Reduction Of Optimization Time Research Articles

Deep learning-based voxel sampling for particle therapy treatment planning

Objective. Scanned particle therapy often requires complex treatment plans, robust optimization, as well as treatment adaptation. Plan optimization is especially complicated for heavy ions due to the variable relative biological effectiveness. We present a novel deep-learning model to select a subset of voxels in the planning process thus reducing the planning problem size for improved computational efficiency. Approach. Using only a subset of the voxels in target and organs at risk (OARs) we produced high-quality treatment plans, but heuristic selection strategies require manual input. We designed a deep-learning model based on P-Net to obtain an optimal voxel sampling without relying on patient-specific user input. A cohort of 70 head and neck patients that received carbon ion therapy was used for model training (50), validation (10) and testing (10). For training, a total of 12 500 carbon ion plans were optimized, using a highly efficient artificial intelligence (AI) infrastructure implemented into a research treatment planning platform. A custom loss function increased sampling density in underdosed regions, while aiming to reduce the total number of voxels. Main results. On the test dataset, the number of voxels in the optimization could be reduced by 84.8% (median) at <1% median loss in plan quality. When the model was trained to reduce sampling in the target only while keeping all voxels in OARs, a median reduction up to 71.6% was achieved, with 0.5% loss in the plan quality. The optimization time was reduced by a factor of 7.5 for the total AI selection model and a factor of 3.7 for the model with only target selection. Significance. The novel deep-learning voxel sampling technique achieves a significant reduction in computational time with a negligible loss in the plan quality. The reduction in optimization time can be especially useful for future real-time adaptation strategies.

Parallel Genetic Algorithm Interface II: A novel computational tool for accelerated simulation-based optimization

The ever increasing power of computational tools encouraged by the general need for development of more sustainable technologies fuels the interest in modern optimization approaches. While simulation-based optimization has been receiving considerable attention in the past decades, it still struggles to overcome some challenges, namely excessive computation time. This study proposes a novel optimization interface, the Parallel Genetic Algorithm Interface II (PAGAN-II), which utilizes parallelization of flowsheet simulations to drastically reduce the optimization time without the need to use clustered CPUs and/or modified optimization algorithms. Results of a detailed performance study showed up to 2100% increase in computation rate when optimizing demanding process flowsheets; and approximately 300% increase when optimizing simple ones. Capabilities of the proposed interface were demonstrated by optimization of a 5 MTPA C3MR LNG technology processing 12 different feedstocks, where a 15–30% decrease in the specific energy consumption was achieved. At the same time, the algorithm increased the optimization speed 13-fold compared to the traditional approach. This translates into a reduction of optimization time from 69 days of non-stop computation to approximately 7 days.

Multi-objective optimization of reservoir development strategy with hybrid artificial intelligence method

Optimization of subsurface hydrocarbon production holds paramount importance for decision-makers as it determines crucial development strategies such as optimal well placement and well control parameters (e.g. injection/production rates of injectors and producers). Despite the availability of numerous established optimization methods in this field, traditional reservoir production optimization methods face challenges in simultaneously addressing multiple development objectives and coordinating the interaction of well control parameters in different control steps. In this work, we construct a hybrid artificial intelligence method to jointly optimize well placement and well control parameters, taking into account two development objectives and dynamic optimization. It consists of two stages. First, a reservoir potential map is generated with the production potential formula that considers reservoir pressure, remaining oil saturation and other reservoir properties (permeability, hydrocarbon column height) etc. The reservoir potential map provides guidance for placing well in medium to high potential areas and engineering constraints for the optimization process. Then, a hybrid artificial intelligence method that couples deep learning method (Long and Short Term Memory (LSTM)) and multi-objective optimization algorithm (Non-dominated Sorting Genetic Algorithm II (NSGA-II)) is established to seek a compromise between the two objectives in water-flooding processes. The LSTM neural network is trained as the surrogate model to replace the high-fidelity simulator to achieve high efficiency of overall optimization workflow. The NSGA-II algorithm is employed for handling the joint optimization problem of well placement and well control parameters by maximizing the cumulative oil production and minimizing the water cut. The performance of the proposed method is tested on one benchmark function and two reservoir models. On the 2D synthetic reservoir model the optimized scheme leads to a notable increase of 3 × 104 m3 in cumulative oil production, accompanied by 17 % reduction in water cut when contrasted with the base scheme. Similarly, within the 3D reservoir model, the optimized scheme results in a substantial enhancement, boosting cumulative oil production by 14 × 104 m3 and reducing water cut by 20% compared to the base scheme. Moreover, the proposed method surpasses alternative multi-objective optimization (MOO) algorithms by demonstrating 82% and 95% reduction in optimization time, respectively. The results demonstrate that this method can provide the optimal water-flooding strategies under the premise of different development objectives. The Pareto front (or optimal solutions) generated by the hybrid method offers a variety of diverse water-flooding strategies to assist subsurface engineers in making informed decisions.

