteries with minimum-charging time while limiting battery degradation, has been receiving increasing attention and is a critical challenge to battery community. Difficulties in this optimization lie in that: (i) The parameter space of charging strategies is high dimensional while the budget of the experimental cost is often limited; (ii) The evaluation of charging strategies' performance is expensive, and (iii) the degradation process of battery is strongly nonlinear and multiple degradation mechanisms occur simultaneously leading to difficulties for establishing accurate first-principles models. Current methods to address these difficulties are mainly electrochemical model-based optimization and grid search, which are rarely adaptive to battery degradation and/or are of low sample efficiency. In this work, we propose an adaptive model-based reinforcement learning (RL) approach for fast charging optimization while limiting battery degradation, in which a probabilistic surrogate model of differential Gaussian process (GP) is adopted to adaptively describe the degradation of cells. The effectiveness of the proposed approach is demonstrated on PETLION, a high-performance PET-based battery simulator. The results show that (i) compared with the model-free RL method, the proposed adaptive GP-based RL approach possesses superior charging performance and high sample efficiency, and (ii) the proposed method performs well in the handling of degradation constraints on voltage and temperature for dynamically aging batteries with its adaptability to the variations of environment.

Graph Learning-Based Voltage Regulation in Distribution Networks With Multi-Microgrids

Microgrids (MGs), as localized small power systems, can effectively provide voltage regulation services for distribution networks by integrating and managing various distributed energy resources. Existing literature employs model-based optimization approaches to formulate the voltage regulation problem of multi-MGs, which require complete system models. However, this assumption is normally impractical due to time-varying environment and privacy issues. To fill this research gap, this paper suggests a data-driven decentralized framework for the cost-effective voltage regulation of a distribution network with multi-MGs. A novel multi-agent reinforcement learning method featuring an augmented graph convolutional network and a proximal policy optimization algorithm is proposed to solve this problem. Furthermore, the techniques of critical bus and electrical distance enhance the capability of feature extractions from the distribution network, allowing for the decentralized training with privacy preserving. Simulation results based on modified IEEE 33-bus, 69-bus, and 123-bus networks are developed to validate the effectiveness of the proposed method in enabling multi-MGs to provide distribution network voltage regulation.

Multi-Scale Spatio-Temporal Memory Network for Lightweight Video Denoising.

Deep learning-based video denoising methods have achieved great performance improvements in recent years. However, the expensive computational cost arising from sophisticated network design has severely limited their applications in real-world scenarios. To address this practical weakness, we propose a multiscale spatio-temporal memory network for fast video denoising, named MSTMN, aiming at striking an improved trade-off between cost and performance. To develop an efficient and effective algorithm for video denoising, we exploit a multiscale representation based on the Gaussian-Laplacian pyramid decomposition so that the reference frame can be restored in a coarse-to-fine manner. Guided by a model-based optimization approach, we design an effective variance estimation module, an alignment error estimation module and an adaptive fusion module for each scale of the pyramid representation. For the fusion module, we employ a reconstruction recurrence strategy to incorporate local temporal information. Moreover, we propose a memory enhancement module to exploit the global spatio-temporal information. Meanwhile, the similarity computation of the spatio-temporal memory network enables the proposed network to adaptively search the valuable information at the patch level, which avoids computationally expensive motion estimation and compensation operations. Experimental results on real-world raw video datasets have demonstrated that the proposed lightweight network outperforms current state-of-the-art fast video denoising algorithms such as FastDVDnet, EMVD, and ReMoNet with fewer computational costs.

Thermal Modeling With Surrogate Model-Based Optimization of Direct Oil Cooling Heat Transfer Coefficient for HEV Motor

Thermal Modeling With Surrogate Model-Based Optimization of Direct Oil Cooling Heat Transfer Coefficient for HEV Motor

Model-based many-objective optimization for control parameters of underwater glider considering long-term high-quality CTD measurements

Currently, the optimization of underwater gliders (UGs) primarily focuses on improving the platform performance, while ignoring the incidental influence on onboard observations. In this study, a model-based many-objective optimization framework is first introduced to acquire the optimal control parameters of UGs, which enables efficient energy utilization and quality-assured conductivity, temperature, and depth (CTD) observations. Different from prior studies only emphasizing platform performance, our framework innovatively takes both CTD measurement errors and sampling errors into the optimization of control parameters, presenting a novel standpoint for achieving energy-efficient observations of interest within the dynamic ocean system. To further improve optimization in local environments, the dynamic model of UG in our framework was also upgraded through the integration of near real-time environmental information, facilitated by the Douglas–Peucker algorithm. Within the context of this framework, we investigated the requirements for conducting CTD observations in the forward and depth directions of glider movement and analyzed the optimal control parameters settings for these scenarios, encompassing both regular and segmented control strategies. The results suggest that conducting CTD observations during UG's climbing phase, rather than its diving phase, yields advantages. Otherwise, upgrading the buoyancy system and implementing segmented control strategy would be necessary. The thorough assessment in our study demonstrates the potential and versatility of the proposed framework in providing the optimal control parameters options for enhancing long-term high-quality CTD measurement by UGs. We believe that it will provide valuable guidance for glider operation in future field experiments.

Surrogate model-based multi-objective design optimization of vibration suppression effect of acoustic black holes and damping materials on a rectangular plate

The acoustic black hole (ABH) is a ground-breaking passive technology for vibration reduction and noise attenuation, characterized by its ability to manipulate flexural waves. When multiple two-dimensional ABHs are embedded in an irregular shaped complex plate structure used in engineering, such as the support plate of a refrigerator compressor, its vibration characteristics often cannot be solved through analytical or semi-analytical methods, resulting in the inability to carry out subsequent ABH optimization design. To address this issue, we propose a high-precision modelling method that combines surrogate models with finite element methods to construct a vibration characteristic database of the compressor support plate embedded with multiple ABHs in all design spaces of the ABH and damping materials geometric parameters using limited sample data. A high-precision surrogate model was constructed based on the vibration response data of the ABH plate under 50 parameter combinations of ABH and damping layer, and the fitting accuracy of the surrogate model was verified through vibration response data under another 25 parameter combinations. Based on the constructed surrogate model, the Multi-objective Genetic Algorithm (MOGA) was employed to optimize the five parameters of ABH and damping layer simultaneously, aiming to maximize the vibration reduction effect of the ABHs. Our investigation reveals that the optimized ABH and damping materials significantly improve the vibration suppression effect in a wide frequency range compared to the unoptimized structures. Finally, experimental verification was conducted to solidify the accuracy and efficacy of our approach in enhancing vibration reduction effect for irregular shaped complex plate structures using acoustic black holes.

Design and Evaluation of an Adjustable Compliant Constant-Force Microgripper.

Precise control of the manipulating force within an appropriate range is crucial to prevent potential damage to the operating object. However, achieving accurate force control through force feedback is challenging in micro-scale applications. This study presents the design of a quasi-zero stiffness-compliant constant-force microgripper with adjustable force output. The parameters of the constant-force mechanism are designed using a model-based optimization method. By utilizing this mechanism, a compliant microgripper capable of providing adjustable constant-force output is developed to overcome the limitation of traditional grippers that offer only a single constant force. Finite element analysis is performed to simulate the behavior and verify the stability of the constant-force output. Furthermore, an experimental platform is constructed to validate the mechanical properties of the developed microgripper. The experimental results demonstrate that the automatically optimized structural parameters enable the microgripper to achieve the desired constant-force value of 2 N with an adjustable range of 0.15 N. These findings provide a further basis for the application and promotion of compliant constant-force structures.

Surrogate empowered Sim2Real transfer of deep reinforcement learning for ORC superheat control

The Organic Rankine Cycle (ORC) is widely used in industrial waste heat recovery due to its simple structure and easy maintenance. However, in the context of smart manufacturing in the process industry, traditional model-based optimization control methods are unable to adapt to the varying operating conditions of the ORC system or sudden changes in operating modes. Deep reinforcement learning (DRL) has significant advantages in situations with uncertainty as it directly achieves control objectives by interacting with the environment without requiring an explicit model of the controlled plant. Nevertheless, direct application of DRL to physical ORC systems presents unacceptable safety risks, and its generalization performance under model-plant mismatch is insufficient to support ORC control requirements. Therefore, this paper proposes a Sim2Real transfer learning-based DRL control method for ORC superheat control, which aims to provide a new simple, feasible, and user-friendly solution for energy system optimization control. Experimental results show that the proposed method greatly improves the training speed of DRL in ORC control problems and solves the generalization performance issue of the agent under multiple operating conditions through Sim2Real transfer.