Optimal design and operation of reactive extrusion processes: Modeling, model validation, and optimization algorithm

AbstractIn this work, a methodology for the optimal design and operation of reactive extrusion processes in twin‐screw extruders is presented. As a process model, a very accurate mechanistic finite volume twin‐screw extruder model is used which is adapted to the reactive extrusion case. In the model, in contrast to other models, the pressure is described as a differential state in order to have a consistent set of equations. The model is compared to other models and advantages and disadvantages with respect to accuracy and computation time are discussed. For optimization, a memetic optimization algorithm (MA) that consists of a combination of a local optimization for the continuous operating parameters and an evolutionary algorithm (EA) for the global optimization of the discrete decision variables is employed to overcome the drawback of the longer model simulation time compared to simpler models. A reduction of optimization time by 90% for the MA compared to the EA was observed. The presented modeling and optimization methodology is validated on the widely investigated reactive extrusion process for ε‐Caprolactone. The methodology is very flexible and the underlying model has good prediction characteristics, thus it is applicable to a wide range of reactive extrusion processes of industrial interest.Highlights Simulation and optimization of reactive extrusion. Improved mechanistic twin‐screw extruder model. Novel approaches for pressure and residence time modeling. Efficient combination of local and global optimization in a memetic algorithm. Benchmarking of the optimization results and performance with literature data.

Use of the mCherry fluorescent protein to optimize the expression of class I lanthipeptides in Escherichia coli

BackgroundLanthipeptides are a rapidly expanding family of ribosomally synthesized and post-translationally modified natural compounds with diverse biological functions. Lanthipeptide structural and biosynthetic genes can readily be identified in genomic datasets, which provides a substantial repository for unique peptides with a wide range of potentially novel bioactivities. To realize this potential efficiently optimized heterologous production systems are required. However, only a few class I lanthipeptides have been successfully expressed using Escherichia coli as heterologous producer. This may be attributed to difficulties experienced in the co-expression of structural genes and multiple processing genes as well as complex optimization experiments.ResultsHere, an optimized modular plasmid system is presented for the complete biosynthesis for each of the class I lanthipeptides nisin and clausin, in E. coli. Genes encoding precursor lanthipeptides were fused to the gene encoding the mCherry red fluorescent protein and co-expressed along with the required synthetases from the respective operons. Antimicrobially active nisin and clausin were proteolytically liberated from the expressed mCherry fusions. The mCherry-NisA expression system combined with in vivo fluorescence monitoring was used to elucidate the effect of culture media composition, promoter arrangement, and culture conditions including choice of growth media and inducer agents on the heterologous expression of the class I lanthipeptides. To evaluate the promiscuity of the clausin biosynthetic enzymes, the optimized clausin expression system was used for the heterologous expression of epidermin.ConclusionWe succeeded in developing novel mCherry-fusion based plug and play heterologous expression systems to produce two different subgroups of class I lanthipeptides. Fully modified Pre-NisA, Pre-ClausA and Pre-EpiA fused to the mCherry fluorescence gene was purified from the Gram-negative host E. coli BL21 (DE3). Our study demonstrates the potential of using in vivo fluorescence as a platform to evaluate the expression of mCherry-fused lanthipeptides in E. coli. This allowed a substantial reduction in optimization time, since expression could be monitored in real-time, without the need for extensive and laborious purification steps or the use of in vitro activity assays. The optimized heterologous expression systems developed in this study may be employed in future studies for the scalable expression of novel NisA derivatives, or novel genome mined derivatives of ClausA and other class I lanthipeptides in E. coli.