Learning From Images: Proactive Caching With Parallel Convolutional Neural Networks

With the continuous trend of data explosion, delivering packets from data servers to end users causes increased stress on both the fronthaul and backhaul traffic of mobile networks. To mitigate this problem, caching popular content closer to the end-users has emerged as an effective method for reducing network congestion and improving user experience. To find the optimal locations for content caching, many conventional approaches construct various Mixed Integer Linear Programming (MILP) models. However, such methods may fail to support online decision making due to the inherent curse of dimensionality. In this paper, a novel framework for proactive caching is proposed. This framework merges model-based optimization with data-driven techniques by transforming an optimization problem into a grayscale image. For parallel training and simple design purposes, the proposed MILP model is first decomposed into a number of sub-problems and, then, Convolutional Neural Networks (CNNs) are trained to predict content caching locations of these sub-problems. Furthermore, since the MILP model decomposition neglects the network resources (such as caching space and link bandwidth) competition among sub-problems, the CNNs' outputs have the risk to be infeasible solutions. Therefore, two algorithms are provided: the first uses predictions from CNNs as an extra constraint to reduce the number of decision variables; the second employs CNNs' outputs to accelerate local search. Numerical results show that the proposed scheme can reduce 71.6% computation time, whose computation time reaches around 28.9 ms, with only 0.8% additional performance cost compared to the MILP solution, which provides high quality decision making in pseudo real-time.

Optimization of diffuse Raman spectroscopy for in-vivo quantification of foreign body response in a small animal model.

Diffuse Raman spectroscopy (DRS) allows subsurface molecular analysis of optically turbid samples. Numerical modeling of light propagation was used as a method for improving the design of an DRS instrument to maximize the signal to noise ratio (SNR) while ensuring safe laser exposure parameters required for in-vivo measurements. Experimental validation of the model was performed on both phantom samples and disks implanted postmortem to mimic the typical response to foreign bodies (formation of a fibrotic capsule around an implant). A reduction of laser exposure of over 1500-fold was achieved over previous studies whilst maintaining the same Raman collection rates and reaching the safe power density of 3 mW/mm2. The validation of this approach in a subcutaneous implant in a mouse cadaver showed a further improvement of 1.5-fold SNR, with a thickness limit of detection for the fibrotic layer of 23 µm, under the same acquisition times. In the animal body, a thickness limit of detection of 16 µm was achieved. These results demonstrate the feasibility of numerical model-based optimization for DRS, and that the technique can be improved sufficiently to be used for in-vivo measurement of collagenous capsule formation as a result of the foreign body response in murine models.

Multi-planner validation of RapidPlan knowledge-based model for volumetric modulated arc therapy in prostate cancer.

To investigate the performance of a model-based optimization process for volumetric modulated arc therapy (VMAT) applied to prostate cancer patients with the multi-planner. The 120 prostate plans for VMAT treatment were entered into the database system of the RapidPlan (RP) knowledge-based treatment planning. The treatment planning data for each plan was used to create and train the RP model. Twelve prostate cancer cases were selected and were used for planning by a manual of 12 planners based on the clinical protocol for dose constraints. Then, the treatment plans for each patient were compared with the RP model plans and analyzed with Wilcoxon tests. On average, the RP models can estimate comparable doses among all planner plans and clinical plans for the PTV, which Dmax , D95% , D98% , HI, and CI were used to evaluate. For the normal organ doses of the bladder, rectum, penile bulb, and femoral head, all RP model plans showed comparable or better dose sparing than all planner plans and clinical plans. Moreover, the average planning time of the RP model was faster than manual plans by about two times. The RP model can significantly reduce the variation dose of the normal organs compared with the manual plans among the planners. The automated plans of the RP model might benefit from further fine-tuning of the dose constraints of the normal organs, although both procedure plans are acceptable and fulfill the clinical protocol goals so that the RP model can enhance the efficacy and quality of plans.

Virtual Patient Simulation Using Copula Modeling.