Optimization of Tuned Mass Damper for Submerged Floating Tunnel with Frequency-Domain Dynamics Simulation

In this study, the Tuned Mass Damper (TMD) optimization is carried out to reduce the resonant motion of Submerged Floating Tunnel (SFT) under wave excitations. The SFT dynamics is evaluated in frequency domain; a new approach to cost-effectively optimizing TMD parameters for a moored system is suggested. Discrete-Module-Beam (DMB) method is used to model the Tunnel; mooring lines are included as equivalent stiffness matrix through static-offset tests by the fully coupled model. Since the frequency-domain dynamics simulation model is employed, a significant reduction in optimization time can be achieved. TMD is installed at the tunnel’s mid-length to mitigate the lateral motion of the Tunnel and coupled with the Tunnel with translational and rotational springs and dampers. The optimization process for TMD parameters is performed through the Genetic Algorithm (GA). The GA generates the TMD mass and spring and damping coefficients. The dynamics simulation is performed under wave conditions and this process is repeated until the stopping criteria is satisfied. Results demonstrate that TMD with optimized parameters significantly reduces the lateral motion, especially near the system’s lowest lateral natural frequency. This frequency-domain optimization also works as intended with significantly decreased optimization time.

Reducing MRI-guided radiotherapy planning and delivery times via efficient leaf sequencing and segment shape optimization algorithms

Objective. Extended treatment session times are an operational limitation in magnetic resonance imaging guided adaptive radiotherapy (MRIgRT). In this study a novel leaf sequencing algorithm called optimal fluence levels (OFL) and an optimization algorithm called pseudo gradient descent (PGD) are evaluated with respect to plan quality, beam complexity, and the ability to reduce treatment session times on the Elekta Unity MRIgRT system. Approach. Ten total patients were evaluated on this Institutional Review Board approved study: three with prostate cancer, three with oligometastases, two with pancreatic cancer, and two with liver cancer. Plans were generated using the clinical Monaco Hyperion optimizer and leaf sequencer and then re-optimized using OFL and PGD (OFL + PGD) while holding all IMRT constraints and planning parameters constant. All plans were normalized to ensure 95% of the PTV received the prescription dose. A paired t-test was used to evaluate statistical significance. Main Results. Plan quality in terms of dosimetric OAR sparing was found to be equivalent between the OFL + PGD and conventional Monaco Hyperion optimizer plans. The OFL + PGD plans had a reduction in optimization time of 51.4% ± 5.0% (p = 0.002) and reduction in treatment delivery time of 10.6% ± 7.5% (p = 0.005). OFL + PGD generated plans had on average 13.2% ± 12.6% fewer multi-leaf collimator (MLC) segments (p = 0.009) and 0.1 ± 0.1 lower plan averaged beam modulation (PM) (p = 0.004) relative to the Monaco Hyperion plans. Significance. The OFL + PGD algorithms more quickly generate Unity treatment plans that are faster to deliver than with the conventional approach and without compromising dosimetric plan quality. This is likely due to a delivery complexity reduction enabled by OFL + PGD relative to the Monaco Hyperion plans.

Fast 3D movement of a laser focusing spot behind scattering media by utilizing optical memory effect and optical conjugate planes

Controlling light propagation intentionally through turbid media such as ground glass or biological tissue has been demonstrated for many useful applications. Due to random scattering effect, one of the important goals is to draw a desired shape behind turbid media with a swift and precise method. Feedback wavefront shaping method which is known as a very effective approach to focus the light, is restricted by slow optimization process for obtaining multiple spots. Here we propose a technique to implement feedback wavefront shaping with optical memory effect and optical 4f system to speedy move focus spot and form shapes in 3D space behind scattering media. Starting with only one optimization process to achieve a focusing spot, the advantages of the optical configuration and full digital control allow us to move the focus spot with high quality at the speed of SLM frame rate. Multiple focusing spots can be achieved simultaneously by combining multiple phase patterns on a single SLM. By inheriting the phase patterns in the initial focusing process, we can enhance the intensity of the focusing spot at the edge of memory effect in with 50% reduction in optimization time. With a new focusing spot, we have two partially overlapped memory effect regions, expanding our 3D scanning range. With fast wavefront shaping devices, our proposed technique could potentially find appealing applications with biological tissues.

A Simplified Model for Predicting the Installed Performance of Circularly Polarized Antennas

We propose a simplified equivalent model for predicting the installed performance of circularly polarized (CP) antennas. We use a set of infinitesimal sources to represent the irradiated fields of an antenna of interest; however, instead of using infinitesimal dipoles, our source is composed of a set of infinitesimal crossed dipoles, the most basic radiator that generates CP waves. This choice facilitates the optimization process and still maintains the analytical aspect of the model; the irradiated fields can be described analytically. We achieve a reduction in computational cost when comparing to direct modeling, without impairing the accuracy of the model during the antenna placement task, and a reduction of optimization time when comparing to classical dipole models. Also, the equivalent model building step has reduced a computational cost for CP antennas. In order to validate our model, we construct a scaled model of a circular cylinder with a top-mounted GPS antenna, and we measure its radiation pattern in an anechoic chamber. The experimental results are in remarkable agreement with those simulated using the proposed equivalent model.

TH‐A‐213AB‐01: A Novel Method to Accelerate Optimization by Employing Approximate Dose Values

Purpose: To introduce a novel method that increases the efficiency of beam and beam let weight and fluence optimization, by employing approximate dose values in the objective function. Methods: A least squares objective function was used, based on clinical dose‐volume objectives. In the initial stage of optimization, the method utilizes dose values determined by the additive dose approximation (ADA) to evaluate the objective function. These dose values are based on pre‐computed dose indices, along with a novel correction scheme. During this stage, no calculation of the true dose is required, resulting in a significant reduction in computation time, since the manipulation of large dose matrices is avoided. The resulting optimized parameters are subsequently used as the starting point in the final optimization step. This method was evaluated on prostate bed, esophageal, and brain patients with varying numbers of treatment fields, by comparing the CPU time required with that of a standard optimization method. The downhill simplex algorithm was used for the evaluations. Results: Nine prostate bed patients with three different beam configurations were used: 4 field 3D‐CRT plans, 16 field 3D‐CRT plans and a 64‐beamlet IMRT‐like plan. These demonstrated an average reduction in optimization time of 33%, 54%, and 46%, respectively. Two 3D‐CRT esophageal patients were evaluated, one with 14 fields and one with 16 fields, resulting in an average improvement of 29%. Finally, a 3D‐CRT 16 field brain patient was studied and exhibited a 57% decrease in CPU time. Conclusions: The proposed method, when used in conjunction with existing optimization algorithms, dramatically decreases the number of standard objective function evaluations and, as a result, computation time. This technique may be applied to any type of optimization algorithm and set of clinical variables requiring pre‐computed dose matrices and an objective function involving dose‐volume objectives (eg. multi‐criteria optimization).

Per-flow optimal service selection for Web services based processes

With the development of the Service-Oriented Computing (SOC) paradigm, flexible business processes can be defined from independently developed services. Multiple services corresponding to the same functionality but characterized by different Quality of Service (QoS) attributes can be offered by different service providers and the best set of Web services can be selected at run-time in order to maximize the QoS for end users. In the literature many approaches have been proposed for the optimal service selection which is usually performed on a per-request basis, i.e., considering a single process invocation. In this paper we propose a broker-based framework which solves the optimal service selection on a per-flow basis. Multiple applications, defined as different BPEL processes are considered at the same time and multiple requests to the same process are optimized concurrently. Service selection is formulated as a constrained non-linear multi-criteria optimization problem and an heuristic algorithm is proposed to determine a scalable and efficient solution. A comparison with top performing state-of-the-art approaches for a number of different scenarios of interest is also provided. Results show that the overall optimization time reduction which can be achieved by our solution is proportional to the system incoming workload. Order of magnitudes optimization time improvements can be obtained if compared to alternative methods. Moreover, our solution is robust to workload prediction uncertainty.

Design of Fast and Efficient Energy-Aware Gradient-Based Scheduling Algorithms Heterogeneous Embedded Multiprocessor Systems

In this paper, we present two heuristic energy-aware scheduling algorithms (EGMS and EGMSIV) for scheduling task precedence graphs in an embedded multiprocessor system having processing elements with dynamic voltage scaling capabilities. Unlike most energy-aware scheduling algorithms that consider task ordering and voltage scaling separately from task mapping, our algorithms consider them in an integrated way. EGMS uses the concept of energy gradient to select tasks to be mapped onto new processors and voltage levels. EGM-SIV extends EGMS by introducing intra-task voltage scaling using a Linear Programming (LP) formulation to further reduce the energy consumption. Through rigorous simulations, we compare the performance of our proposed algorithms with a few approaches presented in the literature. The results demonstrate that our algorithms are capable of obtaining energy-efficient schedules using less optimization time. On the average, our algorithms produce schedules which consume 10% less energy with more than 47% reduction in optimization time when compared to a few approaches presented in the literature. In particular, our algorithms perform better in generating energy-efficient schedules for larger task graphs. Our results show a reduction of up to 57% in energy consumption for larger task graphs compared to other approaches.