Virtual patient simulation is increasingly performed to support model-based optimization of clinical trial designs or individualized dosing strategies. Quantitative pharmacological models typically incorporate individual-level patient characteristics, or covariates, which enable the generation of virtual patient cohorts. The individual-level patient characteristics, or covariates, used as input for such simulations should accurately reflect the values seen in real patient populations. Current methods often make unrealistic assumptions about the correlation between patient's covariates or require direct access to actual data sets with individual-level patient data, which may often be limited by data sharing limitations. We propose and evaluate the use of copulas to address current shortcomings in simulation of patient-associated covariates for virtual patient simulations for model-based dose and trial optimization in clinical pharmacology. Copulas are multivariate distribution functions that can capture joint distributions, including the correlation, of covariate sets. We compare the performance of copulas to alternative simulation strategies, and we demonstrate their utility in several case studies. Our work demonstrates that copulas can reproduce realistic patient characteristics, both in terms of individual covariates and the dependence structure between different covariates, outperforming alternative methods, in particular when aiming to reproduce high-dimensional covariate sets. In conclusion, copulas represent a versatile and generalizable approach for virtual patient simulation which preserve relationships between covariates, and offer an open science strategy to facilitate re-use of patient data sets.

Pushing the boundaries of phosphorylase cascade reaction for cellobiose production II: Model-based multiobjective optimization.

The inherent complexity of coupled biocatalytic reactions presents a major challenge for process development with one-pot multienzyme cascade transformations. Kinetic models are powerful engineering tools to guide the optimization of cascade reactions towards a performance suitable for scale up to an actual production. Here, we report kinetic model-based window of operation analysis for cellobiose production (≥100 g/L) from sucrose and glucose by indirect transglycosylation via glucose 1-phosphate as intermediate. The two-step cascade transformation is catalyzed by sucrose and cellobiose phosphorylase in the presence of substoichiometric amounts of phosphate (≤27 mol% of substrate). Kinetic modeling was instrumental to uncover the hidden effect of bulk microviscosity due to high sugar concentrations on decreasing the rate of cellobiose phosphorylase specifically. The mechanistic-empirical hybrid model thus developed gives a comprehensive description of the cascade reaction at industrially relevant substrate conditions. Model simulations serve to unravel opposed relationships between efficient utilization of the enzymes and maximized concentration (or yield) of the product within a given process time, in dependence of the initial concentrations of substrate and phosphate used. Optimum balance of these competing key metrics of process performance is suggested from the model-calculated window of operation and is verified experimentally. The evidence shown highlights the important use of kinetic modeling for the characterization and optimization of cascade reactions in ways that appear to be inaccessible to purely data-driven approaches.

Adaptive model-based optimization for fusion-based metal additive manufacturing (directed energy deposition)

In this paper, a computationally fast adaptive optimization methodology based on a semi-analytical physics-based model is developed for predicting and altering the processing parameters to avoid overheating and improve the dimensional accuracy and uniformity of the print in fusion-based metal additive manufacturing. A unique model-based algorithm is developed, as a closed-form optimization solution, for multi-track/multi-layer printing. The algorithm uses simulated data of the temperature field and geometrical profile of tracks to back-calculate the optimal power of the thermal energy source such that overheating is neutralized, and each track arrives at the desired dimensions. The proposed methodology is adaptive as it predicts the power of the thermal energy source for subsequent tracks based on historical thermal information from previous tracks during the process, regardless of the initial processing parameters. This abstraction offers required speed-ups for computationally fast optimization. The developed adaptive methodology is experimentally tested and validated for directed energy deposition in the present study, while the methodology can be adapted to other fusion-based additive processes using different thermal energy sources such as laser beam, electron beam, and electric arc.

Developing auto process mapping technique for powder bed fusion using an electron beam

Powder bed fusion using an electron beam offers promise for manufacturing intricate metal parts. However, process optimization for defect-free parts proves costly and time-consuming. Many studies have investigated process optimization and defect prediction methods, but automating process optimization remains a significant challenge. This study developed and validated software to automatically determine i + 1-th trial conditions based on the results of the i-th trial experiment. Two algorithms were implemented and evaluated:—a dynamic programming approach and a selecting boundary conditions approach. The latter method considerably reduced the time required to determine the next conditions compared to the former approach. Considering a process mapping experiment requiring real-time trial condition determination during the build, we chose the selecting boundary conditions approach. The selecting boundary conditions approach was used to conduct a process mapping experiment to validate the software for constructing a process map using machine learning. The model and hyperparameters were optimized using sequential model-based global optimization with a tree-structured Parzen estimator. The process map underwent four updates using the developed software to determine i + 1-th trial conditions and construct a process map from the results of the i-th trial experiment.